Multi-criteria classification for pricing European options

Abstract

This paper builds a novel multi-criteria, non-parametric classification framework in order to improve the accuracy of pricing European options. The proposed approach is based on classifying financial options according to their implied volatility, time to maturity and moneyness. Using a recent data set for the daily S&P 500 index call options traded in 2012, the multi-criteria modular neural network model demonstrates its superior out-of-sample pricing performance relative to competing parametric and non-parametric models. By observing the model's pricing errors across various option types, the analysis provides additional insights into pricing biases and stresses the importance of selecting appropriate classification criteria.

Unlocking the black box: Non-parametric option pricing before and during COVID-19

“…One of the pioneering papers in this area was Hutchinson et al (1994), followed by Qi and Maddala (1996), Garcia and Gençay (2000), Gençay and Qi (2001), Gençay and Altay-Salih (2003), and Gradojevic et al (2009). More recent work can be seen in, for example, Gradojevic (2016) and Jang and Lee (2019). The goal of such scholarly efforts is to harness the learning ability and flexibility of machine learning models to achieve better prediction accuracy than the classical (parametric) financial option models.…”

“…As the literature shows that models based on neural networks are able to price options more accurately than parametric alternatives such as the SV and SVJ approaches, we focus our attention on non-parametric methods(Gençay and Gibson 2009;Andreou et al 2008;Gradojevic 2016;Jang and Lee 2019).6 Ruf and Wang (2020) offer an excellent up-to-date review of the relevant literature on the applications of neural networks for option pricing and hedging.…”