{"slug":"backtest","term":"Backtest","category":"Quant & Stats","definition":"A backtest runs a trading strategy's exact rules against historical data to estimate how it would have performed, producing metrics like win rate, expectancy, and maximum drawdown before real money is risked. A backtest cannot prove a strategy works — it only shows the rules didn't fail in the past under the assumptions tested.","street":"If you haven't backtested it, you're not trading a strategy, you're trading a mood.","depth":["A usable backtest needs rules precise enough that two people would take identical trades, realistic costs, and enough samples. Costs are where edges quietly die: 0.05% taker fees paid twice plus slippage is roughly a 0.15% hurdle per round trip, and on perps funding can add or subtract more. Aim for 100+ trades minimum, test across different regimes — 2021 trend, 2022 bleed, sideways chop — and hold back a chunk of data the rules never saw.","The classic ways a backtest lies: lookahead bias (a rule secretly using information from the candle it trades inside), survivorship bias (testing today's top-50 list on old data skips every coin that died), and fitting parameters to the exact past. Working assumption from the desk: live performance comes in around half the backtest, and if the edge vanishes when a parameter moves 20%, there was no edge — just a curve wearing a strategy costume."],"live":{"label":"How our own backtests are run, nulls included","href":"/methodology"},"related":["overfitting","expectancy","win-rate","drawdown"]}