Applying the Precommitment Approach to Bottom-Up Stress Tests: A New Old Story

Abstract

Stress tests have become a key tool for banks, supervisors and macro prudential authorities. An aspect of these exercises is the need for statistical models to obtain risk measurements under an adverse scenario and a fundamental question is who should develop such models. If models are developed by the authorities (top-down approach), homogeneity of treatment among banks and more control over results are achieved, but the authorities do not necessarily have all the information that individual banks have. Banks’ models (bottom-up approach) may be more accurate. However, banks may have incentives to underestimate the impact of a shock, thus reducing the supervisory reaction. In this paper, we focus on bottom-up stress tests and suggest creating a system of monetary penalties (charges) proportional to the difference between the expected and the realised losses of a portfolio. The charges would aim to induce model developers to reveal their best forecasts. We show that this approach can be seen as an adaptation of the pre-commitment approach (PCA) developed and promoted by the US Federal Reserve in the 1990s but also as an application of the penalty criterion proposed by the Italian mathematician de Finetti as the foundation of the subjectivist definition of probability. We explain how the PCA could be adapted to bottom-up stress testing and provide a practical example of the application of our proposal to the banking book. What emerges is that the PCA can indeed mitigate banks’ incentives to provide underestimated measures of risk under the adverse scenario and thus better align the incentives of banks and supervisors.

Keywords: de Finetti penalty; principal-agent problem; pre-commitment approach; stress test; supervision

JEL Classification: G2