Download Report

Loans and the transmission of monetary policy shocks: A
VAR analysis of the role of commitments.
Andrea Civelli
Nicola Zaniboniy
University of Arkansas
Transparent Value, LLC
Abstract
This paper uses VAR analysis to illustrate that bank loans under commitment behave differently than loans not under commitment in response to a monetary shock. We …nd that …rms
use commitments more intensively after a monetary tightening and argue this helps explain the
puzzling response of C&I loans documented in the monetary VAR literature.
JEL Classi…cation: E50, E51, E44
Keywords: Monetary Policy Transmission, Loan Commitments, Credit Channel
1
Introduction
The basic lending channel of the transmission mechanism of monetary policy predicts a contraction
of loan supply in response to tightening monetary conditions. den Haan, Sumner, and Yamashiro
(2007) (DSY henceforth) …nd this prediction is empirically con…rmed for real estate and consumer
loans, but it is strongly rejected for commercial and industrial (C&I) loans, which increase following
a contractionary shock. In this paper we exploit a contractual feature of C&I loans, namely whether
or not they are issued under commitment, to shed some light on this empirical …nding. To this end,
we use a relatively under-explored dataset, derived from the Survey of Terms of Business Lending
(STBL). We adopt the same empirical framework as DSY, and identify monetary shocks in a VAR
model using the strategy proposed by Christiano, Eichenbaum, and Evans (1999) (CEE hereafter).
Our …ndings suggest that the response of the C&I loans is due to demand-side e¤ects, in large
part explained by a more intensive utilization of commitments by …rms that have access to unused
credit lines, in spite of tighter credit conditions. At the same time, the relative response of spot
contract loans and loans under commitment allows us to identify a supply-side credit rationing that
is consistent with the implications of the lending channel.
Corresponding author: University of Arkansas, Business Building 402, Fayetteville, AR 72701. Email: [email protected]. URL: http://comp.uark.edu/ acivelli/.
y
Transparent Value, LLC, New York, NY 10017. The views expressed here are those of the author and not
necessarily those of Guggenheim Partners or its subsidiaries.
1
The literature has several examples of studies that compare responses of di¤erent …nancial aggregates to better understand the monetary transmission mechanism by disentangling supply-side
from demand-side e¤ects. Kashyap, Stein, and Wilcox (1993) argue they can identify a loan-supply
channel by analyzing shifts in …rms’ …nancing mix from bank loans to commercial paper. More
recently, motivated by the events of the latest …nancial crisis, Becker and Ivashina (2014) use the
substitution between loans and bonds at …rm-level; in a similar context, Adrian, Colla, and Shin
(2013) study the shift in composition of credit between loans and bonds. Our analysis focuses on
bank loans and uses a sample that predates the …nancial crisis of 2008-09. More closely related to our
work, So…anos, Melnik, and Wachtel (1990) and Morgan (1998) exploit the di¤erences between loans
under commitment and spot loans to explore the monetary transmission mechanism, but our data
source di¤ers from theirs. Berger and Udell (1992) conduct an exhaustive study of the importance
of credit rationing deriving pre-1988 commitments data at the bank-level from the STBL.
There is a more recent literature that focuses on commitments. Campello, Giambona, Graham,
and Harvey (2011) extensively describe credit lines management in relation to external funds and
corporate decisions during the …nancial crisis. Black and Rosen (2007) use information on commitments from STBL at the individual-loan level to identify separate bank-lending and balance-sheet
channels (as codi…ed, for example, in Bernanke and Gertler, 1995). Duca and Vanhoose (1990) and
Woodford (1996) study optimal monetary policy in presence of lending commitments; Demiroglu,
James, and Kizilaslan (2012) examine the relation between changes in bank lending standards and
availability and use of credit lines for public and private …rms. While these papers highlight the key
role of credit lines in insulating and protecting …rms from monetary tightening and credit crunches,
they do not explicitly analyze the response of loans to structural monetary shocks.
