1. finance
2. bank

Credit Risk: Why Less Can Be More In Quantitative Models
Webinar: Wednesday 25th June 2014
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ.
Not for distribution to the public. Copyright © 2014 by Standard & Poor’s Financial Services LLC (S&P). All rights reserved.
Welcome
Silvina Aldeco-Marinez, Managing Director, S&P Capital IQ
2
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Overview Of S&P Capital IQ
S&P Capital IQ is a key contributor to
the growth, transparency, liquidity,
innovation and value creation in the
global capital markets.
We are a leading provider of multiasset class and real-time data,
research and analytics to institutional
investors, investment and commercial
banks, investment advisors and
wealth managers, corporations, public
institutions
and
universities
worldwide.
Our broad suite of capabilities are
designed to help track performance,
generate alpha, identify new trading
and investment ideas, and undertake
risk analysis and mitigation strategies.
Through leading desktop solutions,
enterprise solutions and research
offerings, S&P Capital IQ sharpens
financial intelligence into the wisdom
investors need.
3
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
S&P Capital IQ Organizational Context
McGraw Hill Financial
• Corporate &
Government Ratings
• Structured Finance
• Quality
• Ratings Operations
Separation of Businesses
Credit Ratings &
Research
S&P Dow Jones
Indices
S&P Capital IQ
Standard & Poor’s
Ratings Services
Enterprise
Solutions
• Reference, Ratings1 And
Fixed-Income Data Sets
Research &
Analytics
• Research3
• Valuations2
• Deep Global And North
America Fundamentals
• CUSIP® Global Services
• GICS®
Integrated
Desktop
Solutions
• The Platforms
Benchmark
Solutions
Indices for:
– Equities
– Capital
– ETFs
– TheMarkets.com
• Fixed Income
– Global Credit Portal®
• Real Estate
– MarketScope® Advisor
• Commodities
• Global Markets
Intelligence3
• Leveraged Commentary
& Data
IQ®
• Risk Solutions
• Equity
• Emerging & Frontier
• Strategy
• Thematic
• Custom
1. Provided by Standard & Poor’s Ratings Services, which is analytically and editorially independent from any other group at McGraw Hill Financial.
2. Provided by Standard & Poor’s Securities Evaluations, Inc. (SPSE), a part of S&P Capital IQ and a registered investment adviser with the U.S. Securities and Exchange Commission (SEC). SPSE’s advisory services include
evaluated pricing and model valuation of fixed income securities, derivative valuations, and analyses of certain U.S. and European fixed-income securities using its proprietary Risk-to-Price scoring methodology. Products and
Services provided by SPSE may not be available in all countries or jurisdictions.
3. S&P Capital IQ’s research reports are prepared in the U.S. by Standard & Poor’s Investment Advisory Services LLC (SPIAS), a part of S&P Capital IQ and a registered investment adviser with the SEC. Research reports are
issued in the U.S. by Standard & Poor’s Financial Services LLC; in the UK by McGraw-Hill Financial Research Europe Limited, which is authorized and regulated by the Financial Conduct Authority; and by affiliates of SPIAS
registered or licensed in certain other jurisdictions. SPIAS also provides non-discretionary investment advice in the U.S.
McGraw Hill Financial includes other lines of businesses that are not included in the above graphic.
4
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
CREDIT RISK DATA – WHY LESS CAN BE MORE
Giorgio Baldassarri, Global Head, Analytical Development Group, S&P Capital IQ
5
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Topics
• The Eternal Dilemma of Risk: Quantity vs. Quality
• The 4 M’s in “Credit Risk”: Modeling, Measuring, Monitoring (and Managing)
• Striking a balance between quality and quantity in Credit Risk
• When Less Can be More:
– Mathematically (modelling)
– Data-wise (measuring)
– Operationally (monitoring)
6
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Many Possibilities, Not Many Reliable Answers
Banks and Corporations engage in business transactions with counterparties that
present limited or no information and/or unreliable credit assessments
RATED: WEALTH OF
INFORMATION
Banks ~ 6,500 & Corporations ~
3,500
PUBLICLY LISTED COMPANIES
Accuracy
~ 60,000 Banks & Corporations
(Active)
Coverage
PRIVATE: INFORMATION
SCARCITY
Banks (Est.) ~ 50,000 & Corporations
(Est.) Infinite?
