Abstract

The socioeconomic implications of this research are profound, in that if financial institutions had robust risk measurement tools along the lines proposed herein, the massive global wealth loss and concomitant social ills (e.g., the rise of populist political ideologies) may have been far limited in severity. A critical question for banking supervisors is the correct amount of capital or liquidity resources required by an institution to support the risks taken in the course of business. The finan cial crises of the last several years have revealed traditional approaches such as regulatory capital ratios to be inadequate, giving rise to supervisory stress testing and scenario analysis (“ST/SA”) as primary tools. A critical input into this process are macroeconomic variables that are provided by the prudential supervisors to institutions for exercises such as the Federal Reserve’s Compre hensive Capital Analysis and Review (“CCAR”) program. We propose an application to this ex ercise of a unified methodology that incorporates non-linear scenarios from alternative perspec tives in a general non-normal setting for economic and market factors, which is novel in this ap plication to credit risk and a supervisory application of ST/SA. This includes a flexible framework for causal and predictive market scenarios that combines Bayesian networks (“BNs”) and entropy pooling (“EP”), with BN generating a finite set of joint causal views for the relevant risk factors, while EP is used to project each of these stress scenarios over stochastic simulations. The joint view probabilities from BNs are naturally used as weights for the associated EP probability vec tors to compute a single posterior probability distribution. The framework allows us to implement economic scenarios and perform ST/SA conditional on realizations of relevant risk in a purely causal and predictive manner. This procedure provides a tool providing a tool for portfolio, risk and supervisory practitioners to manage credit risk and profitability. We test these methodologies empirically with aggregate banking charge-off data for several lending segments and benchmark the results against a challenger vector-autoregressive model, finding that the BN and EP ap proaches produce scenarios that have superior characteristics. Namely, the proposed approach produces more conservative and more accurately estimated portfolio risk measures for the same severely adverse scenario, with a more parsimonious model that has better fit to the data.

DOI: doi.org/10.63721/26JESD0114

To Read or Download the Article   PDF