Research Archive and Retention Policy
A research archive and retention policy is the set of rules governing which backtests, datasets, and code versions a firm must keep, for how long, and in what recoverable form, so that any past decision can later be reconstructed and defended.
Prerequisites: Audit Trail Requirements for Research
Backtests get rerun, data vendors change their history, and code repositories get refactored — which means that without a deliberate policy, the exact evidence behind a past trading decision can quietly become unreconstructable within a year or two. A research archive and retention policy fixes that by specifying what must be snapshotted and kept: the exact code version used, the exact data snapshot (not just "the vendor feed," which gets restated), the backtest configuration, and the resulting report, all timestamped and tied to whichever strategy or decision they supported.
Retention length is usually set by regulatory or internal audit requirements rather than researcher preference — commonly five to seven years for anything tied to a live trading decision, sometimes longer for material that supported a client-facing claim. The policy also specifies how things are kept, not just that they are: a zipped folder on someone's laptop does not satisfy an archive requirement, whereas a versioned, access-controlled repository with immutable snapshots does.
The payoff shows up rarely but matters enormously when it does: a regulator, an internal audit, or a post-mortem after a loss asking "what did the model actually show at the time" needs an answer that doesn't depend on anyone's memory or on data that has since been restated.
A research archive policy exists so that any past strategy decision can be reconstructed from an immutable, timestamped snapshot of the code, data, and results actually used at the time — not from data or code as they exist today.
Further reading
- SR 11-7, Guidance on Model Risk Management (2011)