Classical Option Pricing Theory and Extensions to Deep Learning

Specialeforsvar: Peter Pommergård Lind

Titel: Classical Option Pricing Theory and Extensions to Deep Learning

Abstract: The concepts for option pricing theory are presented and closed form solutions are provided in special cases. The options with no closed form solution are investigated through numerical methods, where both the binomial lattice model and LSM will be presented assuming the underlying Black-Scholes theory. Deep learning is then investigated to look for improvement of the existing methods, where we look specifically at the MLP regression. Our numerical study did not find any improvement in using MLP I instead of LSM. The MLP II was very fast, but lacks the accuracy of the classical methods. Therefore, the MLP II could be beneficial in some circumstances where speed weighs more than precision. In spite of the low accuracy for both deep learning methods, we believe MLP I and II can become an alternative to the classical methods with further investigation.

Vejleder: David G. Skovmand
Censor:    Mads Stenbo Nielsen, CBS