Replaying a credit policy before activating it
Changing a threshold is the fastest and most consequential act a risk department performs. Replaying it against history turns a bet into an informed decision.
The riskiest gesture in the job
Lowering a debt service threshold by three points takes ten seconds. The effect is measured six months later, on a portfolio that can no longer be reconstructed except by approximation.
That asymmetry between the ease of the gesture and the slowness of its evaluation explains a widespread practice: thresholds stop being touched. Policy freezes, not out of conviction but out of caution — and the institution loses the adaptability that justified the system.
What replay provides
Replaying a candidate policy against the institution's history answers a precise question: among files already handled, how many would change outcome under this policy, in which direction, and for what volume of exposure.
That question has an exact answer, not an estimate. The files exist, so does their data, and the candidate policy is deterministic. This is not predicting the future but measuring a difference on the past.
The limit is real and must be stated: replay does not say whether newly accepted files would have repaid. It says how many there are, who they are, and what they represent. That is already considerably more than nothing.
What replay does not replace
It does not replace judgement. A policy that accepts eighty more files a month may be excellent or catastrophic depending on the nature of those files — and the risk department judges that, not the arithmetic.
Nor does it replace after-the-fact monitoring. An activated policy must be evaluated on its real cohorts and compared with what the replay predicted. It is that comparison, repeated, that builds trust in the tool.
A governance requirement, not a convenience
Replay takes on its full meaning once it is mandatory: no policy activates without having been simulated at least once, and the author of a change cannot approve it themselves.
Those two rules look bureaucratic. They are exactly what an examiner looks to verify, and they cost less than reconstructing an undocumented policy change two years later.
Sources
- NEXCLARA functional specifications — ADMIN module, policy lifecycle.
- WAMU Banking Commission — internal control and separation of duties requirements.