8 Giovanni Majnoni

Access and Risk: Friends or Foes?Lessons from ChilebyOsvaldo Adasme, Giovanni Majnoni, Myriam Uribe: 

Access and Risk: Friends or Foes? Lessons from Chile by Osvaldo Adasme, Giovanni Majnoni, Myriam Uribe Seminario sobre Acceso a los Servicios Financieros Banco de la Republica, Bogota’ 9-10 de noviembre, 2006

Outline of the presentation: 

Outline of the presentation Does better risk assessment lead to less lending? A test based on bank lending in Chile Is credit risk linked to loans size? Preliminary answers

Risk versus Access: 

Risk versus Access A common perception ... “Strict adherence to risk management precepts leads to greater prudence and less lending” ... our hypothesis ... “Better risk assessment moves forward the frontier of eligible bank customers reducing the area of uncertainty where a blurred perception of risk impedes credit decisions” ... and a simple test “Does the accurate measurement of risk of bank loan portfolios show a direct or an inverse relationships between loan size and required economic capital?”

Risk versus Access: 

Risk versus Access The loan loss distribution of a loan portfolio... The loan loss distribution is a key tool to evaluate the level economic capital required to cover a certain level of losses of a bank portfolio ... is not easy to observe ... Important analytical restrictions are necessary to derive the shape of the loan loss distribution of a portfolio of bank loans ... especially in emerging countries As a result, the size and shape of loan loss distributions in emerging countries is largely unknown

Risk versus Access: 

Risk versus Access The contribution of this paper is twofold... It uses a simple statistical methodology seldom applied to the analysis of credit risk to derive the shape and parameters of the loan loss distribution It shows the difference of economic capital required for loans of different size .. and its results support several conjectures Proper risk assessment can reduce the cost of small loans or of loans to small borrowers It suggests some possible explanations of the current explosion of retail lending in most emerging economies The relevance of appropriate calibration of regulatory capital across different loan portfolios

Outline of the presentation: 

Outline of the presentation Does better risk assessment lead to less lending? A test based on bank lending in Chile Is credit risk linked to loans size? Preliminary answers

The test adopted: 

The test adopted A simple measurement approach Montecarlo simulations of a loan portfolio of given size (500 loans) based on the universe of bank loans censored by the Credit Register in Chile. Two sets of data The universe of loans between 1 and 10 million Chilean pesos (between US$2,000 and US$20,000). A yearly average of 178,000 loans. The universe of loans above 10 million Chilean pesos (above US$20,000). A yearly average of 101,000 loans.

The test adopted: 

The test adopted The partition into performing and defaulted loans At the beginning of each given year the loans with more than 12 months maturity are partitioned into defaulted loans (that experienced more than 90 days past due payments) and not defaulted ones (the others). The loss given default As in Basel II, the losses given default were assumed to be arbitrarily equal to 50 percent. The sample period The exercise is repeated for each year over the period 1999-2005.

The simulation procedure: 

The simulation procedure A random sample was extracted in each given year from the two universes of large and of small loans The size of the sample has been conventionally set to 500 (i.e. we assumed a bank portfolio with 500 loans as our benchmark) Loss measurement for each loan portfolio The ratio of defaulted loans over total portfolio value provides the percentage losses as a proportion of the portfolio’s value at the beginning of the year

The simulation procedure: 

The simulation procedure The simulation of 20,000 portfolios. The previous steps has been replicated 20,000 times in order to derive 20,000 values of the incidence of losses on each simulated portfolio Plotting the 20,000 values of percentage losses provides with an estimate of loan loss distributions in any given year From the estimated loan loss distribution we can derive the value of Expected Losses (EL) and Unexpected Losses (UL) that are generally linked to the definition of the appropriate size of Loan Reserves (provisions) and of Economic Capital. The following slide shows the estimated loan loss distribution of large loans in the year 2005

The simulation procedure: 

The simulation procedure

The simulation procedure: 

The simulation procedure The output The simulations have been done separately for the universe of small and of large loans over a period of seven years that goes from the trough of the economic cycle (1999) to its pick (2005) covering the ascending phase of the economic cycle A new evidence The results shown in the following slides are the first evidence, to our knowledge, of the shape that loan loss distributions take in an emerging economy: For loans of different size Over the cycle

Outline of the presentation: 

Outline of the presentation Does better risk assessment lead to less lending? A test based on bank lending in Chile Is credit risk linked to loans size? Preliminary answers

Large loans and loan portfolios’ losses: 

Large loans and loan portfolios’ losses

Large loans and loan portfolios’ losses: 