Ratings are provided by Standard & Poor’s Ratings Services, which is analytically independent and separate from S&P Capital IQ. For Illustrative Purposes Only.
7
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
1
Regulatory Requirements For PD Models
• Long-run average of one year default rates
• Three specific techniques for Probability of Default (PD) estimation, namely
internal default experience, mapping to external data and statistical default
prediction models
• At least five years of historical data / parallel run
• Basel II specifically stipulates that advanced banks (AIRB) must ensure that
all data entering credit risk models is accurate and “fit-for-purpose”
• Basel III goes one step further: “One of the most significant lessons learned from the global
financial crisis that began in 2007 was that banks’ information technology (IT) and data architectures were
inadequate to support the broad management of financial risks. Many banks lacked the ability to aggregate risk
exposures and identify concentrations quickly and accurately at the bank group level, across business lines and
between legal entities. Some banks were unable to manage their risks properly because of weak risk data
aggregation capabilities and risk reporting practices. This had severe consequences to the banks themselves and to
the stability of the financial system as a whole”.2
------------------------------1 Basel Committee on Banking Supervision, International Convergence of Capital Measurement and Capital Standards, A Revised Framework, June 2004.
2 Principles for effective risk data aggregation and risk reporting Basel Committee on Banking Supervision, January 2013.
8
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Analytical Requirements
Analytical models require a sufficient amount of data to be robust and avoid
overfitting
"With four parameters I can fit an elephant, and with five I can make him wiggle his trunk"
John von Neumann (Mathematician and Physicist, 1903 – 1957)
params = 5
Source: Courtesy of Neil Gunther: http://perfdynamics.blogspot.co.uk/2011/06/winking-pink-elephant.html. For illustrative Purposes Only.
9
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Modelling Credit Risk
Goal: Develop a global PD model for non-financial private companies based
on financials and default flags
• S&P Capital IQ (CIQ) database contains more than 600,000 private
company with meaningful financial statements:
• Good coverage of financials (Balance Sheet, Income Statement, etc.)
• Data cleansing
• Homogenization across different accounting standards
• Annualization
• Some thousand default flags are registered, with 20% coming from North
America
10
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Typical Fundamental Credit Risk Analysis
In many expert-judgement driven rating frameworks, business risk and financial risk
factors are evaluated separately and then combined into an overall assessment.
BUSINESS RISK
Country and macroeconomic risk
In a quantitative model
this dimension is often
ignored, due to
challenges in quantifying
it
Industry Risk
Competitive Position
• Market Position
• Diversification
• Operating efficiency
• Management growth and
operating strategy; risk
appetite; track record
• Ownership/governance
Business Risk
Score
Profitability/peer comparisons
Rating
FINANCIAL RISK
Accounting
These risk dimensions
can be easily reflected
by financial ratios
Financial governance and policies/risk
tolerance
Cash flow adequacy
Financial
Risk Score
Capital structure/asset protection
Liquidity/short term factors
For Illustrative Purposes Only.
11
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
How To Avoid Stepping Into Elephants Or Make Them Behave
• Avoid including too many free parameters
− Model parsimony (Akaike information criterion)
• Train in-sample, but optimize on out-of-sample data:
− Robust model performance when new data arrive
• Penalise concentration of weight into single explanatory factors:
− Robust model performance when economic conditions change (Tikhonov
Regularization, etc.)
