Backtest checklist: what to verify before you trust the results
By Overfit Labs · Updated · Educational only, not financial advice.
A backtest is a historical simulation of a trading idea. A smooth equity curve is easy to produce and easy to misread. Use this checklist before you treat any backtest as evidence of an edge.
Backtest checklist
- One clear rule set: entries, exits, size and risk written down before you tune.
- Costs: fees, slippage, and (for perps) funding — not zero.
- No look-ahead: signals use only data available at that bar; no future OHLC leaks.
- Realistic fills: especially around gaps, thin books and stop cascades.
- Out-of-sample (OOS): hold out a final slice of data you do not tune on; report in-sample and OOS separately.
- Walk-forward or rolling re-fit if parameters are optimised — one fixed optimum on the whole history is a red flag.
- Parameter sensitivity: nearby settings should not collapse performance; a single sharp peak often means overfitting.
- Multiple trials: if you tried dozens of ideas, the best one needs a higher bar (deflated or haircut Sharpe thinking).
- Sample size: enough trades and enough different regimes; a bull-only window proves little.
- Drawdown: max DD and time-under-water, not only CAGR.
- Stress cases: what happens if fees double, or fills are worse by 1–2× your assumption?
- Paper or tiny live next: a backtest is a hypothesis. Confirm with paper trading or minimal size — never leap to full size on a chart alone.
Common failure modes
- Optimising until the past looks perfect (overfitting).
- Ignoring that crypto regimes change quickly.
- Comparing strategies without accounting for how many you tried.
Tools from Overfit Labs
Backtest Toolkit is offline Python research software with tearsheets and robustness helpers. Its page has three free example robustness reports (score, verdict, out-of-sample and Monte Carlo checks, run on made-up data). Research software only — not financial advice, and not a prediction of future returns.