A chart candle is not an executable price
Backtests commonly assume that an order fills at an observed open, close or stop level. Live orders meet a queue, bid-ask spread, latency and available liquidity. The difference is slippage, and it tends to become larger when a strategy most wants to trade.
Small assumptions compound
Fees, funding, taxes, missed signals and partial fills can turn a thin historical edge into no edge at all. Test conservative costs and compare results across multiple time periods rather than optimising one perfect interval.
- Include realistic fees and adverse slippage
- Use out-of-sample periods that were not used to tune rules
- Inspect drawdown length, not only maximum drawdown
- Paper trade the unchanged rules before considering live use
Markets change their character
Volatility, correlations, participants and exchange rules evolve. A trend strategy can struggle in a long range; a mean-reversion strategy can fail when a genuine repricing begins. Robust rules should make their weaknesses visible, not hide them.
Treat live performance as new evidence. Monitor deviations from tested behaviour and pause when execution or data quality is unreliable.