• Limit number of sub-models
− Reduce number of “elephants” (cluster analysis of financials/risk drivers
homogeneity via Mahalobian distance of financials’ medians)
12
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
The S&P Capital IQ PD Model Fundamentals Private Corporates
Industry Cluster Specific Models
Risk
Financial
Dimension
x1
Total Equity / Total Assets
x2
Net Income / Total Liabilities
Debt Service Capacity (Proxy)
x3
Current Liabilities / Net Worth
Short-Term Leverage
x4
Return on Capital
x5
Cash / Total Assets
x6
Net Income / Total Liabilities
Dummy (Slope)
1 FR
2 FR
3 FR
Capital Structure
Profitability
Liquidity
Debt Service Capacity (Proxy)
Global Model
Risk
Business
Dimension
x7
Market Share
Competitiveness
x8
Diversification
Competitiveness
x9
Operating Efficiency
Competitiveness
x10
Country Risk Score
Country Risk Score (S&P)
x11
Macroeconomic Aggregated Score
Macro Environment
x12
Industry Risk Score
Industry Risk Score
BR
• The model achieves an granular/optimal industry/country segmentation.
• Financial Risk (FR) is modeled by industry cluster (here: 3: Capital Intensive,)
• Business Risk (BR) is modeled on a global basis, because systemic factors such as country risk and industry risk are taken into account
as powerful discriminatory factors
• A PD (and mapped credit score) is estimated separately for FR and BR. Both resulting PDs are then combined into a (final) standalone PD
(and mapped credit score)
• The final PD median levels are calibrated at the country level, using official historical benchmarks from international sources.
For Illustrative Purposes only.
13
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Model Performance Observation
In our experience, ROC* on actual out-of-sample shows 8% gain (i.e., 16% in accuracy
ratio!) in comparison to the training.
Model Cluster
ROC*
In-Sample
Out-of-Sample
(training)
Out-of-sample
(actual)
1
76.1%
-
82.1%
2
78.8%
-
89.5%
3
83.4%
-
85.4%
Overall
79.4%
79.8%
87.4%
* ROC = Receiver Operating Characteristic, a standard measure for discriminatory power
In addition, the combined model exhibits greater stability as it is less sensitive to the ‘normal’
fluctuations of financial ratios across a business cycle.
Yet, mechanisms are in place to account for financial deterioration, that indicates increasing
credit risk.
For Illustrative Purposes only.
14
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Measuring Credit Risk
Default flags: key ingredients for optimal model performance and calibration
There is a spectrum of definitions of what constitutes a default and the Bank for
International Settlements reference definition reflects this:
• A loan is placed on non-accrual
• A provision is made
• The obligor is more than 90 days past due
• The obligor seeks protection from its creditors
• Mathematically, it is important to have a large number of default flags to robustly train a PD
model, but what about the non-defaulters?
• Common assumption: “no information = no default”.
What are the implications of this assumption?
15
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
“No Assumption” vs. “No Information = No Default”
Default Rate
For Private Companies, globally, in S&P Capital IQ database
Size (in USD Mil.)
No assumption for non-default
Assumption for “no info” = “non-default”
The “No Assumption” case is more aligned with economic intuition and historical default
experience
It is equally important to have good quantity of default flags and high quality of
non-default flags
For Illustrative Purposes only.
16
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
The Typical “Conundrum” Of A Risk Manager
• Fundamental driven models offer a mid to long term view of the credit worthiness of the
counterparty
• A scoring model trained on Ratings offers a long-term view over the creditworthiness of a
public company and can be used to extend the range of rated companies to the unrated
universe
• A long-term view is useful as inputs into longer-lasting strategic decision such as: limit
setting; credit risk origination /underwriting policies; debt pricing for fixed income,
syndicated or private loans (just some examples)
• Globally, there are 2,818 publicly listed companies with a “b-”, 2,648 with a “ccc+” and 1,258
with a “ccc or worse” score.*
• In the experience of Standard & Poor's Ratings Services, B- rated companies show a
historical observed default rate of approximately 10%
But… how to anticipate when, and/or take preventive measures?
