1. law, govt and politics
2. legal issues
3. legislation
4. tax laws

2/2015
39 (2) 2015
ARTUR ŚWISTAK, SEBASTIAN WAWRZAK
and AGNIESZKA ALIŃSKA
In pursuit of tax equity: lessons from VAT rate
structure adjustment in Poland
ANA KUNDID NOVOKMET
Cyclicality of bank capital buffers in South-Eastern
Europe: endogenous and exogenous aspects
MIRNA DUMIČIĆ
Financial stress indicators for small, open, highly
euroized countries: the case of Croatia
ANA PERIŠIĆ and VANJA WAGNER
Development index: analysis of the basic instrument
of Croatian regional policy
Vol. 39, No. 2 | pp. 115-243
June 2015 | Zagreb
udc 336
issn 1846-887x
Publisher
Institute of Public Finance, Smičiklasova 21, Zagreb, Croatia
Editor-in-Chief
Katarina Ott
Production Editor
Marina Nekić
Editorial Board (Institute of Public Finance)
Marijana Bađun
Anto Bajo
Predrag Bejaković
Vjekoslav Bratić
Mihaela Bronić
Martina Fabris
Katarina Ott
Ivica Urban
Goran Vukšić
Editorial Advisory Board
Hrvoje Arbutina (Faculty of Law, Zagreb, Croatia)
Will Bartlett (London School of Economics and Political Science, London, UK)
Helena Blažić (Faculty of Economics, Rijeka, Croatia)
Balázs Égert (OECD, Paris, France)
Edgar L. Feige (Professor of Economics Emeritus at the University of Wisconsin, Madison,
Wisconsin, USA)
Božidar Jelčić (Faculty of Law, Zagreb, Croatia)
Evan Kraft (American University, Washington D.C., USA)
Peter J. Lambert (University of Oregon, Department of Economics, Eugene, USA)
Olivera Lončarić-Horvat (Faculty of Law, Zagreb, Croatia)
Dubravko Mihaljek (Bank for International Settlements, Basel, Switzerland)
Peter Sanfey (European Bank for Reconstruction and Development, London, UK)
Bruno Schönfelder (Technical University Bergakademie Freiberg, Faculty of Economics
and Business Administration, Freiberg, Germany)
Miroslav Verbič (Faculty of Economics / Institute for Economic Research, Ljubljana, Slovenia)
Athanasios Vamvakidis (Bank of America Merrill Lynch, London, UK)
Hrvoje Šimović (Faculty of Economics and Business, Zagreb, Croatia)
Financial Theory and Practice is abstracted and indexed in:
DOAJ (Directory of Open Access Journals, Lund University, Sweden)
EBSCO Publishing Database
EconLit (American Economic Association’s electronic database), JEL (Journal of Economic
Literature, Pittsburgh, Pennsylvania, USA)
HRČAK (Portal of Scientific Journals of Croatia)
IBSS (International Bibliography of the Social Sciences, ProQuest, Cambridge, UK)
RePEc (Research Papers in Economics)
Editorial Office
Institute of Public Finance – Financial Theory and Practice
Smičiklasova 21, Zagreb, Croatia, P.O. BOX 320
phone: +385 (0)1 4886 444; 4819 363; fax: +385 (0)1 4819 365
web-site: www.fintp.hr; e-mail: [email protected]
Subscription
Annual subscription amounts 400 kuna
Payments to account no. HR7024840081100661775, Institut za javne financije, Zagreb;
quoting: subscription to Financial Theory and Practice, 2015
Printed in 115 copies
The journal comes out four times a year
The journal is co-financed by the Ministry of Science, Education and Sport of the Republic
of Croatia
Computer typesetting and printing
Denona d.o.o., Zagreb, Marina Getaldića 1
2/2015
Reviewed scientific journal
Vol. 39, No. 2 I pp. 115-243 I June 2015 I Zagreb
Financial Theory and Practice publishes scientific articles in the field of public sector
economics, and also welcomes submissions of applied and theoretical research papers on
a broader set of economic topics such as economic growth and development, the role of
institutions, transition to the market economy and EU integration. Empirical analysis is
preferably related, but not limited, to the experience of countries in Central and Eastern
Europe and Southeast Europe.
Submission of papers
Submission of a paper will be held to imply that it contains original unpublished work and
is not being submitted for publication elsewhere. All papers have to pass through a double
blind peer-reviewing process.
Guidelines for authors can be found at www.fintp.hr.
Financial Theory and Practice is published in accordance with the highest level of professional ethics and standards. Ethical guidelines for journal publication can be found at
www.fintp.hr.
2/2015
TABLE OF CONTENTS
115
Articles
ARTUR ŚWISTAK, SEBASTIAN WAWRZAK
and AGNIESZKA ALIŃSKA
In pursuit of tax equity: lessons from VAT rate structure adjustment
in Poland
139
ANA KUNDID NOVOKMET
Cyclicality of bank capital buffers in South-Eastern Europe:
endogenous and exogenous aspects
171
MIRNA DUMIČIĆ
Financial stress indicators for small, open, highly euroized countries:
the case of Croatia
205
ANA PERIŠIĆ and VANJA WAGNER
Development index: analysis of the basic instrument of Croatian
regional policy
237
Book review
MAJA VEHOVEC (Ed.)
Health care from an economic perspective (Anto Bajo)
In pursuit of tax equity:
lessons from VAT rate
structure adjustment in
Poland
ARTUR ŚWISTAK, M.A., M.P.S., PhD candidate*
SEBASTIAN WAWRZAK, M.A., PhD candidate*
AGNIESZKA ALIŃSKA, PhD, Assoc. Prof.*
Article**
JEL: D12, H23, H24, K34
doi: 10.3326/fintp.39.2.1
The authors wish to thank Ruud de Mooji, Francisco Vazquez and Robert Sierhej (all IMF), and three anony­
mous referees for their valuable comments. All errors and omissions are the authors’ only. Research project
is funded by National Science Centre, Poland (Grant No. DEC­2013/09/B/HS4/03610).
**
Received: May 20, 2014
Accepted: February 15, 2015
*
A previous version of this paper was presented at the conference Tax Reforms: Experiences and Perspectives
organized by the Institute of Public Finance, Faculty of Economics and Business, Zagreb and Faculty of Eco­
nomics, Rijeka in Zagreb on June 20, 2014.
Artur ŚWISTAK
International Monetary Fund, Fiscal Affairs Department, 1900 Pennsylvania Ave NW, Washington, DC 20431,
USA
e­mail: [email protected]
Sebastian WAWRZAK
Warsaw School of Economics, al. Niepodległości 162, 02­554 Warsaw, Poland
e­mail: [email protected]
Agnieszka ALIŃSKA
Warsaw School of Economics, al. Niepodległości 162, 02­554 Warsaw, Poland
e­mail: [email protected]
116
financial theory and
practice
39 (2) 115-137 (2015)
Abstract
In 2011, in the aftermath of the economic crisis, Poland increased its value added
tax rates. Despite an already large VAT policy gap, further rate differentiation
was used to address distributional concerns and to protect the most vulnerable
households. We find that the changes to the VAT rate structure hardly improved the
overall progressivity of the VAT and the tax system as a whole. While providing
only minimal relief to the poor, taxation of food products at a super reduced rate
greatly subsidized the richer households. With a small change to the income tax
structure, the government could have secured more progressivity at a lower cost
in terms of revenue foregone.
Keywords: value added tax, household consumption, income distribution, tax progressivity, VAT reduced rates, Polish tax reform experience
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
1 INTRODUCTION
The effectiveness of pursuing vertical equity objectives through value added tax
(VAT) has been long questioned in economic literature (Cnossen, 1982; Tait, 1988;
Ebrill et al., 2001; OECD, 2010). Yet policymakers invariably exhibit a particular
liking for VAT rate differentiation. Even if desperate to raise revenue, they seem
to prefer to increase the general VAT rate rather than to eliminate reduced rates
and exemptions. This is the experience of the EU countries that undertook fiscal
consolidation in response to the recent financial crisis (IMF, 2013). Most of them
favored outright VAT rate increases over the much advocated base broadening and
closing the gap between standard and reduced rates (Garnier et al., 2013).
The Polish experience has not been different. While seeking additional revenue
the government opted for a universal VAT increase rather than abolition of re­
duced rates. In 2011 both the standard and the reduced rates were raised by one
percentage point. At the same time, the rate on most food products was further
lowered thus widening the VAT rate gap. Such rate adjustment was driven by the
government desire to protect the most vulnerable and improve the overall progres­
sivity of the tax regime (PRM, 2010). But the question is raised as to whether
pursuing these objectives through VAT is the right policy choice. Can we afford to
help the poor by subsidizing the rich?
The objective of this paper is to evaluate the distributional impact of the 2011
changes to VAT rate structure in Poland and assess whether the reform was suc­
cessful in achieving distributional objectives or perhaps there might have been
better alternatives to mitigate the impact of general VAT increases. We use data
from the household budget survey and estimate the impact of VAT for households
across the income distribution for the actual reform and an alternative scenario
where the VAT rate on food is not reduced. In our analysis we measure VAT inci­
dence in two ways: (1) relative to consumption which we take as a proxy for life­
time income, and (2) against annual income. We use gross rather than disposable
income in order to arrive at a comparable benchmark for VAT and income tax
burden.
2 VAT AND DISTRIBUTIONAL EQUITY
2.1 OBJECTIVES OF VAT RATE DIFFERENTIATION
Value added tax is one of the most popular taxes. In one or another form it has now
been implemented worldwide by an overwhelming majority of countries. The ac­
tual causes and consequences of the spread of VAT have been little investigated
empirically (Keen and Lockwood, 2010). Yet countries find it a good replacement
to their often obscure and less efficient production, trade or retail sales taxes. Once
introduced, VAT becomes an inherent part of tax systems.
VAT owes its remarkable popularity to its design. Many features are undeniably
appealing. By taxing domestic consumption VAT raises significant amounts of
revenue in an efficient and stable manner. On average it raises revenue amounting
to seven percent of GDP in high income countries and five percent of GDP in low­
income countries with an upward trend (Keen, 2013b). It serves as an effective
replacement for trade taxes being now successively curtailed due to regional inte­
gration and ongoing trade liberalization. Most importantly, however, VAT raises
revenue without hampering investment and savings. It is also fairly simple in
There are only a few instances where it was abolished, e.g. Malta, Vietnam, Grenada; and this happened
mostly due to poor planning and implementation, rather than VAT design per se (Grandcolas, 2005). Most of
these countries have now reintroduced a form of VAT.
Although VAT revenue exhibits a pattern of pro­cyclicality (see Ebeke and Vazquez, 2014) its fluctuation is
less pronounced than income taxes.
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
The remainder of the paper is divided into three parts. First, we briefly review
theoretical aspects of VAT distributional impact, focusing on the rationale and ef­
fects of rate differentiation as well as measurement of VAT incidence. Then, we
describe the Polish VAT structure and performance and explain in greater detail
the VAT reform analyzed. Finally, we evaluate changes in VAT redistribution as a
result of the reform.
financial theory and
practice
39 (2) 115-137 (2015)
There is no shortage of Polish literature analyzing the redistributive effects of in­
direct taxes (see for example: Nagel and Neneman, 1995; Radziukiewicz, 2011;
Dobrowolska and Starzynska, 2011). Amongst foreign academics, perhaps the
first insights into the Polish VAT rate structure and its implications were provided
by Cnossen (1998). Only recently, with Poland’s accession to the EU has the cov­
erage of the equity aspects of VAT in Poland improved. Many academics (e.g.
Borselli et al., 2012) and various European institutions (e.g. European Commis­
sion, 2011) embarked on analyzing the structure, performance and redistributive
impact of the Polish VAT, mostly as part of broader and comparative studies. To
the best of our knowledge, however, the incidence of the recent adjustments of the
VAT rates in Poland has not been subject to a rigorous analysis. This paper is
aimed to partially fill this gap.
7
118
terms of compliance. If well designed it has most of the attributes of a “good” tax
(IMF, 2011; Bird and Gendron, 2007).
financial theory and
practice
39 (2) 115-137 (2015)
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
Although there are plenty of reasons to praise VAT, its distributional impact raises
a lot of concerns amongst policymakers.3 Over the years, VAT has gained a reputa­
tion as a regressive tax and this perception seems nowhere close to fading away. It
does persist, at least at policymaker’s level, and results in various attempts to
make VAT a more progressive tax (e.g., Matheson and Swistak, 2014). Rate dif­
ferentiations and exemptions are used for this purpose. On the surface the policy
argument seems to be straightforward – with lower VAT rates on basic products,
e.g. food, the poorer households may consume more and their total VAT payments
represent a smaller share of their incomes than would be the case if all goods were
subject to the standard rate. Yet the richer households also benefit from reduced
rates, in absolute terms even more – all at the cost of foregone revenue, efficiency
loss and increased complexity (Tait, 1988; IMF, 2011; Matheson and Swistak,
2014).
There are only a handful of countries where VAT concessions are hardly used at
all, New Zealand and Australia being commonly cited as flagship examples. These
are countries with relatively new VATs and they managed to rely on economic and
administrative logic rather than cultivating historical and compromised solutions
(Cnossen, 2003). This is the experience of most EU member states, where VAT
concessions continue to be widespread. Apart from a few notable exceptions, in­
cluding Bulgaria and Denmark, most European countries rely on them heavily.
The list of items commonly subject to reduced rates, including a zero­rate, is
lengthy and encompasses food products, medicines, housing, books and newspa­
pers, and many others, not excluding clothing, energy products, and even alco­
holic beverages.
Addressing distributional concerns is not the exclusive reason for the adoption of
reduced VAT rates. They are also used, rightly or wrongly, to change the relative
prices of goods and services and steer consumption in a direction conforming to
other policy objectives. For example, lower taxation of labor intensive services in
the EU, e.g. hairdressers, window cleaning or repair services, was meant to in­
crease demand for these services, mostly self­supplied at home rather than pro­
cured in the market, and thus boost employment (Copenhagen Economics, 2007).
Norway used VAT rate differentiation to promote healthier diets (Gustavsen and
Rickertsen, 2013). Many countries tax merit goods and services, e.g. medicines,
textbooks, sport and cultural services, at a lower level (or exempt them) to encour­
age their consumption, as being in public interest and benefiting the whole society.
Similarly, rate differentiation is used to correct externalities by way of applying
reduced rates to energy­saving appliances (OECD, 2010). The justification for
The other Achilles heel of VAT is the need for refunds of excess credits. In many countries, especially devel­
oping ones where the tax administration capacity is low, this feature of VAT raises a lot of concerns.
3
2.2 VAT INCIDENCE
In theory a uniform and comprehensive VAT imposes a flat burden on all con­
sumption expenditure (Tait, 1988). In this sense, VAT is proportional – all taxpay­
ers give up an equal share of their consumption to pay the tax. This is true regard­
less of their personal characteristics, consumption preferences or even source of
income used to finance their spending. Whether they are rich or poor, single or
with dependents, healthy or disabled all taxpayers forego a fixed percentage of
their private spending to meet the VAT liability. The same holds true whether they
buy staple food or lavish meals in expensive restaurants, rely on private cars or
use public transport, rent or buy a house. Under a broad­based and uniform VAT,
all taxpayers – as long as they spend – are equally burdened with VAT. Neverthe­
less, in practice achieving proportional distribution is hardly possible, for even
under a well­designed VAT a portion of spending escapes taxation.6
The actual distribution of VAT payments is, nevertheless, very sensitive to the rate
structure and exemptions built into the VAT and to the patterns of consumer pref­
erences (Ebrill et al., 2001). Relative to consumption the use of preferences may
make VAT a progressive tax. In such a setting the share of VAT payment rises with
4
The Romanian authorities are currently looking to apply the same reduction to meat products to counter­
act fictitious imports and tax evasion in the meat industry (see news by Simona Bazavan on: http://business­
review.eu/featured/romanian­authorities­consider­cutting­the­vat­for­meat­products­to­9­pct/).
5
The third possibility – at least in theory – would be to test VAT incidence against wealth. This discussion,
however, goes beyond the scope of the paper.
6
This happens for administrative reasons, e.g. exemption of small traders as a result of the VAT threshold. Such
a design benefits mostly low income households as they are more likely to buy from non­taxed small businesses.
They are also more likely to buy from individuals (e.g. used goods) and informal businesses.
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
The focal point of the discussion of the vertical equity of VAT is the question of
who bears the burden. Fairness requires taxes to take into account the taxpayers’
ability to pay, hence implying that the share of income taken in taxes increases as
income rises (progressivity). Even though such a statement may be difficult to
defend on efficiency grounds this is a strong political expectation: taxes should be
progressive or at least not regressive. Measuring progressivity of VAT is challeng­
ing. The notion of income redistribution would imply that VAT incidence should
be tested against income. Yet the VAT base is consumption and not income. Should
the VAT incidence be assessed against consumption, as the nature of the tax im­
plies, or income, as the incidence theory suggests? Below we briefly review con­
ceptual arguments for using both metrics.5
119
financial theory and
practice
39 (2) 115-137 (2015)
VAT reduced rates is also offered on administrative grounds. Some countries, e.g.
France, the Netherlands, and Poland, tax construction services related to renova­
tion and maintenance of private dwellings at a reduced rate, and others, e.g. Ice­
land, effectively zero­rate them through reimbursement of VAT paid on such serv­
ices, with a view, at least partially, to counteracting tax evasion in the construction
sector. In 2013 Romania reduced its VAT rate on bakery products on the very same
grounds – to tackle tax evasion.4
0
the level of consumption as better off households spend more on fully taxed goods
and services than those less affluent, where their spending on lightly taxed or zero­
rated basic products, e.g. food, constitutes a large proportion of their consump­
tion.
financial theory and
practice
39 (2) 115-137 (2015)
An intuitive approach to the measurement of the distributional impact of any tax,
including a VAT, would be to use income as denominator. Early empirical studies,
including those by Musgrave et al. (1974) or Pechman (1985), favored the annual
incidence approach, similar to that used for income tax evaluation. Such an ap­
proach renders VAT a regressive tax. This is inevitable. As poorer households save
little and consume higher shares of their current income, VAT accounts for a larger
proportion of their earnings than the VAT paid by the rich.
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
Some argue, however, that measurement of VAT incidence against annual income
is misleading (Gaspersen and Metcalf, 1994; Metcalf, 1997; Creedy, 1998). This
is because, in line with permanent income hypothesis and life­cycle considera­
tions (Friedman, 1957), income levels tend to vary over time, with young and old
households earning low incomes and middle aged households disproportionally
higher, i.e. high enough to pay back past borrowing and save for future consump­
tion. In anticipation of future changes in income levels households prefer to
smooth their consumption so it is higher than income in the early years of a life­
time cycle, lower in the middle and again higher at the end.
In theory, given the above considerations, one could argue that all income is con­
sumed over the lifetime. From this perspective any consumption tax, including
VAT, is a proportional tax (Gaspersen and Metcalf, 1994; Athreya and Reilly,
2009). When rate differentiation is used the incidence of VAT shifts towards richer
consumers as they tend to consume more goods and services taxed at a standard
rate. Using the lifetime incidence Decoster et al. (2010) find that actual VATs in a
number of European countries are slightly progressive. The same approach has
been used in a number of individual VAT studies. For example, Arsic and Altipar­
makov (2013) show that Serbian VAT is proportional. The same results may be
found in Slintakova and Klazar (2010) for the Czech Republic, and in Braz and
Correia da Cunha (2009) for Portugal.
Proponents of the lifetime approach argue that measuring VAT incidence at any
given point in the life­cycle renders blurred results. In a lifetime sense, neither
young nor old households are poor nor are high­earners rich at their peak. Since
consumption fluctuates less from year to year than income it is a better measure of
household well­being than total annual income and constitutes a good proxy for
its lifetime income (Poterba, 1989). This reinforces the logic, mentioned above,
that incidence of consumption taxes should be measured against their base – con­
sumption.
2.3 VAT AND INCOME TAX
Value added tax, as an in rem and not personal consumption tax, is ill suited to
pursue progressivity. Fortunately, VAT is not the only element of the tax system
and its distributional impact should not be analyzed in isolation. If one is con­
cerned about fairness of taxation as a whole then an assessment of the overall im­
pact of taxes on income distribution should be made. An income tax, whose inci­
dence is usually measured on annual basis, is an obvious candidate. The inherent
regressivity of a broad­based and single­rated VAT may be mitigated by progres­
sive personal income taxes. We show this correlation in a stylized way in figure 1.
All things being equal both taxpayers and governments are indifferent to the form
in which taxes are raised. What is important is the change in disposable income as
a consequence of overall taxation. As income level rises, a smaller proportion of
income is taken in VAT but this is counterbalanced by higher shares of personal
income tax, making the tax system progressive. As argued in Atkinson and Stiglitz
(1976) and Cnossen (1982) as long as a country can flexibly choose the rate struc­
ture under the personal income tax, then it has no reason to choose differential tax
rates on the consumption of different goods and services.7 Such argumentation
7
In addition, progressivity of the tax system may be further enhanced by well­designed property taxes and
government transfers.
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
Finally and most importantly, the political economy argument has to be taken into
account. The lifetime approach has a long horizon which is not attractive to poli­
cymakers and taxpayers. It hinges on the assumption that the incomes of younger
households grow and exceed their needs in middle age allowing for savings to be
used towards the end of the life cycle. This is too optimistic to convince currently
poor individuals, with low skills, productivity and thus little potential for increased
future earnings. It is not difficult to argue that the poor need help when they face
life hurdles. And the governments provide it – either through transfers from the
budget or taxation. Interestingly, in defining the beneficiaries of social programs,
e.g. family allowances, housing benefits, etc., countries tend to rely on an annual
income test. Income taxes are also annual – the tax scale is applicable to annual
income and most of the allowances and credits apply annual ceilings. Examples
where carry­forward or carry­back are used to smooth personal income over years
are scarce. Given the above it is more reasonable to measure the incidence of
consumption taxes against annual income.
financial theory and
practice
39 (2) 115-137 (2015)
Unfortunately, lifetime income or consumption incidence does not seem to be ap­
pealing to policymakers. Indeed, not all households use up all their lifetime in­
come and much of it is passed on to the next generations in bequests. In welfare
states, not all households rely exclusively on their own income and later on their
savings – a system of benefits and state funded pensions plays a significant role.
Also, households’ sizes and compositions change over time and it is hardly pos­
sible to have a household that follows the lifecycle of an earning individual.
financial theory and
practice
39 (2) 115-137 (2015)
implies that income tax is an effective instrument of redistribution, which may not
necessarily be true (Bird, 1987), as many poor individuals do not pay income
taxes and the rich derive their wealth from non­wage income and usually escape
progressive taxation.8 Yet, the same may be true for VAT. The poor, with no or
little income, either do not participate in monetized economy or buy from infor­
mal or small businesses and thus pay little or no VAT. In turn, the rich may avoid
part of the VAT burden by claiming their private consumption as business inputs
(e.g. cars, accommodation, computer equipment and software, etc.) or exploit
cross­country rate differentiations and partly consume in lower­taxed jurisdictions
(e.g. tourism, wellness and beauty, high tech devices, etc.). After all, in the case of
the extremely rich, both income and consumption taxes fail to take a fair share of
their annual incomes.
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
Figure 1
Stylized correlation of PIT and VAT incidence
Tax burden
Household income level
PIT
VAT
Total
Source: Authors.
Although we give due credit to the imperfections of income taxes in addressing
distributional concerns, we argue that it is a better way of attaining progressivity
in the tax system than using rate differentiation under VAT. Our argument for a
uniform VAT rate structure is the poor targeting and the concomitant unnecessary
revenue loss due to rate differentiation; in addition, there are the complexities and
costliness of VAT compliance and administration involved with differentiated
rates. If a specific good is taxed at a lower rate, say, fruit and vegetables, bakery
products or children’s clothing, such a good becomes more affordable to the poor
and they pay less in VAT. But the more affluent households also benefit from this
measure, even more in absolute terms. This translates into significant revenue loss
and/or higher taxes elsewhere to make up for the revenue loss. In order to alleviate
the plight of the poor, governments agree to give even higher tax relief to rich
For example, in Poland, business and capital income are taxed at a flat rate of 19 percent, whereas labor
income is subject to progressive tax rates.
8
Also, efforts required for a proper delineation between standard and lower taxed
goods (e.g. children’s and adult clothing) drive up compliance and administration
costs. Equity gains, once again, are likely to be outweighed by the increased ad­
ministration costs resulting from cost of interpretation, classification rules, issu­
ance of advanced rulings, complex filing, audits, disputes, etc.9
3.1 VAT STRUCTURE AND REFORM OVERVIEW
After reviewing theoretical considerations on the redistributive effects of VAT we
turn now to an incidence analysis of Polish VAT. Our focus will be on the distri­
butional impact of the 2011 adjustments to the VAT rate structure. We begin with
a brief overview of VAT structure and performance, present details of the VAT
reform and follow with our findings.
Value added tax is the main source of government revenues for Poland, account­
ing for almost 44 percent of total tax collection. In 2011 it raised PLN 120.8 bil­
lion (USD 40.8 billion0) which totaled to 8.1 percent of GDP, around the average
for EU member states. The revenue performance, measured as a ratio of actual and
theoretical VAT collections, at only 45 percent in 2011, was one of the lowest in
the EU though (EU, 2013). One of the reasons for the relatively wide VAT gap is
the rate structure. Borselli et al. (2012) estimates that more than 40 percent of
final consumption is subject to reduced rates, bringing down the effective VAT to
less than 15 percent, only two­thirds of the standard rate of 23 percent.
Data from annual tax expenditure budgets prepared by the Polish Ministry of Fi­
nance confirms the generosity of VAT (MF, 2013). In 2011 they amounted to PLN
41.1 billion (or USD 13.9 billion), i.e. 2.7 percent of GDP and 34 percent of ac­
In 2013 alone, the Polish tax administration issued over 2,000 advanced rulings regarding the application of
the VAT rate. Inquiries for interpretation or clarification included the application of a proper VAT rate to take­
away meals, furniture assembling, certain construction services performed in dwellings, drinks with addi­
tion of coffee, latex gloves, magazine and CD/DVD bundles, etc. (Data received from Ministry of Finance,
Poland.)
0
All PLN values cited in this paper were converted into USD at PLN/USD exchange rate at 2.96 (average
for 2011).
Assumes total final consumption taxed at standard VAT rate and perfect tax compliance (no evasion).
The other important reasons are exemptions and tax compliance. The recent European Commission study on
VAT compliance gap found that in 2011 Poland lost 15 percent of their VAT (theoretical tax liability of VAT
as legislated) due to imperfect tax compliance (EC, 2013).
9
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
3 REDISTRIBUTIVE IMPACT OF VAT RATE DIFFERENTIATION IN POLAND
123
financial theory and
practice
39 (2) 115-137 (2015)
taxpayers, an odd and costly manner of injecting progressivity into the tax system.
Adoption of reduced rates could yield equitable results only if applied to goods
and services whose absolute consumption falls as income rises. Given the conver­
gence of consumption patterns across all income levels, there are not many obvi­
ous candidates. Staple food, like rice, potatoes or flour may potentially conform to
such a pattern.
124
tual VAT collections.13 The highest contributors to the overall value of tax expen­
ditures in the VAT were residential construction services, food products and med­
icines. Among the food products the most costly were meat, dairy and bakery,
totaling respectively to PLN 2.9 billion (USD 1 billion), PLN 1.7 billion (USD 0.6
billion) and PLN 1.1 billion (USD 0.4 billion).
financial theory and
practice
39 (2) 115-137 (2015)
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
Unlike revenue performance, the stability of rates has long been a signature fea­
ture of the Polish VAT, putting it at the forefront of the EU member states (see EC,
2014). On its adoption in mid­1993 VAT in Poland was imposed at the standard
rate of 22 percent. A number of goods and services, including certain food prod­
ucts, clothing and accessories for children, energy products for household con­
sumption, telecommunication services and construction materials were taxed at a
lower seven percent rate. In 2000, a second reduced rate of three percent was
adopted to encompass previously exempted basic food products. Upon accession
to the EU the list of goods and services subject to reduced rates as well as to ex­
emptions was further revised. In spite of such reshuffling the main rates, however,
remained unchanged.
Not until 2011, faced with the need to constrain public debt, did Poland decide to
increase VAT rates. The adjustments of VAT rates were twofold. The standard and
reduced rates rose by one percentage point – from 22 to 23 percent and from
seven to eight percent respectively. The VAT increases coincided with the expiry
of the pre­accession derogatory regime. The concessional zero­rate on books and
periodicals (not newspapers) and the three percent rate on certain unprocessed
food items, as being below the EU minimum of five percent required by the VAT
Directive, were doomed for increases. The government, rather than use the re­
duced rate already in place decided to introduce a new one – a super reduced rate
of five percent. Most of the food products, previously subject to the higher re­
duced rate (now at eight percent) were moved to the new lower five percent rate.
Interestingly, the government, even though faced with the need to collect more tax
revenues, did not decide to rely on only one reduced rate, let alone abolishing re­
duced rates altogether. Adoption of the super reduced rate of five percent, compro­
mising VAT revenue productivity, contradicted the objective of general VAT in­
creases. Such a policy choice was geared to address equity concerns and improve
progressivity of VAT. As stated in the justification to the bill introducing new
rates, it was the Government’s desire to alleviate the impact of VAT increases on
the poorer households via lower taxation of food which “constitutes significant
part of spending by the less affluent part of the society” (PRM, 2010). The Gov­
ernment chose to achieve this explicitly by way of rate differentiation, a three
13
The actual cost of revenue foregone may be in fact higher since – in line with the adopted benchmark – only
such VAT concession were considered to be a tax expenditure that deviates from the standard VAT design as
provided in the EU VAT Directive. Hence, for example most of exemptions granted to services provided in
public interest (e.g. public bodies, postal, broadcasting or social services) and financial, including insurance,
services were not seen as VAT concessions and thus not estimated.
percentage points lower taxation of food. No corresponding measures on the reve­
nue side were proposed and adopted.
financial theory and
practice
39 (2) 115-137 (2015)
Such a policy was not conceptually different from the approach taken in the past
when reduced rates on Internet services and construction materials were abol­
ished. In every single case it was an explicit government desire to correct the
distributional impact of VAT rate changes and the only differences was in the
choice of measures for achieving that. Table 1 summarizes the VAT changes and
the corresponding compensating measures.
125
Table 1
Summary of VAT rate changes and compensating measures in Poland
Year
2008
0
Compensating measure
Income tax allowance of PLN 760
(USD 257) per year
Cash rebate equal to 12.3 percent of
Repealing of reduced rate
spending on construction materials
on construction materials
previously taxed at a reduced rate
Standard and reduced rate increased
VAT rate on food products lowered
by one percentage point (from 22
by three percentage points (from eight
and 7 to 23 and 8 percent, respectively) to five percent)
Source: Polish legislation (available at: www.orka.sejm.gov.pl).
In the next two sections, we analyze the 2011 adjustments to the VAT rate struc­
ture. For simplicity we call these adjustments a “reform” though such a policy
measure falls short of a genuine tax reform. Like the other changes to VAT rates
listed in table 1, it appears to be merely a one­off adjustment mitigating the effects
of VAT increases, triggered either by revenue needs or solely by the ongoing tax
harmonization.
3.2 DATA AND METHODOLOGY
We take a very simple approach to our analysis. First, using data from 2011 house­
hold budget survey, compiled by the Central Statistical Office of Poland (Główny
Urząd Statystyczny – GUS), we estimate the VAT burdens for households at differ­
ent income levels.
First VAT burden (Bc) is measured against lifetime income using consumption as
a proxy:
(1)
where:
Tvd denotes monthly amount of VAT paid by households from a given decile
group (d),
Cd denotes monthly consumption spending by households from a given decile
group (d),
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
2004
VAT rate change
Repealing of reduced rate
on Internet access services
126
d represents decile group from 1 to 10.
Then VAT burden (Bi) is measured against annual gross income (GI):
(2)
financial theory and
practice
39 (2) 115-137 (2015)
The data set we obtained presents monthly consumption expenditure compiled
from 37,584 representative households on a quarterly rotation basis. The survey
excludes spending on purchases of residential property14, formally treated as in­
vestment spending.
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
The detailed information on consumption spending allows us to assign a VAT rate
applicable to a given expenditure item and calculate the amount of tax. In cases
where it was not readily apparent what rate (or exemption) should be used we
made necessary assumptions and apportionment. For example, data on bread con­
sumption does not allow for immediate distinction between fresh bread, defined as
bread to be consumed within 14 days and taxed at five percent, and non­perishable
bread with an expiration date longer than 14 days, taxed at eight percent. An im­
portant caveat has to be made here. Since the household budget survey does not
allow us to establish from where all the goods and services were procured we as­
sume that all non­exempt items are taxable. This does not necessarily have to be
true, as part of the spending, especially on grocery, takes place at small establish­
ments and greenmarkets, not to mention informal businesses, and as such is ex­
empt. For the lack of data we did not make an attempt to correct those values and
acknowledge that such an approach may overstate the actual VAT burden for the
poorer households, as they are more likely to buy from non­taxed suppliers. In
cases where consumption items are exempt from VAT, e.g. healthcare services, we
assume that VAT paid on inputs is passed through to consumers in the final price
of the exempt good or service. To derive the amount of VAT, for simplicity, we
assume that taxable inputs constitute 40 percent of the final price and they all
carry VAT at the standard rate.15 Taking these assumptions into account the amount
of tax (Tvd) is derived as follows:
(3)
where:
βd denotes the share of total consumption (Cd) subject to the standard VAT rate
(sr), reduced VAT rate (rr), super reduced VAT rate (srr) or exempt from VAT (e),
r denotes the applicable VAT rate – standard (sr), reduced (rr), or super re­
duced (srr).
Although this category includes spending on newly built houses and apartments, which are subject to VAT
in Poland, it does not critically impact our analysis for the 2011 VAT reform as it entailed changes in taxation
of food products and a few merit goods. Nevertheless, on an aggregate level, it understates the VAT burden
falling mostly on the rich and not the poor who usually self­build.
15
The rigorous approach would dictate to use supply­use (or input­output) tables to approximate for the value
of input VAT.
14
Since incomes reported in the household budget survey for each decile are net
amounts (NI), we calculate the gross (pre­tax) income (GI) that it would be neces­
sary to earn, the difference being the amount of tax paid (Ti).
7
financial theory and
practice
39 (2) 115-137 (2015)
For measurement of the VAT burden against annual income we choose to rely on
gross rather than net income. Gross income is the most sterile benchmark availa­
ble and allows for comparisons of burdens generated by different taxes. Such an
approach is motivated by a belief that taxes are interchangeable and this is the
overall tax burden that matters for taxpayers.16 Reliance on disposable income,
already adjusted for income tax payments, blurs and – in general – inflates the
actual incidence of VAT.
(4)
(5)
where:
SSCd denotes the amount of social security contributions,
HCd denotes the amount of health contributions,
ITd denotes income tax payment.
Since
(6)
and
(7)
and
(8)
where:
rssc denotes the rate of social security contributions,
rhc denotes the rate of health contribution,
rit denotes the applicable rate of income tax,
A is the amount of personal basic tax allowance,
γ denotes the share of health contribution allowed for deduction against in­
come tax,
CTC is the amount of child tax credit.
Naturally, the two general taxes – PIT and VAT – are not the only taxes falling on households. There are
other, including property taxes and excise taxes, but these are ignored. The analysis of the progressivity is
also limited to the tax system only and all government transfers aimed at helping the poor, e.g. family bene­
fits, unemployment benefits, disability benefits, are not taken into account.
