Replay of Valuation Runs
Replay means re-running a past valuation with the same trades, snapshot, models, and code version to recover the original numbers.
Explanation
Replay starts from a run ID, then loads the referenced snapshot, model parameters, and trade set as they existed at the time.
The same library version and configuration are used to recompute PVs and Greeks, which should match stored results within tolerances.
For Monte Carlo methods, fixed seeds or deterministic random generators ensure that replays remain consistent across time.
Replay is essential for debugging unexplained P&L, investigating model changes, and satisfying audit and governance demands.
replayreproducibilityvaluationruns
Interactive visualisation
Replay of valuation runs
Replay should match stored results within tolerance. Seeds turn Monte Carlo into a reproducible diagnostic.
Replay controls
Seed
42
Tol
2 bps
Noise
20
Replay status
1 FAIL
Interpretation
- A replay is: same trades + snapshot + models + library + config.
- Tight tolerances suit deterministic pricers. Monte Carlo needs tolerance tied to noise/SE.
- Fixed seeds make replay a stable check instead of a lottery.
Original vs replay
PV
original 102.3541 · replay 102.2523 · diff -0.1018
FAIL
tolerance: 2 bps · seed: 42
Delta
original 0.5172 · replay 0.5174 · diff 0.0002
PASS
tolerance: 2 bps · seed: 42
Vega
original 0.1826 · replay 0.1825 · diff -0.0001
PASS
tolerance: 2 bps · seed: 42
Gamma
original 0.0123 · replay 0.0123 · diff 0.0000
PASS
tolerance: 2 bps · seed: 42
Deviation plot
bars show |diff|; marker shows tolerance