How to tell if a strategy actually worked

Begin after costs

Count trading fees, the spread paid when crossing quotes, slippage from expected to modeled fill, and infrastructure cost. State which costs are in the trade result and which are separate so they are not subtracted twice. The SEC documents how trading mechanics affect execution and market quality. SEC ↗

Ask what the test knew

A backtest replays past data. It can accidentally use a later value in an earlier decision—lookahead bias—or choose a rule after seeing that history—overfitting. Testing only coins or funds that survived creates survivorship bias. Freqtrade documents a practical lookahead check; published research shows why selecting the best backtest is risky. Freqtrade docs ↗ SSRN ↗

A forward paper test fixes a method and observes new data using simulated orders. It reduces hindsight, but modeled fills are still not real fills. Live execution adds actual orders, operational and financial risk. Do not promote one evidence stage into another.

Compare with doing less

Cash asks whether activity helped at all. Buy and hold asks whether a simple exposure did as well without repeated trades. State differences in capital, exposure and timing. A strategy that beats cash but lags a comparable holding is not the same finding as one that beats both. These are our research comparisons, not universal benchmarks.

On a Watch page, read the written rule and status before any return. Look for the observation window, open positions, realized and open profit/loss, fees, drawdown, missing intervals and halts. Recorded performance remains private while market-data republication rights are reviewed. Missing public performance is not a win.