16
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
Two important qualifications apply. First, all income received by a household is
labor income taxed at progressive rates (18 and 32 percent) with a basic tax allow­
ance. Second, social security and health contributions are treated as taxes and
form part of a tax wedge. This approach is not different from OECD Taxing Wages
(OECD, 2013a) methodology and the European Commission and Eurostat’s Taxation Trends publications (EU, 2013). This implies:
128
Equation (4) may be transformed as follows:
(9)
financial theory and
practice
39 (2) 115-137 (2015)
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
To derive the amount of gross income and tax paid we use a simple Excel­based
micro simulation technique, similar to that employed by the OECD for estimates
of the tax burden in Taxing Wages (OECD, 2013b). In calculation of income tax
we made the following assumptions. Each wage earner has only one job and lives
and works in the same city. Each works throughout the whole year and has no sick
or maternity/paternity absence, nor does he or she benefit from any other social
support. If there are children in a household the earning parent(s) live with them
and have the right to benefit from the child tax credit. A lone­parent allowance, but
not family cash allowance7, is included in the calculation where applicable. If
there are two working adults in a household they are deemed a married couple and
are taxed jointly. Unless specifically indicated, no tax allowances and further de­
ductions are used. All income is earned in Poland and does not escape taxation.
Both data on expenditure and income are reported for one person. To arrive at
household income and expenditure we use the modified OECD equivalent scale,
which attributes a weight of 1 to the first household member, 0.5 to the remaining
members over 14 years old and 0.3 to each child (below 14 years old).
Calculations for VAT and PIT tax burden are performed for a number of different
households: (1) single adult household, (2) a married couple with two children, (3)
a married couple with one child, and (4) a single parent household – one adult and
one child. It is assumed that all the adults work and children do not.
To evaluate the distributional impact of the reform we compare two rate structures
– the actual and a potential one, i.e. one the government could have chosen in the
absence of equity concerns. The actual structure, as implemented in 2011, oper­
ates a standard rate of 23 percent, reduced rate of 8 percent and super reduced rate
of 5 percent, applicable to food products and a few merit goods (actual scenario).
The potential rate structure operates only two rates – 23 percent and the reduced 8
percent, implying that the government chooses not to “improve” VAT incidence
(base scenario). VAT­exempt consumption is kept constant in our calculations.
We use the same consumption dataset for both scenarios. Our analysis is therefore
static and it does not take into account any behavioral response to changes in the
level of taxation. Also, a full pass­through of VAT burden in consumer prices is
assumed.
Family cash allowance is a social subsidy available to families with children, having very low income per
family member. It is not a tax measure.
7
3.3 FINDINGS
The 0 VAT rate structure adjustment only slightly improved the progressivity
of VAT – measured against both life­time and annual income. It did so at a very
high cost to the budget. While providing only minimal relief to the poor it greatly
subsidized the richer households.
18.5
-0.0
-0.1
18.0
-0.2
-0.3
17.5
-0.4
17.0
-0.5
-0.6
16.5
-0.7
16.0
-0.8
-0.9
15.5
<10%
10-
Base scenario
20-
30-
40-
Actual scenario
50-
60-
70-
80-
>90%
Change in effective VAT rate (right axis)
Source: 2011 Household Budget Survey, GUS, Authors’ calculations.
Annual income incidence analysis renders similar results. In line with expecta­
tions our calculations confirm that the Polish VAT is regressive and the 2011 re­
form only slightly improved its income redistribution. The adoption of a 5 percent
rate brought the VAT burden down by 1.2 percentage points for the first income
decile and by 0.2 percentage points for the highest. Although the balance tips to
the poorest the VAT remained a regressive tax. In figure 3 we show the shift in the
overall burden of VAT measured against gross income. We do not plot results for
Interdecile comparisons between first and last income decile may blur the results due to the pronounced
impact of exemptions. The highest income decile consumes disproportionally more exempt services (i.e. health,
recreation, culture), thus benefiting more than the poorer households. Bearing in mind that our analysis regards
reduced rates we do not consider any changes in exemptions and keep it constant in our calculations.
18
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
Figure 2
VAT distribution measured against consumption (%)
financial theory and
practice
39 (2) 115-137 (2015)
In the base scenario, i.e. without lowering the tax rate on food, the average VAT
rate amounts to 17.8 percent. Lower taxation of food brought this rate down to
17.2 percent and benefited mostly the poor – the lowest income decile gained 0.84
and the highest 0.38 percentage points. In this sense the reform was progressive.
However, the reform did not significantly change the overall VAT incidence meas­
ured against annual consumption (as a proxy for lifetime income). As we show in
figure 2 the actual reform only slightly tipped the balance of VAT burden towards
the richer households – the ninth income decile paid 1 percentage point more in
VAT than the first one. In the base scenario this difference would be only 0.6 per­
centage point.18
129
130
the first income decile for presentational purposes only. We will include it in fur­
ther analysis and discussion.
Figure 3
VAT distribution measured against gross income (%)*
financial theory and
practice
39 (2) 115-137 (2015)
15
14
13
12
11
10
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
9
8
10-
20-
30-
40Base scenario
50-
60-
70-
80-
>90%
Actual VAT
* Gross income of a household composed of two working adults and two children below 14
years old.
Source: 2011 Household Budget Survey, GUS, authors’ calculations.
As argued earlier, VAT is just another tax paid by households, and should be ana­
lyzed together with their income tax payments. Our calculations show that distri­
bution of the overall tax wedge, i.e. ratio of tax payments and gross income, to a
large extent hinges on the income tax burden. This implies that in an annual per­
spective VAT regressivity is counterbalanced by income taxes, especially for the
richer part of society (last five income deciles). Since the income tax provides for
a number of family­related tax preferences – joint taxation of spouses, joint taxa­
tion of single parents and their not working children (lone­parent allowance) and
child tax credits – the degree to which the VAT burden is corrected by income
taxation depends on household composition. In figure 4 we present income tax
and VAT distribution for four classes of households: a single adult, single parent
with a child, a married couple with two children, and a married couple with one
child. In addition, as previously, we plot in the effect of 2011 VAT reform as a
hypothetical addition to the actual tax burden.
As can be seen in figure 4, VAT disproportionally burdens households in the first
income decile, regardless of their composition. These results are not surprising,
given that the consumption, as reported in the household budget survey, exceeds
almost twice the income of households in the first decile. There are two important
remarks to make. First, as noted in the literature (e.g. Arsic and Altiparmakov,
2013), data for this group is often subject to measurement errors, implying that
actual income may be higher. Second, as we already noted, a significant portion of
consumption spending by the poorest households is not subject to VAT, implying
that their actual VAT payments may be lower than our calculations indicate.
50
1 Adult
45
45
40
40
35
35
30
30
25
25
50
<10% 10- 20- 30- 40- 50- 60- 70- 80- >90%
1 Adult 1 Child
20
45
40
40
35
35
30
30
25
25
<10% 10- 20- 30- 40- 50- 60- 70- 80- >90%
VAT Addition (Base scenario)
2 Adults 2 Children
50
45
20
<10% 10- 20- 30- 40- 50- 60- 70- 80- >90%
20
<10% 10- 20- 30- 40- 50- 60- 70- 80- >90%
VAT (Actual)
Income Tax
Source: 2011 Household Budget Survey, GUS, authors’ calculations.
Finally, bearing in mind that the objective of this paper is to evaluate the 2011
VAT reform, we should note that regardless of the class of household the gain
from 5 percent VAT rate did not significantly alter the overall tax burden house­
holds face. A small adjustment to the income tax structure could yield better re­
sults in terms of tax distribution than lowering VAT on food consumption.
Whatever the addition to VAT progressivity through lower VAT rates it is clear
from our analysis that they benefit households across all income levels. The 2011
reform aggravated this. In absolute terms (dollar value) most benefits of the re­
form accrued to the rich households. On average the lowest income decile saved
in VAT payments PLN 47.24 (USD 15.96) per annum, whereas the highest income
decile gained more than twice this amount – PLN 101.23 (USD 34.20). We show
the results in figure 5 below.
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
20
2 Adults 1 Child
50
financial theory and
practice
39 (2) 115-137 (2015)
Figure 4
Overall tax wedge faced by different classes of households (%)
131
132
Figure 5
Gains from 2011 VAT Reform (left axis in USD, right axis in percent)
financial theory and
practice
39 (2) 115-137 (2015)
40
1.8
35
1.6
30
1.4
1.2
25
1.0
20
0.8
15
0.6
10
0.4
5
0.2
0
<10%
10-
20-
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
VAT gain (left axis)
30-
40-
50-
60-
70-
Share of income
80-
>90%
0.0
Share of consumption
Source: 2011 Household budget survey, GUS, authors’ calculations.
The cost of the reform, what we estimate to be at around PLN 2.6 billion (USD 0.9
billion) or 0.2 percent of GDP, brought down the VAT revenue by 3.2 percent, as
compared with the base scenario. Most of the revenue the government decided to
forego to help the poor accrued to higher income households. The poorer part of
the society (first three income deciles) benefited by 22 percent and the richer (last
three income deciles) by 39 percent of the total revenue foregone. Bearing in mind
that those poorer households procure their food products from non­taxed suppliers
(greenmarkets, small traders or informal businesses) more often than the richer
ones, their benefits from the reform are even smaller than indicated by our calcula­
tions.
The consumption patterns provide valuable insights into VAT’s inability to ad­
dress distributional equity concerns. In Poland they do not deviate from the pre­
dictions of economic theory. Total consumption of households increases as their
disposable income rises. In line with this the basket of goods and services pro­
cured changes – the share of spending on food and other essentials (e.g. energy,
cold water) declines and expenditure on durable goods, e.g. means of transport
and various services, including tourism, recreation, restaurants, culture, educa­
tion, rises. In figure 6 we show changes in consumption of chosen classes of goods
and services.
In absolute terms, with only a few exceptions, food spending increases. It merely
represents a lower share of their incomes and total consumption basket. The low­
est income decile spent PLN 171.22 (USD 57.85) monthly for food and non­alco­
holic drinks, which represented 36.4 percent of total consumption, whereas the
highest decile spent more than twice this amount – PLN 371.65 (USD 125.56)
accounting for only 16.7 percent of total spending. The interdecile ratio of abso­
lute spending is not uniform for all food items. The sharpest increase in consump­
Figure 6
Share of consumption for selected classes of goods and services (%)
40
133
financial theory and
practice
39 (2) 115-137 (2015)
tion may be observed, amongst other, for fish, tropical fruit, fruit juices, mineral
water, chocolate and butter. Spending on bread, milk and certain vegetables is
fairly constant and there are only a few items where it declined, potatoes and
sugar being the most prominent ones. This dooms any efforts to improve progres­
sivity of taxation through reduced VAT rates to ineffectiveness, at least it terms of
revenue.
35
30
20
15
10
5
0
<10%
10-
20-
30-
40-
50-
60-
70-
80-
>90%
Percentile of adult equivalent net income
Food
Clothing and footwear
Recreation and culture
Transport
Restaurants and hotels
Source: 2011 Household Budget Survey, GUS, authors’ calculations.
4 CONCLUSIONS
Polish society is not egalitarian but income inequality is smaller than in many
other OECD countries. With an after tax Gini coefficient at 0.3 in 2010 equity is­
sues, as it seems, should not be a major concern. Yet Poland in 2011, while in­
creasing VAT by 1 percentage point, as part of its fiscal consolidation effort, de­
cided to lower its rate on food and a few merit goods – from 8 to 5 percent – to
ease the tax burden for the poor. No other mitigating measures, either through in­
come tax or on the spending side, were adopted.
We find that the 2011 VAT reform only slightly improved the distribution of VAT
and the overall progressivity of the tax system. It did so at a very high cost to the
budget. While providing only minimal relief to the poor it greatly subsidized richer
households. The three last income deciles gained as much as 39 percent of the total
benefits of the reform which – in terms of revenue forgone – we estimated at PLN
2.6 billion (USD 0.9 billion) or 0.2 percent of GDP. The three first income deciles
benefited only 22 percent of this amount with all the likelihood for this number to
be even lower as poorer households more often than richer buy from non­taxed
suppliers, including greenmarkets, small traders and informal businesses.
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
25
134
financial theory and
practice
39 (2) 115-137 (2015)
The attempt to alleviate the plight of the poor does not seem to have been neces­
sary. Measured against consumption, as a proxy for lifetime income, VAT would
be anyway mildly progressive and the reform hardly changed that. When its an­
nual incidence is considered – an approach favored by policymakers – VAT with­
out further rate reduction on food would have imposed a higher burden on the
poorer part of the society. Again, the rate adjustment did not change this regressive
property of the VAT. A closer look at the combined income and VAT tax burden
faced by households reveals that the regressive nature of VAT to a large extent is
counterbalanced by progressive income tax. With a small change to the income
tax structure the government could have secured more progressivity at a lower
cost in terms of revenue foregone.
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
The Polish experience with VAT rate structure adjustment confirms that any ef­
forts to pursue progressivity through reduced rates are doomed to be ineffective.
There are several reasons for this, but consumption patterns are the underlying
one, especially in the case of food. In line with economic theory, spending on food
declines as income increases. It declines as a share of income but not in absolute
terms. In such a setting the richer by definition benefit more, at a high cost of rev­
enue forgone and loss of efficiency. There are only a few inferior food products
whose consumption declines as income rises. The Polish household budget survey
reveals only two – the potato and sugar.
Although consumption patterns differ with income, consumption per se does not
constitute a good basis for addressing distributional concerns. This renders VAT
the tax least suited for the pursuit of distributional equity objectives for they inher­
ently relate to the income and personal characteristic of taxpayers. Reliance on a
broad­based and single­rated VAT for revenue generation and use of progressive
income taxes may yield far better results. As noted in Keen (2003a) “even poorly
targeted [government] spending may be a better way to support the poor than a
reduced rate.” Helping the poor by subsidizing the rich is one of the most impru­
dent and deceitful policies and one that no society can ever afford. Pursuing dis­
tributional equity through the differentiation of VAT rates is one such policy and
should be avoided.
135
financial theory and
practice
39 (2) 115-137 (2015)
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
REFERENCES
1. Arsic, M. and Altiparmakov, N., 2013. “Equity aspects of VAT in emerging
European countries: A case study of Serbia”. Economic Systems 37, pp. 171­
186. doi: 10.1016/j.ecosys.2012.12.003
2. Athreya, K. B. and Reilly, D., 2009. “Consumption smoothing and the mea­
sured regressivity of consumption taxes”. Economic Quarterly, 95 (1), pp.
75­100.
3. Bird, R. and Gendron, P., 2007. The VAT in Developing and Transitional
Countries. Cambridge: Cambridge University Press. doi: 10.1017/
CBO9780511619366
4. Bird, R., 1987. “A new look at indirect taxation in developing countries”.
World Development, Vol. 15, No. 9, pp. 1151­1161. doi: 10.1016/0305­
750X(87)90184­7
5. Borselli, F., Chiri, S. and Romagnano, E., 2012. “Patterns of reduced rates in
the European Union”. International VAT Monitor, January/February, pp. 13­
21.
6. Braz, C. and Correia da Cunha, J., 2009. “The redistributive effects of VAT in
Portugal”. Economic Bulletin, Banco de Portugal, Winter, pp. 71­86.
7. Business Review, 2013. Romanian authorities consider cutting the VAT for
meat products to 9 pct. [online]. Available at: <http://business­review.eu/fea­
tured/romanian­authorities­consider­cutting­the­vat­for­meat­products­to­9­
pct/>.
8. Caspersen, E. and Metcalf, G., 1994. “Is a Value Added Tax regressive? An­
nual versus lifetime incidence measures”. National Tax Journal 47, pp. 731­
746.
9. Cnossen, S., 1982. “What rate structure for a value­added tax?” National Tax
Journal, 35, pp. 205­214.
10. Cnossen, S., 1998. “VATs in CEE countries: A survey and analysis”. De Economist, 146 (2), pp. 227­255. doi: 10.1023/A:1003291011259
11. Cnossen, S., 2003. “How much tax coordination in the European Union?”
International Tax and Public Finance, 10, pp. 625­649. doi: 10.1023/
A:1026373703108
12. Copenhagen Economics, 2007. Study on reduced VAT applied to goods and
services in the Member States of the European Union, [pdf]. Available at:
<http://ec.europa.eu/taxation_customs/resources/documents/taxation/vat/
how_vat_works/rates/study_reduced_vat.pdf>.
13. Creedy, J., 1998. “Are consumption taxes regressive?” The Australian Economic Review, 31, pp. 107­116. doi: 10.1111/1467­8462.00057
14. Decoster, A. [et al.], 2010. “How regressive are indirect taxes? A microsimu­
lation analysis of five European Countries”. Journal of Policy Analysis and
Management, 29, pp. 326­350. doi: 10.1002/pam.20494
15. Dobrowolska, B. and Starzynska, W., 2011. Redystrybucyjne efekty opodat­
kowania konsumpcji gospodarstw domowych podatkiem VAT. In: A. Bor­
136
16.
17.
financial theory and
practice
39 (2) 115-137 (2015)
18.
19.
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
20.
21.
22.
23.
24.
25.
26.
27.
28.
29.
30.
kowska and U. Lagnowska­Szczesniak, eds. Spoleczno-Ekonomiczne Determinanty Konsumpcji w Polsce. Uniwersytet Opolski, pp. 113­137.
Ebeke, C. and Vazquez, F., 2014. Cyclical Behavior of VAT Revenues in Eu­
rope. Selected Issues Paper, IMF (Forthcoming).
Ebrill, L. [et al.], 2001. The Modern VAT. International Monetary Fund. doi:
10.5089/9781589060265.07
European Commission (EC), 2011. A retrospective evaluation of elements of
the EU VAT system. TAXUD/2010/CC/104 [pdf]. Available at: <http://ec.eu­
ropa.eu/taxation_customs/resources/documents/common/publications/stud­
ies/report_evaluation_vat.pdf>.
European Commission (EC), 2013. Study to quantify and analyse the VAT
Gap in the EU-27 Member States. Final Report. TAXUD/2012/DE/316 [pdf].
Available at: <http://ec.europa.eu/taxation_customs/common/publications/
studies/vat­gap.pdf>.
European Commission (EC), 2014. VAT Rates Applied in the Member States
of the European Union. Situation at 13th January 2014 [pdf]. Available at:
<http://ec.europa.eu/taxation_customs/resources/documents/taxation/vat/
how_vat_works/rates/vat_rates_en.pdf>.
European Union (European Commission and Eurostat), 2013. Taxation trends
in the European Union. 2013 edition, Publication Office of the European
Union, Luxemburg.
European Union Council Directive 2006/112/EC of 28 November 2006 on the
common system of value added tax.
Friedman, M., 1957. A theory of consumption function. Princeton: Princeton
University Press.
Garnier, G. [et al.], 2013. “Recent Reforms of Tax Systems in the EU: Good
and Bad News”. Taxation Papers, Working Paper, No. 39. Available at:
<http://ec.europa.eu/taxation_customs/resources/documents/taxation/gen_
info/economic_analysis/tax_papers/taxation_paper_39.pdf>.
Grandcolas, C., 2005. “The Occasional Failure in VAT Implementation: Les­
sons for the Pacific”. Asia-Pacific Tax Bulletin, 1­2, pp. 6­13.
Gustavsen, G. and Rickertsen, K., 2013. “Adjusting VAT rates to promote
healthier diets in Norway: A censored quantile regression approach”. Food
Policy, 42, pp. 88­95. doi: 10.1016/j.foodpol.2013.07.00
IMF, 2011. Revenue Mobilization in Developing Countries. Board Paper,
March 2013. Available at: <http://www.imf.org/external/np/pp/eng/2011/
030811.pdf>.
IMF, 2013. Fiscal Monitor: Taxing Times, October 2013.
Keen, M. and Lockwood, B., 2010. “The value added tax: Its causes and con­
sequences”. Journal of Development Economics, 92 (2), pp. 138­151. doi:
10.1016/j.jdeveco.2009.01.01
Keen, M., 2013a. “Targeting, Cascading, and Indirect Tax Design”. IMF
Working Paper, WP/13/57. Available at: <http://www.imf.org/external/pubs/
ft/wp/2013/wp1357.pdf>.
137
financial theory and
practice
39 (2) 115-137 (2015)
artur świstak, sebastian wawrzak, agnieszka alińska:
in pursuit of tax equity: lessons from vat rate structure adjustment in poland
31. Keen, M., 2013b. “The anatomy of the VAT”. IMF Working Paper, WP/13/111.
Available at: < http://www.imf.org/external/pubs/ft/wp/2013/wp13111.pdf>.
32. Matheson, T. and Swistak, A., 2014. “Modernizing the Icelandic VAT”. IMF
Country Report No. 14/291. Available at: <http://www.imf.org/external/pubs/
ft/scr/2014/cr14291.pdf>.
33. Metcalf, G., 1997. “The National Sales Tax: Who Bears the Burden?” CATO
Policy Analysis, No. 289.
34. Ministerstwo Finansów (MF), 2013. Preferencje podatkowe w Polsce, Nr. 3
[pdf]. Available at: <http://www.finanse.mf.gov.pl/documents/766655/8294824/
20140401+preferencje+podatkowe+w+Polsce.pdf>.
35. Musgrave, R. A., Case, K. E. and Leonard, H., 1974. The distribution of fiscal
burdens and benefits. Public Finance Quarterly, (2), pp. 251­311.
36. Nagel, T. and Neneman, J., 1995. Effects of a Reform of Private Consumption
Taxation in Poland. In: M. Belka and H.G. Peterson, eds. Economic Transformation in Poland: reforms of Institutional Settings and Macroeconomic Performance. Frankfurt: Campus Verlag, pp. 212­237.
37. OECD, 2010. “Choosing a Broad Base – Low Rate Approach to Taxation”.
OECD Tax Policy Studies, No. 19. Paris: OECD.
38. OECD, 2013a. Methodology and limitations in Taxing Wages 2013. Paris:
OECD.
39. OECD, 2013b. Taxing Wages 2013. Paris: OECD.
40. Pechman, J. A., 1985. Who Paid the Taxes:1966-85? Washington: Brookings
Institution Press.
41. Poterba, J. M., 1989. “Lifetime incidence and distributional burden of excise
taxes”. American Economic Review, 79 (2), pp. 325­330.
42. Prezes Rady Ministrów (PRM), 2010. Projekt ustawy o zmianie ustawy o finansach publicznych oraz niektórych innych ustaw. Uzasadnienie (Druk nr
3576) [doc]. Available at: <http://orka.sejm.gov.pl/Druki6ka.nsf/0/C9C153E
56DBDADA5C12577D7002BF9B1/$file/3576­uzasadnienie.doc>.
43. Radziukiewicz, M., 2011. Redystrybucja dochodów. Kto zyskuje? Kto traci?
Warszaw: Polskie Wydawnictwo Ekonomiczne.
44. Slintakova, B. and Klazar, S., 2010. “Impact of harmonization on distribution
of VAT in the Czech Republic”. Prague Economic Papers, (2), pp.133­149.
45. Tait, A. A., 1988. Value Added Tax. International Practice and Problems.
Washington: International Monetary Fund.
Cyclicality of bank capital
buffers in South-Eastern
Europe: endogenous and
exogenous aspects
ANA KUNDID NOVOKMET, PhD*
Article**
JEL: G, G8, P34
doi: 10.3326/fintp.39.2.2
*
This paper is based on author’s PhD dissertation. The author would like to thank two anonymous referees
for their useful comments and suggestions.
**
Received: June , 04
Accepted: December 19, 2014
The article was submitted for the 2014 annual award of the Prof. Dr. Marijan Hanžeković Prize.
Ana KUNDID NOVOKMET
University of Split, Faculty of Economics, Department of Finance, Cvite Fiskovića 5, 21 000 Split, Croatia
e-mail: [email protected]
40
financial theory and
practice
39 (2) 139-169 (2015)
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
Abstract
The interdependence between the regulatory capital ratio and macroeconomic indicators,
with reference to the phenomena cyclicality and pro-cyclicality is a widely emphasized
disadvantage of the capital adequacy concept. Redesign of the aforementioned concept
towards the countercyclical capital requirements is a kind of recognition of the creators
of the Basel standards of the previous oversights in its development. This paper aims to
explore empirically the direction, intensity and significance of endogenous and exogenous
determinants of the changes in banks’ capital buffers by taking into consideration both
the impact of the macroeconomic properties and the bank-specific characteristics of
South-Eastern Europe. More than 80 commercial banks in the period from 2000-2010 have
been encompassed by the research. Use of a dynamic panel analysis led to the conclusion
that the bank capital buffers decreased during the observed period, with the exception
of certain years during the economic expansion, which confirms the appropriateness of
regulatory requirements considering the countercyclical capital buffers. Nevertheless, it
might be that capital building and spending in the future will not follow the pattern from
the last decade due to the specificities of the observed period, as well as the banking
sector ownership transformations, economic and credit growth as well as asset prices
growth in the post-transitional period, and finally, the real crisis which spilled over onto
the financial sectors.
Keywords: bank capital buffers, cyclicality, commercial banks, South-Eastern Europe
1 INTRODUCTION
The capital requirements regulation for banks, in common with the practice of referring
to capital adequacy as the ultimate measure of the banking stability, has persisted as the
key instrument of prudential oversight worldwide for a more than two decades, despite
criticisms from both the academic community (Daníelsson et al., 2001; Rodríguez, 2003;
Saidenberg and Schuermann, 003; Benston, 007; Moosa, 00) and the banking lobbies
(Herring, 007; Kane, 007a, b). The phenomena of capital requirements cyclicality and
pro-cyclicality, among other controversies, are rather important points of reference when
the adequacy of this regulatory concept is being disputed (Jackson et al., 1999). While
cyclicality stands for macroeconomic impacts on a bank’s performance, pro-cyclicality
implies a bank’s reaction to the macroeconomic environment which amplifies macroeconomic fluctuations (Marcucci and Quagliariello, 2008:47).
Examinations intent on determining the implications of capital requirements for the size
of banking intermediation and which directly or indirectly tackled the question of the
macroeconomic consequences of the observed regulatory measure long dominated researches on the effects of capital requirements implementation. Moreover, in the first decade
of the capital adequacy standard implementation in practice the relationship between the
capital requirements and the volume as well as the structure of banking activities was at
the centre of research. In fact there was an endeavour to explain the credit contraction
recorded at the beginning of the 1990s in the countries that signed the Basel Accord by the
adoption of the capital requirements. Thus the hypothesis that the bank capital channel (i.e.
For these reasons, this research does not aim to simulate or estimate the macroeconomic
consequences of the implementation of capital requirements, indirectly throughout the
bank credit activity level, but rather concentrates on the cyclicality in the volume of
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
On the other hand, the bank capital channel works out only if the following assumptions
are satisfied (Francis and Osborne, 2009b:1): if banks do not have a sufficient capital buffer
through which they could insulate themselves from the movements in the credit supply
when regulatory changes occur; if capital enlargement is a costly process; if economic
agents are highly dependable upon bank loan financing. If the phenomena of demand-driven
and supply-driven credit rationing are taken into consideration, the macroeconomic effects
of “adequate” capitalization might be an argument for a reasonable or, on the other hand, a
more stringent criticism of the (supra)national prudential authorities. Thus, the importance
of principles and practice of the occurrence of business and economic cycles is temporarily
downgraded, while primacy in the explanations of movements in the aggregate credit and
investment level is given to the effects of the capital requirements or to the supply side of
the process. It is in this sense that the phrase capital crunch is used to indicate the cause
of pro-cyclicality or, to be more precise, the contraction in credit activities, particularly
in the case of more weakly capitalised banks and of loan categories assigned with high
risk weights (for example, loans to small and medium-sized firms, which are anyway
highly dependent on bank financing). However, the capital requirements are not the
only relevant factor of the credit activity level and structure, and recent researches with
inconsistent conclusions brought this issue sharply into focus. Gambacorta and Mistrulli
(2004) empirically confirm that the bank capital might cause shocks in aggregate lending;
Brissimis and Delis (2009) prove that bank specificities cannot be the main reasons for
the aforementioned conclusions; while Berrospide and Edge (00) verify the modest
impact of bank capital on lending. A solution to reconcile these contradictions is modelled
by Miyake and Nakamura (007), who ascribe to capital regulation long-term stabilising
effects that address the macroeconomic consequences of negative shocks to productivity.
On the other hand, in the short term it can have a pro-cyclical effect, because of which the
tightening of capital regulation needs timing precisely; this is the key practical implication
of their research. Generally, there are two groups of conclusions present in this type of
research: () the implementation of capital requirements did not induce the credit shock,
() the implementation of capital requirements combined with another supply-side and
with demand-side determinants of the credit level contributed to the development of the
credit shock.
4
financial theory and
practice
39 (2) 139-169 (2015)
the hypothesis of the financial accelerator induced by bank capital) provided the impulse
that led to the credit crunch, a regulatory mechanism that had pro-cyclical effects, was
examined in many empirical investigations. Most research on the macroeconomic effects
of the implementation of the capital requirements was oriented to the identification of
the relationship between the credit volume fluctuations (and consequently the economic
output) and the more restrictive capital regulation. Likewise, Van den Heuvel (008)
calculated that the compliance with capital requirements leads to a permanent loss in the
volume of consumption in the United States in the range of from 0.% to %.
4
financial theory and
practice
39 (2) 139-169 (2015)
capital requirements (i.e. regulatory indicators of financial leverage, to be more precise,
capital buffers). Additional justification for the exogenous treatment of economic trends
in the analysis of the capital requirements efficiency is found in Quagliariello (2008),
while Saidenberg and Schuermann (003:8) point out that if capital requirements procyclicality even exists, it is not clear how it can be confirmed. A distinction of the direct
effects deriving from the capital requirements prescriptions from the effects induced by
the shifts in the economic cycles remains an empirical challenge, although pro-cyclicality
has always previously been the most tested aspect of the capital requirements effects.
However, if the assumption of pro-cyclicality is a consequence of cyclicality is accepted
(Quagliariello, 2008:103; Marcucci and Quagliariello, 2008), then research that explores
the capital buffers or regulatory capital cyclicality might indirectly serve as a support for a
certain conclusions on the macroeconomic implications of the implementation of capital
requirements i.e. on the pro-cyclicality of capital requirements.
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
The size of banks’ regulatory capital is a volatile category, what might be a consequence
of endogenous and exogenous factors. Thus, it is justified to question the role of economic
cycles in the (non)functionality of capital requirements. Nevertheless, there are only a
few research works that combine microeconomic or bank-specific and macroeconomic
determinants of capital requirements volatility. With reference to the aforementioned, this
research acknowledges the empirical background in the following papers: Bikker and Hu
(2002), Ayuso et al. (2004), Bikker and Metzemakers (2007), Jokipii and Milne (2008),
Stolz (2007), Francis and Osborne (2009a), and Stolz and Wedow (2011). By taking into
consideration cyclicality in the capital itself rather than its cyclical implications, the paper
is in line with the current trends in empirical research into the issues and challenges related
to capital adequacy standards; theoretical and empirical findings have shown that capital
alone cannot be held responsible for a contraction in credit activity, as used to be argued,
irrespective of whether increases in capital requirements or shocks in capital size were
concerned. Moreover, banks pro-cyclical behaviour might be mitigated by maintaining
surpluses of capital above those regulatorily required (i.e. capital buffers) and with their
accumulation in the periods of economic expansion, which is currently being promoted
within the Basel III framework. Whatever the case might be, new versions of the aforementioned regulatory concept should not be designed without empirical evidence on the
endogenous and exogenous determinants of capital buffers cyclicality being provided
and taken into consideration.
2 CYCLICALITY OF BANK CAPITAL BUFFERS: A REVIEW OF EMPIRICAL
RESEARCH
A certain novelty of empirical research into the interdependence between the capital
requirements and macroeconomic trends is found in the empirical analysis of the impact
of macroeconomic tendencies on the volatility of the regulatory capital level. Following
the methodological framework developed by Shrieves and Dahl (1992), the changes in
bank capitalization (ΔCAPj,t) are determined with both exogenous factors (Ẽj,t) and endogenous, controllable or discretionary adjustments (ΔCAPMj,t), which can be summarized
as follows:
()
()
Altogether, the following equation can be written:
(3)
financial theory and
practice
39 (2) 139-169 (2015)
Furthermore, the discretionary changes in capital (ΔCAPMj,t) are defined as the difference between the target capital level (CAP*j,t) and the capital level in the previous period
(CAPj,t-):
43
where α is adjustment speed in the capital level.
Thus, banks have to weigh the costs and benefits from holding a certain level of capital
above the minimally prescribed. In such a manner, when determining the discretionary
capital they have to bear in mind the following costs: the remuneration cost of capital
requested by the shareholders, the costs of the franchise value loss, the costs of reputation
loss, bankruptcy costs, the costs of regulatory interventions and sanctions and the costs of
adjustment to the requirements of the regulator and the market participants, for example,
the credit rating agencies, potential and existing shareholders, uninsured depositors and
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
The aforementioned formula takes into account the fact that banks can have deviations
from the target capital level, i.e. that they are not always in a position to make ad hoc
adjustments to the targeted capital levels. Thus, banks usually maintain a higher level
of regulatory capital than that prescribed, which is also the key conclusion of the capital
buffer theory (Milne and Whalley, 2001). The surplus of capital above the minimally
prescribed level (by the regulators) is replaced by the formulation of capital buffer in
the rest of the paper, with the following note in mind: prudential authorities may request
banks that they perceive to engage in unusually high risk-taking behaviour to maintain
a capital adequacy level higher than that which is minimally prescribed. Therefore, the
capital buffer or the discretionary capital can be an outcome of the bank’s discretion (as
an object of regulation), as well as, the regulator’s discretion. Whatever the case might
be, it is evident that banks usually have the higher capital levels than those minimally
prescribed, while the motives for the maintenance of the capital buffers might be strategic
or reputational (Lastra, 2004:230; Bikker and Metzemakers, 2007:13; Jokipii and Milne,
008:44), i.e. have to be supported by the following considerations:
– cheaper refinancing and borrowing in the future, i.e. the market discipline functionality (practiced by the bank clients, creditors, credit rating agencies, shareholders),
– avoiding the costs of regulatory interventions in case of insufficient capitalization,
– granting loans in a recession, i.e. reduced pro-cyclical effects of bank capital (not
missing the chance for future bank growth),
– financing mergers and acquisitions,
– expansion in the business of banking,
– a more flexible bank management, and
– protection against unexpected losses.
44
financial theory and
practice
39 (2) 139-169 (2015)
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
other, wholesale creditors of banks (Ayuso et al., 2004:253). Consequently, the targeted
capital level is ambiguously determined. Moreover, it is dependent upon the bank specificities, which are proxied by numerous bank-specific variables in empirical researches.
Further, an exogenous changes of capital might be the outcome of regulatory pressures for
an increase of capital or unexpected changes in the volume of earnings caused by operating
income volatility or loans value impacting the level of provisioning, and is connected to
or is, in the first place, originated by the general economic context (Shrives and Dahl,
1992:446-447). Whatever the case might be, due to the manner in which the measure of
capital requirements has been designed, the level of risks taken and the changes in the risk
level ought to be reflected in the capital level. According to Shrieves and Dahl (1992) a
positive relationship between risk and capital is explained by banks’ efforts to mitigate
bankruptcy costs or by the risk aversion of bank managers, while a negative impact of
risk on capital can be the consequence of oversights in the deposit insurance premiums.
The level of risks taken is correlated with the expected or achieved return (which is an
outcome of the size and the structure of bank activities). Altogether, this makes bank
profitability as well as growth indicators the endogenous factors of volatility in capital
requirements. And finally, the cyclicality of capital requirements is determined by bank
characteristics and macroeconomic trends.
The key methodological features and conclusions of the reviewed empirical researches
(encompassed by table ) on the exogenous determinants of capital requirements cyclicality can be summarized in the following points:
– Research methodology selection. Almost all the research works reviewed employ
dynamic panel analysis.
– Data sample unit. Researches usually observe commercial and savings banks or
savings banks and cooperatives (Ayuso et al., 2004; Lindquist, 2004; Stolz, 2007;
Jokipii and Milne, 2008; Stolz and Wedow, 2011), which enables subsamples to
be analysed and conclusions to be made as to how much capital requirements of
various groups of credit institutions are volatile due to cyclical movements and how
much volatility is caused by bank specificities, banking sector characteristics and
by a given bank’s being a certain kind of credit institution. Research that focuses
on savings banks and cooperatives usually has significantly larger data samples
(according to number of observations) than those that take into consideration solely
commercial banks (Boucinha and Ribeiro, 2007; Francis and Osborne, 2009a).
