Elizaveta Sizova's Abstract
This paper investigates how and when banks underreport market risk using internal models. We study hand-collected data on modelling and disclosure choices and examine how they relate to the level of Value-at-Risk (VaR) and the number of VaR violations. We find that more elaborate modelling like using more historical data and Monte Carlo (MC) simulation can correspond to more conservative VaR, but seems to be more punitive in terms of capital requirements. Presumably more sophisticated, but also more opaque internally computed 10-day VaRs, instead, seem not to capture tail risk well, precisely when in-house information should be particularly valuable. We conclude that capital requirements for market risk are indeed compromised by strategic modelling, but that the degree of the implied distortion depends on the specific model characteristic. By providing evidence from inside the black box of internal risk models, our results help to understand how risk measures are manipulated in practice and to assess the de facto performance of specific models. This complements the risk management literature, which typically analyzes model characteristics using conventional statistical methods, but does not consider the robustness to adverse incentives. Our findings can inform regulators, who may prefer to approve theoretically inferior modelling choices, if they are less likely to be abused in practice, and who may want to insist more vigorously on the more robust modelling options.