Uncertainty and Explainability in Deep Rough Volatility: A Neural Information-Theoretic Posterior Approach
深層粗糙波動率模型的不確定性與可解釋性：基於神經資訊理論的後驗分析方法
While deep learning accelerates stochastic volatility model calibration, point estimates fail to capture post-calibration uncertainty from the implied-volatility (IV) surface. This paper proposes a Simulation-Based Inference (SBI) framework using Neural Ratio Estimation (NRE) to learn parameter posteriors. These posteriors propagate through heteroscedastic surrogate pricers to generate uncertainty-aware price intervals for forward-start, barrier, and realized-variance claims. Additionally, the novel 'Hellinger-SHAP' method is introduced to attribute posterior information gains to specific maturity-moneyness regions.
Key points
Simulation-Based Posterior Calibration
Uses Neural Ratio Estimation (NRE) to learn conditional posterior distributions for rough Heston (rHeston) parameters, capturing residual parameter uncertainty.
Uncertainty-Aware Pricing
Propagates parameter posteriors through heteroscedastic surrogate pricers to yield calibrated price intervals for exotic options.
Hellinger-SHAP Explainability
Introduces Hellinger-SHAP by applying Kernel SHAP to posterior-information functionals to identify key maturity-moneyness regions.
How it works
Why it matters
In financial engineering, deep learning is often hindered by its black-box nature and lack of risk quantification. This framework addresses these gaps by not only outputting robust pricing intervals with confidence bounds for exotic options (preventing material mispricing), but also using Hellinger-SHAP to mathematically explain which market data regions drive parameter inference, enhancing transparency for financial institutions and risk management.
Who it affects
- AI Researcher
- AI Developer
- Enterprise Leader
How to use it
- 1Pricing exotic options and path-dependent contracts (e.g., forward-start, barrier, and realized-variance claims)
- 2Calibrating rough stochastic volatility models (rHeston) under explicit parameter uncertainty
- 3Auditing and explaining neural calibration models using information-theoretic attributions
Limitations & caveats
- Computational cost depends on the simulation data required to train high-dimensional heteroscedastic surrogate models
- Inference quality is highly dependent on the correct specification of the prior-predictive model
Related
FurE: 10x Faster 3D Animal Fur Reconstruction Without Animal Datasets
FurE：免用動物毛髮資料集，實現 10 倍加速的 3D 動物毛髮重建技術
FurE is an efficient 3D animal fur reconstruction method that leverages a human-hair trained PCA decoder and Gaussian Frosting to achieve 10x faster, highly detailed, and editable groom reconstruction without animal datasets.
Learning to Stop without Learning to Stop: Self-Supervised Confidence Training Improves Reasoning Efficiency
自我監督信心訓練：免於刻意「學會停止」即可提升 LLM 推理效率
Researchers found that training reasoning models to predict their own confidence at intermediate steps naturally reduces generated tokens by up to 25% at matched accuracy, without explicitly optimizing for length or stopping.
First-Order Stationarity of Reverse Diffusions: Bridging Optimization and Sampling
逆向擴散的一階駐點性：連結優化與生成採樣的數學機制
This research establishes a first-order optimization theory for diffusion models, proving that SDE-based reverse Langevin diffusions contract Fisher divergences exponentially under strongly convex noising.