The Randomized Hull-White Model: Extending Affine Diffusions for xVA Accuracy by Kees Oosterlee, Thomas van der Zwaard, and L. A. Grzelak
The Hull-White model has long been the backbone of interest rate modeling for xVA (Valuation Adjustment) engines. Its appeal lies in speed, tractability, and analytical clarity. However, its limitations have become increasingly evident in modern markets. The classic one-factor Hull-White setup can only generate implied volatility skew, not the full smile. As a result, it fails to fully calibrate to the volatility surface, leaving xVA and exposure calculations misaligned with market realities.
This new paper by Kees Oosterlee, Thomas van der Zwaard, and L. A. Grzelak introduces a major advance: the randomized Hull-White (rHW) model. Their approach preserves the computational efficiency of the Hull-White framework while adding flexibility to match both skew and smile. By introducing a stochastic, state-dependent structure, the model extends the traditional affine diffusion system used in xVA engines.
The core innovation is simple but powerful. Instead of relying on a single parameter set, the authors randomize one of the Hull-White parameters—typically the mean reversion rate or volatility—and approximate it as a convex combination of several calibrated Hull-White dynamics. Each scenario is assigned a probability weight using the Randomized Affine Diffusion (RAnD) technique. This means that pricing European instruments remains semi-analytic: prices are simply weighted sums of standard Hull-White prices. Calibration thus stays efficient and robust, while the model gains the ability to fit the full volatility surface.
For xVA purposes, the rHW model is evaluated through regression-based Monte Carlo simulations. These capture how volatility skew and smile affect potential future exposures (PFEs) and credit/funding valuation adjustments. The authors show that accounting for these features materially changes exposure profiles for both linear and early-exercise interest rate products.
The results are impressive!
The rHW model replicates swaption surfaces much more accurately than the standard Hull-White, without sacrificing speed or numerical stability. It bridges the gap between semi-analytic tractability and realistic volatility dynamics—something that has long eluded affine models.
This paper offers practical implications for risk and model management. Banks and xVA desks can now maintain computational efficiency while incorporating a more realistic treatment of market volatility. The randomized Hull-White framework also provides a natural foundation for scenario-based exposure analysis and model risk benchmarking.
In short, this research keeps the analytical strengths that made Hull-White the industry standard, but enhances it with a deeper understanding of how skew and smile shape exposures. For practitioners focused on xVA, risk analytics, or interest rate derivatives, the randomized Hull-White model provides a meaningful and implementable improvement to the existing toolkit.
Join us:
2025 Generative Ai Investing Summit & 2025 Wall Street Quant Awards
