Post on 14-Dec-2015
Credit Risk Assessment of Corporate Sector in Croatia
Saša Cerovac, Lana Ivičić
Croatian National BankFinancial Stability Department
Structure of the presentation
Intro – motivation and credit risk assessment framework
Data & definitions
Migration matrices
Logit model
Applications and further steps
Objective
Modeling credit risk of non-financial businesses entities: assessment and predicting of the rating
migration probabilities predicting the probability of being in the
default state
A contribution to the development of the CNB's technical infrastructure designed for the credit risk assessment (Figure 1)
Data sources
Two primary databases:
CNB’s database with prudential information on bank exposures and exposure ratings (quarterly frequency)
Financial Agency (FINA): micro data on corporate financial accounts (annual frequency)
Data preparation & cleaning (I)
Detailed CNB’s database available since June 2006 full coverage of the banks and detailed risk classification
Entries for non-residents, non-corporates, non-market based firms, group of activities and unidentified debtors (other debtors and portfolio of small loans) are removed from the population
All exposures towards each single debtor are summed according to their ID number and multiple entries are avoided by prioritizing them according to supervisory actions
Data preparation & cleaning (II)
Exposures towards small debtors – those not exceeding 100,000 kunas (13,500 euros) - are also removed reducing the volatility steaming from group of debtors that
have marginal share in total liabilities of the corporate sector
Negative values (“overpayments”) were treated as no exposure
Sample was stabilized by removal of enterprises entering and/or exiting the database during the period under observation (year, quarter)
Combining the CNB’s and FINA’s databases
Some further data reductions took place in the modeling phase due to errors and omissions in FINA’s database
Merging CNB’s database with annual financial statements of private non-financial companies obtained from FINA reduced sample dataset to 7,719 firms during 2007 and 2008 (covering more than 75% of bank’s exposures towards market-oriented corporates)
Final data set: non-balanced panel of 12,462 observations of binary dependent variable – default state.
Construction of credit rating (I)
The CNB's database provides only information on the risk classification of individual exposures (placements and off-balance sheet liabilities) - no risk classification of debtors
AX - standard A90d – standard, but over 90 days overdue B – substandard (over 90 days overdue) C – delinquent (over 365 days overdue)
Construction of credit rating (II)
The procedure for classifying debtors into distinct risk categories is based on solving a simple optimization problem derived from the risk classification of their total debt to the banking system as a whole
C 50% or more
B
A90d
AX
TOTAL 100% 100% 100%
Rating of debtor C B A90d
50% or more
50% or more
Treshold of 50% maximizes AX rated liabilities to AX rated companies and non-AX rated liabilities to the rest of firms.
Share of exposure of specific risk category
Distribution of rated debtors from June 2006 to December 2008
84,3
6,0
2,1
7,6
AX
A90d
B
C
Definition of default
Following the provisions of the Basel Committee on Banking Supervision (Basel II Accord) and applying general definition of default (Official Journal of the European Union, I.177 p. 113) :
Default state: ratings A90d, B or C
Rating migrations and the probability of default
Migration matrix
• Migration frequency:
• Discrete multinomial estimator:
• Migrations forecast:
• Domestic corporate sector: no absorbing state (reversals are possible); k=4
where
over horizon
Unconditional migration matrices
AX A90d B CAX 95,0 2,0 2,7 0,3A90d 43,0 22,0 32,3 2,6B 10,1 1,8 81,9 6,1C 1,7 0,1 1,3 96,9
AX A90d B CAX 97,5 1,5 0,9 0,1A90d 40,6 43,6 14,9 0,8B 6,0 0,9 90,8 2,3C 1,5 0,2 0,8 97,5
1-Year
1-Quarter
Note: Initial rating in rows, terminal rating in columns
PD
PRDegree of rating stability
Conditional matrices I
Hypothetical distributions of rating upgrades/downgrades
Unconditional distribution
Conditional distribution 1
Conditional distribution 2
Default area
rating change
prob
abili
ty
0
Quarterly conditional migration matrices II
AX A90d B CAX 97,5 1,5 0,9 0,2A90d 34,6 48,2 16,4 0,8B 5,3 0,6 91,9 2,3C 1,2 0,3 0,8 97,7
AX A90d B CAX 97,5 1,5 0,9 0,1A90d 46,5 40,8 12,1 0,6B 8,7 1,5 87,1 2,8C 1,7 0,0 1,3 97,0
AX A90d B CAX 97,5 1,5 0,8 0,1A90d 40,9 42,8 15,4 0,9B 5,6 0,9 91,4 2,1C 1,6 0,2 0,7 97,5
Construction
Non-financial services
a. Migration matrices conditional on economic activity
Industry
AX A90d B C
AX 97,2 1,7 0,9 0,2A90d 45,2 40,2 13,9 0,7B 6,1 1,0 90,3 2,6C 2,3 0,2 0,8 96,7
AX A90d B C
AX 97,8 1,3 0,8 0,1A90d 36,1 47,1 16,0 0,9B 5,9 0,9 91,2 1,9C 0,7 0,2 0,9 98,2
Retardation phase
b. Migration matrices conditional on economic cycle
Acceleration phase
Note: a. Initial rating in rows, terminal rating in columns b. Differences in migration frequencies that are statistically significant (5% level) in relation to the parameters of unconditional matrix are in italic[4].
