More sensitive macro factors

Assessment of default probability in
conditions of cyclicality
Totmyanina Ksenia
Moscow, 2014
Actuality
• Corporate sector represents a significant part of banking business
worldwide.
• Loans to corporates are a significant part of Russian banking
portfolio: to the end of 2013 loans to corporates reached 56% of
total credit portfolio and 39% of total assets of Russian banks.
• Number of researches devoted to corporate credit risk estimation is
strongly limited, especially for emerging market economies.
• Level of non-performing loans in corporate portfolio is increasing this fact can lead to instability of Russian financial and banking
system
• Construction companies are the most widespread among Russian
borrowers and at the same time very exposed to systematic risks
Object of research – Russian contracting companies
Item of research – Assessment of default probability
Purpose of research
The purpose of our research is to develop an empirical model for estimation default
probability of potential corporate clients of Russian banks.
The key steps to achieve this purpose are:
• research the different approaches to default definitions
• represent the classification of existing models to default modeling, review the
advantages and disadvantages of these models
• analyze the nature and sources of the procyclicality effect, represent the review of
available instruments to mitigation of the procyclicality effect
• collect the sample of financial indicators for defaulted and non-defaulted
companies and macro factors for the specified period
• execute a statistical analysis to determine the sensetive financial indicators and
macro factors
• execute a multivariable analysis to build sets of logit models
• analyze the quality and predictive power of final model and represent the
economic interpretation of the observed relationship
Default definitions
There are a lot of approaches:
• Default as non-fulfillment the conditions of the loan
agreement due to inability or unwillingness of the
borrower
• Default as the bankruptcy
• Default based on BIS criteria:
overdue more than 90 days and / or
bank considers that the debtor is unable to repay the loan
Review of default models
Probability of default models
1.2 Market-based models
Structural models
1.1 Fundamental-based models
1.1.1 Models based on financial statements
Reduced forms
Scoring models
1.3 Advanced models
Linear discrimination models
Fuzzy sets models
Multiple discrimination
Neuron networks
Univariate discrimination
Binary choice models
1.1.3 Rating-based models
1.1. 2 Macroeconomics models
Cohort approach
Exogenous factors
Duration approach
Endogenous factors
Procyclicality issue
Procyclical effect - increased business cycle fluctuations
Sources can be different:
1) Prudential control: for example, capital adequacy requirements increase
during periods of recession and reduce during the period of growth
2 ) The behavior of economic agents: for example, lending activity increase
in periods of growth and decrease in periods of recession
3) Expectations of economic agents: for example, the risk is
underestimated in the periods of growth, and overestimated during
recessions
4) The corporate governance system: for example, the KPI systems and
bonuses for managers
Mitigations of procyclicality
Procyclicality mitigation instruments
via inputs data
EAD
conversion
TTC LGD
via outputs data
Other parameters
Time horizon
Quantile
TTC PD
Conter-cyclicality index
Capital buffers
Dynamic provisions
Scalar factor
Stress-testing
Macro factors
Financial parameters that can be
statistically significant
Group of financial factors potentially affecting the
level of credit risk:
• Size
• Profitability
• Turnover
• Financial stability
We formed a long list of financial indicators from
each class above - finally total list consists of 31
indicators
Sample for modeling
• All defaulted companies in constructing industries during 2005-2013 –
159 companies
• Default = bankruptcy
• For each defaulted companies we had 3 analogical (same size and
industry) non-defaulted companies – 477 non-defaulted companies
Defaults dynamics
40
35
30
25
20
15
10
5
0
2005
2006
2007
2008
2009
2010
2011
2012
2013
Univariate analysis: selection of the risk
dominant financial indicators
Instruments:
1) Analysis and normalization of data (Chebyshev’s inequality)
2) Statistical tests to identify the most descriptive variables
(Student's test, Welch tests, ANOVA test)
More risk-dominant factors:
Balance value
Return on sales
Working capital
Share of stocks in current assets
Return on assets
Profitability of expenses
Coefficient of autonomy
Univariate analysis: selection of the risk
sensitive macro indicators
Instruments:
1) Analysis and normalization of data (Chebyshev’s inequality)
2) Regression models between macro factors and average
default rate (based on S&P data)
More sensitive macro factors:
Oil price
Export of goods and services
Imports of goods and services
Current account
Unemployment rate
Loans to individuals
Multivariate analysis: a binary choice
model
Binary logit-models:
where
if the company is default
otherwise
set of financial and macro factors
Multivariate analysis
On the basis of selected financial indicators and
macro variables, all possible multivariate models
were built
The resulting combination was selected based on
following criteria:
• No significant correlation
• Significance of indicators (t-statistic and F-test)
• The highest value Mc Fadden R2
Multivariate analysis: results
Best model (Mc Fadden R2=32%):
• Stocks in current assets, profitability of expenses,
coefficient of autonomy and import are most risk
sensitive
• Share of stocks in current assets was included in
the quadratic form that led to increase of R2 by
2%
Quality of model – classification table
Model results
Observed
Classification table
Non default
Default
Non default
84% (TN)
53% (FP)
Default
16% (FN)
47% (TP)
• Model better predicts no default cases
• With the exclusion of macroeconomic indicators
the quality of the model decreases
Thank you for your attention!