*Based on Standard and Poor’s historical observed default rates, available on www.spcreditpro.com.
** Statistics and observations calculated as of 15th October 2013 using the recently launched S&P Capital IQ CreditModel 2.6, available on www.spcapitaliq.com.
For illustrative purposes only.
17
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
The Typical “Conundrum” Of A Risk Manager
Counterparties in
emerging
markets
Source: Venitism Blogspot, 14 May 2014
Source: topnews.in, 14 May 2014
Private
company
financials
14,000 Suppliers
from around the
world
Poor You
Source: The Economist, 21 April 2012
18
Source: S&P Capital IQ, 14 May 2014
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Long-term vs Short-term Measures of Credit Risk
Distribution of 29,555 listed companies on October, 15th 2013
Probability of Default Buckets
Number of companies
10000
1000
100
10
1
APAC
Europe
North America
CreditModel* score (long-term “DNA”)
PD Model Market Signals** (short-term “DNA”)
“b- or worse” companies with high historical
observed default rate
Current benign economic conditions reflected in
the market view
For illustrative purposes only.
* S&P Capital IQ CreditModel 2.6 scores are calculated using S&P Capital IQ proprietary quantitative model, available on www.spcapitaliq.com.
** S&P Capital IQ Market Signals Probability of Defaults are calculated using S&P Capital IQ proprietary quantitative model, available on www.spcapitaliq.com.
19
Emerging Markets
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Monitoring Credit Risk
Short Term PD (mapped to credit scores)
Long-Term Score
aaa
aa+ to aaa+ to abbb+ to bbbbb+ to bbb+ to bccc+ and below
aaa aa+ to aa1
52
260
349
90
3
0
0
32
242
595
230
39
1
a+ to a-
bbb+ to bbb-
1
26
232
687
337
71
2
0
19
196
568
501
176
4
bb+ to bb- b+ to b- ccc+ and below
0
25
138
488
509
266
11
0
0
47
192
272
275
43
0
0
6
40
95
143
52
In our sample only about 11% of companies fall into a higher risk category
(highlighted). Likely candidates for future default and where to manage exposures
tightly.
Source: Bankruptcy and default data from SP CreditPro and S&P Capital IQ Key Developments for the S&P rated universe. This sample contains 7,316 observations where there are 200 defaulters.
CreditModel Scores are from S&P Credit Analytics’ pre-scored database, Market Signals PD are a subset of companies pre-scored in CreditModel from S&P Capital IQ, from 2001 to 2013.
For illustrative purposes only.
20
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Monitoring Credit Risk
Or… how to “reduce anxiety in credit risk managers”.
Combining outputs from different models enables to filter companies that show
biggest credit risk and focus on a limited number that are more likely to pose
problems both in the short- and long-term
Combined Model Outcome
Observed
Default rate
Both models IG or at least one a- or better
0.00%
One model in bbb range, other NIG
0.29%
CM in bb range, PD MS bb or worse
1.48%
CM in b range, PD MS bb or worse
10.53%
CM in ccc range, PD MS bb or worse
50.00%
Source: Bankruptcy and default data from SP CreditPro and S&P Capital IQ Key Developments for the S&P rated universe. This sample contains 7316 observations where there are 200 defaulters.
CreditModel Scores are from S&P Credit Analytics’ pre-scored database, Market Signals PD are a subset of companies pre-scored in CreditModel from S&P Capital IQ, from 2001 to 2013.
For illustrative purposes only.
21
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Summary
• Quantity / Quality are often antithetic, but both critical in Credit Risk.
