Backtest Validation
Curve fitting, look-ahead leakage, survivorship in the instrument set, slippage assumptions and parameter stability, checked before capital follows the equity line.
A backtest is an argument, not evidence. This section tests whether the argument holds, which matters most in the window between a promising result and the decision to trade it.
Where results usually break
Look-ahead leakage
Signals computed on a bar that had not closed. Indicators calculated across the whole series rather than progressively. Instrument lists assembled with knowledge of which symbols survived. Each of these produces a smooth equity curve and none of them survives live execution.
Curve fitting
Tested by moving each parameter one step in both directions and recomputing. A result that collapses when a setting shifts slightly was fitted to noise. A robust method sits on a plateau, not a spike, and the shape of that surface is more informative than the peak value on it.
Cost assumptions
Most testers default to optimistic fills. Spread at the moment of entry rather than average spread, slippage on stops during fast markets, commission, swap and funding all get reapplied at realistic levels, and the strategy is re-run. Short-horizon methods often stop working at this step alone.
Sample partitioning
Whether an out-of-sample period genuinely existed, or was consulted during development and so became in-sample without anyone deciding that it should. Walk-forward results are reconstructed where the data allows.
What to send
The strategy report, the parameter set, and the data range and source. Source code is useful but not required. Where a platform’s own report is the only artefact available, that is workable, with the limitations stated in the findings.