A0636
Title: Calibrating option pricing models using neural networks and population-based optimization methods
Authors: Antonio Santos - University of Coimbra (Portugal) [presenting]
Jose Luis Esteves dos Santos - University of Coimbra (Portugal)
Margarida Biscaia Caleiras - University of Coimbra (Portugal)
Abstract: In finance, model calibration is a crucial task that ensures that financial models accurately reflect market conditions, reducing the risk of decisions based on unreliable information. However, this calibration process is often computationally intensive and time-consuming, especially when dealing with complex models. To address these challenges, neural networks have emerged as a promising approach for developing more efficient option pricing methods, consequently allowing the utilization of algorithms that enhance the calibration process. The aim is to present a novel implementation and comparative analysis of the performance of two types of neural networks, feedforward neural networks (FNN) and long short-term memory networks (LSTM), in solving the Heston model calibration problem. A two-step calibration approach is employed, using neural networks to approximate the pricing function and significantly reduce calibration time. Numerical experiments demonstrate that LSTM networks, particularly when combined with a convergent variant of the differential evolution algorithm, can improve calibration accuracy compared to FNNs.
