This chapter is similar to Chapter 9 of the third edition, but in keeping with its growing
importance there is more discussion of expected shortfall (ES). The chapter requires 2 to
3 hours of classroom time. It starts by explaining value at risk (VaR). It compares VaR
to ES and shows that the latter has better theoretical properties. (To use the technical
term, it is more “coherent.”) It discusses the choice of parameters for VaR and ES, and
the impact of autocorrelation on estimates.
One thing I found after I wrote the first edition of the book is that many students have
difficulty calculating VaR and expected shortfall when the probability distribution of losses
is discrete. Consequently, I included several extra examples illustrating the calculations.
I find it useful to go through most of these examples in class. Many financial institutions
like the properties of expected shortfall and use it internally as one of their risk measures.
The Fundamental Review of the Trading Book (see Chapter 17) indicates that ES will
soon become the regulator’s measure of choice for market risk even though it is considered
more difficult to backtest than VaR.
Sections 12.7 to 12.9 explain marginal VaR, incremental VaR, and component VaR.
They explain the role of Euler’s theorem in the allocation of VaR and give a way of
aggregating VaRs. The material on back-testing in Section 12.10 gives some standard
tests of statistical significance and includes a relatively powerful two-tailed test proposed
by Kupiec.
Any of Further Questions can be used as hand-in assignments. I usually use 12.13.
Chapter 13: Historical Simulation and Extreme Value Theory
This chapter is similar to Chapter 14 of the third edition. It has more emphasis on
expected shortfall. The sequence of chapters has been improved for the fourth edition with
the calculation of VaR and ES coming immediately after the introduction of the measures
in Chapter 12. This chapter requires about 1.5 hours of classroom time. The advantage
of the historical simulation approach is that it requires no assumptions about probability
distributions and correlations. It assumes that percentage changes in all market variables
over the next day are a random sample from the last Ndays (or in the case of stressed
VaR a random sample from consecutive days during a period of particular stress for the
portfolio). Results are presented for a simple portfolio involving investments in four indices.
Section 13.1 explains the basic historical simulation approach while Section 13.2 ex-
plains how to calculate a standard error for VaR. (The standard error is quite large and
would be even larger if the assumption that the joint distribution of daily changes in market
variables is stationary through time could be relaxed.) Section 13.3 describes some exten-
sions of the historical simulation approach. These involve alternative weighting schemes,
procedures involving the volatility updating procedures of Chapter 10, and the use of
the bootstrap method to determine a confidence interval for VaR. (New material in the
fourth edition includes a simpler application of volatility updating.) Section 15.4 covers
the delta/gamma approximation. Section 13.5 and 13.6 present material on extreme value
theory which extends the material on the power law in Chapter 10. It provides a scientific
way of “smoothing the tails” of an empirically observed distribution.
Excel worksheets that go with the four-index example that is covered in the chapter
can be downloaded from my web site.
8