Comparative analysis of option pricing models: deep neural network versus traditional approaches

Details

Supervisors

Vrins, Frédéric

Faculty

Louvain School of Management

Degree label

Master [120] en sciences de gestion, à finalité spécialisée

Abstract

enThis thesis compares the ability of three option pricing approaches, the Black-Scholes model, the Heston stochastic volatility model and a neural network, to approximate the implied volatility surface of S&P 500 European call options over the period 2007-2026. The models are evaluated in implied volatility space on several million observations, across market regimes and moneyness segments, with performance rankings validated by a statistical test, the Model Confidence Set. The results establish the dominance of Heston: the model records the lowest errors in both calm and stress regimes. The comparison between Black-Scholes and the neural network reveals no overall winner but a clear division across moneyness: the parametric model prevails near-the-money, while the network dominates at the extremes, most clearly under stress, where the misspecification bias of a constant volatility is most pronounced.