[this page | pdf | references | back links]
Return to Abstract
1.1 Portfolio backtesting comes in two main
(a) Backtesting the
return generating potential of a particular investment strategy, and
(b) Backtesting the
forecasting ability of a risk model.
1.2 In either case, backtesting can be thought of
as a short-hand way of seeking working out whether some sort of forecasting
approach might work in the future without actually having to wait for the
future to arrive. In (a) we are forecasting, in effect, the first moment of
the distribution, i.e. the mean drift of the relative return that might
arise were we to follow a particular investment strategy. In (b) we are
forecasting, in effect, second and higher moments, by testing the spread of
returns that should have arisen in the past were the model to be accurate
versus the spread of returns that actually did arise.
1.3 The aim of this and the following pages is to
explore this topic further and to comment on the range of tools that can be
used for such exercises. They build on material on backtesting (of risk models)
contained in Kemp
(2009). These tools need to be slightly more sophisticated than we might
first expect, because in the past we would not have had the same amount of
information as we have now.
1.4 We focus principally in these pages on
backtesting risk models, because in some sense it is a more general
mathematical problem than one focusing mainly on the first moment of the
distribution, and because it also intimately relates to calibration, a
topic that is explored in some detail in Kemp (2009).
Contents | Prev | Next