Machine Learning Methods in Derivative Pricing and Hedging

Specialeforsvar: Pedro Duarte Gomez

Titel:  Machine Learning Methods in Derivative Pricing and Hedging

Abstract: The purpose of this article is to expose in a new light, the top-of the-art machine learning methods in derivative pricing and hedging. The current practice, both in academia as well as in the industry, of using data-driven models to price and hedge derivatives products has gained momentum, particularly relying on the neural networks predicting capabilities. However, there is no exposition on the different models, with a single theory that allows for comparison both from a theoretical as well as from an implementation perspective, the risk-neutral valuation methods and the recent deep hedging method.
In this thesis, a powerful theory is built, supported by functionalanalytical results in order to see how financial models can be constructed and how the machine learning practices, such as neural networks, can be incorporated into that modelling, further showing the latter can be reduced into an understandable mathematical representation. This theory will be illustrated by a simulation experiment, whose results point towards the advantage of deep hedging methods in more challenging contexts.

Vejleder: David Glavind Skovmand
Censor:    Nina Lange, DTU