How do desks manage model risk in exotic derivatives?

Question

AcadiFi · Accepted Answer

There's no 'right' model — only models that are fit-for-purpose given their assumptions. Model risk management means acknowledging uncertainty and pricing it in.

Erika Vandenbosch at Antwerp Model Risk Office runs a model validation framework across three dimensions:

1. Calibration quality — Does the model fit today's market? Compute RMS error on vanilla quotes after calibration; reject if error exceeds threshold (say 0.3 vol points for FX, 1 bp for rates).

2. Forward consistency — Does the model produce sensible forward smile dynamics? Simulate paths, compute smile statistics at t+1 month, compare to historical patterns.

3. Pricing dispersion — Run three candidate models (LSV, SABR, Bates) through the exotic. Max - min spread = model risk buffer.

For a worst-of autocallable, Erika found LSV priced it at 4.12% of notional, Heston at 4.38%, and Bates at 4.51%. She reserves the midpoint (4.35%) for PnL and holds the 0.39% dispersion as a model reserve. When the trade unwinds naturally, the reserve releases; when re-priced, reserves are re-marked.

Regulatory frameworks (SR 11-7 in the US, PRA SS1/23 in the UK) mandate:
- Independent validation before deployment
- Ongoing performance monitoring
- Periodic recalibration and benchmarking
- Documentation of known limitations

Exotic desks often run 'challenger models' — a second pricing engine that shadows production. Material deviations trigger investigation. Finally, trader PnL is risk-adjusted: desk with aggressive model parameters sees lower bonus if model reserves grow. This aligns incentives with prudent model use.

How do desks manage model risk in exotic derivatives?

Master Level III with our CFA Course

Related Questions

Practice Questions