Assessing the Quality of Risk Measures

VaR has been subjected to much criticism. In the previous chapter, we reviewed the sharpest critique: that the standard normal return model underpinning most VaR estimation procedures is simply wrong. But there are other lines of attack on VaR that are relevant even if VaR estimates are not based on the standard model. This chapter discusses three of these viewpoints:

  1. The devil is in the details: Subtle and not-so-subtle differences in how VaR is computed can lead to large differences in the estimates.
  2. VaR cannot provide powerful tests of its own accuracy.
  3. VaR is “philosophically” incoherent: It cannot do what it purports to be able to do, namely, rank portfolios in order of riskiness.

We will also discuss a pervasive basic problem with all models, including risk models: The fact that they can err or be used inappropriately. A further major critique, the putative potential for VaR to exacerbate systemic risk, is discussed in Chapter 14.


In Chapter 10, we focused on the basic modeling problem facing VaR, that the actual distribution of returns doesn't conform to the model assumption of normality under which VaR is often computed. Using a VaR implementation that relies on normality without appreciating the deviations of the model from reality is an example of model risk. Models are used in risk measurement as well as in other parts of the trading and investment process. The term “model risk” describes the possibility ...

Get Financial Risk Management: Models, History, and Institutions now with the O’Reilly learning platform.

O’Reilly members experience books, live events, courses curated by job role, and more from O’Reilly and nearly 200 top publishers.