Backtesting

Backtesting is running a precisely defined set of trading rules against historical data to measure how it would have performed. The output worth keeping is a trade distribution — win rate, average win and loss, expectancy, maximum drawdown — not a single equity-curve number.

Three biases wreck most backtests. Overfitting: tuning parameters until they fit the past perfectly and nothing else. Survivorship bias: testing on today's index members excludes the companies that went to zero. Look-ahead bias: using information (closing prices, restated financials) that was not available at the moment of the simulated trade. Realistic commissions and slippage belong in every run; edges that die from $0.02 per share of friction were never edges.