– Data sample spatial characteristics. Researches that consider the banking sector of a
certain country are the most frequent (Ayuso et al., 004; Lindquist, 004; Boucinha
and Ribeiro, 2007; Stolz, 2007; Francis and Osborne, 2009a; Stolz and Wedow,
2011), while Stolz (2007) and Stolz and Wedow (2011) focus solely on one region
of one observed country, i.e. the western part of Germany, due to the disparities
in the economic development of the two parts after the unification of the country.
Cross-country analyses are usually related to the political or economic affiliation of
a country to a certain association, e.g. the Organization for Economic Co-operation
and Development (OECD) (Bikker and Metzemakers, 2007) or the European Union
(EU) (Jokipii and Milne, 008). An exception to this is constituted by Fonseca and
González (2010) and Fonseca et al. (2010), who analyse the banking sectors of 70 and
92 countries worldwide, respectively. Excluding the last mentioned research works,
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
According to the presented research findings, it is evident that there is a gap in the
empirical literature on the research issue for the South-Eastern European countries. In
fact, as the capital adequacy standard was aimed at the most developed countries, or
rather, at internationally active banks originating in these countries, it is explainable why
research works for those countries outnumber those for developing countries, which
adopted the Basel Committee recommendations in their national regulatory acts at a
later date. However, numerous researchers warn that the effects and implementation of
capital requirements might be significantly different in countries at different economic
development levels (e.g. Caprio and Honohan, 1999; Morisson and White, 2005). This
paper acknowledges that idea.
145
financial theory and
practice
39 (2) 139-169 (2015)
it can be concluded that investigations have been carried out only for developed
countries, which means that there is an urgent need to bridge the research gap with
respect to the banking sectors of developing countries. Interestingly, in this theme,
research into European countries dominates, while there is little or no research related
to the United States (except a part of the research by Jokipii and Milne, 0).
– Data sample time period characteristics. The shortest time period range, that of 7
years, is encompassed by Lindquist (004), while the longest time period is found in
Ayuso et al. (2004), Bikker and Metzemakers (2007), Boucinha and Ribeiro (2007),
Stolz (2007) and Fonseca et al. (2010) with more than 11 years encompassed by the
data sample. Empirical researches at the level of a single country’s banking sector
have not taken into consideration data later than the year 006 (Francis and Osborne,
2009a).
– Variables selection. Most of the research aims to examine the impact of macroeconomic and bank-specific variables on the capital buffers which is set out as the
dependent variable, while some researches, e.g. Bikker and Metzemakers (2007)
and Francis and Osborne (2009a) use also the capital adequacy indicator. Other ones
likewise Stolz (2007) or Stolz and Wedow (2011) use an indicator of regulatory
capital over total assets or equity to total assets ratio, as do Bikker and Metzemakers
(007). Economic trends are usually described by taking into account the real gross
domestic product (GDP) growth.
– Impact of economic cycles on capital requirements volatility. All research (at the
level of the overall sample, as there are some differences in the subsample approach)
confirmed that capital buffers increase in an economic downturn, and that they tend
to decrease in periods of economic expansion.
– Other conclusions. Commercial banks have lower capital buffers than savings banks
and/or cooperatives (Lindquist, 004). In addition, the results reveal a positive
relationship between capital buffers and economic growth in small banks and in
cooperatives (Jokipii and Milne, 008) due to the earnings-retaining policy of these
credit institutions in periods of expansion and a slower growth of placements (and thus
the risk-weighted assets) as they mainly finance themselves with their core deposits.
Jokipii and Milne (2008) confirmed that there is a difference in the cyclicality of
capital buffers between the newly acceded countries and the older member countries
of the European Union; in the older member countries there is a negative correlation
between capital buffers and economic growth, while in the newly acceded countries
there is a positive correlation.
29 OECD
Bikker and
Metzemakers countries; 199000
(007)
Lindquist
(004)
3 savings
banks and 6
commercial
banks/Norway;
1995-2001
Dynamic
panel data
model; GMM
estimator
The return on equity and the ratio of non-performing loans negatively impact the buffers.
Lagged dependent variable has a positive influence. Bank size has a negative effect, as
does growth of loans. Commercial banks’ buffers are more robust to negative influences
of the business cycles than savings banks.
There are differences in the obtained results for the savings and commercial banks. A
negative connection between the risk and the dependent variable is recorded for savings
banks. The debt price and the buffers size are negatively related, which implies that banks
with a lower regulatory capitalization pay higher costs of debt financing. Larger savings
banks have smaller buffers than small-sized savings banks. There is also a negative
To explore the
GLS Random- relationship between the risk and capital for the commercial banks. A more stringent
determinants of capital
buffers with focus on the Effects Model regulatory oversight over the commercial banks increases the buffers. Besides, a negative
impact of the reserves for unidentified losses, and a positive impact of the losses on the
microeconomic variables.
level of capital is confirmed, which might lead to the following conclusions: banks use
reserves as an alternative to increasing capital buffers, and banks build up their buffers
after the period of losses no matter what their price. Bank size negatively drives its buffers.
The main conclusion is that banks do not enlarge their capital because of increased risks.
In the equation in which the equity to assets ratio is the dependent variable, it was found
that the lagged dependent variable has a positive impact. This means that banks gradually
To explore the impact
adjust their capital to the targeted level. Credit risk indicators have a negative impact on
Dynamic
of macroeconomic and
capital, while the effect of the return on assets is a positive one. Results of the equation
panel data
bank-specific variables
with the regulatory capital being a dependent variable are similar to the aforementioned,
on the bank capitalization model; GMM
with a significant difference in the (country level) cost of capital (or the average return on
level and the regulatory estimator
equity of the banking sector) in the previous period being significant and having a negative
capital level.
sign. Subsamples analysis (when the banks’ size was a criterion variable for the
subsampling) reveals differences in the results for the banks according to their size.
Methodology
Results
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
Data sample
characteristics / Research assumptions
Authors
and goals
spatial and time
period attributes
4 commercial
To explore the effects
Ayuso, Pérez
and savings
of economic trends on
and Saurina
banks/Spain;
capital buffers.
(004)
1986-2000
There is no strong evidence of cyclicality in the
equity financing with reference to the economic
trends. There is some evidence of regulatory
capital cyclicality.
Economic growth negatively impacts the capital
buffers of the commercial and savings banks.
The commercial banks have, on average, lower
capital buffers in comparison to the savings
banks.
The business cycle has a negative impact on
the capital buffers.
Main conclusions on the exogenous
determinants of capital buffers cyclicality
financial theory and
practice
39 (2) 139-169 (2015)
Table 1
A review of empirical research on capital requirements cyclicality
46
To explore the
determinants of capital
buffers.
7 banks/
Portugal;
1994-2004
492 savings
banks and 2,159
cooperatives/
Germany
(western part);
1993-2003
468 banks
(commercial,
savings and
cooperatives)/
EU-25;
1997-2004
Boucinha
and Ribeiro
(007)
Stolz
(007)
Jokipii and
Milne
(008)
To explore the impact of
economic trends on the
capital buffers, as well as
on the ratio of regulatory
capital to total assets and
risk-weighted assets to
total assets. Dependent
and independent
variables (without
dummy variables) are
defined as changes, rather
than the level of the
observed indicator.
To identify the
dependence of the capital
buffers on economic
growth. Subsampling
analysis according to the
criterion of the newly
acceded and older
members of the EU, as
well as the bank size and
the bank type criterion are
assumed to be relevant in
explaining the relationship
between economic growth
and capital level.
Research assumptions
and goals
Dynamic
panel data
model; GMM
estimator
Results reveal capital buffers cyclicality in
dependence with the economic growth. Higher
economic growth reduces buffers. Stock markets
growth positively impacts the capital buffers.
Main conclusions on the exogenous
determinants of capital buffers cyclicality
There is a difference in the cyclicality of the capital buffers between the new and the old
members of the EU; whereas, in the old members a negative movement of the capital
buffers and the economic growth is recorded, in the new members that relation is a
positive one. That means that in the old members economic growth follows a reduction in
capital and vice versa. In the new members economic growth follows growth in the capital
buffers. A negative impact of the cost of equity (the return on equity) and of bank size on
the capital buffers is confirmed. Credit risk positively determines the buffers in most of the
subsamples. The lagged dependent variable has a positive impact on the capital buffers, as
it was expected, which is also proven for reinvested earnings. The growth of loans reduces
the buffers.
There is a negative relationship between
economic growth and capital buffers.
Commercial, savings and big banks confirm
a negative relationship, while the data sample
which is composed of small-sized banks and
cooperatives confirms a positive relationship.
A positive relationship of the latter mentioned
banks is a consequence of earnings-retaining
policy, retaining earnings being more frequent
in economic growth periods. Furthermore,
small-sized banks and cooperatives have a
slower growth of placements due to a mainly
core deposit financing policy.
The lagged dependent variable in the equation where the change in the capital buffers is
There is inverse relationship between the
dependent variable has a positive sign, as well as the ratio of liquidity and the ratio of loan
economic cycles and the changes in capital
loss reserves to total assets. The bank size and the return on assets have a negative impact
buffers.
on the changes in capital buffers.
The lagged dependent variable has a positive impact on buffers, which confirms the
assumption that banks gradually increase their capital buffers. A negative impact of the
loan loss reserves proves that reserves are a substitute for the capital surplus. Banks with
larger investments in stocks maintain larger capital buffers, while the bank size negatively
impacts the buffers. Higher profitability and smaller volatility reduce the capital buffers.
Dynamic
panel data
model; GMM
estimator
Dynamic
panel data
model; GMM
estimator
Results
Methodology
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
Data sample
characteristics /
spatial and time
period attributes
financial theory and
practice
39 (2) 139-169 (2015)
Authors
47
Banks with the higher quality of capital have higher capital adequacy ratios. The level of
risk in the current period is negatively connected with capital, while the level of risk in the
previous period is positively connected with capital. The ratio of loan loss reserves to total
assets has a positive impact on capital. Bank size negatively drives the capital adequacy.
Partial
The market discipline proxy with the share of the subordinated debt in total liabilities
Adjustment
increases the capital adequacy. Variables which approximate the adjustment costs of
Model;
capital (lagged capital adequacy and the Tier indicator) positively impact the capital
Random
adequacy. Estimations of the various subsamples revealed the following results: large
Effects Model,
banks increase their capital adequacy more, if they are under the regulatory pressures, than
Arellano-Bond
small banks; banks with lower capitalization increase their capital adequacy more; if banks
and Blundellare under regulatory pressure in periods of economic expansion they increase their capital
Bond GMM
more than in times of recession; banks under market pressures experience a weaker impact
estimator
of the capital requirements, and a higher impact of risk on the capital adequacy is expected
in these banks. All the aforementioned results were obtained for the period 1998-2006.
Similar results were obtained for the period from 1990-1995, with reference to the
estimated parameters signs, but they usually lack statistical significance.
To identify what,
determines the banks’
capital adequacy levels,
how and how much, and
the nature of the role of
prudential regulation in
this process. The bankspecific and
macroeconomic
determinants of the bank
capitalization are thus
explored. Besides, the
research aims to check
whether the regulatory
requirements’ impact
changes depending on
bank characteristics and
the overall economic
activity.
The dependent variable
is the capital adequacy
ratio. The quality of
capital is measured with
the Tier capital. Risk is
proxied by the riskweighted assets to total
assets.
68 banks in the
period 1998006 and 47
banks in the
period 19901995; United
Kingdom
Francis and
Osborne
(2009a)
Results
Methodology
Research assumptions
and goals
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
Authors
A negative connection between GDP growth
and the level of capital is recorded, and thus the
(pro)cyclicality phenomenon asks for an
additional empirical estimations.
Main conclusions on the exogenous
determinants of capital buffers cyclicality
financial theory and
practice
39 (2) 139-169 (2015)
Data sample
characteristics /
spatial and time
period attributes
48
,36 banks/
92 countries;
1990-2007
492 savings
banks and 2,139
cooperatives/
Germany
(western part);
1993-2004
Fonseca,
González
and Pereira
da Silva
(00)
Stolz and
Wedow
(0)
To explore the impact of
economic trends on the
capital buffers, as well as
on the ratio of regulatory
capital to total assets and
the ratio of risk-weighted
assets to total assets.
Source: Author’s presentation.
,337 banks/
70 countries;
1995-2002
Fonseca
and
González
(00)
To explore the
determinants of capital
buffers by taking into
account the bank-specific
variables and the
disparities in the
supervisory, regulatory
and accounting standards
of the observed countries.
To explore the effect of
capital buffers on the
banks’ credit and deposit
price in the developed
and developing countries.
Further, the determinants
of capital requirements
cyclicality are identified
with a presumption of
their lower volatility in
the developing countries
in comparison to the
developed countries.
Research assumptions
and goals
Dynamic
panel data
model; GMM
estimator
Dynamic
panel data
model; GMM
estimator
Dynamic
panel data
model; GMM
estimator
Methodology
There is some evidence of a negative impact
of the GDP growth on the capital buffers.
There is a negative relationship between the
GDP growth and capital buffers.
An economic downturn positively affects the
capital buffers. Weakly capitalized banks do not
increase capital in periods of expansion and
downturn (they decrease it, rather), nor do they
decrease their risk-weighted assets in the
downturn period (on contrary, they increase it).
The cost of deposits in the previous period positively influences the buffers, while bank
size has a negative impact. The credit price in the previous period positively impacts the
buffers.
There is a difference in the buffers’ movements during the business cycles depending on
bank capitalization. Weakly capitalized banks reduce their buffers in periods of economic
expansion and downturns. The lagged dependent variable positively impacts the buffers,
while bank size has a negative sign. Banks with a higher liquidity tend to maintain larger
capital buffers. Well capitalized banks increase their regulatory capital over the total assets
in the economic expansion and downturn, while for weakly capitalized banks the results
are opposite. Well capitalized banks do not change their risk-weighted assets during the
business cycle, while weakly capitalized banks increase them during economic expansion
and economic downturns.
Main conclusions on the exogenous
determinants of capital buffers cyclicality
The lagged dependent variable positively impacts the capital buffers, as well as, market
power (except if it is at a substantial level) and the costs of deposits. The bank size
and the credit risk indicators negatively impact the buffers.
Results
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
Data sample
characteristics /
spatial and time
period attributes
financial theory and
practice
39 (2) 139-169 (2015)
Authors
149
150
3 EMPIRICAL RESEARCH CONCERNING THE BANKING SECTORS OF SOUTHEASTERN EUROPE
3.1. DATA, METHODOLOGY AND MODEL DEVELOPMENT
financial theory and
practice
39 (2) 139-169 (2015)
Empirical research into the endogenous and exogenous determinants of the cyclicality
of capital buffers has been carried out on a data sample of commercial banks from the
9 South-Eastern European countries that were active in the period from 2000-2010 and
whose financial statements and financial indicators (which serve as approximations of
the endogenous aspects of capital buffers cyclicality) were available in the Bankscope
database. A distribution of banks by countries in the selected data set is given in the
appendix (table A).
Table 2
Data sources for the groups of indicators
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
Variable
Explanation
and/or
Data source
Microeconomic indicators
Microeconomic, i.e. financial, indicators of banks in the period
from 000-00 were selected in the data sample according to
the geographical criteria (Balkan States), status (active banks),
type (commercial banks) and financial statements
consolidation code (banks with consolidated (C and C) and
unconsolidated statements (U)).
Banking sector indicators
Minimally prescribed capital adequacy ratio
(for all the countries the officially prescribed
indicator is taken into consideration, while for
Romania in 2009 and 2010 the required rate is
proxied by the IMF recommendations of 0%,
minCAP
due to the non-transparency of this information
on the official website of the central bank of that
country for the observed years). Minimally
prescribed capital adequacy ratios by countries
are encompassed in table A in the appendix.
E_A
Equity to assets ratio for the observed banking
sector
Bankscope, Bureau van Dijk
Official websites of central
banks by countries (various
publications and decisions), the
European Central Bank, annual
publication Transition report in
the period 000-00,
Wisniwski (2005), Barisitz and
Gardó (008), Jokipii and Milne
(008), Athanasoglou (0)
The World Bank (World
Development Indicators &
Global Development Finance)
Macroeconomic indicator
GDP growth
Annual rate of growth of gross domestic
product
The World Bank (World
Development Indicators &
Global Development Finance)
Dummy variables
Dummy variable for an economic cycle (growth – dyEXP,
downturn – dyREC)
Dummy variables for the following years: 007 and 008
Source: Author’s presentation.
Δ GDP (%) £ 0 (downturn –
recession)
Δ GDP (%) > 0 (growth –
expansion)
Years which indicate a shift in
an economic trend
Table 3
Definition of banks financial indicators employed in the econometric model
Variable
NPL_L
ROA
GROWL
LLR_L
dydevBUFLow,
dydevBUFWella
LOWCA,
WELLCAb
Absolute value of capital buffer =
Bank capital adequacy ratio –
Minimally prescribed capital
adequacy ratio
Non-performing (bad debt) and
partially performing loans / Total
loans
Return on assets
Growth of loans
Loan loss reserves (identified and
unidentified losses) / Total loans
Undercapitalized banks in
comparison to the regulatory
prescriptions;
Adequately capitalized banks
Below-average capitalized banks;
Above-average capitalized banks
Group of indicators
Regulatory capitalization ratio
Credit risk (asset risk)
indicator
Overall profitability indicator
Growth indicator
Credit risk indicator
Regulatory pressure variable
Regulatory pressure variable
a
If a bank’s regulatory capital (capital adequacy ratio) is higher than the minimally prescribed plus a standard deviation of the minimally prescribed capital adequacy ratio, a bank is perceived to be a well-capitalized one (dydevBUFWell). In the opposite case, it is held to be under-capitalized (dydevBUFLow).
b
If a bank equity to assets ratio is higher than the average value of the aforementioned indicator for
the banking sector in which a bank operates, a bank is perceived to be well-capitalized (WELLCA).
In the opposite case it is considered undercapitalized (LOWCA).
Source: Author’s presentation.
The empirical research employed the econometric method of dynamic panel models. The
collected secondary data have a time and spatial component, and a suitable data analysis
method is thus an econometric method of panel analysis. Namely, use of the simple multiple
regression is not possible as it cannot be assumed that there is an independence between
the observations of one observed item during a time period (Škrabić, 2009:14). Thus, in
a situation of the analysis of bank financial indicators, the indicators of one period are
dependent on the same indicators in the previous period, i.e. there is a process of the firstorder autoregression. “The dynamic panel models contain the dependent variable which is
being lagged for one or more periods” (Škrabić, 2009:29). Furthermore, the collected data
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
ABBUFF
Explanation
151
financial theory and
practice
39 (2) 139-169 (2015)
Banking sector indicators and macroeconomic indicators are taken from the official websites of central banks of the countries encompassed by the data sample as well as from the
World Bank. Detailed insight into the data sources for the groups of indicators is provided
in table . By taking into consideration the empirical background, an econometric model
which encompassed the microeconomic variables from table 3 was developed. All the
selected variables report annual values. The data were taken in euros, while delta (Δ)
stands for the first difference of the observed variable value in order to cover the absolute
changes in the variable in the two successive periods.
152
financial theory and
practice
39 (2) 139-169 (2015)
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
are characterized with a larger number of groups (N) than the time component (t), which
this method handles very well. The empirical estimation of the panel data was performed
using the dynamic panel model, to be more precise with the GMM (generalized method of
moments) Arellano-Bond two-step estimator as well as the GMM Blundell-Bond two-step
estimator. In the empirical work on the research issue both the “difference” GMM ArellanoBond and the “system” GMM Blundell-Bond estimator as a certain improvement of the
Arellano-Bond in a case in which the autoregressive parameter value is near to one, and
the number of observations is relatively small, were used. The preliminary data analysis
using the Arellano-Bond estimator gave sufficient reasons for employing the improved
estimator. By using the Arellano-Bond estimator the values of the lagged dependent variables were below 0.15 at worst. However, the specificities of the models in which the
dependent variables were the first differences (absolute changes – Δ) of selected financial
indicators, and where lagged absolute values of the same variables were used as independent
variables, ask for an additional analysis if the estimated parameters of the aforementioned
independent variables are high. Additional analysis can be obtained by dropping the variable from the model or by employing the Blundell-Bond estimator when there is large
number of groups. Thus, from this point forward, only the Blundell-Bond estimations of
the econometric model will be presented. The model’s quality is evaluated using the tests
which are usually applied in the dynamic panel analysis likewise Sargan’s test as well as
autocorrelation tests. The data were analysed in the statistical package STATA .
The dynamic panel model for the selected variables is given with the following equation:
(4)
where i denotes an individual and t denotes time, μ is an intercept, γ is a parameter of the
lagged dependent variable, β1, β2,..., βK are the parameters of the exogenous variables, xit
are independent variables, αi is an individual-specific effect and εit the error term.
The basic dynamic panel model on the dependence of changes in the banks’ capital buffers
upon endogenous aspects has the following form:
(5)
The exogenous aspects of the macroeconomic variables influence are approximated with
the dummy variables and various interaction terms. Namely, due to the small number of
groups and observations for certain countries, the traditional use of the GDP growth rate
and other macroeconomic variables was not an advisable solution.
3.2 RESULTS OF EMPIRICAL RESEARCH
The descriptive data of the selected model variables precede the estimations of the econometric models. The mean value of the capital buffers (absolute level) has continuously decreased since the year 2001 (figure 1). This might lead to the conclusion that the
South-Eastern European banks have used their capital buffers, i.e. that they have increased the volume of risk-weighted assets more than they have built up their regulatory
Figure 1
The mean values of capital buffers (ABBUFF), the ratio of non-performing loans
(NPL_L) and the ratio of risk-weighted assets to total assets (RWA_A)
60
24
50
20
40
16
30
12
20
8
10
0
4
2000
2001
2002
2003
ABBUFF (left)
2004
2005
2006
NP L_ L (left)
2007
2008
2009
2010
RWA _ A (right)
Source: Bankscope.
According to the figure 1 a conclusion on the cyclical movement of the ratio of nonperforming loans can be made. A quality improvement of the credit portfolio of the data
sample banks is evident after the year 2002, what represents the first considerable effects
of liberalization and almost completed ownership transformation of the banking sectors
of the data-sample countries. A quality of credit receivables is getting worse in the last
two observed years and reaches the mean value of 4%, as it was in the year 000. The
lowest mean value of the non-performing loans to total loans was below 6% what was
recorded in the years 2005 and 2006. When looking at the figure it is clear that there is
cyclicality in the movement of capital buffers and in the variable which approximates a
credit risk as the potential key bank-specific determinant of capital buffer.
The ratio of loan loss reserves continuously falls down over the period 000-006 from the
level slightly lower than 12% to the level of about 5% (figure 2). The reserves were stagnating, from 006 to 008 and in the last two observed years they rose up to the level of about
7%. These movements correspond to the impaired loans movements. The fact on the most
of the bad loans being originated in good times makes the credit growth rates interesting
for an observation. It is evident that the credit portfolio of the data sample banks was on
average increasing a more than 40% annually until 008, while in the last observed year it
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
28
153
financial theory and
practice
39 (2) 139-169 (2015)
capital. Namely, when looking at the figure it is clear that the mean value of the ratio of
risk-weighted assets to total assets was continuously on the increase until the year 2009,
after which it started to decline. Besides, in certain countries minimally prescribed capital
adequacy ratios increased, which serves as an additional explanation of tendencies for the
capital buffers to decrease. In the year 2001 the mean value of the capitalization for the
data-sample banks was 7 percentage points higher than the minimally prescribed value,
while in the year 2010 the equivalent figure was less than 7 percentage points.
154
financial theory and
practice
39 (2) 139-169 (2015)
is on the level of just 10%. Although on the descriptive level, the aforementioned confirms
the thesis on the credit activity “freezing” or the so called credit crunch as a radical form of
the credit rationing process in the presence of the global financial crisis, and in the SouthEastern European area the shift towards negative economic tendencies with a stronghold in
the serious structural problems. With reference to profitability (figure 3) it is observable that
the mean value of the return on assets for the data-sample banks was higher than 0.5% until
2008, while from 2009 onwards that indicator is being zero or has a negative value.
Figure 2
The mean values of the ratio of loan loss reserves (LLR_L) and the credit growth
(GROWL)
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
12
70
11
60
10
50
9
40
8
30
7
20
6
10
5
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
0
2010
LLR_ L (left)
GROW L (right)
Source: Bankscope.
Figure 3
The mean value of the return on assets (ROA)
1.5
1.0
0.5
0.0
-0.5
-1.0
-1.5
-2.0
2000
2001
Source: Bankscope.
2002
2003
2004
2005
2006
2007
2008
2009
2010
Table 4
Panel data estimation of developed model with changes in capital buffers as
dependent variable
Explanatory
variables
DNPL_Lit
ROAit
ABBUFFi,t–1
LLR_Lit
dy2007
dydevBUFLowit
dydevBUFLowit*
dy2008
dydevBUFWellit*
dy2007
LOWCAit * dy2007
LOWCAit * dyREC
a
No. of
observations
No. of banks
Sargan test
(p-value)
First-order
autocorrelation
(p-value)
Second-order
autocorrelation
(p-value)
Model 3
Model 4
Model 5
Model 6
-0.1139***
(0.048)
0.00**
(0.0105)
.334***
(0.1537)
-0.408***
(0.0175)
-0.0640***
(0.007)
0.5782***
(0.0545)
0.9903***
(0.3)
-0.37***
(0.0149)
0.0229**
(0.0098)
1.0745***
(0.1494)
-0.3972***
(0.076)
-0.0630***
(0.0067)
0.5673***
(0.0544)
0.9608***
(0.300)
-.868***
(0.6899)
-0.1253***
(0.06)
0.04**
(0.0109)
.08***
(0.1652)
-0.3889***
(0.0165)
-0.0545***
(0.0046)
0.5820***
(0.0539)
-0.1139***
(0.048)
0.00**
(0.0105)
.333***
(0.1537)
-0.408***
(0.0175)
-0.0640***
(0.007)
0.5782***
(0.0550)
-0.3***
(0.068)
0.0**
(0.0095)
.0***
(0.1469)
-0.3970***
(0.068)
-0.0635***
(0.006)
0.5780***
(0.0554)
-0.06***
(0.06)
0.0253**
(0.0)
.77***
(0.706)
-0.3907***
(0.068)
-0.0572***
(0.0054)
0.5693***
(0.0548)
–
–
–
–
–
–
–
–
–
–
–
-3.4941***
(0.5903)
–
–
–
–
–
–
0.9903***
(0.3)
–
–
–
–
–
1.4592***
(0.2649)
–
–
-0.5561**
(0.2559)
-0.8738*** -0.7972*** -.033*** -0.8738*** -0.9019*** -0.783***
(0.2589)
(0.2554)
(0.2675)
(0.2589)
(0.807)
(0.2839)
–
–
–
–
–
66
66
66
66
66
66
88
88
88
88
88
88
0.063
0.0913
0.1105
0.063
0.0986
0.07
0.064
0.007
0.00
0.064
0.036
0.044
0.6785
0.6655
0.5902
0.6785
0.7594
0.4495
*** Statistically significant at 1% level, ** statistically significant at 5% level, * statistically
significant at 10% level.
Source: Author’s calculation.
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
GROWLit
Model 2
financial theory and
practice
39 (2) 139-169 (2015)
DABBUFFi,t–1
Model 1
155
156
Table 5
Panel data estimation of developed model with changes in capital buffers as
dependent variable
Explanatory variables
financial theory and
practice
39 (2) 139-169 (2015)
DABBUFFi,t–1
DNPL_Lit
ROAit
ABBUFFit–1
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
GROWLit
LLR_Lit
dydevBUFLowit * dyREC
Model 7
-0.1219***
(0.06)
0.04**
(0.0109)
1.0898***
(0.1657)
-0.3874***
(0.066)
-0.0545***
(0.0047)
0.5706***
(0.0538)
-1.9806***
(0.8179)
dydevBUFWellit * dyREC * ABBUFFi,t–1
–
dydevBUFWellit * dyEXP * ABBUFFi,t–1
–
a
No. of observations
No. of banks
Sargan test (p-value)
First-order autocorrelation (p-value)
Second-order autocorrelation (p-value)
-0.9113***
(0.680)
66
88
0.0966
0.08
0.4980
Model 8
-0.347***
(0.060)
0.036***
(0.08)
1.0551***
(0.67)
-0.3600***
(0.0175)
-0.0649***
(0.0067)
0.5196***
(0.0498)
–
-0.1159***
(0.0)
–
-0.440
(0.2857)
66
88
0.0925
0.076
0.3948
Model 9
-0.07***
(0.0)
0.000**
(0.0095)
0.9977***
(0.64)
–
-0.0501***
(0.0036)
0.5423***
(0.076)
–
–
-0.0952***
(0.0145)
-.663***
(0.44)
66
88
0.3385
0.0040
0.46
*** Statistically significant at 1% level, ** statistically significant at 5% level, * statistically significant at 10% level.
Source: Author’s calculation.
In the empirical estimation of the exogenous determinants of the cyclicality of the capital buffers, already specified the basic model has been supplemented with the dummy
variables and various interaction terms which reflect the economic environment. As the
years 007 and 008 are perceived to be the breaking points between the periods of the
financial and economic stability and distress in the financial system functionality as well as
indicating a shift towards economic downturn (as evident from the previous figures), the
models have been supplemented with dummy variables for the years 007 and 008. Apart
from that, the dummy variables that represent the periods of recession and expansion were
used in the following way: positive GDP percentage changes stand for an expansionary
period, while negative percentage changes denote a period of recession. With reference
to this, and taking into consideration the empirical background, annual GDP growth and
annual GDP decrease proxy the periods of expansion and recession in this paper, while
in practice it is common to use two successive annual (percentage) GDP changes in order
to reach a conclusion of whether an expansion or a recession is involved.
The lagged dependent variable DABBUFFi,t–1, lagged value of capital buffers ABBUFFi,t–1
and credit growth GROWLit are in a negative relationship with the changes in capital buffers, while the changes in the non-performing loans DNPL_L it, the return on assets ROAit
and the ratio of loan loss reserves LLR_Lit have a positive sign of the estimated parameters.
From figure 1, which shows the movement of the capital buffers volume in the observed
period, a negative sign of the lagged dependent variable was somewhat expected: since
the year 00 onwards, a continuous decrease in the mean value of capital buffers was
recorded. Analogously to this, the empirical estimation results confirmed that the lagged
variable ABBUFFi,t–1 has a statistical significance and that the estimated parameter has a
negative sign. The conclusion can be made that the growth of capital buffers in one period
will have a negative impact on the capital buffers in the following period, i.e. the growth
of capital buffers in one period leads to a decrease in the capital buffers in the following
period and vice versa. This might be a consequence of the risk-weighted assets being
increased and/or an increase in the minimally prescribed capital adequacy ratio, which
occurred in certain South-Eastern European countries in the observed period (Bosnia and
Herzegovina, Croatia and Serbia).
All models report a positive influence of the changes in non-performing loans DNPL_L it
on the changes in capital buffers, which might be interpreted in the following way: with
an increase in the credit portfolio riskiness, banks increase capital buffers and vice versa.
The mean value of capital buffers was continuously falling in the observed period, while
the mean value of non-performing loans decreased in the first five years. Since the year
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
3.3. INTERPRETATION AND DISCUSSION OF RESEARCH RESULTS
financial theory and
practice
39 (2) 139-169 (2015)
The empirical results of the estimated models with the changes in the capital buffers set
out as a dependent variable obtained with the Blundell-Bond estimator are presented in
the tables 4 and 5. From the correlation matrices given in the appendix it is clear that
independent variables are not strongly correlated, except for some interaction variables,
which are thus not simultaneously included in the model. In the model, which contains
dy2007, the appearance of the variable dydevBUFWellit instead of dydevBUFLowit does
not cause a sizeable deviation in results. Namely, the difference is visible in the sign of
estimated parameter of this variable which becomes a positive one, while the constant
term has a somewhat lower-estimated parameter and is statistically insignificant. Analysis
shows that well-capitalized banks dominate within the sample, due to the results of the
model with dydevBUFWellit * dy2007 and of the model with dy2007 being almost equal.
Besides the presented models, models with other possible combinations of interaction
terms were estimated, but those results are not presented here, as those variables which
were assumed to improve the basic model were statistically insignificant. In addition,
estimations of the data subsamples in which the economic cycle, i.e. the existence of
expansion or recession was the criterion variable for the data subsampling were made.
The results of the latter approach are given in the appendix (table A3).
157
158
financial theory and
practice
39 (2) 139-169 (2015)
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
008 this has exponentially increased up to the level of the mean value from the beginning of the observed period (figure 1). Thus, according to the graphical presentation, no
conclusion that an increase in the riskiness of assets caused an increase in the capital
buffers can be made. On the contrary, a decrease in the riskiness of assets contributed to
a reduction in capital buffers. In addition, subsamples analysis (table A3 in the appendix)
reveals that in the period of economic expansion there is no significant linkage between
the changes in the non-performing loans and the changes in capital buffers, while in the
recessionary period there is a certain level of significance. Bank profitability measured
by return on assets ROAit positively drives the changes in the bank capital buffers. The
higher the return on assets, the higher the capital buffers and vice versa, a decrease in
bank profitability reduces the capital buffers, which is logical if the bank regulatory
capital structure is borne in mind. These tendencies are even more noticeable during
economic expansion than in recession, and the estimated parameters from the table A3
in the appendix clearly serve as a proof.
In the presented estimations, the sign of the estimated parameter of the lagged value of
capital buffers ABBUFFi,t–1 is negative and significant. During the whole period the mean
value of capital buffers was ., i.e. the banks had capital adequacy that was .
percentage points on average higher than that minimally prescribed. Thus, a gradual
“spending” of the capital surplus is expected. The aforementioned confirms the practice
of having periods of capital accumulation and periods of capital “consumption”. For the
observed data set the accumulation of capital above that regulatorily required occurred in
the years 00 and 00 or even earlier, and after that the process of capital spending, i.e.
a decrease in the capital buffers took place. Thus the value of capital buffers was “only”
6.88708 percentage points on average at the end of 00. Finally, it can be concluded
that banks with higher initial capital buffers endeavour to maintain these levels, while
the banks with the lower capital buffers tend to build up their levels of regulatory capital,
which adds up to an empirical verification of the capital buffer theory.
The ratio of loan loss reserves LLR_Lit has a positive sign, which implies that an increase/
decrease of reserves leads to an increase/decrease of capital buffers. This relationship is
partly explainable by the fact that reserves, in a certain measure, contribute to a buildup
process of regulatory capital (directly throughout the special reserves for unidentified
losses in the supplementary capital I). Nevertheless, most of the loan loss reserves are the
reserves for identified losses, whose costs are a deductible item in the income statement,
and thus, indirectly, through their influence on profit, contribute to the regulatory capital
variations. Therefore, an alternative interpretation is possible. The figure on the mean value
of the ratio of loan loss reserves shows a long period of the ratio being decreased, together
with the continuous fall of the capital buffers (figure 2). That is, a radical decline in the
ratio of loan loss reserves is partially an outcome of the so called process of “cleaning”
the credit portfolios of bad debt (i.e. exclusion of the bad debt from the banks’ balance
sheets) which happened in the years of the observed banking sectors’ restructuration and
rehabilitation (in the years 2000 and 2001). With reference to this, the average value of the
ratio of net charge-off of loans to total loans was 6% in 000; it was below % in 00,
The year 2007 positively influenced the changes in the capital buffers. Until 2007 capital
buffers were, on average, trending down on a yearly basis. A slowdown in the credit
growth that took place after the year 007 as well as a slowdown in the growth of riskweighted assets, due to the shift in lending activity favouring economic agents whose
debt is assigned lower risk weights (e.g. the governmental sector), explain the tendency
for there to be slower usage of (or smaller changes in) the capital buffers. The variable of
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
Obviously, the changes (to be more precise, the droop in the observed data set) in the banks’
capital buffers can be explained in various ways. Besides the aforementioned factors, like
a decrease in profitability or recorded losses, an increase in the risk-weighted assets as well
as the minimally prescribed capital adequacy ratio, there is also the reason of the growth of
loans. Empirical findings confirmed that the growth of loans GROWLit leads to a decrease
in the capital buffers, due to an increase in the banks’ exposures, i.e. the volume of the
risk-weighted assets when there is a speed-up in credit growth. Furthermore, the size of the
banks’ capital buffers may vary due to the changes in the structure of exposures. A threat
from the so called cosmetic adjustments of the capital adequacy ratio (i.e. the unchanged
regulatory capital) or even from the false impressions of an increase in the capital buffers,
due to loans being made to economic agents with lower risk-weights in crisis episodes,
requires additional attention when the dynamics of the capital buffers movements are being
explained. Namely, the maintenance of capital buffers at a certain level, besides by an
increase in the regulatory capital, might be driven by the following factors: a decrease in
the risk-weighted assets, a credit “freezing” process, continuity of recording profits and
their reinvestment as well as by raising the minimally prescribed capital adequacy ratio.
