The Effect of Regression Methods on the Performance of Options Pricing using Machine Learning

The pricing of financial derivatives is one of the most important issues in financial markets. Due to the huge computational complexity, highly dependence of model and parameter of the traditional pricing model based on Monte Carlo simulation, it has brought huge challenges to complex option pricing and risk management. This paper proposed a data-driven pricing method for non-parameter models based on regression methods. By constructing training sets and test sets, the three most common machine learning methods, namely support vector regression (SVR), Gaussian process regression (GPR) and kernel ridge regression (KRR), were employed to price barrier options and lookback options. Which were compared with pricing with Heston model. It was found that the dependence of the option price on the parameters predicted by the non-parametric model was much lower than that of the Heston model, and the convergence of the pricing accuracy and time cost were very excellent and smaller. A cross-comparison with different machine learning methods shown that the predictive performance of the GPR model was better than the other two models for the pricing of barrier and lookback options. This option pricing, which does not depend on parameter conditions, has important significance for a wider range of option pricing.