Finally, we view our results as complementary to those in DSY: their work stresses the importance
of looking at di¤erent categories of loans to properly account for their aggregate behavior in response
to monetary shocks. Our analysis provides a potential explanation for their …ndings on the dynamics
of one such category, namely the C&I loans.
2
Empirical Methodology
Let Yt indicate the vector of macroeconomic and …nancial variables of interest in the analysis. As is
standard in the monetary literature (see DSY for a recent application), we adopt a pth order VAR
to model the reduced-form dynamics of Yt
Yt =
p
P
Bi Yt
i
+ "t
i=1
where the VAR residuals "t have covariance matrix Et ("t "0t ) = . The relation between the reducedform residuals of the VAR, "t , and the fundamental structural innovations of the model, ut , is assumed
2
to be linear
"t = A0 ut
where the structural shocks are orthogonal and Et (ut u0t ) = V is diagonal. We follow the framework
of CEE and identify the monetary policy shock by imposing a block-recursive structure to the impact
matrix A0 . First, the variables in Yt are sorted into three blocks
2
3
Xt
6
7
Yt = 4 S t 5
Zt
where Xt is an n1 -vector of macroeconomic aggregates, St represent the policy instrument of the Fed
(n2 = 1), and Zt is an n3 -vector of monetary and …nancial variables that inform the policy decision.
Second, the macroeconomic variables are assumed to respond with a lag to the other variables of the
model, including St . At the same time, the contemporaneous values of Xt , but not those of Zt , are
assumed to be part of the information set of the Fed. Correspondingly, the impact matrix is blocktriangular
2
3
A11 012 013
6
7
A0 = 4 A21 A22 023 5
A31 A32 A33
where Aij and 0ij are matrices of parameters and zeros, respectively, with dimensions ni nj . The
results presented in Section 4 are based on a Bayesian VAR model estimated with Minnesota priors.
We estimate a VAR with four lags (p = 4) using quarterly data over a sample that starts with
Paul Volcker’s chairmanship at the Fed and ends just before the onset of the …nancial crisis of
2008 (1979:1-2008:2). We will say more about the sample choice in the next section. The impulse
response functions are then reported for the lower triangular normalization of A0 . The results are
then invariant to the ordering of the variables within Xt and Zt , since we are only interested in
identifying the monetary shock (see CEE). What becomes important, then, is the selection of the
variables to be included in the two non-policy blocks of Yt ; we turn to the discussion of this point
next.
3
Variables De…nition and Identi…cation Strategy
The set of variables in Yt for the most part corresponds to those commonly used in the VAR literature
following CEE. Speci…cally, the policy instrument, St , is measured by the federal funds rate (FFR) and
the real macroeconomic block, Xt , includes the logs of real GDP, GDP de‡ator, and a commodity
price index. These variables are obtained from the Federal Reserve Economic Data (FRED), the
online dataset maintained by the Federal Reserve Bank of St. Louis; all variables are seasonally
adjusted.
3
9
x 10
4
commitment
spot
8
real-2005 million dollars
7
6
5
4
3
2
1
0
1980
1985
1990
1995
2000
2005
year
Figure 1: C&I loans under commitment and spot loans.
In addition to the logs of real total bank reserves and non-borrowed reserves (which are, once
again, standard variables in the framework above and obtained from FRED), we include in the
…nancial/monetary block, Zt , variables that aim to better characterize a "bank-lending channel".
Depending on the speci…cation under consideration, this subset of variables comprises a weighted
market loan return rate and either a series for overall loans made by domestic banks or some partition
between loans issued under commitment and not under commitment. As above, we take the logtransformation of these variables in real terms.