Thus, in the banking sectors that experienced periods of extremely large recapitalizations
(e.g. the large banks in Croatia after an introduction of the marginal obligatory reserve) it is
realistic to expect a further decrease in the capital buffers. Nevertheless, the credit crunch
caused by the accumulated structural problems of the observed economies contributed to
the capital buffers being enlarged in the recession period.
159
financial theory and
practice
39 (2) 139-169 (2015)
after which it remained at the same or a somewhat lower level. At the same time, after a
sizeable growth of the capital buffers (in the first two observed years), a long period in
their decline followed. Thus, in a period of a small frequency of the occurrence of risk
events (loans charge-offs, costs recognition and value adjustments of reserves, due to an
increase in non-performing loans), banks reduce their capital buffers, aimed at diminishing
the effects of the unexpected losses. By taking into consideration a synchronization of the
variables of the non-performing loans and the loan loss reserves as well as the appearance
of a credit expansion in the period of their decrease (from 00-007), and on the contrary
the appearance of the credit crunch in the period of their increase (from 007-00), a
conclusion on the banks’ perceptions of credit risk and the pro-cyclicality of their credit
activity can be made. Thus, a decline in the banks’ capital buffers is the consequence of
a long period of expansive credit policies on the part of the banks, which are connected
to their perceptions of reduced credit risks and the occurrence of risk events, and which
are altogether reflected in their provisioning policies and loans classifications into groups
according to their quality.
60
financial theory and
practice
39 (2) 139-169 (2015)
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
the regulatory pressures for an increase in capital is proxied by the two types of dummy
variables that are interchangeably employed in the models, as well as the interaction terms
which contain those dummy variables. The first approximation is set out in form of the
banks with a lower or a higher capitalization in comparison to the minimally prescribed
capital adequacy ratios for the country (dydevBUFLow and dydevBUFWell). The second
one differentiates banks with a lower or a higher equity to assets ratio in comparison to
a mean value of a certain banking sector in which the bank operates (LOWCA and WELLCA). Most of the banks from the South-Eastern European area were well-capitalized,
which is supported by the fact of the sizeable capital buffers. That caused a number of
options for a definition and analysis of the regulatory pressure impact as being rather
small. Thus, the methodology that is regularly applied in the empirical background is
borrowed in this paper. With reference to this, banks with a capitalization that is lower
than or equal to that which is regulatorily required are treated as the banks under regulatory
pressures, while banks that according to that criterion were well-capitalized were banks
without regulatory pressures. Although this logic may be absolutely justified soon after
the capital requirements standard was put into effect, when there was a substantial number
of undercapitalized banks, a familiarity with the characteristics of banking sectors of the
observed countries serves as a support for the conclusion that weakly capitalized banks are
usually large banks or those that are “too big to fail”, while the banks with extremely high
capitalization are usually small-sized banks (for which there was a poor data continuity
in the available financial indicators), which do not serve prime customers, but most often
riskier clients. However, that knowledge might contribute to the economic interpretation of
the research findings. As was earlier pointed out, there were more well-capitalized banks
(dydevBUFWell) than weakly capitalized banks (dydevBUFLow) in the data set, when the
benchmark is the minimally prescribed capital adequacy ratio. In the case of the second
dummy variables group (LOWCA and WELLCA) the differences between the subsamples’
size with reference to capitalization were not so remarkable. The obtained results confirm
that the banks which are assumed to be under the regulatory pressure (dydevBUFLow)
had a greater drop in the capital buffers. A literal and isolated approach would lead to a
conclusion that bank capital requirements regulation did not add to the banks’ capitalization, i.e. that it did not fulfill its purpose. On the other hand, if there is an understanding
that accumulation and consumption of capital are carried out in phases or cycles, and that
in the observed data sample a decreasing trend in the capital buffers was recorded, then
the obtained results are expected. If lower capitalization is linked to the larger banks, the
estimated direction of influence is even clearer. This implies the following conclusion: the
weakly capitalized banks at least maintained a certain capitalization level, as the volume
of loans and the risk-weighted assets was continuously increasing during the observed
period. Thus, capital adequacy regulation is not irrelevant, as might be concluded from
the research results. Furthermore, the weakly capitalized banks could be perceived as the
most efficient ones in terms of the cost of capital, as they keep the required capitalization
at the minimum level, as do those whose market position obviously ensures them fast
and successful recapitalization if necessary in periods of economic expansion, which is in
contrast to the understanding that those banks are under regulatory pressures. Thus, higher
bank capital buffers may not necessarily have positive connotations, if riskier operations
and volatile secondary financial funds of the small-sized banks are borne in mind, which
altogether has to be compensated with substantial capital surpluses.
Empirical research into South-Eastern Europe confirmed that there is a certain amount
of evidence for the cyclicality of changes in the capital buffers, which is an outcome
of the financial characteristics of the observed banks, as well as the economic environment. In periods of economic expansion, banks increase their capital the most often as
a consequence of market pressures and their appetites for risk taking, while in economic downturns or crisis periods, banks usually have to increase their capital due to the
regulatory pressures. Meanwhile, the risk undertaken in the previous periods is already
significantly being materialized. The research into South-Eastern Europe reveals that the
changes in the level of the credit risks taken, as well as an increase in the profitability and
the loan loss reserves, positively determine changes in the banks’ capital buffers, while
the initial levels of capital buffers and credit growth negatively drive the changes in ca-
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
4 CONCLUSION
The initial introduction of capital requirements was followed by a radical decrease in the
aggregate credit level, and thus the problem of the capital requirements pro-cyclicality
was heavily exploited and empirically examined in the 1990s. Nevertheless, when it is
learned that the changes in the aggregate volume of loans could be also explained by
some other effects, apart from the effect of regulatory restrictions, the cyclicality of the
capital requirements begins to be more at the focus of research. Requirements for the
better capitalization of banks regularly occur in periods of economic distress. In such
times, the volume of the partially collectable and non-performing loans increase and
with the drop in banks’ credit and investment activity, their profitability declines. In the
aforementioned circumstances, credit and market risks enlarge, recapitalizations are less
available, while deposits stagnate or in an even worse case they decline. All this leads
to credit and, most often, equity rationing as well as fire sales of securities portfolios. In
that case, the volume of the “free” regulatory capital or the capital buffers may neutralize
any impairment of the key elements of the banking stability. Thus, it is rather important
empirically to identify what drives the volume of the banks’ capital buffers. The discovery
that there has been no empirical research into the determinants of changes in the capital
buffers of South-Eastern European banks inspired this research.
financial theory and
practice
39 (2) 139-169 (2015)
Using the interaction terms LOWCAit * dyREC and dydevBUFLowit * dyREC it is found
out that the capital buffers of the weakly capitalized banks decrease in the recession
period. This is additionally confirmed for well-capitalized banks (dydevBUFWellit *
dyREC * ABBUFFi,t–1). A particularly negative influence on the changes in capital buffers
was recorded in the year 008 (variable dydevBUFLowit * dy2008). On the other hand,
well-capitalized banks, which dominate the sample, in the expansion periods reduce
their capital buffers (dydevBUFWellit * dyEXP * ABBUFFi,t–1). To sum up, it might be
concluded that at the data set level, capital buffers are continuously trending downwards,
which is driven not only by the banks’ specificities, but also by the general economic
conditions.
6
6
financial theory and
practice
39 (2) 139-169 (2015)
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
pital buffers. Furthermore, the capital buffers increase in only certain years of economic
expansion, while during the recession they tend to decrease continuously. Although the
applied methodology is not completely comparable to the earlier empirical estimations
in the part of an approximation of the economic cycle, the conclusion can be made that
the obtained results are comparable to the empirical background. Notwithstanding the
initial capitalization level, banks mainly reduced their capital buffers over the observed
period. Thus, an affirmation of the counter-cyclical capital buffers seems to be an adequate
direction in the capital adequacy standard development. Nevertheless, a question can be
asked as to the repercussions from the implementation of counter-cyclical capital buffers,
as well as from the changes of the regulatory capital structure towards a higher share of
the Tier capital. Namely, the cost of capital is an integral part in loan pricing, and thus
it remains questionable what the implications of the mentioned regulatory restrictions
will be in addition to those from the selected or regulatory required capital buffers to the
risk and return of banks with a time lag. The importance of getting an empirical answer
to this question for South-Eastern European countries is additionally supported by the
high risk premiums of the observed countries, which enlarge the cost of capital and other
financing sources of banks in the crisis periods. Thus, there is a great challenge for the
prudential authorities in their attempts to maintain the banking sectors’ stability, which
can be summarized in the following: with the discussed regulatory changes, they must
not encourage banks to adjustments in their operations such as to reveal the countereffectiveness of the regulatory actions in the long-term.
Examinations show that the risk-free rate (which reflects the country risk premium) in the estimations of
the cost of equity capital, using the CAPM model, represents up to /3 of the overall cost of the bank equity
capital (e.g. King, 2009).
APPENDIX
63
Table a1
Distribution of banks by countries in the selected data sample
Number of banks (N=88)
Croatia
15
Romania
3
Bulgaria
Greece
Slovenia
7
Macedonia
7
Bosnia and Herzegovina
7
Albania
4
Source: Author’s presentation.
Table a2
Required regulatory minimum in the capital adequacy ratio in the South-Eastern
European countries in observed years
Country
2000 2001 2002 2003 2004 2005 2006 2007 2008 2009 2010
Albania
Bosnia and
Herzegovina
0
0
0
Bulgaria
Greece
8
8
8
8
8
8
8
8
8
8
8
Croatia
0
0
0
0
0
0
0
0
0
0
8
8
8
8
8
8
8
8
8
8
8
8
8
0
0
8
8
8
8
0
8
8
8
8
8
8
8
8
8
8
Macedonia
Romania
Serbia
Slovenia
8
Source: Official websites of central banks across countries (various publications and decisions), ECB,
annual publication Transition report in the period from 2000-2010, Wisniwski (2005), Barisitz and
Gardó (2008), Jokipii and Milne (2008), Athanasoglou (2011).
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
Serbia
financial theory and
practice
39 (2) 139-169 (2015)
Country
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
1.0000
0.0572
0.1819
-0.4745
-0.3953
-0.0160
-0.0660
-0.0348
-0.0344
-0.0660
0.0016
0.0892
-0.0339
-0.1554
-0.4054
1.0000
-0.1516
-0.0799
-0.1655
0.0838
-0.0986
0.0376
-0.0124
-0.0986
-0.0612
0.1314
0.0478
0.0355
-0.1044
Source: Author’s calculation.
deltaABBUFF
deltaNPL_L
ROA
lagABBUFF
GROWL
LLR_L
dy2007
dydevBUFLow
BUFLow_~2008
BUFWell~2007
LOWCA_dy2007
LOWCA_dyREC
BUFLow_dyREC
B~REC_lagA~F
B~EXP_lagA~F
deltaA~F deltaN~L
1.0000
-0.0464
0.0939
-0.0318
0.1252
-0.1117
-0.0436
0.1252
0.0598
-0.1887
-0.1289
-0.1733
0.0525
1.0000
0.1417
0.1703
0.0371
-0.0756
-0.0504
0.0371
-0.0684
-0.1970
-0.0683
0.3504
0.8433
ROA lagABB~F
1.0000
-0.1584
0.3881
-0.0493
-0.0145
0.3881
0.2198
-0.1987
-0.0448
-0.1483
0.2338
GROWL
1.0000
-0.1344
-0.0030
-0.0159
-0.1344
-0.0816
-0.0237
0.0059
0.0751
0.1347
LLR_L
1.0000
-0.0445
-0.0280
1.0000
0.7027
-0.1684
-0.0397
-0.1343
0.1165
1.0000
0.6297
-0.0445
-0.0312
0.1999
0.8931
-0.0449
-0.0688
1.0000
-0.0280
-0.0197
0.0654
0.3483
-0.0283
-0.0433
1.0000
0.7027
-0.1684
-0.0397
-0.1343
0.1165
1.0000
-0.1184
-0.0279
-0.0944
-0.0167
1.0000
0.2358
0.0909
-0.2605
1.0000
-0.0401
-0.0614
1.0000
-0.2077
1.0000
dy2007 dydevB~w BUF~2008 BUF~2007 LOW~2007 LOWCA_~C BUFLow~C B~REC_~F B~EXP_~F
. correlate deltaABBUFF deltaNPL_L ROA lagABBUFF GROWL LLR_L dy2007 dydevBUFLow BUFLow_dy2008 BUFWell_dy2007 LOWCA_dy2007 LOWCA_dyREC BUFLow_dyREC BUFWell
> _dyEXP_lagABBUFF
(obs=341)
financial theory and
practice
39 (2) 139-169 (2015)
PrinT ouT a1
From STATA 12: Correlation matrix
64
1.0000
B~EXP_~F
1.0000
0.3895*
-0.0510
0.0065
0.1447*
0.6764*
-0.0268
-0.0634
1.0000
-0.0575
-0.0299
-0.1524*
0.0250
-0.0789*
-0.0216
0.0764*
0.0180
-0.0744*
-0.1163*
-0.2385*
0.0497
1.0000
0.2192*
0.0343
-0.0089
-0.0553
-0.0295
-0.0147
-0.0699
-0.1454*
-0.0493
0.3956*
0.8715*
1.0000
-0.1673*
0.1352*
-0.0459
-0.0162
0.2117*
0.0996*
-0.1381*
-0.0573
-0.0945*
0.2987*
1.0000
-0.0716*
0.1241*
-0.0138
-0.0577
-0.0900*
-0.0184
0.0401
0.1404*
-0.0035
LLR_L
1.0000
-0.0199
1.0000
-0.0131
0.6425* 1.0000
0.0625 -0.1395* -0.0683* 1.0000
0.2858* -0.0345 -0.0227
0.2474* 1.0000
-0.0119 -0.0771 -0.0520
0.0623 -0.0207
-0.0282
0.0261 -0.0503 -0.1904* -0.0489
Source: Author’s calculation.
B~EXP_lagA~F
dydevBUFLow
BUFLow_~2008
BUFWell~2007
LOWCA_dy2007
LOWCA_dyREC
BUFLow_dyREC
B~REC_lagA~F
B~EXP_lagA~F
1.0000
-0.1529*
-0.0903
-0.1661*
0.0400
-0.0556
0.0162
-0.0105
-0.0996
-0.0203
0.1687*
0.0496
0.0361
-0.1083*
GROWL
1.0000
-0.0228
-0.0200
0.9948*
0.6585*
-0.1037*
-0.0347
-0.0771
0.0261
dy2007
1.0000
-0.1051*
dydevB~w BUF~2008 BUF~2007 LOW~2007 LOWCA_~C BUFLow~C B~REC_~F
1.0000
0.0595
0.0968*
-0.7010*
-0.3680*
0.0648
0.0025
-0.0187
-0.0016
0.0025
0.0128
0.0565
-0.0119
-0.2319*
-0.6339*
ROA lagABB~F
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
deltaABBUFF
deltaNPL_L
ROA
lagABBUFF
GROWL
LLR_L
dy2007
dydevBUFLow
BUFLow_~2008
BUFWell~2007
LOWCA_dy2007
LOWCA_dyREC
BUFLow_dyREC
B~REC_lagA~F
B~EXP_lagA~F
deltaA~F deltaN~L
financial theory and
practice
39 (2) 139-169 (2015)
. pwcorr deltaABBUFF deltaNPL_L ROA lagABBUFF GROWL LLR_L dy2007 dydevBUFLow BUFLow_dy2008 BUFWell_dy2007 LOWCA_dy2007 LOWCA_dyREC BUFLow_dyREC BUFWell_dyREC_lagABBUFF
> EXP_lagABBUFF, star(5)
PrinT ouT a2
From STATA 12: Correlation matrix (statistically significant at the 5% level)
165
66
Table a3
Subsamples results when economic trend is criterion variable for the data
subsampling (REC is for recession, EXP is for expansion)
Explanatory variables
financial theory and
practice
39 (2) 139-169 (2015)
DABBUFFi,t–1
DNPL_Lit
ROAit
ABBUFFi,t–1
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
GROWLit
LLR_Lit
a
Number of observations
Number of banks
Sargan test (p-value)
First-order autocorrelation (p-value)
Second-order autocorrelation
(p-value)
REC=1
-0.1191***
(0.0404)
0.086**
(0.04)
0.3983**
(0.844)
-0.674***
(0.06)
-0.073**
(0.034)
0.5085***
(0.30)
0.9302
(0.8258)
105
75
0.487
0.0658
0.9297
EXP=1
-0.70***
(0.030)
0.0091
(0.0192)
1.7256***
(0.373)
-0.3400***
(0.0154)
-0.044***
(0.0073)
0.4982***
(0.060)
-1.8949***
(0.3477)
6
73
0.487
0.0658
0.9297
*** Statistically significant at 1% level, ** statistically significant at 5% level, * statistically significant at 10% level.
Source: Author’s calculation.
67
financial theory and
practice
39 (2) 139-169 (2015)
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
REFERENCES
1. Ayuso, J., Pérez, D. and Saurina, J., 2004. Are Capital Buffers Pro-cyclical? Evidence from Spanish Panel Data. Journal of Financial Intermediation, 3(), pp.
249-264. doi: 10.1016/S1042-9573(03)00044-5
. Athanasoglou, P., 0. Bank Capital and Risk in the South Eastern European
Region. [pdf] Munich Personal RePEc Archive, 300 (May). Available at:
<http://mpra.ub.uni-muenchen.de/300//MPRA_paper_300.pdf>.
3. Barisitz, S. and Gardó, S., 2008. Financial Sector Development in Serbia: Closing
Ranks with Peers. Focus on European Economic Integration, 2, pp. 94-119.
4. Benston, G. J., 007. Basel II and Bankers’ Propensity to Take or Avoid Excessive
Risk. Atlantic Economic Journal, 35(4), pp. 373-382. doi: 10.1007/s11293-0079093-7
5. Berrospide, J. M. and Edge, R. M., 2010. The Effects of Bank Capital on Lending:
What Do We Know, and What Does it Mean?. [pdf] Federal Reserve Board, Finance and Economics Discussion Series, 44 (August). Available at: <http://www.
federalreserve.gov/pubs/feds/00/0044/0044pap.pdf>.
6. Bikker, J. A. and Hu, H., 2002. Cyclical Patterns in Profits, Provisioning and
Lending of Banks. [pdf] De Nederlandsche Bank, Staff Reports, 86 (May). Available at: <http://www.dnb.nl/binaries/sr086_tcm46-46863.pdf>.
7. Bikker, J. A. and Metzemakers, P. A. J., 2007. Is Bank Capital Procyclical? A
Cross-Country Analysis. Kredit und Kapital, 2 (August), pp. 1-29.
8. Boucinha, M. and Ribeiro, N., 007. The Determinants of Portuguese Banks’
Capital Buffers. [pdf] Banco de Portugal Financial Stability Report, pp. 153-164.
Available at: <https://www.bportugal.pt/en-S/BdP%0Publications%0Research/
AR0070_e.pdf>.
9. Brissimis, S. N. and Delis, M. D., 2009. Identification of a Loan Supply Function:
A Cross-Country Test for the Existence of a Bank Lending Channel. Journal of
International Financial Markets, Institutions & Money, 19(2), pp. 321-335. doi:
10.1016/j.intfin.2008.01.004
10. Caprio, G. and Honohan, P., 1999. Restoring Banking Stability: Beyond Supervised Capital Requirements. The Journal of Economic Perspectives, 3(4), pp.
43-64. doi: 10.1257/jep.13.4.43
. Daníelsson, J., Embrechts, P., Goodhart, C., Keating, C., Muennich, F., Renault,
O. and Song Shin, H., 00. An Academic Response to Basel II. [pdf] Special
Paper, London School of Economics Financial Markets Group and Economic &
Social Research Council Research Centre, 30 (May). Available at: <http://www.
bis.org/bcbs/ca/fmg.pdf>.
12. Fonseca, A. R. and González, F., 2010. How Bank Capital Buffers Vary Across
Countries: The Influence of Cost of Deposits, Market Power and Bank Regulation.
Journal of Banking & Finance, 34(4), pp. 892-902. doi: 0.06/j.jbankfin.2009.09.020
13. Fonseca, A. R., González, F. and Pereira da Silva, L., 2010. Cyclical Effects of
Bank Capital Buffers with Imperfect Credit Markets: International Evidence. [pdf]
68
14.
financial theory and
practice
39 (2) 139-169 (2015)
15.
6.
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
7.
18.
19.
0.
.
.
23.
4.
25.
26.
Banco Central do Brasil, 6 (October). Available at: <http://www.bcb.gov.br/
pec/wps/ingl/wps216.pdf>.
Francis, W. and Osborne, M., 2009(a). On the Behaviour and Determinants of
Risk-Based Capital Ratios: Revisiting the Evidence from UK Banking Institutions. [pdf] Financial Services Authority, 3 (March). Available at: <http://www.
fsa.gov.uk/pubs/occpapers/op31.pdf>.
Francis, W. and Osborne, M., 2009(b). Bank Regulation, Capital and Credit Supply: Measuring the Impact of Prudential Standards. [pdf] Financial Services Authority, 36 (September). Available at: <http://www.fsa.gov.uk/pubs/occpapers/
op36.pdf>.
Gambacorta, L. and Mistrulli, P. E., 004. Does Bank Capital Affect Lending Behavior?. Journal of Financial Intermediation, 13(4), pp. 436-457. doi: 0.06/
j.jfi.2004.06.00
Herring, R. J., 007. The Rocky Road to Implementation of Basel II in the United
States. Atlantic Economic Journal, 35(4), pp. 411-429. doi: 10.1007/s11293-0079094-6
Jackson, P., Furfine, C., Groeneveld, H., Hancock, D., Jones, D., Perraudin, W.,
Radecki, L. and Yoneyama, M., 1999. Capital Requirements and Bank Behaviour:
The Impact of the Basle Accord. [pdf] Bank for International Settlements, Basle
Committee on Banking Supervision, (April). Available at: <http://www.bis.org/
publ/bcbs_wp1.pdf>.
Jokipii, T. and Milne, A., 2008. The Cyclical Behaviour of European Bank Capital
Buffers. Journal of Banking & Finance, 32(8), pp. 1440-1451. doi: 0.06/j.
jbankfin.2007.12.00
Jokipii, T. and Milne, A., 0. Bank Capital Buffer and Risk Adjustment Decisions. Journal of Financial Stability, 7(3), pp. 165-178. doi: 0.06/j.jfs.
00.0.00
Kane, E. J., 007(a). Basel II: A Contracting Perspective. Journal of Financial
Services Research, 32(1-2), pp. 39-53. doi: 10.1007/s10693-007-0020-5
Kane, E. J., 007(b). Connecting National Safety Nets: The Dialectics of the Basel
II Contracting Process. Atlantic Economic Journal, 35(4), pp. 399-409. doi:
10.1007/s11293-007-9095-5
King, M. R., 2009. The Cost of Equity for Global Banks: A CAPM Perspective
from 1990 to 2009. [pdf] Bank for International Settlements Quarterly Review,
September, pp. 59-73. Available at: <http://www.bis.org/publ/qtrpdf/r_qt0909g.
pdf>.
Lastra, R. M., 004. Risk-Based Capital Requirements and Their Impact upon the
Banking Industry: Basel II and CAD III. Journal of Financial Regulation and
Compliance, 12(3), pp. 225-239. doi: 10.1108/13581980410810803
Lindquist, K.-G., 2004. Banks’ Buffer Capital: How Important is Risk. Journal of
International Money and Finance, 23(3), pp. 493-513. doi: 0.06/j.jimonfin.2004.01.006
Marcucci, J. and Quagliariello, M., 2008. Is Bank Portfolio Riskiness Procyclical?
Evidence from Italy Using a Vector Autoregression. Journal of International Fi-
27.
29.
30.
32.
33.
34.
35.
36.
37.
38.
39.
ana kundid novokmet:
cyclicality of bank capital buffers in south-eastern europe:
endogenous and exogenous aspects
31.
169
financial theory and
practice
39 (2) 139-169 (2015)
8.
nancial Markets, Institutions & Money, 8(), pp. 46-63. doi: 0.06/j.intfin.2006.05.00
Milne, A. and Whalley, A. E., 2001. Bank Capital Regulation and Incentives for
Risk-Taking. [pdf] Cass Business School Research Paper, December. Available
at: <http://papers.ssrn.com/sol3/papers.cfm?abstract_id=299319>.
Miyake, A. and Nakamura, T., 007. A Dynamic Analysis of an Economy with
Banking Optimization and Capital Adequacy Regulations. Journal of Economics
and Business, 59(1), pp. 14-27. doi: 10.1016/j.jeconbus.2005.12.00
Moosa, I. A., 2010. Basel II as a Casualty of the Global Financial Crisis. Journal
of Banking Regulation, 11(2), pp. 95-114. doi: 10.1057/jbr.2010.
Morrison, A. D. and White, L., 2005. Crises and Capital Requirements in Banking. The American Economic Review, 95(5), pp. 1548-1572. doi: 10.1257/000282
805775014254
Quagliariello, M., 2008. Does Macroeconomy Affect Bank Stability? A Review of
the Empirical Evidence. Journal of Banking Regulation, 9(2), pp. 102-115. doi:
10.1057/jbr.2008.4
Rodríguez, L. J., 2003. Banking Stability and the Basel Capital Standards. Cato
Journal, 23(1), pp. 115-126.
Saidenberg, M. and Schuermann, T., 003. The New Basel Capital Accord and
Questions for Research. [pdf] Wharton Financial Institutions Center, 03-4
(May). Available at: <http://fic.wharton.upenn.edu/fic/papers/03/0314.pdf>.
Shrieves, R. E. and Dahl, D., 1992. The Relationship Between Risk and Capital in
Commercial Banks, Journal of Banking & Finance, 16(2), pp. 439-457. doi:
10.1016/0378-4266(92)90024-T
Stolz, S. M., 2007. Bank Capital and Risk-Taking: The Impact of Capital Regulation, Charter Value, and the Business Cycle. Kiel Studies. Berlin Heidelberg:
Springer-Verlag.
Stolz, S. and Wedow, M., 2011. Banks’ Regulatory Capital Buffer and the Business Cycle: Evidence for Germany. Journal of Financial Stability, 7(2), pp. 980. doi: 10.1016/j.jfs.2009.09.00
Škrabić, B., 2009. Determinante razvoja financijskog sustava zemalja središnje i
istočne Europe. Specijalistički poslijediplomski rad. Zagreb: Ekonomski fakultet.
Van den Heuvel, S. J., 2008. The Welfare Cost of Bank Capital Requirements.
Journal of Monetary Economics, 55(2), pp. 298-320. doi: 0.06/j.jmoneco.007..00
Wisniwski, S., 2005. Bankers’ Perspectives – Dynamic Banking in a Changing
Market. In: I. Matthäus-Maier and J. D. von Pischke, eds. EU Accession – Financial Sector Opportunities and Challenges for Southeast Europe. Germany: Springer, pp. 3-6. doi: 10.1007/3-540-26963-0_1
Financial stress indicators
for small, open, highly
euroized countries:
the case of Croatia
MIRNA DUMIČIĆ, MSc*
Review article**
JEL: E44, E50, G0
doi: 10.3326/fintp.39.2.3
Special thanks go to Ana-Maria Čeh and two anonymous referees for exceptionally useful suggestions and
comments. The views stated in this work are the viewpoints of the author and do not necessarily express the
stance of the Croatian National Bank.
**
Received: June , 04
Accepted: February 0, 05
*
The article was submitted for the 2014 annual award of the Prof. Dr. Marijan Hanžeković Prize.
Mirna DUMIČIĆ
Croatian National Bank, Trg hrvatskih velikana 3, 10000 Zagreb, Croatia
e-mail: [email protected]
7
financial theory and
practice
39 (2) 171-203 (2015)
Abstract
The main objective of this paper is to construct high-frequency composite indicators of financial stress for Croatia that will enable the monitoring of the total
level of financial stress and its components on the domestic financial market. Emphasis is put on the choice of variables appropriate for small, open, highly euroized economies characterised by bank-centric financial systems dominantly
owned by foreign banks, shallow financial markets and dependence on foreign
capital.
Keywords: financial stress, financial stability, financial markets, systemic risk,
composite index, Croatia
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
1 INTRODUCTION
Many research papers refer to the strong and negative relation between stress episodes on the financial markets and financial and macroeconomic stability, and
emphasize their adverse impact on overall economic activity (Hubrich and Tetlow,
2012; Kliesen et al., 2012). In developing an analytical framework for monitoring
financial markets, the objective is to encompass all of the most important stress
sources that might cause the materialization of systemic risks. For this purpose,
individual indicators could be used. Nevertheless, in order to reduce the volume
of data from financial markets, while obtaining information about the total level
of stress in the financial system, these indicators are often aggregated into composite indices. The stronger the initial shock, the higher the correlation among the
different segments of the financial system. This makes the use of aggregated data,
reflecting developments in various segments of the financial system, the starting
point for an analysis of financial stress and systemic risk (Kota and Saqe, 2013).
The main goal of this paper is to construct high-frequency financial stress indicators for Croatia, which will enable the monitoring of the level of overall financial
stress in the domestic financial system, as well as in particular segments of the financial market, and which will provide timely indication of possible stress episodes and systemic risks materialization. In addition to the analysis of stress periods, there is an emphasis on the importance of information that indices can provide in tranquil periods on the financial markets.
In a review paper about the measurement of financial stress, Kliesen et al. (2012)
show that most of the composite financial stress indicators are constructed for
highly developed financial markets with diversified financial instruments and numerous financial indicators. However, by adjusting the set of variables included in
the index, they can also be useful for the analysis of developments in less developed financial systems. The main focus of this paper is thus not on development
of a methodology for computing composite indices, but on the identification of
variables that reflect financial stress in small, open, highly-euroized economies,
characterized by bank-centric financial systems dominated by foreign banks, shal-
2 FINANCIAL STRESS AND THE CHANNELS THROUGH WHICH
IT AFFECTS FINANCIAL AND MACROECONOMIC STABILITY
There is no universally accepted definition of financial stress. Dufrénot et al.
(0) describe it as a situation in which there is an enhanced probability of turbulence in the financial markets accompanied with a currency or balance of payments crisis, a sudden stop in capital inflows or capital outflows, stock market
crashes or the inability of a government to meet its liabilities. Holló (0) considers financial stress as disturbances in the financial system that unexpectedly
affect the price and turnover of financial instruments, which can be accompanied
by the collapse of systemically important financial institutions and the inability of
the financial system to carry out its main role and allocate financial resources, resulting in a considerable economic slowdown. It can be concluded that financial
stress implies disturbances in the normal functioning of financial markets and in
the process of financial intermediation that can spill over onto the real sector
(Hakkio and Keeton, 2009; Balakrishnan et al., 2009).
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
The paper is divided into five main parts. After an introduction there is a detailed
explanation of the theoretical concept of financial stress, and the specific features
of stress episodes are described, together with the channels through which they
can influence financial and macroeconomic instability from the perspective of a
small, open and highly euroized economy. The third section presents the problems
involved in constructing FSIs, and a review of the methods of aggregating individual indicators into a composite index is given. Part four explains the variables
assessed as being appropriate for the construction of the FSIs for Croatia, considering the characteristics of the financial system, and the country’s economic and
monetary characteristics. In order to check the robustness of the results, the indices are calculated by three methods. The main trends in financial stress in the period from early 00 to the end of 03 are also described, as well as the stress
episodes identified by the Markov switching model. The final part briefly summarizes the main results and contributions of this paper.
73
financial theory and
practice
39 (2) 171-203 (2015)
low financial markets and dependence on foreign capital because of het shortfall
in domestic savings. Apart from that, the focus is on the transmission mechanism
of financial market shocks to these countries. Particular attention is directed to
cases in which the main transmission channel of monetary policy is the exchange
rate, and in which there is no classical reference interest rate. In such situations,
the domestic money market is not necessarily a relevant source of bank funding,
and the interest rates in this market segment have a negligible impact on the borrowing costs for domestic sectors as they depend on trends in world financial
markets and levels of risk premiums for the country and the parent banks of domestic banks. Therefore, the analysis of variables that reflect financial stress in a
small open economy is the most important contribution of this paper. Its practical
contribution inheres in the potential use of the FSIs for economic policy makers
and financial market participants.
74
financial theory and
practice
39 (2) 171-203 (2015)
Although all stress episodes have their own specific features, Sinenko et al. (2012)
accentuate their common attributes, such as increased uncertainty related to the
value of financial assets and to the expectations of future economic developments,
high expected financial losses, increased risk aversion and the general tendency of
investors to keep less risky and more liquid financial assets. All of this increases
the instability and volatility of prices on financial markets and results in rising risk
premiums. Therefore, apart from the level, as a measure of financial stress the
volatility of given variables is also often used.
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
Financial stress is an inherent characteristic of a financial system. However, even
when it is not high, such information should not be neglected. Sinenko et al. (2012)
emphasize that prolonged periods of relatively low financial stress compared to
the long-term average have often been accompanied by an exaggerated optimism
on the part of market participants. This has resulted in excessive credit activity, a
rise in the prices of different asset classes and the accumulation of macroeconomic imbalances due to increased current account deficits and external debt. All
of this has increased risks to financial stability. If disturbances indicated by highfrequency indicators endure, there is a great probability that they will shortly affect low-frequency indicators through financial or trade channels.
Increased instability on financial markets usually results in higher risk aversion,
causing a rise in risk premiums and in the financing costs of domestic sectors. If
combined with reduced liquidity or even completely frozen financial markets, this
may result in the strong slowdown of capital flows. The research has shown that
emerging markets are more affected by this than developed countries because of
their increased vulnerability to potential sudden changes in capital flows (Catao,
006).
Financial institutions’ interlinkages are also an important source of financial stress.
In most European countries banks are the most important financial intermediaries,
and their exposure to financial stress can result in high macroeconomic costs
(Schou-Zibell et al., 2010). Because of their role, problems induced by the financial stresses that they have can rapidly spill over to other segments of the financial
system, such as the interbank money market or the payments system. The links
among financial institutions in the money market present potential channels for
the rapid spillover of risks and difficulties among financial institutions in a short
period. The speed of this process is proportional to the level of uncertainty and
information asymmetry.
An enhanced level of uncertainty related to financial stress can reduce the banks’
willingness to make loans. This motivates them to tighten their lending standards,
which reduces the demand for loans as borrowers might postpone or give up on
investments (Hubrich and Tetlow, 2011). In less developed financial systems the
role of bank loans in the financing of the economy is usually higher than in devel-
oped countries, in which the securities market is a noteworthy source of funding.
This makes the influence of financial stress on loan conditions and the loan supply
via this channel much more important for developing countries.
