Once a trading idea is formed, it needs to be tested on historical data. The testing process is known as a back test. This chapter describes the key considerations for a successful and meaningful back test.
The purpose of back tests is twofold. First, a back test validates the performance of the trading model on large volumes of historical data before being used for trading live capital. Second, the back test shows how accurately the strategies capture available profit opportunities and whether the strategies can be incrementally improved to capture higher revenues.
Optimally, the trading idea itself is developed on a small set of historical data. The performance from this sample is known as "in-sample" performance. One month of data can be perfectly sufficient for in-sample estimation, depending on the chosen strategy. To draw any statistically significant inferences about the properties of the trading model at hand, the trading idea should be verified on large amounts of data that was not used in developing the trading model itself. Having a large reserve of historical data (at least two years of continuous tick data) ensures that the model minimizes the data-snooping bias, a condition that occurs when the model overfits to a nonrecurring aberration in the data. Running the back test on a fresh set of historical data is known as making "out-of-sample" inferences.
Once the out-of-sample back-test results have been obtained, they must be evaluated. ...