No-Transaction Band Network: A Neural Network Architecture for Efficient Deep Hedging

Abstract

Deep hedging is a versatile framework for computing the optimal hedging strategy of derivatives in incomplete markets. However, it is subject to the action-dependence problem impeding efficient training because the appropriate hedging action at the next step depends on the current action. To overcome this issue, the authors leverage a no-transaction band strategy, an existing technique that provides optimal hedging strategies for European options and exponential utility. The authors theoretically argue this strategy to be optimal for a wider class of utilities and derivatives, including exotics. Based on the result, the authors propose a no-transaction band network, namely, a neural network architecture that facilitates fast training and precise evaluation of the optimal hedging strategy. Moreover, the authors experimentally demonstrate that, for European and lookback options, their architecture rapidly attains a better hedging strategy compared with a standard feed-forward network. The findings thus have important implications for the practical applications of deep hedging.