Set out and contrast the stages and procedures involved when a strategy is backtested

Backtesting follows three broad steps. First, specify the strategy, including the universe, signals, rebalancing frequency and benchmark. Second, gather point-in-time data, and third, run the simulation to produce performance and risk statistics. The usual design is a rolling window: at each rebalancing date the model is estimated only on data available up to that date, the portfolio is formed, and its return over the following period is recorded, before the window moves forward. This mimics real-time investing and gives out-of-sample results. Data must reflect realistic reporting lags, for example using annual accounts only after they are published. The trap is estimating signals over the whole sample and then testing them on the same period, which builds knowledge of the future into the result.