Large loans and loan portfolios’ losses Few remarks on the loss distribution of large loans The distribution changes very strongly over the cycle It becomes more skewed to the rights as the economic cycle improves (lower losses are more frequent) The fat right tails lowers as the cycle improves Per capital GDP is a statistically significant determinant of the distribution expected losses, of the distributions’ skewness and of its percentiles (the different VaR levels)

Small loans and loan portfolios’ losses: 

Small loans and loan portfolios’ losses

Small Loans and loan portfolios’ losses: 

Small Loans and loan portfolios’ losses Few remarks on the loss distribution of small loans The distribution does not show a clear dependence from the economic cycle It has a clearly symmetric shape and in a few cases the hypothesis of normality cannot be rejected Per capital GDP is not statistically associated with the distributions’ parameters (expected losses, skewness and percentile levels)

Loan size and loan portfolios’ losses: 

Loan size and loan portfolios’ losses

Loan size and loan portfolios’ losses: 

Loan size and loan portfolios’ losses Few remarks on the comparisons of loss distributions of loans of different size The previous chart plots together the two loan loss distributions for large and small loans that Chilean banks faced in the year 2001 It provides a visual evidence that: Expected losses are larger for small loans Unexpected losses are smaller for small loan

Loan portfolios’ losses over the cycle : 

Loan portfolios’ losses over the cycle

Loan size and loan portfolios’ losses: 

Loan size and loan portfolios’ losses Few remarks on the cyclical pattern of loan loss distributions over the cycle The previous chart shows the 14 distributions estimated over the whole sample period It provides a visual evidence that: Smaller loans do not show the dependence from the economic cycle shown by larger loans The shape of the loan loss distribution for large and smaller loans (asymmetric versus symmetric) remains different over the whole cycle

Loan size and loan portfolios’ losses: 

Loan size and loan portfolios’ losses The statistical reason of the different shape Only similarly sized losses over a large sample converge to a normal distribution The similarity of the size of losses is much larger for small loans (between US$2,000 and US20,000) than for the larger loans (between US$20,000 and US$334million) As loans get larger their number tends to become smaller making it impossible to obtain for large loans the same results that we observe for (very numerous) small loans What policy implications can we draw from this statistical evidence?

Outline of the presentation: 

Outline of the presentation Does better risk assessment lead to less lending? A test based on bank lending in Chile Is credit risk linked to loans size? Preliminary answers

Preliminary answers : 

Preliminary answers What policy implications? The average value of Expected Losses and Unexpected Losses (99.9% level) over our seven years sample period is respectively: 2.08% (EL) and 12.07% (UL) for large loans 4.23% (EL) and 2.62% (UL) for small loans The previous evidence suggests that for small loans the higher costs of more frequent (and expected) losses can be largely offset by the lower need of economic capital (a level almost 10% lower that than required for larger loans)

Preliminary answers : 

Preliminary answers What policy implications? Our evidence supports the notion that, when the cost of capital exceeds 20%, it may be cheaper for Chilean banks to lend to small borrowers than to larger ones (the additional cost of higher provisions is fully offset by lower capital requirements) Of course our results refer to risk adjusted exposures and do not take into account the transaction costs. These results indicate that risk considerations do not support a restriction of access to credit of smaller borrowers.

Preliminary answers : 

Preliminary answers What policy implications? Our evidence indicates that smaller loans are almost insensitive to economic fluctuations. These results may depend on the lack of observations for the downside phase of the cycle. But if confirmed – at least in relative terms – they would provide an additional strong incentive for banks to lend to smaller borrowers.

Preliminary answers : 

Preliminary answers Other policy implications Our evidence is coherent with the need to integrate capital regulation with loan portfolios’ concentration limits. In fact only reducing the different size of individual exposures can the loan loss distribution be prevented from becoming exceedingly asymmetric (as we saw only similar exposures led to a symmetric distribution) “Arms’length” lending is justified for small loans (no need to know the customer) while “relationship” lending is required for large loans where inside information is required to limit the cost of lending.

Conclusions : 

Conclusions Our evidence indicates that risk considerations alone do not affect the entry point to the provision of bank credit and therefore constraint on access to bank credit must be sought only on the transaction cost side Our evidence is consistent with outburst of retail lending that we observe all over the emerging economies (and in the LAC region in particular). We conjecture that, thanks to the reduction of transaction costs, banks have overcome the only binding constraint in lending to small borrowers and are already taking advantage of the statistical regularity described in this paper. Regulators, in turn, should support this trend imposing regulatory capital not higher than the economic capital

Slide29: 

Thanks (The paper is available as the Word Bank Policy Research Working Paper # 4003)