Backtests replay rules on history. They do not automatically include microstructure change, fee evolution, regime shifts, or your future discipline unless you model them.
Overfitting is tuning until the in-sample equity curve looks heroic. Indicators pile up; parameters multiply; filters proliferate. The model memorizes noise and calls it alpha.
Symptoms: performance collapses on a holdout period; small parameter changes destroy results; the rule set only works on one instrument or one decade; turnover is implausibly high after realistic costs.
Mitigations help but do not cure: simpler rules; pre-register hypotheses; walk-forward evaluation; true out-of-sample holds; stress tests on spreads and gaps; reporting drawdowns and tail days, not only CAGR screenshots.
Publishing research ethically means showing where the idea failed. VegaDeck does not rank or sell strategies in v1.x.
Educational only—not investment advice.
中文版：/academy/backtest-overfitting-zh
Educational only · not investment advice · Risk disclosures