Deep Optimal Stopping

Specialeforsvar: Andreas Langhoff

Titel: Deep Optimal Stopping 

Abstract: The aim of this thesis is to use Deep Learning theory to solve optimal stopping problems, where the underlying process is a discrete Markov chain. The optimal stopping rule is learned by several feedforward neural networks, where the neural networks are trained using batches ofMonte Carlo simulations of the underlying process. Consequently, when new Monte Carlo samples are obtained, the samples will directly adapt the learned stopping rule, and give an approximated value of the associated value function. The thesis will cover the theory concerning Deep Learning and Deep Optimal Stopping. Lastly the thesis will cover examples of determining the arbitrage-free price of the Bermudan maxcall option, the American put option and the surrender option in a life insurance contract using the deep stopping algorithm. Keywords: Deep learning, Optimal stopping, Bermudan option, American put option, Surrender option, Life insurance.

Vejleder: Jesper Lund Pedersen
Censor: Jeppe Woetmann Nielsen, Akademiker Pension