This paper presents a new modelling framework for seasonal implied volatility surfaces derived from European commodity options. The proposed model extends the dynamic arbitrage-free pricing function by Carr and Wu (2016) in multiple directions. Specifically, it is a high-dimensional nonlinear state-space model that integrates a forward-looking stochastic seasonality component and incorporates the matrix-variate normal (MXVN) distribution to address measurement errors. This distribution captures varying signal-to-noise ratios across different maturity and moneyness combinations, thus assigning a unique data-driven weight for each quote during surface fitting. We develop computationally efficient approximate filtering recursions based on the extended Kalman filter and demonstrate that the resulting quasi-maximum likelihood estimator performs well in Monte Carlo simulations. We apply the new surface model to daily Henry Hub natural gas futures contracts between 2020-2024. Estimation results reveal that inclusion of time-varying seasonality significantly enhances in-and-out-of-sample fit, while the MXVN distribution improves the stability of dynamic factors. Furthermore, we showcase three practical applications of the proposed framework: forecasting, scenario generation, and profit and loss attribution to seasonal risk premiums. This framework offers a powerful new approach for pricing and risk management of complex commodity derivatives.

Yonas Khanna is a Quantitative Analyst in the xVA-CCR risk model development team at ING Bank. He joined ING in 2019, initially as an MSc thesis intern within the same team. Alongside his role at ING, he is also pursuing a part-time PhD in Financial Econometrics at the Vrije Universiteit Amsterdam. His PhD research is conducted in close collaboration with ING and focuses on high-dimensional filtering theory for generating synthetic financial market data for pricing and risk management applications in presence of missing and illiquid data.