Beyond ‘Bayesian vs. VaR’ Dilemma to Empirical Model Risk Management : How to Manage Risk (After Risk Management Has Failed) for Hedge Funds

In aftermath of the Financial Crisis, some risk management practitioners advocate wider adoption of Bayesian inference to replace Value-at-Risk (VaR) models for minimizing risk failures (Borison & Hamm, 2010). They claim reliance of Bayesian inference on subjective judgment, the key limitation of Bayesian methodology as underscored by statisticians (Kass & Raftery, 1995; Kruschke, 2011; Lynch, 2007), as the most significant advantage compared with VaR (Christoffersen, 2012). Despite its well-known limitations, just like all other quantitative models (Derman, 1996; Morini, 2011), VaR – [mostly] non-Bayesian and [increasingly] Bayesian – continues to be a key methodological foundation of risk management and regulation related risk modeling practices in global Finance (Danielsson et al., 2014; Zangari, 1996). Bayesian inference modeling and VaR modeling frameworks are outlined to facilitate model risk management (Derman, 1996; Morini, 2011; US Fed & OCC, 2011) for minimizing risk of any model – Bayesian, VaR, or Bayesian VaR. VaR frameworks are empirically applied for hedge fund risk modeling (Darbyshire & Hampton, 2012, 2014) of a multi-asset fund of funds portfolio of a large Wall Street investment bank. Multiple risk models and measures with transparent assumptions to cross-validate convergent findings across multiple levels of risk analysis are examined for empirical model risk management