3 THE CONSTRUCTION OF A COMPOSITE INDEX
Specific features of small open economies with shallow financial markets and lack
of indicators used for the calculations of FSIs for developed countries imply a
great challenge in the choice of variables reflecting financial stress. According to
the literature, the most important potential sources of stress are the credit market,
foreign exchange market, inter-bank money market and capital market (Oet et al.,
2011; Jakubik and Slačik, 2013). In addition to domestic data, due to the increased
liberalization of financial flows and the fact that in most countries movements on
the domestic financial market are heavily dependent on international developments, data from foreign financial markets are also used. But, even if the created
indicators cover at a given moment all important potential sources of financial
stress, it is necessary continuously to re-examine the appropriateness of the set of
selected variables and to adjust it to the development of and trends in the financial
system. Therefore, FSIs calculated for different countries are not always directly
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
In highly euroized countries, one of the most important channels for possible spillovers of financial stress onto financial and macroeconomic indicators is the exchange rate. In such cases a significant part of the economy is exposed to a currency-induced credit risk deriving from the currency mismatches in debtors’ assets and liabilities. If a strong depreciation of the domestic currency occurs, there
is an increased likelihood of a considerable deterioration of the loan quality. In
these countries, the main monetary policy transmission channel usually is not the
interest rate, but the exchange rate. Therefore, the information contained in movements in domestic interest rates differs from those on developed markets, in which
the interest rate transmission channel is functioning. In addition to that, in some
banking systems that are dominantly foreign owned and that are reliant on parent
bank funding, domestic money market interest rates can have an almost negligible
impact on the cost of domestic sector borrowing. This cost primarily depends on
the perception of the country risk and parent banks’ risks, as well as on the liquidity of international financial markets.
financial theory and
practice
39 (2) 171-203 (2015)
In dominantly foreign owned banking systems, the parent banks of domestic
banks represent an important transmission channel of financial stress from international to domestic financial markets. Parent banks’ problems measured in their
increased risk premium affect not only the funding costs for their subsidiaries, but
also their strategy related to the operations of the subsidiary banks. This can
strongly influence credit activity and real economic developments in the countries
where they have large exposures. Balakrishnan et al. (2009) confirm that the most
pronounced channel for the transmission of financial stress to European emerging
markets during the recent crisis was through the western European banks.
75
76
comparable, but they can provide useful information about the financial stress
dynamics in different markets. Kliesen et al. (2012) present a detailed review of
indicators and methods used for the construction of FSIs, while Jakubik and Slačik
(2013) provide a useful example of the construction of financial instability indices
for CEE countries, including Croatia.
financial theory and
practice
39 (2) 171-203 (2015)
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
In the literature about early warning systems, use is often made of binary methods,
in which tranquil periods are marked with a zero and a crisis period with a one.
But in the construction of the FSIs, the objective is to create a continuous measure
that will show the level and the development of financial stress (Balakrishnan et
al., 2011). Financial markets’ conditions are never either absolutely good or absolutely bad, as might be concluded according to binary indicators, and they should
be observed relatively over a period of time (Oet et al., 2011). Therefore, Illing
and Liu (2006) observe financial stress through a continuous variable, where extreme values mark a crisis episode. Such an approach emphasizes the index dynamics, rather than precise definitions of the beginning or end of a crisis episode.
Commonly used methods for aggregating individual indicators into a composite
index include aggregation via variance-equal weights, principal component analysis, aggregation with the use of variance-equal or chained weights of variables
transformed with the use of the empirical cumulative distribution function and
aggregation in which the shares of given markets in the total financing of the
economy are used as weights (Jakubik and Slačik, 2013; Illing and Liu, 2003;
Sinenko et al., 2012; Puddu, 2008; Holló, 2012).
Although similar results are obtained irrespective of the method used, each one of
them has certain drawbacks. The shortcoming of variance-equal weights aggregation derives from the initial assumption that all the variables included in the index
are equally important. In this manner, greater importance is given to those market
segments represented in the index with more variables (Puddu, 2008). On the
other hand, weighting based on a single component in factor analysis results in a
fixed set of weights for the whole analysed period (Oet et al., 2011). In the transformation of variables with the cumulative distribution function the assumption is
that the gap between neighbouring variables is equal, which is usually not the
case, because during long stable periods the relatively small volatility of the original variables can seem greater after transformation than it actually is (OeNB,
03). At the same time, the weights determined by all of these three methods
have no economic significance, unlike the method where shares of individual markets in total loans in the economy calculated by aggregating bank loans, corporate
and government bonds and shares are used as weights (Illing and Liu, 003). The
potential use of the latter approach is relatively limited in less developed countries. Hence, for calculating the FSIs for Croatia the first three methods are used.
Calculation of the FSIs with various methods is the first test of the robustness of the
results.
3.1 WEIGHTING BASED ON vARIANCE-EqUAL WEIGHTS (FSI_vEW)
Weighting based on variance-equal weights is the most often used method for the
calculation of composite indices (Kliesen et al., 0). It implies the aggregation
of standardized variables into a single index, in which every variable has an equal
weight:
in which k is the number of variables included in the index, is the sample arithmetic mean for the variable Xi and σi is the sample standard deviation for the variable Xi.
financial theory and
practice
39 (2) 171-203 (2015)
()
77
3.2 AGGREGATION OF vARIABLES TRANSFORMED BY THE CUMULATIvE
DISTRIBUTION FUNCTION (FSI_CDF)
.
()
In this manner every observation is turned into the corresponding percentile of the
cumulative distribution function and takes a value between 0 and 1. Following
Sinenko et al. (2012), for the aggregation of transformed variables in the index,
weights for every variable are determined as a share of the transformed variable in
the sum of total transformed variables:
(3)
and the total index is obtained as:
.
(4)
3.3 PRINCIPAL COMPONENT ANALYSIS (FSI_PCA)
In practice, this method is used for easier interpretation of a large number of variables which are transformed into a smaller number of uncorrelated variables or
principal components (Anh and Mägi, 2009). This technique reveals the main
drivers behind data variation and the interlinkages between variables that are not
necessarily obvious. The correlations of the variables in the groups identified are
greater within the groups than among the groups. FSIs are determined as the first
principal component that explains the greatest part of the joint movement of the
variables used for the construction of the index:
FSI_pcat = xtα
(5)
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
Every variable included in the index is initially transformed with the use of the
cumulative distribution function. The greatest value of a given variable has the
highest rank and indicates the greatest degree of financial stress, while rank one
refers to the lowest recorded value of the indicator (Oet, 2011). Values around the
median correspond to the average level of stress. After the rank of every observation within the time series has been determined, the empirical cumulative distribution function is calculated as:
178
in which α is the weight vector (of the dimension number of individual variables
x 1) and xt is the vector of the values of the indicator (of the dimension number of
individual variables x 1) on the basis of which the indices are evaluated. The loadings determine the variables that have the greatest contribution to the explanation
of the joint movement of all the components of the aggregated index (table A1).
financial theory and
practice
39 (2) 171-203 (2015)
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
4 FINANCIAL STRESS INDEX FOR CROATIA
In choosing variables for FSI construction, the objective was to take into consideration as many as possible of the relevant segments of the domestic and foreign
financial systems that might affect the level of financial stress in Croatia. Focus has
been put on the specific features of the domestic financial markets, the availability
of high frequency data, the economic and monetary characteristics of the country
and relevant developments in foreign financial markets that might affect the stability of the domestic financial system. Daily data from 30 January 2001 to 18 December 2013 were used and risks were divided according to the stress origin – domestic or foreign, and also according to market segments (table 1). Apart from such
a division, the main difference compared to the financial instability indices constructed by Jakubik and Slačik (2013) is that these FSIs include more variables for
each market segment, as well as additional market signals and sources of risks such
as risks related to the mother banks, bid-ask spreads and country risk premiums.
Due to the high degree of euroization and the fact that in Croatia the main transmission mechanism of monetary policy is the exchange rate, several variables that reflect developments on the domestic foreign exchange market were used. Regardless of the direction in which the exchange rate moves, the increased EUR/HRK
bid-ask spread and its volatility indicate higher instability and uncertainty related
to the behaviour of market participants and signals an increased stress level. The
forward exchange rate is an indicator of market expectations about future movements in the EUR/HRK exchange rate. Since this figure is not available for every
day, particularly in the initial part of the observed period, the five-day moving average of this variable is used instead.
The index also includes the level of the weighted exchange rate of the kuna to the
euro, the Swiss franc and the US dollar. Although the EUR/HRK exchange rate is
the most important for the Croatian economy as the external debt and most of the
loans of domestic banks are in euros or indexed to the euro, Swiss franc movements
have become an important potential source of stress because of the strong credit
activity in this currency from 2005 to 2008. In this way the impact of the CHF/
HRK exchange rate has also been covered. At the end of 2008, Swiss franc loans
accounted for 16% of total loans, or 24% of all loans denominated in or indexed
to a foreign currency (CNB, 0).
In spite of the importance of the exchange rate for financial stability and notwithstanding the fact there is no typical reference interest rate on the domestic money
The spread between interest rates for short-term and long-term maturities reflects
the liquidity risk premium and is often used as an indicator of money market developments. Since trading in maturities longer than a week in the domestic money
market is negligible and there are many days when such transactions are not executed at all, it is not a reliable indicator for this market segment. In spite of this and
the mentioned limitations of ZIBOR, the FSI includes the spread between the interest rate on the three-month treasury bills of the Ministry of Finance (MF) and the
three-month ZIBOR. For ZIBOR there is a daily series of quotations, while the
interbank interest rates for a three month maturity are available only when such
transactions are executed, which significantly reduces the number of observations.
This spread is the most commonly used variable for FSI calculations as it indicates
the liquidity of the system by measuring the short-term credit risk and the premium
on risk-free government treasury bills (Kliesen et al., 2012; Illing and Lieu, 2006).
When money market liquidity is reduced or there is an increased risk that banks are
unable to repay their liabilities it is expected to increase (Cardarelli et al., 2009).
Another money market liquidity indicator is the use of Lombard loans. These are
overnight collateralized loans available to the banks every day up to the prescribed
amount of the nominal value of MoF treasury bills, at an interest rate set by the
CNB. They are granted at the bank’s request at the end of the working day. An
unpaid intraday loan is automatically considered a request for a Lombard loan.
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
There are several overnight interest rates on domestic money market: the overnight interest rate in interbank trading, the interest rate on the Zagreb Money
Market (ZMM) and Zagreb Interbank Interest Rate (ZIBOR). For calculating the
FSI, the interbank interest rates have been used for the period from September
2002, while for the previous period the ZMM interest rates are used. The coefficient of correlation for these two series in the coinciding period exceeds 0.9, confirming this is a reliable time series for the price of overnight borrowing. The advantage of this variable over ZIBOR is that it is at this rate that transactions are
really executed, while ZIBOR is based on the banks’ quotations which are not
obligatory and do not necessarily represent the rate at which transactions are executed. Although ZIBOR largely tracks the movement of the interbank interest
rates, it has not been perceived as the reference interest rate, which is confirmed
by the CNB bank survey according to which the most important money market
interest rate is the interbank interest rate (Ivičić et al., 2008).
179
financial theory and
practice
39 (2) 171-203 (2015)
market, which is also not a primary funding source for domestic banks, interest
rate movements and their volatility can nevertheless indicate the (in)stability of
the overall financial market. The dynamics and level of short-term interest rates in
the observed period were primarily determined by the surpluses or deficits of
banks’ kuna liquidity. Apart from the banks’ operations, the liquidity of the system
depended on the CNB’s activities, which, when necessary, maintained the exchange rate stability by restricting kuna liquidity.
180
The use of Lombard loans can indicate liquidity problems in individual banks, and
can also be used as an indicator of banking sector stability.
financial theory and
practice
39 (2) 171-203 (2015)
Table 1
Variables used for the calculation of the FSI divided according to market
segments
Financial market
segment
Domestic markets
variable
EUR/HRK bid ask
spread
EUR/HRK volatility
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
Foreign exchange
market
EUR/HRK forward
exchange rate
Exchange rate weighted
according to
commercial banks’ asset
structure
Level of O/N interbank
interest rates
Volatility of O/N
interbank interest rates
Money market
Volatility of turnover
on overnight market
Spread between 3M
MoF T-bills and 3M
ZIBOR
Lombard loans used
Securities market
Returns on CROBEX *
(–1) and its volatility
CROBIS return
volatility
Croatian kuna –
denominated
government bond
maturing in 2019,
bid-ask spread
Measure
Source
Market liquidity, uncertainty,
information asymmetry
Uncertainty, information
asymmetry, pressures on
exchange rate stability
Market expectations,
appreciation or depreciation
of domestic currency
Bloomberg, author’s
calculation
Depreciation or appreciation
of domestic currency
CNB, author’s
calculation
Liquidity of banking system
CNB
Uncertainty, information
asymmetry
Uncertainty, information
asymmetry, increased need
for liquidity
CNB, author’s
calculation
Market liquidity, credit risk
Bloomberg, MF,
author’s calculation
Problems with liquidity of
some participants of financial
markets, indicator of stability
of banks
Uncertainty and information
asymmetry on capital market
Uncertainty and information
asymmetry on bonds market
Measure of market liquidity,
reflects liquidity of bonds of
a given issuer
CNB, author’s
calculation
CNB
CNB, author’s
calculation
CNB
ZSE, author’s
calculation
ZSE, author’s
calculation
Bloomberg, author’s
calculation
Foreign markets
LIBOR – OIS spread
Money market
EONIA volatility
EURIBOR 6M and
EONIA spread
Uncertainty on global money
market, risk of interbank loans
on money market
Uncertainty on euro money
market, market liquidity
Liquidity premium
Bloomberg, author’s
calculation
Bloomberg, author’s
calculation
Bloomberg, author’s
calculation
Financial market
segment
Foreign markets
variable
EMBI for Croatia
181
Bloomberg
Stability of operations of parent
Bloomberg, author’s
banks, costs of borrowing,
calculation
capital withdrawal
Liquidity measure, reflects
liquidity of securities
Bloomberg, author’s
calculation
Measure of the expected shortterm volatility on the capital
market
Bloomberg
It is not very likely that developments in the domestic capital market will significantly influence the overall financial stability, because it is still relatively underdeveloped and is not significant funding source for domestic corporates. Nevertheless, enhanced volatility in the stock prices induced by the instability in other
market segments or macroeconomic developments can indicate an increased degree of risk and instability in the financial system. Hence the FSI includes return
on Zagreb Stock Exchange index CROBEX and its volatility, since a fall in the
prices of shares and their increased volatility can indicate stress in this market
segment. Return on CROBEX is calculated as the annual change in the index
multiplied by minus one so that a fall in the price of shares indicates an increased
financial stress (Balakrishnan et al., 2009).
The domestic debt securities market is characterized by low liquidity where government bonds account for more than 90% of bond market capitalization. CROBIS is a bond price index calculated on the basis of market capitalization calculated at the end of each trading day as average daily price weighted by the quantity for all bonds included in the index and its volatility is included in the FSI. The
FSI also includes the bid-ask spread for the kuna-denominated government bond
maturing in 2019, which measures the liquidity risk. A low level of this spread
characterizes liquid markets with low transaction costs (Holló, 2012).
From the variables representing the foreign financial markets, the index includes
data from global money market, share prices and various forms of risk premiums
calculated on the basis of data from the bond market, which affect the borrowing
costs of domestic sectors. Particularly important for the domestic financial system
is the liquidity of the euro money market as it affects the funding costs of parent
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
VDAX
Investor perception about risk
of investing in Croatian
government bonds,
macroeconomic outlook of
country
Source
financial theory and
practice
39 (2) 171-203 (2015)
Securities market
CDS of parent banks
of domestic banks
weighted according to
their share in banking
sector assets
Difference between bidask spreads for Croatian
and German
government eurobonds
(04)
Measure
182
financial theory and
practice
39 (2) 171-203 (2015)
banks. Developments in this market segment can have both direct and indirect
impacts on the availability and price of borrowing for domestic sectors and capital
flows. In the extreme case of market illiquidity, when there is a considerable fall
in turnover and prices collapse beyond values justified by fundamentals, there is a
large probability that domestic sectors’ funding will be considerably constrained
or even made impossible.
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
The average daily interest rate on overnight unsecured loans in euros (EONIA)
shows the liquidity of the euro interbank market, and partially reflects movements
on world financial markets. Its increased volatility suggests an amplified level of
instability as well as information asymmetry among market participants. The sixmonth EURIBOR is among the most often used interest rates, as it represents the
basis of the determination of many other interest rates. An increase in the spread
between the six-month EURIBOR and EONIA implies an enhanced uncertainty
level on the euro interbank market and an increase in liquidity risk premium. Although levels of these rates do affect the borrowing costs of domestic sectors, they
are not included in the FSI because their growth is not necessarily linked to an
increased stress. They generally closely follow the ECB reference interest rate,
which is used to influence all other interest rates and depends on economic activity, inflationary expectations and the developments in the eurozone financial system.
The spread between LIBOR and the overnight index swap (OIS) measures the
stress on the international money market. OIS is an interest contract swap that
reflects the expected level of the Fed’s reference interest rate, as well as the risk
and liquidity on the money market. Because of the importance of the money market for the banks’ financing, as a measure of risk of interbank lending it also indirectly measures the health of the banking system. An increased LIBOR-OIS spread
implies that banks could profit by borrowing from the Fed and lending to other
banks, which makes sense only in cases of a more pronounced increase in credit
risk.
Because of the importance of the German capital market and the high correlation
between Deutsche Boerse AG German Stock Index (DAX) and the CROBEX in
the pre-crisis period, the VDAX index is used as variable reflecting movements on
the European stock market. This indicator measures expected volatility of prices
on the German stock market, and together with the Chicago Board Options Exchange Volatility Index (VIX), which measures the implicit volatility of prices of
options on the S&P 500, it is often used as an indicator of risk aversion.
The J. P. Morgan Emerging Market Bond Index (EMBI) reflects the risk on investment in Croatian securities and measures the country’s risk premium. It is related
not only to global risk appetite, but also to the specific features of the domestic
economy. It can be seen as a synthetic macroeconomic indicator as it presents in-
vestor perception of the country’s macroeconomic perspective. A rise in this index
reflects an increased level of financial stress and leads to a rise in domestic sectors’
borrowing costs.
4.1 THE MARkOv SWITCHING MODEL
Although this research is primarily focused on FSI dynamics, in order to calculate
the contribution of individual market segments to stress episodes and enable a
better analysis of monetary policy reactions, their dates are calculated with the use
of the Markov switching model. Crisis episodes could have also been defined exogenously, for example, as a period in which the value of the index exceeds a
certain number of standard deviations or some boundary value set on the basis of
a well informed assessment.
The Markov switching model is suitable if the data dynamics changes through
time (Yuan, 2011; Kuan, 2002). This model endogenously finds the boundary values for the determination of a stress episode and divides the sample into periods
of enhanced and reduced stress. It also determines the likelihood of a transition
from one regime to another (Dufrénot et al., 2011).
It is assumed there are two regimes with different FSI dynamics. In the standard
Markov switching model with two states in which yt is the FSI at the moment t and
the arithmetical mean and variance are described by an unobserved state variable
runs:
(7)
from which it follows that
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
The risk premium for the parent banks of the largest domestic banks also considerably impacts the domestic financing costs. It is calculated by aggregating the CDS
premiums for bonds of five parent banks – Unicredit S.p.A., Intesa Sanpaolo
S.p.A., Société Generale, Erste Group Bank and Raiffeisen Zentralbank. Weights
used were the shares of each individual domestic bank in the total assets of these
five banks. This premium directly affects the price of borrowing for parent banks,
which in the next step spills over onto the funding price for the subsidiary banks,
and in the third step affects the borrowing costs for other domestic sectors through
increased lending interest rates. Difficulties in parent bank could also reduce their
available sources for financing domestic banks.
financial theory and
practice
39 (2) 171-203 (2015)
Considering the importance of government borrowing on foreign financial markets, the index also includes the bid-ask spread for Croatian government eurobonds. In order to obtain long time series, the ten-year bond maturing in 2014
was used. The influence of general liquidity risk on global markets is excluded by
reducing the bid-ask spread for the German government bond with comparable
maturity from this spread for Croatian government bond.
183
184
financial theory and
practice
39 (2) 171-203 (2015)
where st is the current state of stress on the financial markets, μ1 and μ2 are the
expectations, and σ1 and σ2 are standard deviations for the two regimes, while ε1
represents white noise.
In this case st =1 can be seen as a steady state on the financial markets, while st =2
designates a state of increased financial stress. It is assumed that this variable follows a Markov process with the following transition matrix:
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
where
designates the likelihood of a transition from one regime
to another, that is, the probability that the process is at time t in regime j, with the
assumption that it was previously in regime i, and that
(Yuan, 0).
Calculation of dates and contributions to the episodes of financial stress was based
on FSI_vew as it enables a more intuitive interpretation of the contributions of the
individual components to the movement of the total FSI (Sinenko et al., 2012).
4.2 RESULTS OF THE MODEL
The remainder of the paper presents indices obtained by different aggregation
methods and describes the FSIs during the observed period, as well as the stress
episodes and the CNB reactions related to financial market developments.
Independently of the aggregation method, the indices are strongly positively correlated and result in similar information about the stress episodes (table ).
Table 2
Coefficients of correlation between indices calculated by various methods
IFS 1_vew
IFS 2_pca
IFS_3_cdf
IFS 1_vew
IFS 2_pca
IFS_3_cdf
0.75
0.94
0.70
Note: Due to the smaller number of variables included in the FSI_pca it is not entirely comparable with the other two indices, which partially explains the lower correlation among them.
Source: Author’s calculation.
In order to additionally check the robustness of the results, FSI_vew and FSI_cdf
are also calculated with the use of only those indicators included in the FSI_pca
FSIs_vew are presented in the text itself (figures 1, 2, 3 and 4), while FSI_cdf and FSI_pca are presented in
appendix, as well as in figures A1 and A3. The list of variables included in the calculation of the FSI-vew and
the FSI_cdf is shown in table 1, while indicators included in the FSI_pca are shown in table A1 in appendix.
(table 3). The correlation coefficients show even greater positive correlation
among the indices and confirm the robustness of the results.
IFS 1_vew
IFS 2_pca
IFS_3_cdf
IFS 1_vew
IFS 2_pca
IFS_3_cdf
0.94
0.88
0.78
financial theory and
practice
39 (2) 171-203 (2015)
Table 3
Coefficients of correlation between indices calculated with the use of the same
variables
185
Source: Author’s calculation.
In order to obtain more detailed information from individual market segments,
sub-indices have been additionally divided into several components. The domestic component FSI-D was divided into indices that describe the domestic foreign
exchange market, money markets and securities market (FSI-D-FXM; FSI-DMM; FSI-D-SM – figure 3). In the same manner, the FSI-F was divided into subindices for the international money market (FSI-F-MM) and the international securities market (FSI-F-SM) (figure 4).
Following the financial and macroeconomic developments that tended to bring
about financial stress in Croatia, the period under observation is divided into five
sub-periods:
) from 00 to the end of 00,
2) from 2003 to end-2004,
3) from 2005 to mid-2007,
4) from mid-2007 to the beginning of 2010, and
5) from 2010 to end-2013.
These sub-periods, movements of total FSI and its components and the causes of
identified stress episodes are described in the rest of the chapter. A comparison of
movements of total FSI and its components with financial, macroeconomic and
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
Useful data about financial stress were obtained by a calculation of sub-indices. In
the first step the total FSI was divided into components related to the origin of the
shocks that might lead to financial stress (figure 2, table 1). The sub-index FSI-D,
which describes trends in the domestic components of overall financial stress, includes variables from the foreign exchange market, the money market and the
securities market. The sub-index FSI-F, which reflects movements on foreign financial markets, includes indicators from the international money market and securities markets that are considered to affect the financial stress in the country
through various channels.
186
monetary developments presents another verification of the robustness of the calculated composite indicators.
Figure 1
Total FSI_vew
Number of standard deviations
1.0
0.0
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
-1.0
FSI – total
Source: Author’s calculation.
Figure 2
Domestic and foreign components of total FSI_vew
2.5
2.0
Number of standard deviations
financial theory and
practice
39 (2) 171-203 (2015)
2.0
1.5
1.0
0.5
0.0
-0.5
-1.0
-1.5
FSI-D – domestic component
Source: Author’s calculation.
FSI-F – foreign component
Figure 3
FSIs divided according to market segments: domestic money market, domestic
securities market and domestic foreign exchange market
187
4.5
4.0
financial theory and
practice
39 (2) 171-203 (2015)
3.5
Number of standard deviations
3.0
2.5
2.0
1.5
1.0
0.5
0.0
-0.5
-1.0
FSI – total
FSI-D-MM – domestic money market
FSI – total
FSI-D-SM – domestic securities market
4.5
4.0
Number of standard deviations
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
-0.5
-1.0
-1.5
4.5
4.0
Number of standard deviations
3.5
3.0
2.5
2.0
1.5
1.0
0.5
0.0
-0.5
-1.0
-1.5
FSI – total
Source: Author’s calculation.
FSI-D-FXM – domestic foreign exchange market
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
-1.5
financial theory and
practice
39 (2) 171-203 (2015)
Number of standard deviations
Figure 4
FSIs divided according to market segments: international money market and
international securities market
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
FSI – total
FSI-F-MM – international money market
Number of standard deviations
188
FSI – total
FSI-F-SM – international securities market
Source: Author’s calculation.
Figure 5 shows the periods determined as episodes of increased turbulences on the
financial markets for an FSI_vew according to the Markov switching model. In
order to verify the robustness and to confirm if the stress episodes have been correctly identified, they were also determined with the use of the FSI_cdf (appendix,
figure A2). The stress periods are very similar, implying the robustness of the results. Nevertheless, it needs to be pointed out that the main objective of the identification of stress episodes is the analysis of the underlying developments and the
market participants’ reactions, rather than determining their precise dates.
The contributions of individual market segments and of individual sources of risks
to the stress episodes were calculated for all eight stress episodes (figures 6 and 7).
To simplify this analysis, episodes that were shorter-lasting but frequent in time
are observed as a single stress episode.
In the pre-crisis period up to the second half of 2008, there were five stress episodes. They were mostly affected by domestic financial market variables, but did
not have pronounced systemic consequences. In the subsequent period, there were
three stress episodes, two of which were extremely powerful not only for their
intensity but also for their length, and for their strong negative impacts on the real
economy. They were primarily initiated by shocks from international financial
markets, although domestic indicators also contributed to them significantly.
2.0
1.0
0.6
0.5
0.5
0.4
0.3
0.0
0.2
-0.5
0.1
-1.0
1/01
4/01
7/01
10/01
1/02
4/02
7/02
10/02
1/03
4/03
7/03
10/03
1/04
4/04
7/04
10/04
1/05
4/05
7/05
10/05
1/06
4/06
7/06
10/06
1/07
4/07
7/07
10/07
1/08
4/08
7/08
10/08
1/09
4/09
7/09
10/09
1/10
4/10
7/10
10/10
1/11
4/11
7/11
10/11
1/12
4/12
7/12
10/12
1/13
4/13
7/13
10/13
0.0
FSI - total
FSI
– total
Probability - right
Probability – right
Note: The shaded areas mark the stress episodes.
Source: Author’s calculation.
Figure 6
Contributions of individual FSI components to stress episodes: according to
market segments (in %)
100
Foreign SM
80
Foreign MM
Domestic SM
Domestic SM
40
Domestic FXM
Domestic FXM
Domestic SM
20
-20
-40
Foreign SM
Foreign MM
Fomestic SM
Domestic FXM
Domestic FXM
Domestic MM
Domestic FXM
0
Domestic SM
Domestic MM
Domestic MM
Domestic MM
Domestic MM
Domestic MM
Foreign MM
Foreign MM
Foreign MM
Domestic FXM
Foreign MM
Domestic SM
Foreign MM
Domestic FXM
Domestic SM
Domestic FXM
Domestic MM
Domestic SM
Foreign MM
Foreign SM
Foreign SM
Foreign SM
Domestic MM
Foreign SM
Foreign SM
Foreign SM
-60
7/01-12/01
3/02-5/02
6/03-4/0
Source: Author’s calculation.
7/04-2/05
9/07-2/08
9/08-1/10
7/11-11/12
6/13-9/13
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
Number of standard deviations
0.8
1.0
Probability of transition to stress regime
0.9
1.5
60
financial theory and
practice
39 (2) 171-203 (2015)
Figure 5
Episodes of increased turbulence on the financial markets determined by the
Markov switching model
189
190
Figure 7
Contributions of individual FSI components to stress episodes: according to
sources and types of risk (in %)
100
financial theory and
practice
39 (2) 171-203 (2015)
Costsof for. borr.
80
For. banks' cred. risk
60
Uncert. SM
Uncert. & liq. D-MM
Curr.
induced
cred. risk
40
Uncert. SM
20
Curr. induced cred. risk
Uncert. & liq. D-FXM
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
0
For. banks' cred. risk
Dom. banks' cred.risk
Uncert. & liq. D-MM
Uncert. & liq. F-MM
Uncert. SM
Dom. banks' cred.risk
Dom. banks' cred.risk
Uncert. & liq. D-MM
Uncert. & liq. D-FXM
Uncert. & liq. D-MM
Uncert. & liq. D-FXM
Uncert. & liq. D-MM
Uncert. & liq. F-MM
Uncert. & liq. F-MM
For. banks' cred. risk
For. banks' cred. risk
-20
-40
Uncert. & liq. D-MM
Costsof for. borr.
Costsof for. borr.
Uncert. & liq. D-FXM
Costsof for. borr.
Uncert. & liq. F-MM
Dom. banks' cred.risk
Curr. induced cred. risk
Costsof for. borr.
Curr. induced cred. risk
Liq.of govt. bonds
Liq.of govt. bonds
For. banks' cred. risk
Uncert. & liq. F-MM Uncert. & liq. F-MM
Uncert. SM
Curr. induced cred. risk
Curr. induced cred. risk
Uncert. & liq. D-FXM Uncert. & liq. D-FXM
Curr. induced cred. risk Uncert. & liq. D-FXM Uncert. & liq. D-FXM
Uncert. SM
Uncert. & liq. D-MM
Uncert. & liq. F - MM Curr. induced cred. risk Liq.of govt. bonds
Dom. banks' cred.risk Uncert. & liq. D-MM
For. banks' cred. risk For. banks' cred. risk
Dom. banks' cred.risk
Uncert. SM
Uncert. SM
Liq.of
govt.
bonds
For. banks' cred. risk
Liq.of govt. bonds
Costsof for. borr.
Costsof for. borr.
Costsof for. borr.
Uncert. SM
Uncert. & liq. F-MM
-60
7/01-12/01
3/02-5/02
6/03-4/04
7/04-2/05
9/07-2/08
9/08-1/10
7/11-11/12
6/13-9/13
Source: Author’s calculation.
4.2.1 First period: from 2001 to 2003
The total FSI in the period from 2001 to 2003 was relatively volatile, but there
were only two, fairly mild, stress episodes. The first one is related with the 9/11
attacks, which led to a rise in uncertainty and volatility on world financial markets
and increased risk premiums. The FSI-F reveals that these turbulences were partially transmitted to the domestic financial market via the money market channel
(figures 1, 2 and 4, column 1). The total FSI in this period was mostly affected by
a speculative attack on domestic currency (figures 1 and 2, column 1). In the middle of August 2001, depreciation pressures on the EUR/HRK exchange rate arose,
showing up much earlier than would usually have been expected, considering the
seasonal inflow of foreign capital from tourism. This was encouraged by the speculative activities of some banks that in expectation of a considerable kuna depreciation started vigorously buying euros. In the shallow domestic market this additionally enhanced depreciation pressures (figure 3, column 1). The central bank
intervened three times in the period between 9 August and 20 August and sold
EUR 408m to the banks, which halted the depreciation pressures. In this period, a
strong rise in the money market interest rates was recorded (figure 2, column 1).
This encouraged some banks to use Lombard loans that they, in spite of their high
rate of interest, considered as favourable within the market expectations regarding
the exchange rate, so the central bank increased the interest rate for Lombard
loans from 9.5% to 10.5%.
The contributions calculated for this stress episode, which lasted from July to December 2001, confirm the importance of events and uncertainties on the foreign exchange
market, and to a lesser extent of movements on the domestic money market, for the
development of overall financial stress in the system (figures 6 and 7, column 1).
The period from early 2003 to the end of 2004 was marked by a relatively low
level of financial stress. This was an introduction into a quite long period of low
risk premiums and low volatility of international market indicators (figures 1, 2, 3
and 4, column ).
Developments on the domestic financial market in 2003 were influenced by the
tightening of the CNB’s policy aimed at slowing down the credit expansion and
banks’ foreign borrowing. Sanctions were introduced on the growth of domestic
loans greater than 16% a year or 4% quarterly and banks were obliged to hold the
minimum required amount of foreign currency claims of 35% in order to ensure
appropriate foreign currency liquidity. The latter measure resulted in depreciation
pressures on the kuna because of the augmented demand for foreign exchange and
a reduction in primary liquidity in the banking system, as well as in a surge in the
level and volatility of money market interest rates in the second half-year (figures
and 3, column ).
In September 2003, the CNB increased the percentage of the reserve requirement
on foreign currency obligations that must be held in kuna from 25% to 35%, and
additionally to 4% in November 003. These changes resulted in temporary appreciation pressures on the EUR/HRK exchange rate and increased volatility of
money market interest rates (FSI-D-FXM; FSI-D-MM – figure 3, column 2). In
February 2004, the minimum percentage of reserve requirements that were set
aside in a special account at the CNB was increased from 40% to 60%; and by
foreign exchange interventions the banking system liquidity was increased, which
temporarily reduced the money market interest rates.
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
4.2.2 Second period: from 2003 to 2004
financial theory and
practice
39 (2) 171-203 (2015)
Domestic banking system liquidity in 2002 was relatively high and the year was
marked by strong appreciation pressures due to large foreign capital inflows from
tourism, enhanced foreign borrowing by the government and the commercial
banks and the privatization of domestic enterprises. Through several foreign exchange interventions the CNB net created HRK4.8bn, which was partially sterilized by increasing the base for the kuna reserve requirement and by increased
issuance of central bank T-bills. This ensured stability on the money and foreign
exchange markets, which had been disrupted only for a short period due to the
internal fraud in the treasury of Riječka banka (figure 3, column 2). The rapid
CNB reaction ensured the necessary domestic and foreign currency liquidity for
the normal operations of Riječka banka in the period until the new owner took it
over, so this turbulence was only temporary and the crisis did not spill over to
other financial institutions. This stress episode lasted from March to May 2002.
191
192
The most important sources of stress in the third stress episode from June 003 to
April 2004 were developments and uncertainty on the domestic foreign exchange
and domestic money market (figures 6 and 7, column 3).
financial theory and
practice
39 (2) 171-203 (2015)
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
In July 004, interest rates surged again when the CNB imposed a marginal reserve requirement on an increase in foreign liabilities of banks, which was initially set at 24%, but has additionally been raised several times (figure 3, FSI-DMM, column 2). This resulted in considerable changes in banks’ liquidity and in
oscillations of money market interest rates until early 2005. This strongly affected
the FSI-D, particularly the FSI-D-MM (figures 2 and 3, column 2). Overnight interest rates went up to as high as 10%, but after several foreign exchange interventions they dropped to a relatively low %. The fourth stress episode lasted from
July 2004 to February 2005 and was marked by developments on the domestic
money market (figure 3, column 2; figures 6 and 7, column 4).
4.2.3 Third period: from 2005 to mid-2007
The third period was the longest period of low-level of financial stress in which
there were no major turbulences in a single segment of the financial market and
not a single stress episode was recorded (figures 1, 2, 3 and 4, column 3; figure 5).
It was marked by high level of global liquidity, low risk appetite and the beginning
of a gradual increase in reference interest rates by the Fed and the ECB prompted
by exceptionally propitious economic trends.