A couple of comments are in order here. First, our identi…cation strategy re‡ects the underlying
assumption that credit markets clear immediately after the observation of the policy rate, but their
feedback to the monetary policy decision takes place with a lag. This ordering is consistent with
the theoretical features of a model in which the policy rate is set by the Central Bank according
to a Taylor rule that primarily responds to in‡ation and output gap. It is also consistent with the
VAR speci…cation in Morgan (1998) and the baseline speci…cation of DSY.1 While this assumption
is fair for "normal" economic times, it may be questionable when the economy is subject to large
…nancial shocks and monetary policy promptly reacts to prevent the spreading of liquidity crises, an
obvious case being the latest "Great Recession" of 2008-09. To avoid this issue, we exclude data
after the second quarter of 2008 from the estimation sample. While other periods are potentially
a¤ected by the same problems (for instance, the aftermath of the 2001 stock market crisis), they
are arguably shorter and less dramatic in nature. Furthermore, the use of the federal funds rate as
policy instrument when it is stuck at the zero lower bound (as it has been the case since late 2008) is
not suitable to properly characterize the e¤ects of monetary policy (see, for instance, Keating, Kelly,
and Valcarcel, 2014). Finally, in light of the recent literature that …nds strong evidence of a lending
channel during the …nancial crisis, the exclusion of that period from our sample characterizes our
1
DSY use data at monthly frequency and they assume a baseline ordering where the FFR does not contemporaneously respond to any variable of the model. Their results are robust to the polar opposite speci…cation too.
4
analysis as a study over regular business cycles, and thus complementary to that literature.2
The second comment pertains to a more detailed description of our loan data. We use information
on loans and commitments from the Survey of Terms of Business Lending conducted by the Federal
Reserve Board of Governors. The survey collects data on lending practices for loans issued during
a representative business week in the second month of each quarter, for a sample of about 350
banks since 1977.3 We consider total C&I loans disbursed by domestic commercial banks, for which
the share issued under bank commitments (formal and informal lines of credit) and the weightedaverage e¤ective loan rate are consistently available since the beginning of the survey.4 Loans under
commitment were about 50% of total loans at the beginning of the sample and have substantially
grown to more than 75% in 2008. Figure 1 plots the loan series under the two contractual forms
and suggests that accounting for commitments is important, if we want to correctly characterize a
lending channel for C&I loans.
DSY derive C&I loan series from the Call Reports; these are stock measures of loans in the
banks’ balance sheets, while our series are within-period loan ‡ows. Commitment data from the
STBL aggregated at the bank-level is used in Berger and Udell (1992) for the period 1977-1988. The
conclusion in their work is that credit rationing is not signi…cant in that sample. Black and Rosen
(2007) also rely on the STBL, but they exploit the micro characteristics of the data at individual loan
level. These two studies are based on panel regression models. So…anos, Melnik, and Wachtel (1990)
and Morgan (1998) use a VAR framework to analyze the credit channel. However, their commitment
data comes from the Survey of C&I Loan Commitments at Selected Large Commercial Banks, which
is limited to the sample 1975-1987.
4
Empirical Results
We …rst want to verify that we are able to replicate the puzzling responses of total C&I loans to
monetary shocks in DSY using our data and setup. We report the results in Figure 2 for a one s.d.
contractionary shock. To save space, we only plot the impulse response functions for output, prices,
FFR, loan rate, and total loans. The dashed bands represent the 14=86th percentiles of the posterior
distribution of the responses. Total loans increase on impact and remain signi…cantly positive for a
prolonged period, in line with the …ndings in DSY. The loan rate follows the fed funds rate, but it
exhibits some stickiness at the very start of its response.
2
We did check the robustness of our results and extended the sample to include more recent years. While qualitatively similar, our main results are less clear-cut than in the baseline speci…cation. This is not surprising, given the
structural breaks in the properties of some of the variables that we include in the VAR, such as the federal funds rate,
as well as banks’reserves.
3
Data is collected from participating banks through form FR 2028a, and the aggregate data is published in report
E.2 by the Board.
