

The Strategy Handoff Test: Can Someone Else Reproduce Your Backtest?
Nov 19 | 3 min read MIN | Strategy Research

By
Kvants AI
A backtest is reproducible when a colleague can rerun a frozen strategy on the same inputs and recover the same trades and metrics within a stated tolerance. Give them the strategy version, a run configuration, data provenance and a reference trade ledger. Ask them to rebuild the result from the package before sharing the headline return.
Package the inputs that determine the run
Save the exact script or strategy export and record the engine version, parameters and signal timing. Specify the venue, instrument, timeframe, date boundaries, time zone, session and warm-up. Identify the data source and snapshot or revision. Where data can be shared, a checksum lets reviewers confirm they are using the same file.
Record initial capital, position sizing, fee schedule, spread or slippage, financing, tick size, order timing and the treatment of open trades. Include dependencies and the random seed if applicable. Store a sample trade ledger with timestamps, quantities, fill prices and costs, plus expected trade count, net return and drawdown. Define time-zone normalization and any rounding tolerance in advance.
Run the handoff in a clean workspace
Ask the reviewer to follow the package with a fresh setup. Compare data coverage, first qualifying signal, first submitted order, first fill and final ledger. Two runs can show the same total return while executing different trades. Trade-by-trade comparison catches differences that aggregates hide.
For example, one run might enter at the signal bar close and another at the next bar open. Both identify the same crossover, but the ledger reveals different fill timestamps and prices. Compare execution timing before changing the strategy or guessing at a data issue.
Document what a reviewer can repeat
Tell reviewers how to obtain licensed data if redistribution is restricted. Record version and run dates so later data revisions can be identified. Keep the run manifest next to the reference outputs and preserve the two files when a strategy changes. Once the original run is reproducible, test a new period or venue as a separate robustness exercise.
Frequently asked questions
Is a chart screenshot enough to reproduce a backtest? A screenshot is a reference for the result. The rerun needs the strategy artifact, data, configuration and execution assumptions.
Where should I start if results differ? Find the first timestamp where the runs disagree, then compare input data, signal generation and fills in that order.

Read more


















