Matejka Kavcic, Ph.D.

1
Multinomial Ordered Probit Model for Assessing
Banking Sector Credit Risk

• Matejka Kavcic, Ph.D.
• Eurobanking 2008
• Maribor, Slovenia May 18-21
  
  

2
Agenda

• Structure of credit portfolio
• Model for credit risk
• Piecewise approach to stress testing

3

• Structure of credit portfolio
• Model for credit risk
• Piecewise approach to stress testing

4
Structure of credit portfolio
Structure of credit worthiness
5
Structure of credit portfolio

• Rapidly rising share of the best rated business
  entities among new exposures in 2003-2004 and
  2006 period as the sign for eased credit
  standards.
• Improving quality of new clients improve the
  total credit portfolio rating structure (with
  some delay).
• In 2006 66.8 of newly exposed business entities
  were rated A, while the share of business
  entities rated B decreased.

6
Structure of credit portfolio

• Business entities worthiness during the
  four-year period

7
Structure of credit portfolio

• Considering clients in business relationship with
  a bank in 2003 and still remaining in that
  relationship in 2006 a significant fall (more
  than 3.5 p.p.) in the share of A rated clients is
  observed.
• For more than 1 p.p. decreased the share of B
  rated clients.
• The share of C, D and E rated clients increased.

8

• Structure of credit portfolio
• Model for credit risk
• Piecewise approach to stress testing

9
Purpose

• To develope a model that enable us to
• estimate and analyse risks in banking sektor,
• determine credit worthiness of business entities
  and their changes,
• improve understanding of credit risk in Slovenian
  banking sector.

10
Contribution

• The model can analyse the expected distribution
  of business entities among more than five credit
  classes using one latent variable (credit
  worthiness of a business entity).

11
Data

• Data description
• Business entities in Slovenia that borrowed from
  at least one of the banks between 1995 and 2006.
• The sample of business entities was restricted to
  companies and sole traders.
• The unit of analysis is uniformly determined by
  the financial year-business entity-bank troika.

12
Dependant variable in the model

• The credit worthiness of the business entity with
  the relevant bank in the relevant year

• Categorical variable
• Ordinal and naturally ordered

13
Ranking of credit worthiness
14
Independent variables in the model
15
The model
A random-effect multinomial ordered probit model
for panel data.

• Employs a method based on calculating the
  probability of a particular business entity being
  in a certain credit worthiness class, based on
  the value of the selected indicators.

16
Expected results

• A higher capital distribution rank, greater
  proportion of cashflow from operations in
  revenue, good liquidity and higher demand improve
  credit worthiness of a business entity.
• Higher short-term indebtedness of a business
  entity in the previous year and an excessive
  increase in liquidity produce worse credit
  worthiness of a business entity.

17
Results of the model
18
Distribution of latent variable
19
Characteristics of the model

• The model takes into account not only selected
  microeconomic indicators (business entity
  specific) but also time component and possible
  different credit worthiness of relevant business
  entity at different banks.

20

• Structure of credit worthiness
• The model
• Piecewise approach to stress testing

21
Actual vs. Model forecast data
Comparison of the actual credit portfolio
structure with model forecast for 2006

• According to 2005 Model the expected share of D
  and E rated clients in 2005 should be 1.36
  percentage point higher compared to the actual
  banks data.
• Even according to the 2006 Model and using the
  same critical value the expected share of D and E
  rated clients would be higher by 0.77 percentage
  point.

22
Conclusions

• The model forecasts higher share of
  non-performing loans.
• Banks aggressiveness in attracting new clients
  and thus a greater market share result in
  over-optimistic credit ratings given to business
  entities.
• In giving credit rating to business entities,
  banks only take account of the current situation.
  
• The model also takes into account a time
  component.

23
Models results in time
24
Models results in time

• Banks are lowering their criteria for approving
  loans.
• If this were not the case, no change in the limit
  critical value would be expected.
• Given the same criteria for approving loans,
  credit ratings would improve if business
  entities indicators improved, which would entail
  a shift of distribution of the latent variable
  towards the left-hand side.

25
Risk factor shocks

• Defining the type and size of the risk factor
  shocks
• Plausible, but not very probable
• Analysis of historical variability (1995-2006)
• Limited to the historicallly largest shocks with
  a statistical probability of 5.

26
Defining risk factor

• Risk factors
• An increase in the short-term debt ratio
• Short-term financial and operating liabilities as
  a proportion of assets in the previous year
  increases by 2.1.
• Liquidity impairment
• Net cashflow per sale unit decreases by 61.3.

Effects of individual shocks on the share of
non-performing loans
27
Conclusions of stress testing

• The shock of the impairment of the liquidity of
  business entities has a greater impact on the
  structure of banks credit portfolio.
• For banks the shock of liquidity impairment means
  a greater probability that their clients will not
  be settling their liabilities on a regular basis.

28

• Thank you for your attention
• For further information please contact

matejka.kavcic_at_sid.si