4
This data is publicly available only starting in 1997; in response to our request, the Board of Governors kindly
assembled the full sample dataset for us. We are in particular grateful to Thomas Spiller and Sam Haltenhof at the
Division of Monetary A¤airs - Banking Analysis of the Board for their outstanding assistance in obtaining the data
when it was collected in 2013.
5
Output
Prices
FFR
0
.006
.001
0
-.002
.003
-.001
-.004
-.002
1
3
5
0
1
Loan Rate
3
5
1
3
5
Total C&I Loans
.004
.024
.016
.002
.008
0
0
1
3
5
1
3
5
Figure 2: Responses to a one s.d. monetary shock - VAR with total loans. Years from the shock.
We then turn to a speci…cation where we replace total loans with loans under commitment,
along with the ratio of spot loans to loans under commitment. Although we do not develop an
explicit model, this speci…cation choice can be justi…ed by a theoretical argument analogous to that
in Kashyap, Stein, and Wilcox (1993), among others. Intuitively, taking common movements across
di¤erent types of loans as re‡ecting demand-side factors, a drop in the ratio of the two types of loans
would suggest the presence of a supply-side credit rationing channel, which a¤ects spot loans but not
loans under commitment. The results for this speci…cation are illustrated in Figure 3. The responses
of the standard variables display only minor changes; one interesting note is that the so-called price
puzzle, which disappeared with the inclusion of a commodity price index in the CEE speci…cation
without loans, resurfaces to some extent here. There is a clear similarity between the response of
total loans in Figure 2 and loans under commitment in Figure 3. This is not surprising since loans
under commitment represent the majority of C&I loans. We then document a signi…cant drop of
the ratio in response to a contractionary monetary shock over a prolonged horizon. We conclude
that, in spite of the existence of a lending channel, the (seemingly) puzzling positive response of C&I
loans to tightening monetary shocks (as found by DSY with more general loan data) is likely to be
explained by demand-side rather than supply-side factors (such as shifts in banks’loan portfolios).
These conclusions and our interpretation of the credit rationing are consistent with previous studies
that use commitment data. In particular, Morgan (1998) strengthens the credit rationing argument
by showing that bank lending standards are concurrently tightened during a monetary restriction,
while Black and Rosen (2007) disentangle the e¤ects of commitments in relation to explicit changes
in the structure of loan supply after monetary slowdowns.5
We consider a few modi…cations of our baseline speci…cation to check for the robustness of the
results. We replace the spot to commitment loans ratio with either the share of spot loans in
total loans or the simple di¤erence between the two types of loans, and we obtain results that are
5
The stickiness displayed by the loan rate is potentially consistent with credit rationing as well, but to test this
implication we would need to analyze the responses of the loan rates for the two categories of loans separately.
Unfortunately, disaggregate data for the two rates could not be provided to us before 1997.
6
Output
Prices
.001
FFR
.006
0
0
-.001
-.001
-.002
-.002
-.003
1
3
5
.003
0
1
Loan Rate
3
5
1
Loans under commitment
3
5
Loans ratio
.004
0
.024
.002
.016
-.002
.008
0
-.004
0
1
3
5
1
3
5
1
3
5
Figure 3: Responses to a one s.d. monetary shock - VAR with loans under commitment. Years from the shock.
qualitatively very similar in both cases. We also include the two types of loans separately. The
response of loans under commitment does not change, while the response of spot loans is positive
but smaller in comparison and not as signi…cant, especially over a longer horizon. These responses
are once again consistent with our baseline speci…cation and the underlying identifying assumption
of credit rationing. We also consider di¤erent orderings of the variables in the VAR. As a main
alternative identi…cation scheme, we move all the loan variables to the Xt block, forcing them to not
respond to the monetary shock on impact. This change does not a¤ect the shape of the response
of total loans (as in DSY) and loans under commitment. The response of the ratio is now positive,
but quite small and not signi…cant for a few periods; it then turns negative after 6 quarters and
marginally signi…cant in the medium term. Finally, the conclusions are robust to changes in the
number of lags in the VAR, as the same qualitative results are obtained with two or six lags.
