Pricing American Derivatives by Machine Learning

Specialeforsvar: Leander Rosenkranz Klarskov

Titel: Pricing American Derivatives by Machine Learning

Abstract: In this thesis, we consider pricing American options as an optimal stopping problem. We introduce the necessary optimal stopping theory to solve the problem. To numerically approximate the solution, two simulation methods are introduced and replicated, the least squares Monte Carlo and the machine learning approach of neural networks. We test the approaches on two problems: the pricing of an American option in a BlackScholes market and a Heston market. Finally, we test the flexibility and high-dimensional capabilities of the neural network approach by employing it to price a Bermudan max call option in a multi-dimensional Black-Scholes market. In all three cases, the applicable methods produce accurate results. In high-dimensional problems, the neural network shines, albeit at the expense of some speed.

Vejleder: Jesper Lund Pedersen
Censor: Mads Stenbo Nielsen, CBD