However, precisely in that period of benign market conditions a considerable deterioration in internal and external imbalances was recorded because of strong
foreign borrowing, excessive credit activity and overheating of the domestic economy. Therefore, the CNB continued tightening its monetary, or rather macroprudential, policy. The marginal reserve requirement rate was gradually increased
and reached 55% by the end of 2005. In early 2006, a special reserve requirement
on newly issued bank debt securities, and increased capital requirements for currency-induced credit risk were introduced, the capital adequacy ratio was increased to %, at the beginning of 007 the annual credit growth was restricted
to 12%, and in 2008 capital requirements for banks with higher than permitted
credit growth were increased. These measures increased the overall level of financial system resilience by creating buffers against possible shocks. But in spite of
CNB efforts aimed at slowing down systemic risk accumulation, it was precisely
in this period of calm and stable conditions on the financial markets that systemic
risks significantly increased. These risks materialized during the following stress
episodes.
4.2.4 Fourth period: from mid-2007 to end-2009
The beginning of the fourth period was also the onset of the world financial crisis.
In the second half of 2007 the first difficulties associated with sub-prime mortgages in the USA appeared. When issuers of securities based on these loans faced
problems with their re-financing, the crisis spilled over from the mortgage market
onto the interbank money market and spread globally (FSI-F – figure 2, column 4;
FSI-F-MM – figure 4, column 4). Initially, these developments only slightly affected domestic FSIs (figures 1 and 2, column 4). The FSI-F increased, but its
level was similar to the level recorded before the long-lasting tranquil period. It
had stabilized at the beginning of 2008 and even started decreasing.
Temporary stabilization on the international financial markets lasted until March
2008 and the collapse of the investment bank Bear Stearns. Although this crisis
situation was solved promptly and there were no significant negative consequences
for the rest of the financial system as the failed bank was taken over by the J. P.
Morgan, the reduced confidence among market participants stimulated a new surge
in volatility and nervousness on the international money market (figure 4, column
4). In spite of that, the influence of the FSI-F and FSI-F-MM on total FSI was not
significant (figures 1 and 2, column 4). After the initial rise, the FSI-F did not change
much until the collapse of Lehman Brothers in September 2008 (figures 2 and 4,
column 4). This resulted in a previously unrecorded rise in global risk aversion and
an increase in price volatility with a simultaneous plunge in liquidity and a rise in
distrust among market participants. Therefore, both components of the FSI-F surged
(figures 2 and 4, column 4). A sudden jump in the Croatian risk premium and a
frozen international money market hindered the access of domestic sectors to foreign capital, which was reflected in all segments of the FSI-D (figure 3, column 4).
In the fourth quarter of 2008, level and volatility of overnight interest rates rose
considerably, and in November 2011 the FSI-D-MM reached a record level (figure
3, FSI-D-MM, column 4). This was induced by transactions on the capital market
related to the takeover of INA d.d. by Hungary’s MOL. This led to a division of
banks into those with considerable surpluses and those with notable liquidity deficits, leading to a surge in money market interest rates. Apart from that, the insta-
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
The fifth identified stress episode lasted from September 2007 to February 2008
and was mostly attributable to uncertainty and volatility on the domestic and international money markets due to the increased credit risk and the fall in confidence among market participants (figures 6 and 7, column 5).
financial theory and
practice
39 (2) 171-203 (2015)
A much greater impact on the total FSI came from the increased volatility in the
domestic money market caused by the IPO of the domestic telecom company THT in October 2007 (figures 1, 2 and 3, column 4). At that time, an imbalance
between money supply and demand arose as a small group of banks generated a
considerable part of the demand for kuna, while participants with liquidity surplus
required interest rates higher than usual. Although the system liquidity was at the
usual level, the shallow money market had difficulties in adjusting to the large
inflows or outflows of money, which contributed to the elevated interest rates
levels and their increased volatility until the end of the year (figure 3, FSI-D-MM,
column 4).
193
194
financial theory and
practice
39 (2) 171-203 (2015)
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
bility on world financial markets resulted in a partial withdrawal of bank deposits
in October 2008. Banks were encouraged to retain more liquid assets, which increased the demand for foreign currency liquidity (figure 3, FSI-D-FXM, column
4). In order to maintain overall financial stability, the CNB was forced to restrict
kuna liquidity (figure 3, FSI-D-MM, column 4). The CNB started releasing previously accumulated reserves and the first step was the abolition of the marginal
reserve requirement in October 2008, which improved foreign currency liquidity
in the banks and ensured the payment of international liabilities. This resulted in
the preservation of a stable exchange rate and high level of international reserves,
as well as in the highest recorded short-term money market interest rates. Due to
the restricted kuna liquidity, the pressures on the foreign exchange market were
reflected on the money market (FSI-D-MM), so in the initial phase of the escalation of the crisis it was only slightly reflected in the developments of the FSI-DFXM (figure 3, column 4).
Instability on the money market at the end of the year and efforts to make it easier
for government to finance within the country led in December 2008 to the lowering of the reserve requirement rate from 17% to 14%. System liquidity was improved and interest rates decreased noticeably. This was reflected in a considerable, but temporary, reduction of FSI-D-MM (figure 3, column 4). As well as
through reverse repo auctions, banks obtained kuna liquidity by the intensive use
of Lombard loans, even though their interest rate had increased from 7.5% to
9.0% and the regulations concerning the amount of securities needed as collateral
had been strengthened.
The first half of 2009 was influenced by global turbulences, frozen international
money markets and a sudden stop in capital flows. Due to renewed depreciation
pressures, in the first quarter of 2009 the CNB intervened three times in the foreign exchange market, at the first two auctions selling and, in the last auction at
the end of February, buying euros. In this period a marked increase in the FSI-DFXM was recorded (figure 3, column 4). In order to stabilize the exchange rate, in
January 2009 the CNB increased the percentage of the foreign currency reserve
requirement set aside in kuna from 50% to 75%, and in order to ensure adequate
foreign liquidity of the system in February the rate of minimum foreign currency
claims was reduced from 28.5% to 20%.
The developments on the foreign exchange market resulted in the first quarter of
2009 in a temporary, but a significant rise in the level and volatility of overnight
interest rates and increased turnover in the money market because of the lower kuna
liquidity (figure 3, FSI-D-MM, column 4). This stabilized at the end of February
after the weakening of depreciation pressures, and both interest rates and volatility
on the money market were reduced (figure 3, FSI-D-MM, FSI-D-FXM, column 4).
By the end of the year, banking sector liquidity was satisfactory and the interest
rates and the exchange rate remained stable (figures 1, 2 and 3, column 4).
The end of 2008 and most of 2009 were marked by a major decline in the CROBEX
and increased volatility of returns on both CROBEX and CROBIS, which were
strongly reflected in the FSI-D-SM (figure 3, column 4).
In the fifth period, the crisis in the government debt market in peripheral eurozone
countries deepened in mid-2011. Apart from threatening banking sector stability,
this crisis adversely affected the expectations of market participants, consumers
and corporate sector related to the economic recovery. The renewed decline in
investor risk appetite increased risk premiums, and stress spilled over onto the
domestic financial system through a rise in the FSI-F (figures 1, 2 and 4, column
5). The risk premium for Croatia rose, absolutely and relatively, much more than
the premiums for European emerging markets, and exceeded the record level
reached in early 2009. These developments considerable enlarged the FSI-F-SM
(figure 4, column 5).
The increased reliance of subsidiaries on their parent banks during the fifth stress
episode reflected a support of the owners to domestic banks, but also increased exposure to parent banks’ liquidity, their needs for capital, financing strategies and to
developments in their home countries, as well as in those in which they had considerable exposures. The worsening of the international debt market conditions in the
second half of 2011 led to a rise in the CDS premiums on the bonds of the parent
banks of the five largest domestic banks. Their average level at the end of 2011
ranged about 500 basis points, almost twice as much as in the period after the escalation of the crisis, primarily because of the exposures to peripheral eurozone countries and concerns regarding the sustainability of their fiscal positions. Adverse developments in CDS premiums for Italy additionally increased the risk perception of
Italian parent banks compared to those from Austria or France. This resulted in the
partial withdrawal of the parent banks’ funding from domestic banks and increased
pressures on the foreign currency liquidity (figures 3 and 4, column 5).
During the seventh stress episode from July 0 to November 0 the major
contribution to movement of total FSI and the increased level of stress came from
the international securities and money market due to high volatility and strong
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
4.2.5 Fifth period: from the beginning of 2010 to end-2013
financial theory and
practice
39 (2) 171-203 (2015)
Unlike previous stress episodes, when some market segments alleviated stress
disruptions, in the sixth episode from September 2008 to January 2010 all of them
contributed to the increase of financial stress (figures 6 and 7, column 6). The
greatest influence on these movements was made by the events on the international and domestic money markets. Since the stabilization of the EUR/HRK exchange rate was a key precondition for the preservation of overall financial stability in the country, interest rates on the domestic money market were, at that time,
“sacrificed” in order to achieve that goal.
195
196
growth of uncertainty and risk aversion observed in an increase of FSI-F (figures
1, 2 and 4, column 5; figures 6 and 7, column 7).
financial theory and
practice
39 (2) 171-203 (2015)
In spite of the negative macroeconomic trends in 2013 and the fall in the country’s
credit rating, the EUR/HRK exchange rate remained stable, the liquidity of the
system was high due to the CNB measures and the money market interest rates
were low and stable. Components of FSI-D were quite low and decreased throughout the year (figures 2 and 3, column 5).
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
The eighth stress episode lasted from mid-June to September 2013 and was driven
by the cost of foreign borrowing, particularly trends in the international securities
markets which assessed Croatia as a rather risky investment. This period was
characterized by the Croatian credit rating downgrade to below investment level
by all three major agencies (Standard and Poor’s, 13 December 2012; Moody’s, 1
February 2013; and Fitch Ratings, 20 September 2013). Notwithstanding this and
the lack of economic recovery, Croatia still had access to the financial markets.
The total FSI reacted relatively strongly to the first downgrade, while the negative
reactions to the two further credit downgrades were somewhat milder (figure 1,
column 50). Such moderate reactions were partially caused by the stabilization on
the international financial markets. Government took advantage of favourable circumstances and issued two new bonds in the USA (in April, USD 1.5bn, yield at
issue of 5.62%; November, USD 1.75bn, yield at issue of 6.20%).
The end of the period was marked by improved world financial market conditions
and a fall in general risk aversion (figure 4, column 5). Nevertheless, the Croatian
risk premium remained elevated and FSI-F negatively contributed to the total FSI
(figure 2, column 5). The reason for this were investors’ concerns regarding the
lack of economic recovery, deterioration of fiscal indicators and absence of structural reforms that would create conditions for sustainable economic growth. Due
to its inherent weaknesses, Croatia did not use the period of stable and unexpectedly favourable conditions in the domestic and international financial markets to
ensure cheaper funding for the private sector, which would have been an important step towards economic recovery. Croatia therefore remained extremely vulnerable to any possible tightening of financial conditions, meaning that in the
event of a more pronounced increase in risk aversion it could face a prohibitively
high price of foreign capital.
4.3 COMPARISON OF FSIs AND INDICES OF FINANCIAL CONDITIONS
Following Kliesen (2012) and in order to additionally check the robustness and
usefulness of the results obtained, total FSI was compared with the financial conditions index (FCI) for Croatia. FCI is calculated by using the principal component analysis, on the basis of 28 macroeconomic and financial variables that reflect financing conditions in Croatia (for details see Dumičić and Krznar, 2013).
FCI is available at a quarterly level, so the FSI was adjusted by calculating its
quarterly averages.
The coefficients of correlation imply a strong positive link between FCI and FSIs
calculated by various methods. This correlation is even stronger if the level of the
FSI of the previous quarter is used, suggesting a great influence of financial stress
on overall financial conditions and the possible use of the FSI for predicting trends
in financing conditions for domestic sectors in the forthcoming period.
financial theory and
practice
39 (2) 171-203 (2015)
Figure 8
Comparison of total FSI with FCI
1.5
1.0
0.5
0.0
-0.5
-1.0
Financial conditions index
FSI – total
2013Q3
2013Q1
2012Q3
2012Q1
2011Q3
2010Q3
2011Q1
2010Q1
2009Q3
2009Q1
2008Q3
2008Q1
2007Q3
2007Q1
2006Q3
2006Q1
2005Q3
2005Q1
2004Q3
2004Q1
2003Q3
-2.0
2003Q1
-1.5
FSI – total (t-1)
Source: Author’s calculation (FCI is calculated on the basis of the data and methodology presented in Dumičić and Krznar, 2013).
Table 4
Coefficients of correlation between FCIs and FSIs
IFS_vew
IFS_pca
IFS_cdf
FCI
0.57
0.58
0.6
FCI (t+1)
0.70
0.69
0.76
Source: Author’s calculation.
5 CONCLUSION
Due to the adverse effects that financial market stress episodes have on financial
and macroeconomic stability, the main objective of this paper has been to construct high-frequency financial stress indicators that will in good time inform economic policy makers of possible disruptions in the financial markets and be a
useful tool for the analysis of risks that might jeopardize the financial and macroeconomic stability of the system. Another objective was to use the Croatian example to create an index that, in spite of the relatively limited accessibility of daily
data, would still cover the most important specific features of a small, open, highlyeuroized country with shallow financial markets and majority foreign-owned
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
Number of standard deviations
197
198
financial theory and
practice
39 (2) 171-203 (2015)
banks. Namely, most of the financial stress indicators were created for developed
countries that are at a different level of economic and financial development and
are characterized by institutional and regulatory arrangements different from those
in emerging markets. Special attention has been devoted to the particular channels
through which financial stress spills over from the financial markets to other segments of the financial system and the real sector in such countries. The identification of sources of financial stress and the factors that affect them, the understanding of the channels through which disruptions spill over to the remainder of the
financial system and real sector, as well as the analysis of the policy makers’ activities in these periods can enable more effective preventive actions and better
reactions in stress episodes.
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
For the aggregation of individual indicators into a composite index, three methods
were used – aggregation with the use of weightings based on equal-weight variances, aggregation of variables transformed with the use of the cumulative distribution function, and the principal component analysis. The indices constructed are
highly correlated, which means that the aggregation method does not essentially
affect the information contained in the index, confirming the robustness of the
results. The Croatian example shows that the calculated FSIs present a useful tool
for describing events on the financial markets as well as monetary and macroeconomic trends, also suggesting the robustness of results.
By analysing the FSIs and the CNB reactions to the observed financial stress episodes, it can be concluded that the CNB was successful in the stabilization of the
financial markets and the preservation of overall financial stability. Although FSIs
did not exist in the observed period, they could have been useful in the process of
identifying sources of stress disruptions, particularly in communication with public and market participants when explaining the preventive measures. Market participants and the public were often unaware of the possible threats to threatening
financial stability, particularly during the stable periods on the financial markets.
It can be expected that these indicators will continue to be developed and that their
components will adjust to financial markets’ developments. In the next step, the
indices might be used as an early warning tool for predicting trends in the financing conditions for domestic sectors or forecasting real economic developments
(i.e. like Jakubik and Slačik, 2013). In combination with other techniques, like
stress testing, various systems of early warning, and other composite indicators
created for the analysis of financial stability in Croatia, such as FCI or indices of
the accumulation and materialization of systemic risks, this indicator should enable a better monitoring of risks and ensure a prompt reaction of economic policy
makers to possible stress episodes.
FSI_cdf_domestic component
0.9
1.0
0.8
0.9
0.7
0.6
0.8
0.7
0.5
0.6
0.4
0.5
0.3
0.4
0.3
0.2
0.2
0.1
0.1
0.0
0.0
FSI_cdf_total
Note: The shaded areas mark the stress episodes.
Source: Author’s calculation.
Probability – right
Probability of transition to stress regime
1/01
4/01
7/01
10/01
1/02
4/02
7/02
10/02
1/03
4/03
7/03
10/03
1/04
4/04
7/04
10/04
1/05
4/05
7/05
10/05
1/06
4/06
7/06
10/06
1/07
4/07
7/07
10/07
1/08
4/08
7/08
10/08
1/09
4/09
7/09
10/09
1/10
4/10
7/10
10/10
1/11
4/11
7/11
10/11
1/12
4/12
7/12
10/12
1/13
4/13
7/13
10/13
1/01
4/01
7/01
10/01
1/02
4/02
7/02
10/02
1/03
4/03
7/03
10/03
1/04
4/04
7/04
10/04
1/05
4/05
7/05
10/05
1/06
4/06
7/06
10/06
1/07
4/07
7/07
10/07
1/08
4/08
7/08
10/08
1/09
4/09
7/09
10/09
1/10
4/10
7/10
10/10
1/11
4/11
7/11
10/11
1/12
4/12
7/12
10/12
1/13
4/13
7/13
10/13
0.9
0.8
0.7
0.3
0.2
FSI_cdf_total
0.9
0.8
0.7
0.6
0.5
0.4
0.3
0.2
0.1
FSI_cdf_foreign component
Source: Author’s calculation.
Figure a2
Episodes of increased turbulence on the financial markets determined by the
Markov switching model
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
1/01
4/01
7/01
10/01
1/02
4/02
7/02
10/02
1/03
4/03
7/03
10/03
1/04
4/04
7/04
10/04
1/05
4/05
7/05
10/05
1/06
4/06
7/06
10/06
1/07
4/07
7/07
10/07
1/08
4/08
7/08
10/08
1/09
4/09
7/09
10/09
1/10
4/10
7/10
10/10
1/11
4/11
7/11
10/11
1/12
4/12
7/12
10/12
1/13
4/13
7/13
10/13
Figure A1
FSIs calculated with the use of the cumulative distribution function
financial theory and
practice
39 (2) 171-203 (2015)
Number of standard deviations
APPENDIX
199
Indices of financial stress calculated with the use of variables transformed
with the cumulative distribution function
0.6
0.5
0.4
00
Financial stress indices calculated with the principal component analysis
method
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
FSI_pca_total
Number of standard deviations
financial theory and
practice
39 (2) 171-203 (2015)
Number of standard deviations
Figure a3
FSIs calculated with the principal component analysis method
FSI_pca_domestic component
FSI_pca_foreign component
Note: Number of variables included in the FSI obtained by the PCA method is smaller than in
indices calculated by other methods because not all data are available since January 2001 and
the exclusion of non-stationary variables. The usual methods of transforming variables with daily
frequency to avoid this problem result in very volatile series.
Source: Author’s calculation.
Table a1
Loadings of parameters for variables included in the FSI_pca
0.0
0.3
0.09
0.68
0.5
0.69
0.50
0.57
0.55
0.6
0.28
0.36
0.34
0.4
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
Source: Author’s calculation.
FSI_pca_total FSI_pca_domestic FSI_pca_foreign
0.45
0.3
financial theory and
practice
39 (2) 171-203 (2015)
variables/Index
Returns on CROBEX
Volatility of overnight
interbank interest rates
Exchange rate weighted
according to banks’ assets
structure
Forward exchange rates
CDS of parent banks of
domestic banks weighted
according to their share in
banking sector assets
EMBI risk premium
for Croatia
EONIA volatility
LIBOR – OIS spread
0
0
financial theory and
practice
39 (2) 171-203 (2015)
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
REFERENCES
1. Anh, T. and Mägi, S., 2009. Principal Component Analysis – Final Paper in
Financial Pricing. National Cheng Kung University.
2. Balakrishnan, R. [et al.], 2009. “The Transmission of Financial Stress from
Advanced to Emerging Economies”. IMF Working Paper, WP/09/133.
3. Cardarelli, R., Elekdag, S. and Lall, S., 2009. “Financial Stress, Downturns,
and Recoveries”. IMF Working Paper, WP/09/100.
4. Catao, L. A. V., 2006. “Sudden Stops and Currency Drops: A Historical Look”.
IMF Working Paper, WP/0/33.
5. CNB, 0. Financial Stability, No. 9. Zagreb: Croatian National Bank.
6. CNB. Annual Reports 2000-2012. Zagreb: Croatian National Bank.
7. Dufrénot, G., Furceri, D. and Zdzienicka, A., 2011. Financial globalization
and financial stress in Latin America: the case of Argentina, Brazil and Mexico. Preliminary version of the paper. Available at: <http://www.defi-univ.org/
img/pdf/paper_march2011pdf>.
8. Dumičić, M. and Krznar, I., 2013. “Financial Conditions and Economic Activity”. CNB Working Paper, W-37. Zagreb: Croatian National Bank.
9. Grimaldi, B. M., 2010. “Detecting and interpreting financial stress in the euro
area”. ECB Working Paper, No. 4.
10. Hakkio, C. S. and Keeton, W. R., 2009. “Financial stress: what is it, how can
it be measured, and why does it matter?”. Economic Review, Federal Reserve
Bank of Kansas City, (q ii), pp. 5-50.
. Hubrich, K. and Tetlow, R. J., 0. “Financial stress and economic dynamics: the transmission of crises”. Finance and Economics Discussion Series,
Federal Reserve Board. Washington, D.C.
12. Illing, M. and Liu, Y., 2003. “An Index of Financial Stress for Canada”. Bank
of Canada Working Paper, No. 003-4.
13. Illing, M. and Liu, Y., 2006. “Measuring financial stress in a developed country: An application to Canada”. Journal of Financial Stability, 2 (3), pp. 24365. doi: 0.06/j.jfs.006.06.00
14. Ivičić, L. [et al.], 2008. “Results of the Fifth CNB Bank Survey”. Survey,
S-15. Zagreb: Croatian National Bank.
15. Jakubik, P. and Slačik, T., 2013. “Measuring Financial (In)Stability in Emerging Europe: A New Index-Based Approach”. Oesterreichische Nationalbank,
Financial Stability Report, No. 6.
16. Kliesen, K. L., Owyang, M. T. and Vermann, E. K., 2012. “Disentangling
Diverse Measures: A Survey of Financial Stress Indexes. Federal Reserve
Bank of St. Louis, Review, 94 (5), pp. 369-397.
17. Kota, V. and Saqe, A., 2011. “A Financial Systemic Stress Index for Albania”.
Bank of Albania Working Paper, WP03/4/03.
18. Kuan, C. M., 2002. Lecture on the Markov Switching Model. Academia Sinika: Institute of Economics.
03
financial theory and
practice
39 (2) 171-203 (2015)
mirna dumičić:
financial stress indicators for small, open, highly euroized countries:
the case of croatia
19. OeNB, 2013. Financial Stabilisty Report, No. 26. Oesterreichische Nationalbank.
20. Oet, M. V. [et al.], 2011. “The Financial Stress Index: Identification of Systemic Risk Conditions”. Federal Reserve Bank of Cleveland Working Paper,
No. /30.
21. Puddu, S., 2008. Optimal Weights and Stress Banking Indexes. Lausanne:
HEC-Université de Lausanne.
22. Schou-Zibell, L., Albert, J. R. and Song, L. L., 2010. “A Macroprudential
Framework for Monitoring and Examining Financial Soundness”. ADB Working Paper Series on Regional Economic Integration, No. 43.
23. Sinenko, N., Titarenko, D. and Arinš, M., 2012. “Latvian Financial Stress Index”. Discussion Paper, No. 1-2012. Riga: Latvijas Banka.
24. Yuan, C., 2011. “Forecasting Exchange Rates: The Multi-State MarkovSwitching Model with Smoothing”. International Review of Economics &
Finance, 20 (2), pp. 342-362. doi: 10.1016/j.iref.2010.09.00
25. Zagreb Stock Exchange. Available at: <www.zse.hr>.
Development index: analysis
of the basic instrument of
Croatian regional policy
ANA PERIŠIĆ, mag. math.*
VANJA WAGNER, mag. math.*
Preliminary communication**
JEL: R1, C18, C43
doi: 10.3326/fintp.39.2.4
The authors would like to thank Dubravka Šišak Jung, PhD as well as to two anonymous referees for their
useful comments and suggestions.
**
Received: June 1, 2014
Accepted: November 21, 2014
*
The article was submitted for the 2014 annual award of the Prof. Dr. Marijan Hanžeković Prize.
Ana PERIŠIĆ
Polytechnic of Šibenik, Trg Andrije Hebranga 11, 22000 Šibenik, Croatia
e-mail: [email protected]
Vanja WAGNER
University of Zagreb, Faculty of Science, Department of Mathematics, Bijenička cesta 30, 10000 Zagreb,
Croatia
e-mail: [email protected]
206
financial theory and
practice
39 (2) 205-236 (2015)
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
Abstract
The development level assessment and categorization of Croatian local and regional units is based on the value of the development index which is the main instrument of Croatian regional policy. The development index is a composite indicator calculated as a weighted average of five socio-economic indicators. The
goal of this paper is to analyze the uncertainty and sensitivity of the development
index that arise from the procedures and indicators used in its construction. This
analysis is then used to propose useful guidelines for future impovements. The
methodology of the Croatian regional development index has been critically reviewed, revealing problems of multicollinearity and the existence of outliers. An
empirical and relatively more objective multivariate approach for weight selection has been proposed. The uncertainty and sensitivity analysis were conducted
using Monte Carlo simulations and variance-based techniques. Instead of a
unique point estimate for the development level of territorial units an alternative
confidence interval approach was considered.
Keywords: development index, composite indicators, multivariate analysis, uncertainty analysis, sensitivity analysis, Croatia
1 INTRODUCTION
The State has a constitutional obligation to promote the economic development of
all the regions in Croatia and to promote economic progress and the social welfare
of all citizens. Accurate assessment of the level of development of territorial units
is crucial for regional planning and development policy, and is a key criterion for
the allocation of various structural funds and national subsidies (Cziraky et al.,
2005). Poor economic situations in local units entail low fiscal capacities, making
them dependent on state subsidies and incapable of pursuing their own development (Puljiz, 2009). This prompted several groups of authors to propose that the
criteria for allocating state grants and subsidies are unclear and non-transparent
(Ott and Bajo, 2001; Bronić, 2008, 2010).
Therefore, in 2009, a new Law on Regional Development inaugurated a new categorization of local (LGU) and regional (RGU) government units based on the
value of a development index. The development index is a composite indicator
calculated as the weighted average of five basic normalized socio-economic indicators: (1) income per capita (X1), (2) budget revenue per capita (X2), (3) unemployment rate (X3), (4) change in population number (X4), (5) educational attainment rate (X5), relative to the national average. The value of the development index for a territorial unit c is calculated using the formula
(1)
where xic, i = 1,2..., 5 represent the normalized value of a sub-indicator Xi for a
territorial unit c. The indicators, their corresponding weights and other issues are
determined by a government decree (Decree on Development Index, 2010). According to the values of their development indices, LGUs are categorized into 5
groups, while RGUs are classified into 4 groups. LGUs classified in groups I or II,
and RGUs classified into group I are rated as lagging behind in development, and
thus have a special status and are granted special support from the state level. The
evaluation and classification of LGUs and RGUs on the basis of the development
index are conducted every three years.
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
The main goal of this study is to examine the existing methodology of the construction of the development index and to propose an improved method for its
financial theory and
practice
39 (2) 205-236 (2015)
Poorly constructed or misinterpreted composite indicators may send misleading
policy messages. Therefore, the validation of such models is of major importance,
and many methods and approaches of model validation have been devised. Since
models cannot be validated in the sense of being proven true (Oreskes et al., 1994),
Rosen (1991) proposed that the justification for a composite indicator lies in its
fitness for the intended purpose and in peer acceptance. However, a group of authors (Saisana and Saltelli, 2008) report that it is more defensible and correct to
say that a model can be corroborated if it passes tests that assess the model’s capacity to explain or predict the “system” in a convincing and parsimonious way.
In order to maximize their utility and minimize their misuse, composite indicators
need to be developed using the best available evidence, documented transparently,
and validated using appropriate uncertainty and sensitivity analyses. Uncertainty
analysis focuses on how uncertainty in the input factors propagates through the
structure of the composite indicator and affects the values of the composite indicator (Saisana et al., 2005). Sensitivity analysis examines the effects of variability
and interactions of indicators as possible sources of instability. While in most
cases the sensitivity and the uncertainty analysis are conducted separately, a synergistic use of an uncertainty and a sensitivity analysis during the development of
a composite indicator could improve the structure (Nardo et al., 2005; Saisana et
al., 2005; Tarantola et al., 2000; Gall, 2007). For example, Saisana et al. (2005)
conducted uncertainty and sensitivity analyses using Monte Carlo simulation and
variance-based techniques. Furthermore, Paruolo et al. (2013) suggest Pearson
correlation coefficient as a measure of variable importance. Also, they conducted
sensitivity analysis using variance-based techniques and applied these methods to
various composite indicators. On the other hand, Hoyland et al. (2012) criticize
the approach in aggregating variables into a composite index without incorporating uncertainty. They suggest a Bayesian approach when assessing uncertainty
and apply this method to three composite indicators. Another group of authors
(Nardo et al., 2008) provide directions for the construction of composite indicators and give recommendations for implementation of sensitivity and uncertainty
analyses. Apart from the papers analyzing regional disparities in Croatia and the
classification of Croatian LGUs and RGUs according to socio-economic development (e.g. Rimac et al., 1992; Grčić and Filipić, 2002; Cziraky et al., 2003, 2005;
Perišić, 2014) there is a lack of related work regarding sensitivity of categorization and uncertainty in the construction of the development index.
207
208
financial theory and
practice
39 (2) 205-236 (2015)
calculation by taking into account the uncertainty and sensitivity of the categorization of LGUs and RGUs. Uncertainty of the development index is assessed
considering the construction methodology and sensitivity is evaluated with respect to the indicators involved in its construction. Uncertainty and sensitivity
analyses play an important role in the construction of composite indicators and
because of their generality these methods can be applied in wide-ranging fields.
The study also includes the analysis of data structure using multivariate methods.
The methods presented in this paper can be useful to policymakers and experts in
assessing the reliability of the composite indicators.
2 The meThODOlOgy Of mODellINg RegIONAl DevelOpmeNT
IN CROATIA
2.1 The DevelOpmeNT Of The RegIONAl DevelOpmeNT pOlICy
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
In its beginning, Croatian regional policy was focused on the renewal of areas affected by war. Some progress was made when the Law on the Areas of Special
State Concern (ASSC) was implemented. These areas were established in order to
encourage their more rapid development and thus they have a special status in the
financing regime (Ott and Bajo, 2001). According to the Law on ASSC adopted in
1996, three categories of units were formed. The first two categories were defined
on the basis of damage caused by the war in Croatia. The third category, introduced in 2002, comprises local units that are lagging behind in development according to four criteria: economic development, structural difficulties, demographic, and a special criterion (Bronić, 2008). The introduction of socio-economic criteria was a step forward in developing regional policy. Eleven indicators
were used for the designation of the third category of ASSC: income per capita,
share of population earning an income, municipality (direct) income per capita,
unemployment rate, employment rate, social aid per capita, change in population
number, educational attainment rate, population density, age index and vitality
index (Puljiz et al., 2005). In addition to the four aforementioned socio-economic
criteria, two special criteria related to border position and mine area status were
used. In total, 185 LGUs enjoyed a special status; of these, 50 were classified in
the first category, 61 in the second category, and 74 in the third category of ASSC.
However, it appeared that the criteria for designation of ASSC were not related to
the real socio-economic indicators. Bajo and Primorac (2013) report that preferential status was also given to some economically more developed (above average) LGUs. Furthermore, a group of authors (Maleković et al., 2011) conclude
that the regional development approach, inherited from central planning, was inconsistent and lacking clearly defined goals, actors, and instruments. The regional
policy has been often conducted within the scope of fiscal equalization and vice
versa, whereby the instruments and the objectives of these policies often clashed,
and the effects were neutralized at a high cost to the state budget. Consequently,
laws related to regional policy needed to be changed. The introduction of a simple
and transparent system of fiscal equalization based on appropriate criteria and fis-
cal instruments remains a challenge for the Government (Bajo and Primorac,
2013).
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
Supported areas were supposed to replace the ASSC, but instead, a parallel existence of different categories of underdeveloped areas occurred. This perennial
problem will be solved by the adoption of a new Law on Regional Development.
Furthermore, this new Law will introduce a uniform system for development
level assessment of territorial units applicable to the whole state. The application
of measures introduced by the new law is expected in 2015. Some of the novelties
of the new Law are introduction of the category of urban area (and additionally,
among such areas, agglomeration centers), the formation of the Council for regional development and binding all future measures to development index, which
acts as the basic instrument for development level assessment of LGUs and RGUs
(MRRFEU, 2013c). By changing the Law on ASSC, the Law on Regional Development, the Law on Hill and Mountain Areas and the Personal Income Tax Act,
the Government endeavors to harmonize tax reliefs and distribution of grants to
LGUs with their development level measured by the development index. The
main changes in the grant distribution to LGUs involve redirecting the grants from
the state budget to LGUs, according to the development index, in the amount of
corporate income tax generated in their area, and changes in the basic personal
allowance (Bajo and Primorac, 2013). By this, LGUs classified in categories I and
II will obtain the right to the entire (100%) corporate income tax revenue collected
in their area, in category III to aid in the amount of 75%, in group IV 60%, and in
category IV 40%. A limitation of this distribution can be presented by following
example: Tovarnik municipality, having the development index value of 74.82%,
will obtain the right to Governmental support in the amount of 100%, while the
City of Ozalj, with a development index value of 75.08%, will obtain the right to
aid in the amount of 75%. These two marginal examples show that more legitimate support is needed. For example, introducing partially linear deductions
would contribute to an equitable support system.
financial theory and
practice
39 (2) 205-236 (2015)
The establishment of a comprehensive and coherent regional development policy
system based on the principles of contemporary regional policy started along with
the preparatory activities for the accession of Croatia to the European Union
(Sumpor et al., 2012). The basis for the implementation of regional policy was the
adoption of the Law on Regional Development in 2009. This Law regulates regional policy by introducing the development index as a basis for the socio-economic development assessment and categorization of territorial units. In addition
to that, the development level of LGUs was assessed for the first time. According
to the Ministry of Regional Development and EU Funds (MRRFEU), the single
system based on the development index contributes to the simplicity and transparency of the whole system, enabling the better direction of incentives and the acquisition of a high-quality basis for the inclusion or exclusion of units from the
supported areas (MRRFEU, 2012).
209
210
financial theory and
practice
39 (2) 205-236 (2015)
Adopting the new Law on Regional Development will stop there being two laws
that define supported areas. However, ASSC again have a special status during
assessment of the development level. The Decree on Amendments to the Decree
on the Development index implies that the development index value of the LGUs
in the ASSC will be reduced by 10 points (the adjusted index) if their original index value is higher than 75% (Decree on Amendments to the Decree on Development index, 2013). In this paper, two indices will be obtained for the local level
(LGU): the unadjusted development index IV (without the reduction by 0 points
for ASSC) and adjusted development index IKV (with the reduction by 0 points for
ASSC).
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
The methodology for construction of the development index for the period 20102012 is different from that used for the period 2006-2008. In 2013, the Decree on
Amendments to the Decree on the Development Index introduced two main modifications: when calculating budget revenue per capita, revenues from the disposal of non-financial assets are excluded; also, the value of the development index is reduced by 10 points for LGUs from the first or second category of ASSC
that have a value of the development index greater than 75%. The first modification prevented the impact of short-term funding sources in budget revenues on the
value of development index. The second modification was introduced in order to
assure the retention of the status of supported area for undeveloped LGUs that
were affected by the war. By this modification, 24 LGUs from the first or second
category of ASSC are, according to the adjusted development index, categorized
as supported areas, although according to the unadjusted development index they
belong to categories of middle or higher development level.
2.2 The CATegORIZATION Of lgUs AND RgUs ACCORDINg TO
DevelOpmeNT INDeX
According to the value of the development index, LGUs are classified into 5
groups (table 1). Table 1 also provides the number of LGUs classified into each
group (N), average values of indicators X1,.., X5 along with standard deviation (in
brackets). Large standard deviations indicate heterogeneity within the group.
According to the development index, RGUs are classified into 4 groups (table 2).
Table 2 also provides the number of LGUs classified into each group (N), average
values of indicators X1,..., X5 along with standard deviation (in brackets).
