Deep Calibration With Artificial Neural Network: A Performance Comparison on Option Pricing Models

Since the seminal work of Black and Scholes (1973) and Merton (1973), the Black-Scholes model has remained the most fundamental model for option pricing. However, its restrictive assumptions, such as constant volatility or Geometric Brownian Motion (GBM), have been criticized for not reflecting the empirical characteristics of financial markets. Many subsequent models have since been proposed to relax the assumptions of the Black-Scholes model. One successful approach is employing stochastic volatility under the generalized autoregressive conditional heteroskedastic (GARCH) framework. The early attempt was introduced by Engle and Mustafa (1992) focusing on implied conditional volatilities. Subsequently, Duan (1995) developed a more rigorous framework of the GARCH option pricing model using the locally risk-neutral valuation relationship that one-period ahead conditional variance remains constant under both the risk-neutral measure and the physical measure.