5
Conclusions
We argue that the puzzling responses of the C&I loans to a tightening monetary shock found by
DSY can more likely be explained by shift in the demand of loans, despite a tightening of the credit
conditions. We exploit the contractual di¤erences between loans under commitment and spot loans
to illustrate two main points: First, these e¤ects are in large part due to …rms drawing down their
commitment credit lines; Second, these e¤ects o¤set an expected supply-side credit rationing that
can be identi…ed comparing the responses of spot loans and loans under commitment. This evidence
provides a valuable explanation of the responses of C&I loans in DSY, reconciling their results with
the basic implications of the bank-lending channel of monetary transmission.
7
References
Adrian, T., P. Colla, and H. S. Shin (2013): “Which Financial Frictions? Parsing the Evidence
from the Financial Crisis of 2007-9,” in NBER Macroeconomics Annual, ed. by D. Acemoglu,
J. Parker, and M. Woodford, vol. 27, pp. 159–214. University of Chicago Press.
Becker, B., and V. Ivashina (2014): “Cyclicality of credit supply: Firm level evidence,”Journal
of Monetary Economics, 62(3), 76–93.
Berger, A. N., and G. F. Udell (1992): “Some Evidence on the Empirical Signi…cance of Credit
Rationing,”Journal of Political Economy, 100(5), 1047–1077.
Bernanke, B. S., and M. Gertler (1995): “Inside the black box: the credit channel of monetary
policy transmission,”Journal of Economics Perspectives, 9(4), 27–48.
Black, L. K., and R. J. Rosen (2007): “How the Credit Channel Works: Di¤erentiating the Bank
Lending Channel and the Balance Sheet Channel,”Federal Reserve Bank of Chicago Working Paper
2007-13.
Campello, M., E. Giambona, J. R. Graham, and C. R. Harvey (2011): “Liquidity Management and Corporate Investment During a Financial Crisis,” Review of Financial Studies, 24(6),
1944–1979.
Christiano, L., M. Eichenbaum, and C. L. Evans (1999): “Monetary policy shocks: what have
we learned and to what end?,”in Handbook of Macroeconomics, ed. by J. Taylor, and M. Woodford.
North-Holland, Amsterdam.
Demiroglu, C., C. James, and A. Kizilaslan (2012): “Bank Lending Standards and Access to
Lines of Credit,”Journal of Money, Credit, and Banking, 44(6), 1065–1089.
den Haan, W. J., S. W. Sumner, and G. M. Yamashiro (2007): “Bank loan portfolios and the
monetary transmission mechanism,”Journal of Monetary Economics, 54(3), 904–924.
Duca, J. V., and D. D. Vanhoose (1990): “Loan Commitments and Optimal Monetary Policy,”
Journal of Money, Credit, and Banking, 22(2), 178–194.
Kashyap, A. K., J. C. Stein, and D. W. Wilcox (1993): “Monetary policy and credit conditions:
evidence from the composition of external …nance,”American Economic Review, 83(1), 78–98.
Keating, J. W., L. J. Kelly, and V. J. Valcarcel (2014): “Solving the price puzzle with an
alternative indicator of monetary policy,”Economics Letters, 124(2), 188–194.
Morgan, D. P. (1998): “The Credit E¤ects of Monetary Policy: Evidence Using Loan Commitments,”Journal of Money, Credit, and Banking, 30(1), 102–118.
Sofianos, G., A. Melnik, and P. Wachtel (1990): “Loan Commitments and Monetary Policy,”
Journal of Banking and Finance, 14(4), 677–689.
Woodford, M. (1996): “Loan commitments and optimal monetary policy,” Journal of Monetary
Economics, 37(3), 573–605.
8