LGUs classified into categories I or II, and RGUs classified in category I are rated
as lagging behind in development and thus have a special status. In Croatia, there
are 556 LGUs, composed of 429 municipalities and 127 towns, with the City of
Zagreb, the capital of Croatia, having the status of both county and town. More
than one third of LGUs are located on ASSC, while according to the development
index, almost one half of LGUs and more than a half of RGUs are categorized as
areas lagging behind in development.
211
Table 1
The categorization of LGUs according to development index
Category
N
I
<50
47
II
<50, 75>
217
III
<75, 100>
173
IV
<100, 125>
93
V
>125
26
X2
X3 (%)
X4
X5 (%)
13,049.28
(2,146.69)
18,095.3
(3,333.85)
23,355.31
(3,668.74)
28,557.11
(4,426.46)
29,938.73
(6,000.14)
748.87
(288.98)
1,020.29
(472.4)
1,912.6
(1,064.74)
3,556.06
(1,179.26)
7,060.54
(1,359.8)
37.67
(6.59)
22.89
(6.3)
14.89
(4.5)
10.84
(3.6)
8
(3.1)
86.20
(11.01)
92.83
(8.91)
99.92
(9.27)
108.1
(12.15)
119.48
(28.22)
54.07
(8.7)
62.17
(8.1)
72
(6.78)
80.35
(5.1)
83.32
(4.9)
Source: Authors’ calculation.
Table 2
The categorization of RGUs according to development index
Category
Development
index value (%)
N
I
< 75
12
II
<75, 100>
3
III
<100, 125>
3
IV
>125
3
X1
X2
X3 (%)
X4
X5 (%)
22,798
(2,494)
25,460
(785)
27,156
(2,746)
35,662
(5,655)
1,912
(472)
2,597
(506)
3,166
(337)
5,213
(682)
20.47
(4.78)
16.43
(3.58)
13.20
(1.95)
10.37
(2.46)
94.1
(2.67)
100.3
(3.57)
106.53
(2.55)
101.77
(2.4)
69.38
(3.93)
78.84
(4.32)
79.51
(3.71)
84.23
(3.14)
Source: Authors’ calculation.
3 meThODOlOgICAl ASpeCTS Of The CONSTRUCTION
Of COmpOSITe INDICATORS
In the indicators literature there is a fundamental division between those who
choose to aggregate variables into a composite indicator, the aggregators, and
those who do not, the non-aggregators. The aggregators believe that composite
indicators can capture reality and are meaningful. Moreover, they believe that
composite indicators are extremely useful in garnering media interest and hence
the attention of policy makers. On the other hand, the non-aggregators emphasize
the arbitrary nature of the weighting process by which the variables are combined
(Sharpe, 2004). Certain disagreements also exist as to how to select indicators and
techniques for measuring disparities. Relying solely on economic indicators is
questionable due to the fact that development either leaves behind, or in some
ways even creates, large areas of poverty, stagnation, marginality, and actual ex-
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
X1
financial theory and
practice
39 (2) 205-236 (2015)
Development
index value (%)
212
clusion from economic and social progress is too obvious and too urgent to be
overlooked (United Nations, 1969). The need for including more indicators for the
purpose of measuring socio-economic development resulted in a considerable
number of composite indicators.
financial theory and
practice
39 (2) 205-236 (2015)
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
The main pros of using composite indicators are the ability of summarizing complex, multi-dimensional realities, an easy and direct interpretation, facilitation of
communication with general public and reducing the amount of a set of indicators
without dropping the underlying information base. The main cons are related to a
problem of misleading policy messages if the indicators are poorly constructed or
misinterpreted, they can invite simplistic policy conclusions, the selection of indicator and weight could be the subject of political dispute and lead to inappropriate
policy (Nardo et al., 2005). If the construction process is not transparent and/or
lacks sound statistical or conceptual principles, composite indicators can be misused and index values can be manipulated in order to support preferred policy. The
basic steps in the construction of composite indicators are: theoretical framework
construction, data selection, imputation of missing data, multivariate analysis, normalization, weighting and aggregation, uncertainty analysis, sensitivity analysis,
linking to other variables, presentation and visualization (Nardo et al., 2005).
3.1 INDICATORS SeleCTION
The strengths and weaknesses of a composite indicator are derived from the quality of the underlying variables. Variables should be selected on the basis of their
relevance, analytical soundness, timeliness, accessibility, etc. The choice of indicators must be guided by the theoretical framework, limiting the developer’s subjectivity. The use of multivariate analysis helps to identify the data structure, and
can increase both the accuracy and the interpretability of final results. Multivariate
methods can also help to investigate the presence of multicollinearity. Multicollinearity causes an unequal variable position because multicollinear variables
measure the same phenomenon, which in turn has a multiple contribution to the
value of the composite indicator. An additional problem is the difficulty of interpretation, while it is hard to assess the impact of the individual variable. If two
variables are highly correlated, the inclusion of both of them in the model may be
redundant (Salzman, 2003).
In the process of the construction of new composite indicator, the development index, the main criterion for variable selection was the estimation of the variable’s
contribution to the objective estimation of socio-economic differences among territorial units. Thereby, income per capita and unemployment rate became key indicators of socio-economic disparities, while some indicators were excluded on the basis
of their weak credibility, and too high or too low correlation with the key economic
indicators (Puljiz et al., 2005). For instance, infrastructural indicators were rated as
not reliable; the indicators “share of persons earning an income” and “income per
capita” were highly correlated (coefficient of correlation was 0.93); the indicators
Table 3
Indicators LGU
Indicator
min.
max.
xˉ
skew
X
7,105
42,175
21,609.14
6,088.6
0.38
X2
223
10,115
1,981.59
1,695.36
1.96
X3 (%)
4.5
54.48
18.94
9.24
0.86
41.3
247.8
98.3
13.8
31.65
90.41
68.58
11.07
X4
X5 (%)
a

2.72a
-0.36
The effect of an outlier, when the outlier is excluded, asymmetry coefficient is 0.65.
Source: Authors’ calculation.
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
Minimum value (min), maximum value (max), mean value ( xˉ ), standard deviation
() and asymettry coefficient (skew) are presented in tables 3 and 4 for the LGUs
and RGUs respectively. Variable X2 has the largest dispersion both on local and
regional level.
213
financial theory and
practice
39 (2) 205-236 (2015)
“vital index” and “age index” had a too low correlation coefficient with the indicators “income per capita” and “unemployment rate” (Puljiz, 2007). Due to some
methodological difficulties, gross domestic product per capita (GDP per capita) was
not involved in the construction of the development index. However, the GDP per
capita is still considered as a potential indicator while it is expected that the reliability of the estimation of GDP will grow (Puljiz, 2007). However, it is not possible to
use the GDP per capita as an indicator on the local level because it is calculated on
the regional and state level. The Decree on the Development Index prescribes the
indicators used in the construction of the development index: income per capita,
budget revenue per capita, unemployment rate, change in population number, and
educational attainment rate. The indicator values have been acquired from MRRFEU (2013a, 2013b, 2010a, 2010b). Local/regional income per capita is calculated
by dividing the total income in a tax period (one calendar year), generated by taxpayers, persons domiciled or habitually resident, in the territory of the local/regional
unit, by the total population number of a local/regional unit. Budget revenue per
capita is calculated by dividing the realized reduced local/regional budget revenue
by the population number of a local/regional unit. Budget revenue is reduced by:
receipts from domestic and foreign aid, subsidies and transfers, receipts derived pursuant to special contracts (co-financing by citizens), additional share in income tax,
equalization aid for the decentralized functions and disposal of non-financial assets.
Unemployment rate is determined by dividing the total number of unemployed persons by the sum of unemployed and employed persons in the area of the local/regional unit. Change in population number is expressed as the ratio of the number of
the population in a local/region unit according to the last two consecutive censuses.
Educational attainment rate is calculated as the number of persons with secondary
education and higher, aged over 15 years, as a proportion of the total number of
population aged 16-65 in the area of the LGU/LRU.
214
Table 4
Indicators RGU
financial theory and
practice
39 (2) 205-236 (2015)
Indicator
X
X2
X3 (%)
X4
X5 (%)
min.
19,455
1,368
7.80
90.7
62.49
max.
42,175
5,997
25.90
109.1
86.93
xˉ
25,638.19
2,660.48
17.41
97.86
74.30

5,262.95
1,248.45
5.54
5.42
7.01
skew
1.66
1.36
0.17
0.50
0.15
Source: Authors’ calculation.
3.2 NORmAlIZATION
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
Data normalization is often required because the indicators in a data set have different measurement units or large differences in their means or standard deviations. The normalization phase can be crucial for the accuracy of a composite indicator and accordingly for the ranking of units. The simplest normalization technique is ranking. This method is not affected by outliers. Standardization, or converting to z-scores, on the other hand, is one of the most popular normalization
methods. It is affected by outliers since indicators with extreme values have a
greater effect on the composite indicator value. The third widely used normalization method is the min-max method, which normalizes the indicators so that they
have an identical range. Distance to a reference measures the relative position visà-vis a reference point (the reference could be the average value, minimum value,
maximum value or some other value determined by the expert decision).
Table 5
Evaluated normalization methods
Normalization methods
The calculation
Standardization (z-score)
(2)
Min-max
(3)
Distance to a reference
(4)
Remark: xci is the value of indicator Xi for a local unit c; i represents the average value of indicator Xi across all units c; i is the standard deviation of indicator Xi; xi,min is the minimum value
of indicator Xi across all units c; xi,max is the maximum value of indicator Xi across all units c; xi0
is the reference value for indicator Xi.
By the Decree on the Development Index, the min-max method was selected as a
suitable normalization method. The min-max normalization of an indicator Xi is
conducted by subtracting the minimum value and dividing by the range of the indicator values. Furthermore, this normalized value is divided by the normalized
state average value of the indicator Xi (formula 5)
215
(5)
financial theory and
practice
39 (2) 205-236 (2015)
Since the extreme values are usually outliers, this normalization method should be
revised (in the sense of excluding outliers or even changing the normalization
method). Changing the minimum value of an indicator can affect the composite
indicator values of all units. The normalization defined in formula 5 reduces the
nominal value of each indicator to a distance to the minimum value relative to the
distance of the national average to the minimum value.
3.3 WeIghTINg AND AggRegATION
An a priori determination of weights for various components implies the existence of a universally acceptable human welfare/development function, which is
not the case (Noorbakhsh, 1998). Thus, the empirical approach, as an alternative
approach to weight setting, is worth pursuing. In order to generate a set of nonsubjective weights, PCA has been applied for the calculation of development index both for the period 2008-2010 and for 2010-2012. PCA enables the reduction
of the variables by a small number of their linear combinations. The objective is
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
Perhaps the most difficult aspect of constructing a multidimensional index is
choosing weights for the components (Tofallis, 2013). A number of weighting
techniques exist. Some are derived from statistical models, such as factor analysis,
principal component analysis (PCA), data envelopment analysis, or from participatory methods that incorporate various stakeholders – experts, citizens and politicians, like budget allocation processes, analytic hierarchy processes and conjoint
analysis (Nardo et al., 2008). The weights are often selected in order to reflect the
relative importance of the indicator for the phenomenon to be measured. This approach is often criticized as too arbitrary since weighting can have a significant
effect on composite index value and ranking of units. On the other hand, multivariate techniques present an empirical and relatively more objective option for
weight selection, allowing for no control over the selection of weighting scheme.
This is due to the fact that the weights are selected based on the data themselves.
The problem of selecting an indicator’s weights is often bypassed by equal weighting, i.e. all variables are given equal weights. Composite indices generally seem
to be additive, with equally weighted components (Booysen, 2002). This essentially implies that all variables are “worth” the same in the composite, but it could
also disguise the absence of a statistical or an empirical basis, e.g. when there is
insufficient knowledge of causal relationships or a lack of consensus on the alternative (Nardo et al., 2005). When a few highly correlated indicators exist and
equal weighting is applied, double counting may be introduced into the index.
Thus, it is desirable to perform correlation analysis prior to weight selection and
to adjust weights according to results of correlation analysis.
216
to explain the variance of the data through a few linear combinations of the original data minimizing the information loss. Therefore, k ≤ 5 linear combinations Yj
of variables Xi are chosen, maximizing the variance of the original data.
k.
(6)
financial theory and
practice
39 (2) 205-236 (2015)
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
The first principal component, Y, is the linear combination of the variables that
has the greatest possible variance. Various guidelines have been developed on the
issue of how many linear combinations k should be retained. For instance: the latent root criterion (only components that have an eigenvalue greater than are
retained), the a priori criterion, scree test and proportion of variance criterion
(Hair et al., 1995). In practice, very few composite indices use PCA weights, in
part because it is difficult to explain the process to non-statisticians, in part because the weights themselves change as the data changes over time, but mainly
because the results using equal weights and PCA weights tend not to differ substantially (Foa and Tanner, 2011).
Because the composite indicators describe multidimensional phenomena by a unidimensional construct, it is necessary to find an appropriate aggregation method.
Additive aggregation is the simplest aggregation method and is independent of outliers. Linear aggregation, the most widespread method, is the summation of weighted
individual indicators. An undesirable feature of linear aggregation is that it implies
full compensability, such that poor performance in some indicators can be compensated for by sufficiently high values in other indicators. Geometric aggregation is
better suited if the modeller wants some degree of non-compensability.
3.4 UNCeRTAINTy AND SeNSITIvITy ANAlySIS
The absence of an “objective” way to determine weights and aggregation methods
does not necessarily lead to rejection of the validity of composite indicators, as
long as the entire process is transparent. A combination of uncertainty and sensitivity analysis can help gauge the robustness of the composite indicator and improve transparency (Nardo et al., 2008). The methodological choices, such as the
treatment of missing data, indicator selection, choosing the normalization method
as well as the aggregation method, and the weighting scheme have a major influence on the composite indicator. The goal of the uncertainty analysis is to evaluate
how the subjective choices taken by the modeller are reflected in the confidence in
the model. Since the development level of a territorial unit cannot be directly
measured, it is not possible exactly to determine how well the composite index
describes this phenomenon. However, a composite indicator needs to be validated.
Thus, each construction of a composite index needs to involve uncertainty analysis and sensitivity analysis. Generally, uncertainties during the development of a
composite indicator are associated with a number of factors: data error, choice of
the mechanism for indicator inclusion and exclusion, the transformation of indicators, missing data, choice of normalization method, aggregation method and especially of the weighting scheme (Nardo et al., 2005). The uncertainty of a compos-
ite index can be assessed by exploring the main sources of uncertainty, with the
aim of capturing all possible synergy effects among uncertain input factors (Nardo
et al., 2008), directly or by Monte Carlo simulation.
financial theory and
practice
39 (2) 205-236 (2015)
High sensitivity to small changes in input information makes the composite index
unstable and unreliable. Thus, when a composite index is being created, it is necessary to determine the effect of the variation in the individual indicator on the
overall index value. Therefore, methods that are not based on the assumptions of
the model, such as variance-based techniques, are used (Satelli et al., 2008).
217
For the composite index Y and indicators X1,…, Xk, the first-order sensitivity index
is assessed as the contribution of an individual indicator X1 to the total output
variance V(Y),
where
(8)
is the variance of Y due to the uncertainty in Xi. In order to compute a variancebased sensitivity measure, the first factor Xi needs to be fixed to a specific value
xi0 in its range. Then the mean of the output is computed by averaging over all factors but factor Xi,
and the variance of the resulting function of xi0 is
averaged over all possible values of the indicator Xi. Higher values of the sensitivity index Si indicate a higher dependance between the index Y and the individual
indicator Xi.
Analogously, it is possible to compute conditional variance to more than one factor (interactions) and then to compute high-order sensitivity indices. When an
additive model is concerned (a model without interactions) the contribution of an
individual indicator to the variance of a composite is entirely obtained by the firstorder sensitivity index and therefore formula (9) is valid.
.
(9)
Si is a good model-free sensitivity measure, and it always gives the expected reduction of the variance of the output that one would obtain if one could fix an individual indicator (Saltelli et al., 2004). In general,
.
(10)
For a non-additive model, instead of computing higher-order sensitivity indices,
the total effect sensitivity index STi is obtained; this is determined by their sum.
,
(11)
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
(7)
218
financial theory and
practice
39 (2) 205-236 (2015)
where
represents the variance of the composite index Y due to the
uncertainty of all indicators but indicator Xi. As in the first-order sensitivity indices, high values of the total effect STi indicate a higher dependence of the index Y
on the variability of an individual indicator Xi. However, unlike the first-order
sensitivity indices, this approach considers all possible interactions with other indicators. Therefore, by comparing the pairs (Si, STi) it is possible to compute the
intensity of indicator interaction.
4 ReSUlTS
4.1 mUlTICOllINeARITy AND OUTlIeR pReSeNCe TeSTINg
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
Correlation matrices of indicators on the local and regional level are presented in
tables 6 and 7. The correlation of indicators is high and statistically significant (at
significance level 1%). On the regional level, correlation coefficients of indicators
income per capita and budget revenue per capita, and also budget revenue per
capita and educational attainment rate are problematic in the sense of too high
correlation.
Table 6
Correlation matrix LGU
X
X2
X3
X4
X5
X1
X2
X3
X4
X5
0.92
-0.68
0.44
0.79
-0.67
0.55
0.81
-0.59
-0.57
0.66
X2
X3
X4
X5
Source: Authors’ calculation.
Table 7
Correlation matrix RGU
X1
X
X2
0.51
X3
-0.67
-0.48
X4
0.25
0.51
-0.39
X5
0.73
0.55
-0.55
0.51
Source: Authors’ calculation.
Multicollinearity was assessed using variance inflation factor and condition index.
It can be concluded that multicollinearity is present since the value of variance
inflation factor is higher than 7 and value of condition index is 13.8 (Bahovec and
Erjavec 2009; Gujarati, 2004). The problem of multicollinearity can be solved by
excluding one of the indicators or by replacing it with another indicator. For ex-
Figure 1
Detection of outliers in LGUs using the squared Mahalanobis distance (d) for the
period (a) 2006-2008 and (b) 2010-2012
d
d 12
0
8
0
6
4
2
0
0
0
100
200
300
LGU
400
(a) Period 2006-2008
500
600
0
100
200
300
LGU
400
500
600
(b) Period 2010-2012
Source: Authors’ calculation.
In both periods one outlier can be detected in the upper right side of the figure. For
the period 2006-2008 the outlier is the municipality Dugopolje which has the largest development index and extremely high budget revenue per capita. In that period, the second largest development index corresponds to Kostrena municipality.
This municipality, however, has a three times smaller budget revenue per capita.
Furthermore, Dugopolje municipality has more than twenty times the budget revenue per capita of the average of all LGUs. Dugopolje municipality is an example
of the failure of aggregation methods, since the deficit in one dimension can be
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
Outliers were detected using Mahalanobis distance, where observations with large
Mahalanobis distances were indicated as outliers (Ben Gal, 2005). Squared Mahalanobis distances (d) corresponding to LGUs for the period (a) 2006-2008, and (b)
2010-2012 are presented in figure 1.
219
financial theory and
practice
39 (2) 205-236 (2015)
ample, for the development index calculated for the period 2006-2008, the problem of multicollinearity was solved by excluding the indicator budget revenue per
capita, and also by replacing this indicator with alternative indicators of regional
budget strength (Perišić, 2014.). A high correlation of budget revenue per capita
and income per capita is expected. Furthermore, local revenues depend mostly on
tax revenues, especially on income tax and surtax on income tax (Ott, 2009).
Therefore, the inclusion of both indicators may be redundant, especially when the
key economic indicator GDP per capita is available on a regional level. This issue
can be circumvented by introducing realized local/regional budget revenue in the
total income of LGU/RGU: an alternative indicator of strength and independence
of local budgets. In that case, the realized local/regional budget revenue should be
reduced by receipts from domestic and foreign aid, subsidies and transfers, receipts derived pursuant to special contracts (co-financing by citizens), additional
share in income tax, equalization aid for the decentralized functions and disposal
of non-financial assets.
220
financial theory and
practice
39 (2) 205-236 (2015)
compensated by a surplus in another and thus create a false image on the unit’s
development level. The extremely high budget revenue of Dugopolje municipality
is the result of short-term funding sources, such as sale of land. A similar situation
occurred with other LGUs. For the period 2010-2012 the municipality of Vir
stands out as an outlier with an extremely high change in the population number
(247.8). These examples show that outliers create a false image of the development level of a territorial unit, and also affect the value of the development index
of all units. This effect, as well as the effect of different data range, is removed
with transformation and normalization of the data.
4.2 pRINCIpAl COmpONeNT ANAlySIS ReSUlTS
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
Principal component analysis is conducted for the development index, and calculated for both periods, 2006-2008 and 2010-2012. A priori only one linear combination was selected (k = 1), representing an alternative to the development index.
It is statistically justified to retain only one component since only one principal
component has the eigenvalue greater than 1, which satisfies the latent root criterion. Indicator weights have been calculated on the basis of PCA, and afterwards
normalized. The resulting weights are represented in table 8. It is important to note
that outliers have been excluded prior to performing the PCA.
Table 8
Normalized indicator weights according to PCA, LGU and RGU level
Normalized Income
weights
w1pCA
pCA
w
Index LGU
2010-2013
Index LGU
2006-2008
Index RGU
2010-2013
Index RGU
2006-2008
Budget Unemployment Change in
revenue
rate
population
w2pCA
w3pCA
number
w4pCA
educational
attainment
rate
w5pCA
variance
explained
(%)
0.21
0.2
0.2
0.17
0.22
62
0.23
0.18
0.21
0.15
0.23
55
0.21
0.22
0.19
0.17
0.21
74
0.24
0.24
0.2
0.
0.22
66
Source: Authors’ calculation.
One of the weaknesses of the weights derived from PCA is the minimization of the
contribution of individual indicators which do not move along with other individual indicators. In case of development index, the indicator change in population number has the smallest weight determined by PCA because this indicator
exhibits the weakest correlation with other indicators. However, weights derived
from PCA assigned to demographic indicators are higher than those determined
by the government decree, at both a local and a regional level.
221
financial theory and
practice
39 (2) 205-236 (2015)
According to the index computed with the use of PCA-derived weights (IPCA), in
total 57 LGUs are classified into category I, and 211 LGUs are classified into
category II. When using the adjusted development index (reducing the index value
for 10 points for the undeveloped LGUs on the ASSC), category II contains 239
LGUs. Also, only three LGUs classified into categories I or II by the development
index are now, according to PCA-derived index, classified into groups with a
higher development level. On the regional level, there is no difference between
the classification of RGUs according to government decree and the PCA derived
index. The values of PCA derived indices are presented in table 0.
4.3 AlTeRNATIve NORmAlIZATION meThOD
Development index, Ic, of a local unit c is calculated as the weighted average of
five basic normalized socio-economic indicators
(12)
Section 3.2 gives a critical review of the normalization method used in the construction of the development index. The direct influence of the maximum value of
an individual indicator on the LGU/RGU development index can be removed by
combining the min-max normalization and averaging relative to the development
index of Croatia (formula 13). The calculation of the development index as a
weighted average of the same indicators, but on the state level, would result with
an indicator less sensitive to extreme values.
.
(13)
4.4 UNCeRTAINTy ANAlySIS Of The DevelOpmeNT INDeX
4.4.1 Uncertainty analysis of the lgU development index
Taking into account analogous indices of countries in the region, one can recognize weights and normalization method selection as the main sources of uncertainty. In total, 12 models have been obtained: 3 normalization methods are presented in table 5 and the normalization method determined by the government
decree (formula 5); 3 weighting methods (weighting method determined by the
government decree wV; weights derived from PCA for the period 2010-2012 wPCA
(table 8); and equal weights wE = (0.2, 0.2, 0.2, 0.2, 0.2)). Indices computed with
different normalization methods are not mutually comparable. Thus, the uncertainty was obtained due to the corresponding LGU ranks. For each model
i = 1,...,12 and for each LGU c (c = 1,..., 556), the development index Ici and corresponding rank Rci were calculated. Figure 2 shows the smallest, greatest and the
median rank value for a LGU c relative to the rank RcNV determined by the unadjusted development index Icv.
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
.
222
Figure 2
Uncertainty analysis due to LGU rank, LGU level
500
financial theory and
practice
39 (2) 205-236 (2015)
400
300
max Rj
min Ri
200
00
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
0
0
100
200
300
RjNV
400
500
Source: Authors’ calculation.
Analysis of the data presented in figure 2 shows good agreement in low and high
regions of RNV and certain jumps of the median rank for the middle region. This
behavior may indicate a possible overestimation, or underestimation of certain
LGUs, especially LGUs with medium development level. On the same set of
units, it can be observed that a wider range of indicator values follow this behavior. This indicates the inability of a model to concisely estimate the development
level of these units. Furthermore, this leads to significant disparities in the categorization based on development index compared to the categorization based on
other models.
A key issue in the construction of a composite indicator is determining the relative
importance of indicators and by that choosing appropriate indicator weights. This
is especially important because the sensitivity of the composite indicator, due to
change in weights, opens up the possibility of manipulation of the categorization
of units lagging behind in development. Thus, uncertainty due to change in weights
is analyzed.
For every LGU c, the minimum Icmin and maximum Icmax possible values of the
index are calculated. Weights were chosen from the interval [0.1, 0.6] and scaled
to a unity sum.
(14)
weights determined by the government decree.
(c) Third simulation S3 was conducted by generating 2500 weights from the
uniform distribution,
where wiPCA represent
weights determined from PCA.
After being generating, all weights were scaled to unity sum. For each simulation,
and for every LGU c, 500 values of development index Ic,k, k = 1,2,...,500 were
calculated The first simulation is almost nonrestrictive, while the second and third
simulation generate weights around values determined by the government decree
wv, and PCA, wPCA.
The percentage of simulations resulting in a classification corresponding to the
development index IV is calculated for each simulation and each unit. The average
percentage for each category is provided in table 9, as well as the average range of
simulated indices,
, where
and
are the maximum and minimum value in each simulation. Additionally, table 9
provides the frequency of incorrectly classified units (nw) which have the probability of wrong classification pw(c) higher than 0.45:
(15)
For simulations S1 and S2, the probability of incorrect categorization was estimated
as the relative frequency of incorrect categorization of LGUs due to the categorization based on IcV. Also, for simulations S1 and S3, the probability was estimated
due to a categorization based on IcPCA.
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
Uncertainty analysis on the LGU level was performed using the Monte Carlo
simulation. Three types of simulations were performed due to the distribution of
weights:
(a) The first simulation S1 was conducted by generating 2500 weights from the
uniform distribution,
. Afterwards, weights were scaled to
a unity sum. The latter ensured for weights to range between 0.1 and 0.35,
preventing domination of an individual indicator over others.
(b) The second simulation S2 was conducted by generating 2500 weights from
uniform distribution,
, where wiv represents
223
financial theory and
practice
39 (2) 205-236 (2015)
Results obtained by formula (14) were compared to the results obtained by using
the development index. The comparison showed that 56% of units classified into
categories I or II by the development index are classified into categories III, IV or
V when Icmax was used, i.e. 56% of the units from the supported areas are not rated
as lagging behind in development. Similar results occur when the classification
according to minimum possible index value, Icmin is analyzed. More than 43% of
units classified into categories III, IV and V by the development index are now
classified into categories that are lagging behind in development.
224
Table 9
Simulation result, LGU level
Simulation
financial theory and
practice
39 (2) 205-236 (2015)
a
b
Category retention (%)
nw
Imax–min
I
II
III
Iv
v
S1
87
92
80
82
98
71a
31b
0.14
S2
96
94
94
96
96
9
0.08
S3
96
94
94
95
98
6
0.08
Number of incorrect categorizations according to IV.
Number of incorrect categorizations according to IPCA.
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
Source: Authors’ calculation.
The simulation S1, which is the least restrictive in the choosing of weights, resulted in 31 LGUs being incorrectly categorized due to the categorization based
on the IPCA. This is a significantly lower number than the number of incorrect categorizations resulting from the categorization based on the development index IV
(71 LGUs incorrectly categorized). Thus, it can be concluded that the categorization according to the development index IV is less confident than the PCA categorization. In addition to that, a greater confidence of the PCA-categorization can be
obtained by considering smaller perturbations of weights around the values of wV
or wPCA.
Monte Carlo simulations open up the possibility of employing confidence intervals for the estimation of the development level of territorial units. Moreover,
when using simulations it is possible to estimate the probability of incorrect categorization for every unit. This is particularly beneficial in the case of marginal
units. For instance, Lećevica municipality, according to development index value
(75.36%), is not categorized in the group of supported units. However, 84% of
simulations S1 resulted in the categorization of this municipality in the group of
supported units. A similar situation occurs with more than 20 LGUs (see appendix). Due to the large number of LGUs, interval estimations are presented only at
a regional level (table 10).
4.4.2 Uncertainty analysis of RgU development index
The Decree says that the methodology for the construction of the development
index on the local level is the same as that for the regional level (Decree on the
Development Index, 2010). However, one can pose the question whether one
methodology can be reliably applied to two different cases? In particular, is it
justified to select the same weights at a local and a regional level?
For every RGU c, two types of development indices were computed: the development index IcV using weights determined by the government decree and the index
IcPCA, computed using weights derived from PCA. Based on indices values, corresponding ranks were determined. The results are presented in table 10, where
the ranks are presented in brackets.
Icv (%)
7.18 (1)
5.56 (1)
7 (1) <0.06, 0.08>
<0.05, 0.06> <0.06, 0.08>
24.03 (4)
18.43 (2)
22.42 (4) <0.19, 0.29>
<0.16, 0.21> <0.20, 0.25>
22.65 (3)
18.73 (3)
21.74 (2) <0.19, 0.26>
<0.17, 0.21> <0.20, 0.24>
21.96 (2)
23.29 (4)
21.84 (3) <0.21, 0.23>
<0.23, 0.24> <0.21, 0.22>
30.25 (5)
33.81 (5)
29.01 (5) <0.26, 0.34>
<0.31, 0.36> <0.27, 0.31>
39.75 (6)
38.70 (6)
41.54 (6) <0.35, 0.44>
<0.36, 0.42> <0.39, 0.44>
49.11 (7)
46.07 (7)
49.21 (7) <0.47, 0.52>
<0.44, 0.48> <0.48, 0.51>
Karlovac
KoprivnicaKriževci
Lika-Senj
53.28 (9)
56.34 (8)
54.47 (9) <0.49, 0.57>
<0.54, 0.59> <0.52, 0.57>
52.63 (8)
59.19 (9)
51.64 (8) <0.46, 0.59>
<0.55, 0.64> <0.48, 0.56>
59.61 (10)
64.82 (10)
61.04 (10) <0.54, 0.65>
<0.61, 0.68> <0.58, 0.64>
Međimurje
KrapinaZagorje
Šibenik-Knin
64.79 (12)
69.65 (11)
62.03 (12) <0.58, 0.72>
<0.64, 0.75> <0.58, 0.66>
62.78 (12)
73.24 (12)
61.45 (11) <0.56, 0.69>
<0.68, 0.78> <0.58, 0.66>
83.04 (14)
80.93 (13)
82.14 (14) <0.80, 0.87>
<0.79, 0.83> <0.80, 0.84>
Varaždin
78.81 (13)
86.34 (14)
77.39 (13) <0.73, 0.84>
<0.82, 0.90> <0.74, 0.81>
IcpCA (%)
90% CI
for IcS1
90% CI
for IcS2
90% CI
for IcS3
Split-Dalmatia
102.55 (15)
93.75 (15) 101.2 (15) <0.97, 1.09>
<0.90, 0.98> <0.98, 1.05>
Zadar
DubrovnikNeretva
Zagreb
PrimorjeGorski kotar
Istria
112.62 (16) 106.39 (16) 109.13 (16) <1.04, 1.23>
<1.01, 1.12> <1.04, 1.14>
123.62 (18) 120.84 (17) 122.31 (18) <1.2, 1.28>
<1.18, 1.23> <1.20, 1.25>
121.21 (17) 124.23 (18) 118.03 (17) <1.15, 1.28>
<1.21, 1.28> <1.14, 1.22>
139.69 (19) 139.21 (19) 141.71 (19) <1.35, 1.44>
<1.37, 1.41> <1.39, 1.44>
154.36 (20) 156.80 (20) 154.20 (20) <1.5, 1.59>
<1.54, 1.6>
185.45 (21) 186.44 (21) 188.5 (21) <1.76, 1.93>
<1.81, 1.92> <1.83, 1.93>
City of Zagreb
CI – interval estimation.
Source: Authors’ calculation.
<1.51, 1.57>
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
¯IcS1 (%)
County of
c
ViroviticaPodravina
Slavonski
Brod-Posavina
VukovarSirmium
BjelovarBilogora
PožegaSlavonia
SisakMoslavina
Osijek-Baranja
financial theory and
practice
39 (2) 205-236 (2015)
Table 10
Comparison of the index value and corresponding rank (in brackets)
225
226
financial theory and
practice
39 (2) 205-236 (2015)
Due to the selection of weights, uncertainty analysis was conducted using Monte
Carlo simulations. Analogously to the analysis conducted on the local level, 5,000
samples of weights from three types of uniform distributions were generated. For
each simulation Sj and every RGU c, 1,000 values of development index
k = 1,2,..., 1000, j = 1,2,3 were calculated. For 21 RGUs, interval estimations
of development index, based on percentile index values, are computed with 90%
confidence level.
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
The comparison of lower bounds of the interval estimations obtained by simulations S1 and S3, shows that the development index IV underestimates the development level for following counties: Virovitica-Podravina, Slavonski Brod-Posavina, Vukovar-Sirmium, Bjelovar-Bilogora, Osijek-Baranja and Split-Dalmatia.
These counties have a development index value lower than the lower bounds of
interval estimations provided by simulations S1 and S3. Similarly, the comparison
of the development index IV and the upper bounds of the interval estimations indicates a possible overestimation of the development level for the following counties: Bjelovar-Bilogora, Koprivnica-Križevci and Varaždin since the development
index is higher than the upper interval bounds. The selection of weights doesn’t
significantly affect the categorization of RGUs determined by the government
decree, but it has an impact on the units ranking. According to the development
index, the county of Split-Dalmatia is categorized in the second group, while according to the average simulated index and PCA derived index is categorized in
the group having a higher development level.
4.5 SeNSITIvITy ANAlySIS Of The DevelOpmeNT INDeX
4.5.1 Sensitivity analysis of the development index, lgU level
First order sensitivity indices Si and total effect sensitivity indices ST are calcui
lated for indicators Xi, i=1, 2,..., 5. The results on the LGU level are presented in
table .
Table 11
Sensitivity indices (LGU level)
I
Si
1
2
3
4
5
0.084
0.145
0.087
0.021
0.022
STi
0.394
0.485
0.390
0.155
0.216
Source: Authors’ calculation.
It can be concluded that a change in the value of a single indicator affects the values of the development index. This is due to the fact that a single indicator interacts with other indicators and thus the cumulative effect is observable on the value
of development index. Significantly higher values of total effect sensitivity index
indicate stronger interactions between economic indicators: budget revenue per
capita, income per capita and unemployment rate.
Figure 3
Interval estimates
227
financial theory and
practice
39 (2) 205-236 (2015)
The effect of small changes in the values of economic indicators on the sensitivity of
the development index was examined directly by using the Monte Carlo simulations.
For every regional unit c, 100 values of the indicator Xi, i = 1,..., 5 were generated from
the normal distribution with a mean equal to the observed indicator value xci and variance i2 equal to the quarter of the sample variance. This way the simulation is limited
only to small perturbations of the parameters, which can appear as the result of imprecise measurement or the fact that indicator values tend to change over time.
1.4
Development index
.0
0.8
0.6
0.4
0.2
0
100
200
300
400
500
LGU
Source: Authors’ calculation.
Tabel 12
Category changed (%)
Category according
to Iv
I
II
III
IV
V
falling into lower
category (%)
Advancing into higher
category (%)
–
9
15
15
8
19
12
12
7
–
Source: Authors’ calculation.
