Modeling extreme but plausible losses for credit risk: a stress testing framework for the Argentine Financial System

While not being widespread, stress tests of credit risk are not new in the Argentine financial system, neither for financial intermediaries nor for the Central Bank. However, they are more often based on rule-of-thumb approaches than on systematic, model based methodologies. The objective of this paper is to fill this gap. With a database that covers the 1994-2006 period we implement a three staged approach. First, we use bank balance sheet data to estimate a dynamic panel data model, with different statistical methodologies, to explain bank losses for credit risk with bank-specific and macroeconomic variables. In a second step, the macroeconomic drivers of bank losses, real GDP growth and cost of short term credit, are modeled with a Vector Autoregression (VAR). The VAR shows the effect of the variables (i.e. risk factors) that we find dominate the domestic business cycle: the price of commodities, the sovereign risk and the federal funds rate. Finally, we use this toolkit to perform deterministic and stochastic scenario analysis. In the first case we use the behavior of the risk factors during the crisis of 1995 (Tequila contagion) and 2001 (Currency Board collapse), and we implement a subjective scenario as well. The stochastic scenarios are performed by Monte Carlo with two alternative methodologies: a non-parametric bootstrapping approach and drawing repeatedly from a multivariate normal distribution. When comparing the estimated unexpected losses to available capital, we find that currently the Argentine financial system is adequately capitalized to absorb the higher losses that would take place in a stress situation.