Filtered Historical Simulation Value-at-Risk Models and Their Competitors

Abstract

Financial institutions have for many years sought measures which cogently summarise the diverse market risks in portfolios of financial instruments. This quest led institutions to develop Value-at-Risk (VaR) models for their trading portfolios in the 1990s. Subsequently, so-called filtered historical simulation VaR models have become popular tools due to their ability to incorporate information on recent market returns and thus produce risk estimates conditional on them. These estimates are often superior to the unconditional ones produced by the first generation of VaR models. This paper explores the properties of various filtered historical simulation models. We explain how these models are constructed and illustrate their performance, examining in particular how filtering transforms various properties of return distribution. The procyclicality of filtered historical simulation models is also discussed and compared to that of unfiltered VaR. A key consideration in the design of risk management models is whether the model’s purpose is simply to estimate some percentile of the return distribution, or whether its aims are broader. We discuss this question and relate it to the design of the model testing framework. Finally, we discuss some recent developments in the filtered historical simulation paradigm and draw some conclusions about the use of models in this tradition for the estimation of initial margin requirements.