The values of the development indices on the LGU level, together with corresponding interval estimations, are presented on figure 3. The interval estimations
are assessed via simulations at the 95% confidence level. The LGUs that shift to
categories with lower or upper development levels are determined by intersections of level boundaries according to IV and interval estimations. Table 12 provides the percentages of shifts between groups. LGUs switching between catego-
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
1.2
228
ries should be analyzed in more detail due to the possibility of incorrect classification.
4.5.2 Sensitivity analysis of the Development index, RgU level
Table 13
Sensitivity indices (RGU level)
1
2
3
4
5
Si
0.084
0.039
0.119
0.021
0.022
0.439
0.322
0.481
0.239
0.221
ST
i
Source: Authors’ calculation.
Similar to the results obtained with the sensitivity analysis of LGUs, the interaction
of economic indicators emerges as the main source of variability of the development index on the RGU level. The values of the development indices on the RGU
level together with corresponding interval estimations are presented on figure 4.
Figure 4
Interval estimations (RGU level)
1.25
XXI
XVIII
I
VIII
XIII
XIX
V
XVII
II
XV
IX
XX
IV
VI
III
XIV
XI
VII
XVI
X
0.75
XII
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
I
Development index
financial theory and
practice
39 (2) 205-236 (2015)
First order sensitivity indices Si and total effect sensitivity indices ST are calcui
lated for indicators Xi, i=1, 2,...,5. The results at the RGU level are presented in
table 13.
RGU
Source: Authors’ calculation.
Interval estimations of the development index for the counties Krapina-Zagorje,
Split-Dalmatia, Dubrovnik-Neretva and Zagreb are partially placed in the categories of higher development level according to the category determined by the index IV. This indicates a possible underestimation of the specified counties. Thus,
these counties should be analyzed in more detail. Using interval estimates allows
for a briefer analysis of these marginal units considering the estimation of an incorrect classification of territorial units.
The researchers’ subjective decisions, such as choosing normalization and aggregation methods and selecting indicators and corresponding weights, play an important role in the construction of a composite index. Thus, it is necessary to detect all sources of uncertainty and to analyze their consequences. One of the main
features of the construction of composite indicators is the lack of consensus concerning relative weights selection. In this study, a development index derived directly from the data by principal component analysis was proposed as a reference
index. The uncertainty of the development index, especially due to the weight
selection, was analyzed using Monte Carlo simulations. This analysis indicated
the inability of the development index to rank LGUs uniformly, particularly for
LGUs categorized in groups II and III. Because weight selection is the main source
of uncertainty, categorization on the basis of the development index was revealed
to be less confident compared to the PCA-categorization.
The analysis of the model determined by the Government decree indicated strong
interactions of economic indicators which dominantly affect the variability of the
development index, both on the local and the regional level. Because even small
perturbations of input variables can significantly affect the categorization outcome, a special emphasis was placed on the inclusion of these methods in the
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
The main goals of this paper were to assess the uncertainty of development index
considering the construction methodology and to evaluate sensitivity considering
the indicators involved in its construction. By taking this to account, useful guidelines for the improvement of the index can be proposed. Furthermore, the methodology of the construction of the development index is presented. In order to examine the data structure, multivariate analysis was conducted, indicating the presence of outliers. In particular, the results of multivariate analysis carried out at the
regional level indicate the presence of collinearity between indicators income per
capita and budget revenue per capita. This problem can be bypassed by excluding
or replacing one of the collinear indicators or by correcting indicator weights.
financial theory and
practice
39 (2) 205-236 (2015)
5 CONClUSION
Prior to the implementation of the development index, the categorization process
of territorial units was often criticized as unbalanced and fragmented. A turnaround started with the employment of composite indices as a basis for the development level assessment of LGUs and RGUs. The development index is calculated as the weighted average of normalized indicator values relative to the state
average. For many local units that are lagging behind in development, the categorization based on the development index is crucial for further development as it
provides it different benefits and incentive intensity.
229
230
financial theory and
practice
39 (2) 205-236 (2015)
construction of composite indicators. The inclusion of uncertainties in the model
is needed since the measurement errors are frequent, and for some indicators, data
are often not provided on time. For this purpose, interval estimations are proposed
instead of point estimation. Therefore, it is possible to incorporate uncertainties in
the model. This also enables some units to be closely analyzed (for example, territorial units having the development index values 5 points away from the marginal values of the categories).
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
Due to the lack of scientific papers on this subject, the intention of the authors was
to emphasize the need for analyzing the sensitivity and the uncertainty of the development index and of composite indicators in general. Since composite indicators are increasingly used, this study provides useful guidelines for the application, as well as for development, of composite indicators in a wide variety of academic and professional fields.
In three years a new calculation of the development index is expected. It will be
interesting to observe the development of the construction methodology, especially the indicator selection process and quality, as well as the assessment of the
relative importance due to the socio-economic conditions in the state. With the
aim of the general acceptance of the development index, it is necessary to improve
the transparency of the construction process. The methodology of the new calculation should follow the expected changes in the dynamics of indicators and their
interaction in order to ensure a modern approach to the assessment of the development level of territorial units.
AppeNDIX
231
Basic guidelines for improving the methodology of the development index
financial theory and
practice
39 (2) 205-236 (2015)
1) The correlation of indicators income per capita and budget revenue per capita
should be corrected in one of the following ways:
a) correction of weights,
b) excluding an individual indicator,
c) replacement with an alternative indicator.
On the regional level, the inclusion of the indicator GDP per capita, as a key
economic indicator, should be considered.
3) Alternative weighting methods should be considered. Furthermore, uncertainty
analysis and sensitivity analysis should be conducted in order to improve the
transparency and the quality of the development index.
4. Interval estimations of the development index should be provided prior to the
categorization of LGUs and RGUs. LGUs/RGUs with the interval estimates
overlapping partially with one of the supported groups should be more closely
analyzed in order to avoid incorrect categorization.
5) An inclusion/exclusion of marginal units in/from supported areas should be
considered. This can be achieved by using Monte Carlo simulations and estimating the probability of (incorrect) categorization (see section 4.4).
5.1) In particular, regarding simulation results, LGUs listed in table A1
should be considered for inclusion into group II, i.e. into supported areas. Table A also provides values of the development index and the
percentage of the categorization in groups II and III from MC simulations S (see section 4.4.1).
5.2) Furthermore, LGUs listed in table A2 should be considered for inclusion into group III, i.e. exclusion from supported areas. Table A2 also
provides the development index values and the percentage of the categorization in groups II and III from MC simulations S (see section
4.4.1).
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
2) The outliers should be excluded when using the min-max normalization. As an
alternative normalization method we propose normalization relative to the
state’s index value (given with formula 13).
232
Table a1
Proposal for the inclusion of LGUs into supported areas
lgU
financial theory and
practice
39 (2) 205-236 (2015)
Ozalj
Development
index
0.7508
Categorization into group (%)
II
III
0.876
0.124
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
Donji Vidovec
0.7517
0.902
0.098
Jesenje
0.7522
0.860
0.140
Sveti Martin na Muri
0.7525
0.760
0.240
Lećevica
0.7537
0.980
0.020
Petrijanec
0.7543
0.788
0.212
Donji Miholjac
0.7612
0.546
0.454
Krašić
0.7624
0.942
0.058
Klenovnik
0.7671
0.742
0.258
Vratišinec
0.7681
0.676
0.324
Hrašćina
0.7683
0.862
0.138
Pregrada
0.7688
0.872
0.128
Đelekovec
0.7689
0.926
0.074
Lepoglava
0.7695
0.584
0.416
Kraljevec na Sutli
0.7700
0.802
0.198
Mače
0.7728
0.766
0.234
Jalžabet
0.7775
0.550
0.450
Đurmanec
0.7829
0.570
0.430
Breznički Hum
0.7856
0.756
0.244
Bedenica
0.7901
0.562
0.438
Veliki Bukovec
0.8231
0.526
0.474
Table a2
Proposal for the exclusion of LGUs from the supported areas
lgU
Vrsi
Imotski
Nova Gradiška
Beli Manastira
a
Development
index
0.728
0.735
0.741
0.743
Categorization into group (%)
II
III
0.176
0.824
0.206
0.794
0.456
0.544
0.392
0.608
AASC, not categorized into supported areas on the basis of unadjusted development index.
6) Although this study did not involve the analysis of alternative aggregation
methods, this would also be worth pursuing. In particular, one could consider
the geometric aggregation as it delimitates the compensability between variables.
RefeReNCeS
233
financial theory and
practice
39 (2) 205-236 (2015)
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
1. Bahovec, V. and Erjavec, N., 2009. Uvod u ekonometrijsku analizu. Zagreb:
Element.
2. Bajo, A. and Primorac, M., 2013. “Package of legislative amendments for a
better Croatian regional policy”. Press Relase, No. 61. doi: 10.3326/
pr.2013.6
3. Ben Gal, I., 2005. Outlier detection. In: O. Maimon and L. Rockach, eds. Data
mining Knowledge Discovery Handbook: A Complete Guide for Practitioners
and Researchers. Kluwer Academic Publishers. doi: 10.1007/0-387-25465X_7
4. Booysen, F., 2002. “An overview and evaluation of composite indices of development”. Social indicators research, 59 (2), pp. 115-151. doi: 10.1023/
A:1016275505152
5. Bronić, M., 2008. “The Croatian Ministry of Finance’s Current Grants to Municipalities and Cities in the Areas of Special National Concern”. Newsletter,
No. 35 [pdf]. Available at: <http://www.ijf.hr/eng/newsletter/35.pdf>.
6. Bronić, M., 2010. “Evaluating the current equalization grant to counties in
Croatia”. Financial Theory and Practice, 34 (1), pp 25-52 [pdf]. Available at:
<http://fintp.hr/upload/files/ftp/2010/1/bronic.pdf>.
7. Cziraky, D. [et al.], 2003. An econometric model for development level assessment with an application to municipality development classification. In:
Z. Reić, ed. Enterprise in transition: proceedings of the 5th Conference on
Enterprise in transition. Split: Faculty of Economics.
8. Cziraky, D. [et al.], 2005. Regional development assessment: A structural
equation approach. European Journal of Operational Research, 174 (1), pp.
427-442. doi: 10.1016/j.ejor.2005.03.012
9. Foa, R. and Tanner, J. C., 2012. “Methodology of the Indices of Social Development“. Institute of Social Studies, Working Paper Series [pdf]. Available at:
<http://www.indsocdev.org/resources/Methodology%20of%20the%20Social
%20Development%20Indices_%20jan11.pdf>.
10. Gall, M., 2007. Indices of social vulnerability to natural hazards: A comparative evaluation. Doctoral thesis. University of South Carolina.
11. Grčić, B. and Filipić, P., 2002. “Evaluation of regional development and
(im)balances: the Croatian case”. The Journal of Management Sciences &
Regional Development, 12 (7), pp. 21-36.
12. Gujarati, D. N., 2004. Basic econometrics. New York: McGraw-Hill Companies.
13. Hair, J. F. [et al.], 1995. Multivariate data analysis: with readings. New Jersey: Prentice Hall.
14. Hoyland, B., Moene, K. and Willumsen, F., 2012. “The tyranny of international index rankings”. Journal of Development economics, 97 (1), pp. 1-14.
doi: 10.1016/j.jdeveco.2011.01.007
234
financial theory and
practice
39 (2) 205-236 (2015)
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
15. Maleković, S., Puljiz, J. and Tišma, S., 2011. “New opportunities for regional
and local actors in Croatia in supporting socio-economic development”.
Southeastern Europe, 35, pp.168-190. doi: 10.1163/187633311X585360
16. MRRFEU, 2010a. Vrijednosti indeksa razvijenosti i pokazatelja za izračun
indeksa razvijenosti na lokalnoj razini [pdf]. Available at: <http://www.mrrfeu.hr/default.aspx?id=405>. Zagreb: Ministry of Regional Development and
EU Funds.
17. MRRFEU, 2010b. Vrijednosti indeksa razvijenosti i pokazatelja za izračun
indeksa razvijenosti na županijskoj razini [pdf]. Available at: <http://www.
mrrfeu.hr/default.aspx?id=405>. Zagreb: Ministry of Regional Development
and EU Funds.
18. MRRFEU, 2012. Strateški plan za razdoblje 2012.-2014. Zagreb: Ministry of
Regional Development and EU Funds.
19. MRRFEU, 2013a. Vrijednosti indeksa razvijenosti i pokazatelja za izračun
indeksa razvijenosti na lokalnoj razini [pdf]. Available at: <http://www.mrrfeu.hr/default.aspx?id=405>. Zagreb: Ministry of Regional Development and
EU Funds.
20. MRRFEU, 2013b. Vrijednosti indeksa razvijenosti i pokazatelja za izračun
indeksa razvijenosti na županijskoj razini [pdf]. Available at: <http://www.
mrrfeu.hr/default.aspx?id=405>. Zagreb: Ministry of Regional Development
and EU Funds.
21. MRRFEU, 2012. Strateški plan za razdoblje 2012.-2014. Zagreb: Ministry of
Regional Development and EU Funds.
22. MRRFEU, 2013c. Glavne novine koje donosi Nacrt prijedloga Zakona o regionalnom razvoju Republike Hrvatske [online]. Available at: <www.mrrfeu.
hr/default.aspx?id=1674>. Zagreb: Ministry of Regional Development and
EU Funds.
23. Nardo, M. [et al.], 2005. Tools for Composite Indicators Building. Ispra: Joint
Research Centre.
24. Nardo, M. [et al.], 2008. Handbook on constructing composite indicators
[pdf]. Available at: <http://www.oecd.org/std/42495745.pdf>.
25. Noorbakhsh, F., 1998. “The human Development index: Some technical issues and alternative indices”. Journal of International Development, 10 (5),
pp. 589-605. doi: 10.1002/(SICI)1099-1328(199807/08)10:5<589::AIDJID484>3.0.CO;2-S
26. Oreskes, N., Shrader-Frechette, K. and Belitz, K., 1994. “Verification, validation and confirmation of numerical models in the earth sciences”. Science, 263
(5147), pp. 641-646. doi: 10.1126/science.263.5147.64
27. Ott, K. and Bajo, A., 2001. “Lokalne financije i lokalni proračuni u Republici
Hrvatskoj”. Financijska teorija i praksa, 25 (3), pp. 311-434.
28. Ott, K., 2009. Proračunski vodič za građane. Zagreb: Institut za javne financije.
235
financial theory and
practice
39 (2) 205-236 (2015)
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
29. Paruolo, P., Saisana, M. and Saltelli, A., 2013. “Ratings and rankings: voodoo
or science?”. Journal of the Royal Statistical Society: Series A (Statistics in
Society), 176: pp 609-634. doi: 10.1111/j.1467985X.2012.01059.x
30. Perišić, A., 2014. “Multivariate Classification of Local and Regional Government Units According to Socio-Economic Development”. Društvena
istraživanja, 23 (2), pp. 211-231. doi: 10.5559/di.23.2.0
31. Puljiz, J. [et al.], 2005. Novi pristup ocjenjivanju i kategorizaciji teritorijalnih
jedinica prema razvijenosti. Zagreb: Institut za međunarodne odnose.
32. Puljiz, J., 2007. Proposal for new development categorisation on regional and
local units in Croatia. Croatian international relations review, 12 (44/45), pp.
125-138.
33. Puljiz, J., 2009. Čimbenici regionalnog razvoja i regionalnih nejednakosti u
Republici Hrvatskoj. Doctoral thesis. Split: Ekonomski fakultet.
34. Rimac, I., Rihtar, S. and Oliveira Roca, M. 1992. “Multivarijatna klasifikacija
općina Hrvatske kao moguća metoda regionalizacije Republike”. Društvena
istraživanja, 1 (1), pp. 87-99.
35. Rosen, R., 1991. Life Itself: A Comprehensive Inquiry into Nature, Origin,
and Fabrication of Life. Columbia University Press.
36. Saisana, M. and Saltelli, A., 2008. Expert Panel Opinion and Global Sensitivity Analysis for Composite Indicators. In: F. Graziani, ed. Computational
Methods in Transport: Verification and Validation. Springer Berlin Heidelberg, pp. 251.-275. doi: 10.1007/978-3-540-77362-7_1
37. Saisana, M., Saltelli, A. and Tarantola, S., 2005. “Uncertainty and sensitivity
analysis techniques as tools for the quality assessment of composite indicators”. Journal of the Royal Statistical Society – A, 168 (2), pp. 307-323.
38. Saltelli, A. [et al.], 2004. Sensitivity Analysis in Practice: A Guide to Assessing Scientific Models. New York: John Wiley & Sons.
39. Saltelli, A. [et al.], 2008. Global Sensitivity Analysis. The Primer. New York:
John Wiley & Sons.
40. Salzman J., 2003. Methodological choices encounterd in the construction of
composite indices of economic and social well-being [pdf]. Available at:
<http://www.csls.ca/events/cea2003/salzman-typol-cea2003.pdf>.
41. Sharpe A., 2004. Literature Review of Frameworks for Macro-indicators.
Ottawa: Centre for the Study of Living Standards.
42. Sumpor, M., Jurlina Alibegović, D. and Đokić, I., 2012. Ocjena sustava
strateškog planiranja i mogućnosti financiranja razvoja županija i lokalnih
jedinica u kontekstu provođenja politike regionalnog razvoja Republike Hrvatske. Zagreb: Ekonomski institut Zagreb.
43. Tarantola, S., Jesinghaus, J. and Puolamaa, M., 2000. Global sensitivity analysis: a quality assurance tool in environmental policy modelling. In: A. Saltelli,
K. Chan and M. Scott, eds. Sensitivity analysis. New York: John Wiley &
Sons, pp. 385.-397.
236
financial theory and
practice
39 (2) 205-236 (2015)
ana perišić, vanja wagner:
development index: analysis of the basic instrument of croatian regional policy
44. Toffalis, C., 2013. “An automatic-democratic approach to weight setting for
the new human Development index”. Journal Population Economy, 26 (4),
pp. 1325-1345. doi: 10.1007/s00148-012-0432-x
45. United Nations, 1969. Social policy and planning in national development.
Report of the Meeting of Experts on Social Policy and Planning. Geneva:
United Nations Economic and Social Council Commision for Social Development.
46. Uredba o indeksu razvijenosti, NN 63/10. Zagreb: Narodne novine.
47. Uredba o izmjeni i dopunama Uredbe o indeksu razvijenosti, NN 158/2013.
Zagreb: Narodne novine.
48. Zakon o brdsko-planinskim područjima, NN 12/02, 32/02, 117/03, 42/05,
90/05, 80/08 and 148/13. Zagreb: Narodne novine.
49. Zakon o regionalnom razvoju, NN 153/09. Zagreb: Narodne novine.
O zdravstvu iz ekonomske
perspektive
Health care from an economic perspective
MAJA VEHOVEC (Ed.)
The Institute of Economics, Zagreb, 2014, pp. 356
Book review by ANTO BAJO*1
doi: 10.3326/fintp.39.2.5
*
Received: December 10, 2014
Accepted: December 11, 2014
Anto BAJO, PhD
Institute of Public Finance, Smičiklasova 21, 10000 Zagreb, Croatia
e-mail: [email protected]
238
financial theory and
practice
39 (2) 237-243 (2015)
The book “Health care from an economic perspective” is the result of a two-year
effort by a team of economists from the Institute of Economics and their external
associates. The book, edited by Professor Maja Vehovec, PhD, includes contributions of thirteen researchers. It has 356 pages, including a preface, the authors’
CVs, 56 tables and 92 figures. The nice cover illustration suggests the conclusions
of the book regarding the need of economy, prudent behaviour and more effective
healthcare system management. The book is divided into five parts featuring thirteen thematic contributions, which attempt to answer five key questions regarding
the Croatian health system economics.
maja vehovec (ed.):
health care from an economic perspective
Answers to the first question: How is health care financed and what are the expectations for the future? are offered by Dubravko Mihaljek in his article “How
to Finance Health Care in Times of Economic Crisis?” and Tanja Broz and Sandra
Švaljek in the “Healthcare Financing in Croatia: from Reform to Reform”. Mihaljek points out that while it is impossible to stop the growth of health care spending in the long run, healthcare financing can be made more effective. Since 1960,
life expectancy has been extended by 11 years, to almost 80 years, which will affect cost control and efficiency improvement. Politicians have ignored this fact due
to a short-term thinking horizon. From 2000 to 2010 alone, health care spending in
advanced economies trended upwards from 9.9% to 12.4% of GDP. The share of
public spending rose, while the share of private spending declined. The role of
private health insurance schemes strengthened, but personal health care expenses
decreased. Health care spending in Croatia stagnated over the last ten years (at
7.8% of GDP). However, government accounted for an above-average share in
health care expenditures, as high as 85% (as compared with 72% in Central Europe
and 73% in Western Europe). The government budget share increased, while the
share of health insurance dropped to 67% in 2010. There are many unknowns regarding the structure of the Croatian health care financing: discrepancies exist between domestic statistics and those published by the World Health Organisation.
Mihaljek deems that Croatia should take macroeconomic and microeconomic measures to curb the growth of government health care spending. Such measures have
already been implemented in developed economies and they could be beneficial to
Croatia as well. At the macroeconomic level, budget constraints and stricter quantity and price controls for health care services should be imposed. Microeconomic
measures should include: partial restructuring of the health care system, increasing
transparency in the health care billing practices (the payment of services by diagnosis-related groups, a wider choice of doctors and medical institutions when contracting with suppliers for health care services) and redressing the imbalance between primary and secondary health protection. Mihaljek also stresses the need for
greater private sector participation in supplying goods to medical institutions, as
well as for stronger competition among insurance companies.
Broz and Švaljek analyse the implemented health care system reforms, providing
information for evaluating their effects. There have been numerous rehabilitations
maja vehovec (ed.):
health care from an economic perspective
Replies to the second question: What ought we to know about health insurance
and private health expenditure? are given by Danijel Nestić and Ivica Rubil in
the article “Private Health Expenditure in Croatia”, and Mario Puljiz in the contribution titled: “Voluntary Health Insurance: Croatia vs. Europe”. Nestić and Rubil
argue that total private health expenditure in Croatia (about 1.2% of GDP) is extremely low and that little is known about its socio-economic characteristics. The
information they used was based on a Household Budget Survey. Private expenditures account for about 15% of total health expenditures. Private insurance expenses account for just a small share in total private health expenditures in Croatia
(0.5% of GDP). Hospitalisation costs are mainly covered from public sources. A
survey on a sample of Croatian households corroborates the previously known
findings of similar international surveys. In addition to income levels, the amount
of private health expenditures is influenced by the level of education (high vs.
extremely low educational levels), age, gender (women spend more on average
for health) and settlement type (the urban population spends more on health).
Propensity to buy supplemental insurance depends on gender (women are more
prone to it), educational level (highly educated persons are more likely to buy
supplemental insurance) and income (persons with higher income are more likely
to opt for it). There are no plans in Croatia for any major health care system restructuring in terms of the sources of funding, and it is mainly considered in the
context of the government budget. Puljiz points out that in Croatia, private insurance cannot be a substitute for compulsory insurance. It is necessary to define the
basket of services covered by the public insurance and that provided under the
239
financial theory and
practice
39 (2) 237-243 (2015)
in the health care area (mostly involving hospitals) in Croatia. From 1994 to 2007
alone, their costs reach about HRK 9.7bn. Apart from the rehabilitation processes,
health care co-payments have been increased and supplemental health insurance
has been introduced. However, despite the applied measures, medical institutions
continued to accumulate losses. In 2012 and 2013 alone, expenditures for the settlement of debts incurred by the Croatian Health Insurance Fund (HZZO) and hospitals amounted to HRK 6.5bn. Since 2004, the prices of drugs have gone up. Regrettably, the numerous reforms were not based on any elaborate strategy documents. The reform launched in 2008 was aimed at curbing public expenditures and
the growth of drug prices by introducing international public tenders for particularly expensive drugs, and reducing benchmark prices of prescription drugs. Moreover, medical equipment procurement has been centralised, and the control of sickness benefits and hospital payments according to diagnosis and therapy-related
groups has been stepped up. Under the reform, all additional sources of the health
care system funding have been tapped. Croatia needs further health insurance market liberalisation and cuts in the number of those insured at the expense of the
government budget. While the bulk of reforms are still carried out at the revenue
side, expenditure reforms are also needed, along with an increase in the number of
private health insurance schemes, given the exhausted public revenue capacities. A
similar conclusion arises from an analysis of private health care expenses.
240
financial theory and
practice
39 (2) 237-243 (2015)
supplemental insurance. Voluntary health insurance accounts for as little as 2.44%
of the total written premium of all insurance companies, which is indicative of an
underdeveloped market. Further liberalisation is called for, as well as cuts in the
number of budget-financed health care users. The use of supplemental and additional health insurance products by citizens could be encouraged by introducing
various forms of tax relief.
maja vehovec (ed.):
health care from an economic perspective
The third question: What are the key elements of health care spending and
what are their specific qualities? is answered by Ivana Rašić Bakarić in the
chapter “Primary Health Protection in between Efficiency and Availability”, and a
group of co-authors (Maja Vehovec, Ivana Rašić Bakarić and Sunčana Sljepčević)
who point to the key problem of the Croatian health care system in the article
“Challenges of Hospital Restructuring”. Sunčana Sljepčević provides an “Evaluation of the Technical Efficiency of Hospitals”, and Tanja Broz analyses “Drug
Consumption and the Specifics of Functioning of the Drug Market”.
According to Rašić Bakarić, the primary health care expenditures have stood at
HRK 2,95bn. Medical centres should be the main primary health protection providers. The problem lies in too few preventive medical examinations, the growing
number of patient referrals to specialists and a relatively large number of patient
visits per day. Every seventh patient is referred to a specialist medical examination. Thus, a general practitioner has 53.1 patient visits a day, and he/she spends 8
minutes on each of them. The basket of services included in the general/family
medicine is not defined, which leads to differences in financial calculations. In
most EU member states, primary health care handles 70-80% of health cases (as
compared to only 50% in Croatia). Referring from primary to specialist health
services is due to weaknesses in the current funding model which is not based on
any standards.
Vehovec, Rašić Bakarić and Sljepčević explain an old malady of the Croatian
health care system, i.e. the non-transparency of costs, especially in hospitals
which account for the largest share of total health expenditures, exceeding the EU
member states’ average. The ratio between the number of hospital doctors and the
number of population is unsatisfactory and is below the EU average (163 doctors
per 100,000 inhabitants). The same is true for the numbers of hospital beds (56 per
10,000 inhabitants) and hospitals (1.3 per 100,000 inhabitants). Croatia is the
leader among the EU member states in terms of duration of treatment (7.2 days).
Below-average coverage of doctors has been observed in as many as 14 counties,
with unequal access to medical services. Hospital expenditures stand at about
HRK 11.3bn. The bulk of them (53%) are expenditures for employees, followed
by material expenditures (for drugs and hospital energy consumption). One of the
key challenges is the status of the HZZO, which only transfers budget funds to
hospitals, without having any autonomy in deciding on the health system’s insurance policy based on the funds raised from compulsory health insurance contribu-
Broz analyses “Drug Consumption and the Functioning of the Drug Market”, concluding that spending on drugs in Croatia exceeds GDP capacity. The drug market
has no detailed and transparent statistical databases. Public spending on drugs
accounts for about 80% of the total spending, which ranks Croatia among the EU
member states with above-average shares of public spending on drugs in total
health care spending. Drug expenses went up from HRK 3.8bn in 2004 to HRK
5.1bn in 2013.
Against the backdrop of uncontrolled consumption and non-existent standards, an
important question arises, namely How far are users involved in health care
service evaluation? The answers are provided by Jelena Budak and Edo Rajh in
an interesting part titled “Corruption in Croatia’s Health Sector: Myth or Reality”.
Jelena Budak’s article deals with “Patients’ Assessment of the Quality of Work of
maja vehovec (ed.):
health care from an economic perspective
In the part “The Technical Efficiency Evaluation of Hospitals“, Sljepčević analyses the efficiency of 54 hospitals using the 2010 data. Some of her findings are
very interesting. Most of the specialised and clinical hospitals employ aboveaverage numbers of administrative and technical staff. The number of nurses
grows, but is still below the EU average. Data on the number of discharged patients per hospital and on the rate of cured patients are unavailable, although they
might be useful for performance analysis and hospital efficiency measurement.
According to the author’s calculation, seven hospitals (a clinical hospital, three
general and three specialised hospitals) operated at marginal efficiency in 2010.
As much as 9.8% of these hospitals use 40% to 50% more inputs than the efficient
hospitals, and 3.9% hospitals are capable of cutting the user of resources by 30%
to 40%, which would represent considerable savings within the system. Four clinical hospital centres show different levels of efficiency, despite their similar business characteristics. The same is true for other types of hospitals (clinics, clinical
hospitals, general and specialised hospitals).
241
financial theory and
practice
39 (2) 237-243 (2015)
tions. The general impression (confirmed by an interview) is that hospital managers know the solutions and even offer recommendations for reforms. The question
remains, however, how to turn proposals into practice and does the medical profession actually want reforms. Hospital managers generally have a negative attitude towards the outsourcing of non-medical activities and do not feel responsible
for salary and material cost overruns. They rather consider them a problem of the
system. The relation between hospital revenues and the cost presentation model is
non-transparent. Limits make up the bulk of hospitals’ budgets, but it is not clear
how they are determined: on the basis of a historical cost model or some additional unknown criteria. So far, hospitals have been known for good doctors rather
than successful management. Hospital financing should depend on performance
indicators. It is necessary to improve the computerisation of hospitals, prescribe
the appropriate standards, procedures and protocols and adopt strategic plans for
each medical institution.
242
Health Care Personnel”. Maja Vehovec writes about “An International Comparison of Health Care Quality from a User Perspective”.
financial theory and
practice
39 (2) 237-243 (2015)
Budak and Rajh hold that corruption exists and it results in unequal treatment of
patients. Corruption is more prevalent in hospitals, due to their inefficiency (long
waiting lists) and poor health system organisation. The most easily enforceable
measure to combat corruption in the patient-doctor relationship is providing patients with transparent information on public health care. The knowledge of patients’ rights, especially to the information on the scope of services covered by
health insurance and the official prices of health services, helps reduce informal
costs. The privatisation of health care, breaking up a monopoly in the health services market, may be an effective step towards rooting out corruption. The key
purpose of the research is to raise awareness of the problem.
maja vehovec (ed.):
health care from an economic perspective
Budak emphasizes that there are no public opinion polls in Croatia reflecting citizens’ opinions on the quality of the public health system. Useful tools in this regard are surveys on the public perception of the quality of doctors’ work.
Vehovec has interpreted the results of the European Health Care Index, making an
international comparison of the health care quality from a user perspective. According to this index, Croatia is in the middle of the list of countries by health care
quality rankings assigned by users.
Is economic evaluation in health care an option or a necessity? Answers to this
questions have been provided by Ana Bobinac in the article “An Introduction to
Economic Evaluation in Health Care” and Dubravka Jurlina Alibegović, who discusses “The Role of Public-Private Partnership in the Economic Assessment of
the Rational Use of Resources in Health Care”. Economic evaluations (cost-effectiveness analysis, cost-benefit analysis and cost and benefit analysis) are tools to
improve the system’s efficiency and are necessary for carrying out the reforms. Of
course, taking decisions on the implementation of evaluations and reforms continues to be the responsibility of politics. A useful means to improve efficiency could
also be public and private partnership – a model that is yet to find widespread
application in the health care system.
The authors of the book have recognised the key macroeconomic challenges facing the health care system in Croatia, and opened the way to dealing with macroeconomic issues in order to review and improve efficiency. The main topics for
future research into the health care economics in Croatia are the following: (a)
evaluation of the success of hospital rehabilitation; (b) efficiency analyses of the
completed decentralisation; (c) the manner of determining minimum financial
standards; (d) asset management and valuation; (e) amount, structure and maturities of liabilities; (f) capital investments (an analysis for the period ending 2015,
and plans by 2020); (g) financial analysis of the operations of medical centres,
pharmacies and hospitals (by name and type of institution); and (h) organisation
and practice of the financial management of hospitals.
financial theory and
practice
39 (2) 237-243 (2015)
This book is, without doubt, a valuable piece of work which is yet to find its
proper place in expenditure analyses. It could also serve as a practical guide in
planning health care system reforms. An examination and review of efficiency can
certainly increase accountability in the health care system. Croatia should step up
the role of economy in health care and have more health professionals who are
skilled in economics.
243
maja vehovec (ed.):
health care from an economic perspective
Publisher
Institute of Public Finance, Smičiklasova 21, Zagreb, Croatia
Editor-in-Chief
Katarina Ott
Production Editor
Marina Nekić
Editorial Board (Institute of Public Finance)
Marijana Bađun
Anto Bajo
Predrag Bejaković
Vjekoslav Bratić
Mihaela Bronić
Martina Fabris
Katarina Ott
Ivica Urban
Goran Vukšić
Editorial Advisory Board
Hrvoje Arbutina (Faculty of Law, Zagreb, Croatia)
Will Bartlett (London School of Economics and Political Science, London, UK)
Helena Blažić (Faculty of Economics, Rijeka, Croatia)
Balázs Égert (OECD, Paris, France)
Edgar L. Feige (Professor of Economics Emeritus at the University of Wisconsin, Madison,
Wisconsin, USA)
Božidar Jelčić (Faculty of Law, Zagreb, Croatia)
Evan Kraft (American University, Washington D.C., USA)
Peter J. Lambert (University of Oregon, Department of Economics, Eugene, USA)
Olivera Lončarić-Horvat (Faculty of Law, Zagreb, Croatia)
Dubravko Mihaljek (Bank for International Settlements, Basel, Switzerland)
Peter Sanfey (European Bank for Reconstruction and Development, London, UK)
Bruno Schönfelder (Technical University Bergakademie Freiberg, Faculty of Economics
and Business Administration, Freiberg, Germany)
Miroslav Verbič (Faculty of Economics / Institute for Economic Research, Ljubljana, Slovenia)
Athanasios Vamvakidis (Bank of America Merrill Lynch, London, UK)
Hrvoje Šimović (Faculty of Economics and Business, Zagreb, Croatia)
Financial Theory and Practice is abstracted and indexed in:
DOAJ (Directory of Open Access Journals, Lund University, Sweden)
EBSCO Publishing Database
EconLit (American Economic Association’s electronic database), JEL (Journal of Economic
Literature, Pittsburgh, Pennsylvania, USA)
HRČAK (Portal of Scientific Journals of Croatia)
IBSS (International Bibliography of the Social Sciences, ProQuest, Cambridge, UK)
RePEc (Research Papers in Economics)
Editorial Office
Institute of Public Finance – Financial Theory and Practice
Smičiklasova 21, Zagreb, Croatia, P.O. BOX 320
phone: +385 (0)1 4886 444; 4819 363; fax: +385 (0)1 4819 365
web-site: www.fintp.hr; e-mail: [email protected]
Subscription
Annual subscription amounts 400 kuna
Payments to account no. HR7024840081100661775, Institut za javne financije, Zagreb;
quoting: subscription to Financial Theory and Practice, 2015
Printed in 115 copies
The journal comes out four times a year
The journal is co-financed by the Ministry of Science, Education and Sport of the Republic
of Croatia
Computer typesetting and printing
Denona d.o.o., Zagreb, Marina Getaldića 1
2/2015
39 (2) 2015
ARTUR ŚWISTAK, SEBASTIAN WAWRZAK
and AGNIESZKA ALIŃSKA
In pursuit of tax equity: lessons from VAT rate
structure adjustment in Poland
ANA KUNDID NOVOKMET
Cyclicality of bank capital buffers in South-Eastern
Europe: endogenous and exogenous aspects
MIRNA DUMIČIĆ
Financial stress indicators for small, open, highly
euroized countries: the case of Croatia
ANA PERIŠIĆ and VANJA WAGNER
Development index: analysis of the basic instrument
of Croatian regional policy
Vol. 39, No. 2 | pp. 115-243
June 2015 | Zagreb
udc 336
issn 1846-887x