CHAPTER 12

Back-Testing ofTrading Strategies

It is hard to make predictions, especially about the future,” reportedly noted Mark Twain. Asset management institutions translate this wisdom into the standard disclaimer: “Past investment performance does not guarantee future returns.” Yet, forecasting has been widely used in economics, finance, and natural sciences (e.g., climate modeling)—arguably in every field where time series analysis is involved. The term back-testing implies that the forecasting models are fitted and tested using past empirical data. Usually, the entire available data set is split into two parts, one of which (earlier data) is used for in-sample calibration of the predicting model while the other is reserved for out-of-sample testing of the calibrated model. In simple models, it is usually assumed that the testing sample is variance stationary; that is, the sample volatility is constant.

If it is expected that the optimal strategy parameters may evolve in time, walk-forward testing can be used. In this case, moving-window sampling is implemented. For example, if a 10-year data sample is available, the first five-year data are used for in-sample calibration and the sixth-year data are used for out-of-sample testing. Then, the data from the second to the sixth year data are used for in-sample calibration and the seventh year is tested out-of-sample, and so on.

In general, the choice of the data sample is not always determined by the rule more is better, as the ...

Get Financial Markets and Trading: An Introduction to Market Microstructure and Trading Strategies 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.