[4] The t-statistics is derived from binominal standard error.
Empirical regularities
0
20
40
60
80
100
120
AX A90d B C
%
Distribution of debtors according to their rating
Empirical probability of default (1-Y Matrix)
Empirical probability of default (1-Q Matrix)
0,0
1,0
2,0
3,0
4,0
q3/2
006
q4/2
006
q1/2
007
q2/2
007
q3/2
007
q4/2
007
q1/2
008
q2/2
008
q3/2
008
q4/2
008
Def
au
lt r
ate
s, %
Industry
Construction
Non-financial services
Probability of default (reversal) in correlation with credit rating
Historical evolution of PDs across sectors
One-year forecasts
AX A90d B CAX 91,4 2,4 5,4 0,8A90d 53,6 6,3 34,8 5,2B 18,7 2,1 68,0 11,0C 3,4 0,2 2,4 94,0
AX A90d B CAX 93,1 2,5 3,9 0,5A90d 69,6 5,4 22,0 2,8B 21,8 1,6 68,9 7,8C 6,2 0,4 2,9 90,5
Annual forecast of migartion probabilities
Annual Forecast based on 1-Y Migration Matrix
Annual Forecast based on 1-Q Migration Matrix
Note: Initial rating in rows, terminal rating in columns
Modeling default state
Multivariate logit regression
Binary dependent variable yi,t explained by the set of factors X
The probability that a company defaults is
Using the logit function:
Share of firms in default across sectors
0,00
0,02
0,04
0,06
0,08
0,10
0,12
0,14
0,16
0,18
0,20
Agriculture andmanufacturing
Construction andreal estate
Non-financialservice
Total
2007
2008
Selection of explanatory variables
Initial set:
Financial ratios: liquidity (16), solvency (23), activity (12), efficiency (7), profitability (27) and investment indicators (1)
Size variables Sectoral dummies
Time lag: t-1
Correction of outliers: winsorization
Selection of explanatory variables
Univariate analysis
Mean equality test Graphical analysis: scatterplots Univariate logit models: ROC
Boxplots
-1
0
1
2
3
4
5
6
0 1
Cas
h to
tota
l ass
ets
Default
-0.2
0.0
0.2
0.4
0.6
0.8
1.0
1.2
0 1
Sh
are
ho
lde
rs' e
qu
ity to
tota
l ass
ets
Default
0
5
10
15
20
25
0 1
365
/ ac
coun
ts re
ceiv
able
turn
over
Default
0
1
2
3
4
5
6
0 1
(sa
les
+ d
ep
reci
atio
n)
/ to
tal a
sse
ts
Default
Scatterplots
Cash to total assets
00.10.20.30.40.50.60.7
-0.05 0 0.05 0.1 0.15
Percentile range average
Ave
rage
def
ault
rate
Shareholders' equity to total assets
0.00
0.05
0.10
0.15
0.20
0.25
0.30
0 0.1 0.2 0.3 0.4
Percentile range avergae
aver
age
defu
alt r
ate
365 / accounts receivable turnover
0.00
0.20
0.40
0.60
0.80
0 0.2 0.4 0.6
Percentile range average
Ave
rage
def
ault
rate
Sales + depreciation to total assets
0.00
0.10
0.20
0.30
0.40
0.50
0 2 4 6
Percentile range average
Ave
rage
def
ault
rate
ROC
The predictive power of a discrete-choice model is measured through its:
Sensibility (fraction of true positives): the probability of correctly classifying an individual whose observed situation is “default”
Specificity (fraction of true negatives): the probability of correctly classifying an individual whose observed situation is “no default”
ROC curves in univariate analysis
Profitability indicators seem to have highest univariate classification ability: AUCs ranging from 0.69 to 0.75
Among liquidity indicators, the best performing is the ratio of cash to total assets
Funding structure appears to be a good individual predictor of default too: ratios of equity capital to total assets and to total liabilities reach AUC values of above 0.70
Multivariate models
Intermediate choice: 28 financial ratios
Numerous models including different groups of variables were tested
Final multivariate model was chosen among best performing combinations of 3, 4, 5 and 6 explanatory variables + economic activity dummy
Best performing competing modelsModel 3_1 Model 4_1 Model 5_1 Model 6_1 Model 6_4
C 4.41 -0.41 -0.30 -0.17 -0.06(0.22) (0.17) (0.22) (0.22) (0.22)
Construction and real -0.45 -0.26 -0.24 -0.28 -0.30estate dummy (0.06) (0.07) (0.07) (0.07) (0.07)Cash to short-term -0.29liabilities (0.01)Cash to total assets -0.67 -0.67 -0.63 -0.65