• Modelling Credit Risk:
Too many explanatory variables can affect the model robustness
• Measuring Credit Risk:
Default flags and non-defaulters are critical to measure risk appropriately
• Monitoring Credit Risk:
Combining outputs from models with shorter and longer time horizons helps
focusing on the companies that pose significant Credit Risk
22
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
APPENDIX
23
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
How To Avoid Stepping Into Elephant: Cluster Analysis
The distances of the median values of financial ratios for each pair of industries is calculated in order to create
the dendrogram below in order to give an idea of the homogeneity of the industries and
Industry ID
Industry
Cluster
1
Aerospace & Defense
1
3
Automotive
1
7
Capital Goods
1
10
Chemicals
1
12
Consumer Products
(Non-Durable)
1
13
Consumer Products (Other)
1
Cluster Analysis by industries
Cluster
Analysis by Industries
14
Retail
1
2
Airlines
2
4
Energy
2
15
Construction Materials +
Forest Products
2
16
Metals & Mining
2
17
Utilities
2
20
Transport (ex Airlines)
2
5
Information Technology
3
6
Hotel & Gaming
3
8
Media
3
3
9
Healthcare
11
Pharmaceuticals
3
18
Telecoms
3
19
Services for Businesses,
Industries or Individuals
3
21
Real Estate
3
Red Circles: Final Cluster 1
Green Circles: Final Cluster 2
Blue Circles: Final Cluster 3
Mahalanobian
distances
Mahalanobian distances
2
1.5
1
0.5
0
7
14
13
10
12
1
19
3
4
16
15
industries
17
20
6
9
8
18
11
21
2
Industries are grouped together based on a combination of degree of
homogeneity of financial profiles, assumed common default risk drivers
and data availability by industry
For Illustrative Purposes only.
24
5
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Model Performance (II)
ROC on final model shows up to 2% gain in comparison to the financial model only.
Model
ID
Model
Business Risk
Model
ID
(global model,
ID
Financial Risk
GMP
ROC
GMP
GPR
ROC
GPR
Final Model
GMP
ROC
1
0.81
82%
300
1
0.79
76%
252
1
0.82
82%
440
2
0.92
88%
230
2
0.92
89%
205
2
0.93
90%
298
3
0.85
83%
310
3
0.84
80%
188
3
0.86
85%
439
TOTAL
0.87
86%
311
TOTAL
0.86
84%
202
TOTAL
0.88
87%
412
GMP = Geometric Mean Probability;
GPR = Growth Pick-Up Rate (over a naïve model); both are maximum likelihood
(calibration) measures, that were used for performance optimization
ROC = Receiver Operating Characteristic, a standard measure for discriminatory power
For Illustrative Purposes only.
25
GPR
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
www.spcapitaliq.com
Copyright © 2014 by Standard & Poor’s Financial Services LLC (S&P). All rights reserved. No content (including ratings, valuations, credit-related analyses and data, model, software or other application or output therefrom) or
any part thereof (Content) may be modified, reverse engineered, reproduced or distributed in any form by any means, or stored in a database or retrieval system, without the prior written permission of Standard & Poor’s
Financial Services LLC or its affiliates (collectively, S&P). The Content shall not be used for any unlawful or unauthorized purposes. S&P, its affiliates, and any third-party providers, as well as their directors, officers,
shareholders, employees or agents (collectively S&P Parties) do not guarantee the accuracy, completeness, timeliness or availability of the Content. S&P Parties are not responsible for any errors or omissions, regardless of the
cause, for the results obtained from the use of the Content, or for the security or maintenance of any data input by the user. The Content is provided on an “as is” basis. S&P PARTIES DISCLAIM ANY AND ALL EXPRESS OR
IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, ANY WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE OR USE, FREEDOM FROM BUGS, SOFTWARE ERRORS OR
DEFECTS, THAT THE CONTENT’S FUNCTIONING WILL BE UNINTERRUPTED OR THAT THE CONTENT WILL OPERATE WITH ANY SOFTWARE OR HARDWARE CONFIGURATION. In no event shall S&P Parties be
liable to any party for any direct, indirect, incidental, exemplary, compensatory, punitive, special or consequential damages, costs, expenses, legal fees, or losses (including, without limitation, lost income or lost profits and
opportunity costs) in connection with any use of the Content even if advised of the possibility of such damages.
