Spot flaws in the backtest of a strategy

Several problems commonly spoil backtests. Survivorship bias arises when the data include only firms that still exist, which lifts returns. Look-ahead bias uses information not available on the date, such as accounts published weeks after the period end, or later restated figures. Data snooping comes from testing many variations and keeping the best; the more trials, the higher the hurdle for significance, for example a t-statistic above 3 rather than 2. Ignoring transaction costs, shorting constraints and capacity inflates results, especially for high-turnover strategies. Remedies include point-in-time data, reporting lags, out-of-sample tests and cost models. The trap is missing look-ahead bias when data are dated by fiscal period rather than publication.