Backtesting

Backtesting runs a strategy over past data to see how it would have behaved, and it is only ever as trustworthy as the history and assumptions behind it. Two failures dominate: using data that would not have been available at the time (look-ahead), and ignoring the costs of actually trading - spreads, slippage, and market impact - which quietly turn a paper edge into a real loss.

The subtler danger is overfitting: with enough parameters and enough tries, any researcher can find a rule that fits the past perfectly and predicts nothing. Guarding against it means out-of-sample testing, honest transaction-cost assumptions, and, underneath it all, point-in-time and survivorship-free data. A backtest is a hypothesis test, not a promise, and treating it as the latter is how good-looking strategies fail live.