Credit-related and other analyses, including ratings, and statements in the Content are statements of opinion as of the date they are expressed and not statements of fact or recommendations to purchase, hold, or sell any
securities or to make any investment decisions. S&P assumes no obligation to update the Content following publication in any form or format. The Content should not be relied on and is not a substitute for the skill, judgment and
experience of the user, its management, employees, advisors and/or clients when making investment and other business decisions. S&P’s opinions and analyses do not address the suitability of any security. S&P does not act
as a fiduciary or an investment advisor except where registered as such. While S&P has obtained information from sources it believes to be reliable, S&P does not perform an audit and undertakes no duty of due diligence or
independent verification of any information it receives.
S&P keeps certain activities of its business units separate from each other in order to preserve the independence and objectivity of their respective activities. As a result, certain business units of S&P may have information that is
not available to other S&P business units. S&P has established policies and procedures to maintain the confidentiality of certain non–public information received in connection with each analytical process.
S&P may receive compensation for its ratings and certain credit-related analyses, normally from issuers or underwriters of securities or from obligors. S&P reserves the right to disseminate its opinions and analyses. S&P's
public ratings and analyses are made available on its Web sites, www.standardandpoors.com (free of charge), and www.ratingsdirect.com and www.globalcreditportal.com (subscription), and may be distributed through other
means, including via S&P publications and third-party redistributors. Additional information about our ratings fees is available at www.standardandpoors.com/usratingsfees.
26
Permission
toIQ
reprint
or distribute
anyPORTAL
content from
this presentation
requires
the prior
written
approval
of S&P
Capital IQ. Not for distribution to the public.
STANDARD
& POOR’S, S&P,
CAPITAL
and GLOBAL
CREDIT
are registered
trademarks
of Standard
& Poor’s
Financial
Services
LLC.
KNOWLEDGE AND DATA – A WINNING COLLABORATION
Fernando Moreira, Lecturer in Business Economics, University of Edinburgh
Business School
27
Permission to reprint or distribute any content from this presentation requires the prior written approval of S&P Capital IQ. Not for distribution to the public.
Incorporating the knowledge of credit
analysts into quantitative credit risk
analyses
Fernando Moreira
Lecturer in Business Economics
University of Edinburgh Business School
S&P Capital IQ Webinar
25 June 2014
Agenda
 Reasons for considering the knowledge of credit
analysts
 Some ideas about how to incorporate analysts’
knowledge into quantitative credit risk assessment
 Challenges of using qualitative information in credit
risk analyses
 An (incipient) example of a European bank
29
Motivations (I)
 As seen before ...
 Most counterparties present limited or no (public)
information
 A PD model that combines Business Risk and Financial
Risk outperforms a model based solely on Financial Risk
…
Financial Risk
(e.g. accounting data,
cash flows, capital
structure)
Business Risk
(e.g. market position,
diversification,
management quality)
Final Model
(combining quantitative
and qualitative
information)
 … but Business Risk is often ignored in quantitative PD
models due to difficulty of representation
30
Motivations (II)
 So, because of the difficulty in interpreting
unstructured (non-numerical) data, some subjective
information is ignored in credit analyses
 It is estimated that 85% of corporate data is
unstructured! (Robb, 2004)
“We are drowning in information but
starved for knowledge”
John Naisbitt
31
Motivations (III)
The 5 Cs of Credit
 The first two Cs typically depend on intuitive
judgement of loan officers
32
Key points (I)
 Credit assessment can be based on the heart
(preference), on the head (analytical) or on “gut
feelings” (experience)
 “I have generally found that if my heart overrules my
head, the loan has almost uniformly been a poor
one. If my head overrules my gut instinct, the
resulting loan may sometimes be a poor one. In
looking back at poor loans, I should have followed
my gut instinct more often”
Stephen Salisbury in “Failures of My Lending Career”,
Journal of Commercial Lending, 67(2), 1984.
