A credible backtest prevents future data from leaking into past decisions, uses executable prices, includes fees, slippage, latency and funding, and evaluates out-of-sample periods. Profit alone is insufficient; inspect drawdown, trade count, regime dependence and sensitivity to parameters.
Transparent formula
Net strategy return = gross trade PnL − fees − slippage − funding − execution failures
Worked example
A strategy showing 8 bps gross edge per trade but paying 5 bps round-trip fees and 2 bps slippage has only 1 bp before funding and errors. Small assumptions can erase the apparent edge.
Use this sequence
- Lock the signal definition before testing.
- Use time-ordered train and test periods.
- Model all execution costs.
- Paper trade before live deployment.
Common mistakes
- Optimizing hundreds of parameters on one sample.
- Using closing price for an unavailable fill.
- Ignoring exchange outages and rate limits.
Verify next
Frequently asked questions
What is look-ahead bias?
It occurs when a historical decision uses information that was not available at that time.
How much out-of-sample data is enough?
Enough to include multiple market conditions and a meaningful number of trades; there is no universal minimum.
When should a bot go live?
Only after stable paper results, operational monitoring, risk limits and a small controlled live test.
This page provides calculation and research frameworks, not investment, legal or tax advice.
