Fast Approximate Modular Inference for Option Price Prediction

This study proposes a novel approximate Bayesian inference framework that can rapidly compute probabilistic forecasts of future option prices using three types of information sources: daily spot returns, high-frequency spot data, and option prices. The framework operates directly on theoretical option pricing models, without requiring explicit statistical models or likelihood functions for observed option prices. Empirical results based on the Heston option pricing model show that this method achieves prediction accuracy comparable to likelihood-based modular inference, with significantly improved computational efficiency, enabling real-time fast updates of predictive distributions.