33
Key points (II)
 The knowledge (impressions, perceptions) of risk
analysts (or other personnel involved in credit
management) should be used to complement
(typically) limited statistical models
Sometimes people
(analysts) know
more than they think
they know
34
Mimicking the
decision-making process
 Expert systems
 Computer programmes that simulate the judgement
(decision-making process) of human experts (e.g. credit
analysts)
 To build these systems, several cases and their respective
solutions are reviewed; experts are interviewed
 Guide decisions of staff (especially those who do not have
much experience in the area) and contribute to keep
homogeneous decision criteria
 Used in credit risk since 1980’s. Examples: Citicorp
Mortgage, Xerox, and American Express
35
Mimicking the
decision-making process
 Neural Networks
 Try to mimic information process in the human brain
 Use past information on loan performance associated to
respective debtor data
 Try to identify the best answer given particular inputs
 Particularly useful when explicit decision rules are not
available
 Overcome some limitations of statistical models
36
Using the
knowledge of analysts
 Using subjective variables in statistical analyses
 For instance:




management quality
market position
quality of bank-debtor relationship
financial conditions (apart from what is shown in
accounting data)
 These variables are scaled by credit analysts (e.g.
from 1 to 10) and used as independent variables in,
e.g., logistic regressions
37
Using the
knowledge of analysts
 Text mining ⇒ to extract relevant information from
unstructured data (text)
 based on semantics
 conversion of unstructured data into structured features
that can be used, for instance, to perform statistical
analyses
 in contrast to … data mining ⇒ to find patterns in
structured (numerical) data
 A popular use: search engine technology
 Several software packages available (some of them
are free of charge)
38
Exploring the
knowledge of analysts
 Text mining in credit risk
 Converts unstructured data (e.g. transcription of credit
analysts’ impressions on borrowers) into structured
features that can be used in statistical analyses (loan
applications, PD calculations, etc.)
 Will likely reveal important information not represented in
traditional (numerical) databases
 Remember …
Sometimes people (analysts) know
more than they think they know
39
Challenges
 fraud and collusion (including pressure from highlevel managers)
 Example: scandal in Brazil
 However, this possibility is reduced if analysts do not have
complete information on how their knowledge impacts the
outcomes
 potential mistakes due to bias of analysts
 high costs
 difficulty in objectively comparing and combining
credit assessments run by different analysts
40
An (incipient) example
 Ongoing project in a European bank
 Objective: to identify characteristics of the largest
customers (debtors) that help explain profitability and
risk
 Context:
 few customers represent ~ 80% of income
 regressions are used
 numerical data (typically accounting data) are used as
explanatory variables
 profitability, risk, and relationship profit/risk are dependent
variables (one regression for each)
 However ...
41
An (incipient) example
 ... regressions (i.e. quantitative analyses) have not
given satisfactory results; and ...
 ... a considerable amount of information about
customers is in text format (memos written by credit
analysts) and has not been used
 The idea: to explore this information to try to identify
customers’ attributes associated with their respective
profit and risk
42
Summary
 Main goal of the ideas presented: to use some
knowledge of credit analysts and still to take
advantage of objective (quantitative) analyses
 Importance:
 information on many counterparties (private companies) is
scarce
 complete model (including qualitative information)
outperforms model based solely on numerical data
 Another objective: to motivate the audience to think
about how to explore analysts’ knowledge
 Moving towards a breakthrough in credit analyses?
43
Some references
 Jankowicz, A., R. Hisrich (1987). Intuition in Small
Business Lending Decisions. Journal of Small
Business Management.
 Lehmann, Bina (2003). Is It Worth the While? The
Relevance of Qualitative Information in Credit
Rating. Working paper.
