FX Option Hedging with Deep Learning

Abstract

The foreign exchange market, beaming with frequent trade and international risk mitigation tactics, is victim to derivative pricing and hedging disparities. Given the advantageous applications of forex options in minimizing foreign exchange rate risk, academics and investors are looking to minimize portfolio hedging error. In this thesis, we study neural networks as an estimation tool for forex option hedging, and develop our own network hedging model. This network is trained to minimize hedging error and directly outputs the hedging ratio, rather than the pricing error. Applied to GBM-simulated G10 foreign exchange spot rates, the network is able to, in most cases, reduce the absolute hedging error of the Garman-Kohlhagen benchmark significantly. The findings of this research offer insight into the application of feed-forward ANNs to the forex option hedging problem, and yield promising results for the future of hedging foreign exchange rate risk.