(0.04) (0.04) (0.04) (0.04)Shareholders' equity to -1.87 -1.96 -2.17total assets (0.19) (0.19) (0.20)Shareholders' equity to -0.23 -0.27total liabilities (0.01) (0.01)After tax profit + -0.04depreciation to debt/365 (0.00)365 / accounts 0.10 0.11 0.09 0.09receivable turnover (0.01) (0.01) (0.01) (0.01)EBIT to total liabilities -0.17 -0.14
(0.01) (0.01)Sales + depreciation to -0.75 -0.51 -0.37 -0.41total assets (0.04) (0.05) (0.05) (0.05)Sales -0.01 -0.01 -0.01
(0.00) (0.00) (0.00)R2
0.18 0.19 0.19 0.20 0.20AUC 0.79 0.79 0.79 0.80 0.80% of correct 0 71.57 72.37 71.29 74.89 75.90% of correct 1 73.21 71.20 72.99 71.20 69.50% of total correct 71.80 72.22 71.51 74.41 75.05
Sector
Liquidity
Financial leverage
Profit
Activity
Size
Indicator
Marginal effects at the means of independent variables
Variable Coefficient Marginal effect (M) 1 Std.dev.*M
Constant -0,17
Construction and real estate dummy -0,28 -0,02 -0,009
Cash to total assets -0,63 -0,048 -0,020
Equity to assets -1,96 -0,149 -0,032
365/accounts receivable turnover 0,09 0,007 0,003
EBIT to liabilities -0,14 -0,011 -0,011
Sales + depreciation to total assets -0,37 -0,028 -0,029
Sales -0,01 -0,001 -0,0004
Kernel density estimate of default probabilities distribution for defaulted and non-defaulted
companies
0
100
200
300
400
500
600
700
800
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Nodefault=1
Nodefault=0
0.01
0.1
1
10
100
1000
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9
Nodefault=1
Nodefault=0
Cross-border lending effects on credit risk distribution
"In the presence of the effective credit limits, foreign banks help arrange direct cross-border borrowing for their clients, typically for the most creditworthy large corporates, leaving the Croatian banks mostly with customers with no other sources of financing.”
IMF (2008): Republic of Croatia: Financial System Stability Assessment—Update
Model application I (debt)Cumulative distribution of debt according to the origin of a creditor
a. Cumulative distribution of debt, 2007 b. Cumulative distribution of debt, 2002
0,0
10,0
20,0
30,0
40,0
50,0
60,0
70,0
80,0
90,0
100,0
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96
Percentiles
Domestic creditors only
Dominantly domestic creditors
Dominantly foreign creditors
0,0
10,0
20,0
30,0
40,0
50,0
60,0
70,0
80,0
90,0
100,0
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96
Percentiles
Domestic creditors only
Dominantly domestic creditors
Dominantly foreign creditors
Model application II (debtors)Cumulative distribution of debt according to the origin of a creditor
c. Cumulative distribution of debtors, 2007 d. Cumulative distribution of debtors, 2002
0,0
10,0
20,0
30,0
40,0
50,0
60,0
70,0
80,0
90,0
100,0
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96
Percentiles
Domestic creditors only
Dominantly domestic creditors
Dominantly foreign creditors
0,0
10,0
20,0
30,0
40,0
50,0
60,0
70,0
80,0
90,0
100,0
1 6 11 16 21 26 31 36 41 46 51 56 61 66 71 76 81 86 91 96
Percentiles
Domestic creditors only
Dominantly domestic creditors
Dominantly foreign creditors
Further steps Refinements of the approach:
Searching for alternative definitions of default Applying alternative estimators and modeling conditionality
of ratings dynamics Examining alternatives for the selection of explanatory
variables Correcting for selection bias using multinomial logit Modeling the event of default (PD) Modeling the event of reversal (PR) Improving explanatory power using macroeconomic
variables (contingent on longer data series)
Model applications: Forecasts of EAD Stress-testing of the corporate sector
Credit risk assessment in the Croatian National Bank
Macro approach
Sectoral approach
EWS
Macroeconomic risk model
Corporate credit risk models
Households credit risk
Bank failiure model
CAMELS downgarde model
Linear probability model (LOGIT)
Migration matrices
Sensitivity of NPLR's
Credit deafult
Sensitivity of financial margin
Capital adequacy