 Robb, Drew (2004). Text mining tools take on
unstructured data. Computerworld 21.
 Weiss, S., N. Indurkhya, T. Zhang, F. Damerau
(2005). Text Mining: Predictive Methods for
Analysing Unstructured Information. New York:
Springer
44
Q&A
45
www.spcapitaliq-credit.com
Copyright © 2014 by Standard & Poor’s Financial Services LLC (S&P). All rights reserved. No content (including ratings, valuations, credit-related analyses and data, model, software or other application or output therefrom) or
any part thereof (Content) may be modified, reverse engineered, reproduced or distributed in any form by any means, or stored in a database or retrieval system, without the prior written permission of Standard & Poor’s
Financial Services LLC or its affiliates (collectively, S&P). The Content shall not be used for any unlawful or unauthorized purposes. S&P, its affiliates, and any third-party providers, as well as their directors, officers,
shareholders, employees or agents (collectively S&P Parties) do not guarantee the accuracy, completeness, timeliness or availability of the Content. S&P Parties are not responsible for any errors or omissions, regardless of the
cause, for the results obtained from the use of the Content, or for the security or maintenance of any data input by the user. The Content is provided on an “as is” basis. S&P PARTIES DISCLAIM ANY AND ALL EXPRESS OR
IMPLIED WARRANTIES, INCLUDING, BUT NOT LIMITED TO, ANY WARRANTIES OF MERCHANTABILITY OR FITNESS FOR A PARTICULAR PURPOSE OR USE, FREEDOM FROM BUGS, SOFTWARE ERRORS OR
DEFECTS, THAT THE CONTENT’S FUNCTIONING WILL BE UNINTERRUPTED OR THAT THE CONTENT WILL OPERATE WITH ANY SOFTWARE OR HARDWARE CONFIGURATION. In no event shall S&P Parties be
liable to any party for any direct, indirect, incidental, exemplary, compensatory, punitive, special or consequential damages, costs, expenses, legal fees, or losses (including, without limitation, lost income or lost profits and
opportunity costs) in connection with any use of the Content even if advised of the possibility of such damages.
Credit-related and other analyses, including ratings, and statements in the Content are statements of opinion as of the date they are expressed and not statements of fact or recommendations to purchase, hold, or sell any
securities or to make any investment decisions. S&P assumes no obligation to update the Content following publication in any form or format. The Content should not be relied on and is not a substitute for the skill, judgment and
experience of the user, its management, employees, advisors and/or clients when making investment and other business decisions. S&P’s opinions and analyses do not address the suitability of any security. S&P does not act
as a fiduciary or an investment advisor except where registered as such. While S&P has obtained information from sources it believes to be reliable, S&P does not perform an audit and undertakes no duty of due diligence or
independent verification of any information it receives.
S&P keeps certain activities of its business units separate from each other in order to preserve the independence and objectivity of their respective activities. As a result, certain business units of S&P may have information that is
not available to other S&P business units. S&P has established policies and procedures to maintain the confidentiality of certain non–public information received in connection with each analytical process.
S&P may receive compensation for its ratings and certain credit-related analyses, normally from issuers or underwriters of securities or from obligors. S&P reserves the right to disseminate its opinions and analyses. S&P's
public ratings and analyses are made available on its Web sites, www.standardandpoors.com (free of charge), and www.ratingsdirect.com and www.globalcreditportal.com (subscription), and may be distributed through other
means, including via S&P publications and third-party redistributors. Additional information about our ratings fees is available at www.standardandpoors.com/usratingsfees.
46
Permission
toIQ
reprint
or distribute
anyPORTAL
content from
this presentation
requires
the prior
written
approval
of S&P
Capital IQ. Not for distribution to the public.
STANDARD
& POOR’S, S&P,
CAPITAL
and GLOBAL
CREDIT
are registered
trademarks
of Standard
& Poor’s
Financial
Services
LLC.