Deep hedging; The future in the field of hedging

Specialeforsvar: Troels Sabroe Ebbesen

Titel:  Deep hedging;The future in the field of hedging

 Abstract: This thesis will investigate the field of hedging financial derivatives and how the use of deep neural network to approximate the optimal hedging strategy can improve previous results in this area. This work will thus mainly be build on the same framework as in Buehler et al. (2019). This is a very versatile framework as it does not assume anything about the underlying model and the same framework can thus be used for a variety of models for the underlying asset as well as for different types of derivatives.
In our particular case we will mainly work in the setup of the Heston model which incorporates stochastic volatility. The performance of these new deep hedging strategies will be tested in a variety of different scenarios including different types of derivatives as well as dealing with the case of market frictions, where no analytical solution to the hedging problem exists. This thesis shows that the deep hedging framework performs similarly to the analytical hedging strategies where these exist, whilst it is also able to find a useful hedging strategy in settings where no analytical solution exists.
The performance of the deep hedging strategies will be explored in more complicated scenarios including setups where the available data is of poor quality and for more complex derivatives. The deep hedging framework performance is reasonably good in many of these more complicated cases, but it does struggle in some of them. This suggest that further research in this area is still needed to refine the method.

Vejleder:  David Skovmand
Censor:    Mads Stenbo Nielsen, CBS