ALTERNATIVE APPROACHES TO THE LEAST SQUARE METHOD IN AMERICAN OPTION PRICING

This paper introduces alternative methods to least square method (LSM) implemented by Longstaff-Schwartz in 2001 to enhance the pricing of American options with Monte Carlo Simulation. The goal is to provide evidence that alternative methods provide more precise pricing compared to least square method. The alternative methods include various machine learning (ML) algorithms classified as regression models: artificial neural networks (ANN), decision trees, support vector machines (SVM), stochastic gradient descent (SGD), isotonic regression, and Gaussian Process Regression (GPR). As a part of the calibration process, real market data used for the Merton jump diffusion model and CIR model. Finally, the paper compares errors with each ML algorithms to arrive to the most appropriate algorithm.