Deep Hedging of Derivatives Using Reinforcement Learning

  • Prof. John Hull, Senior Fellow, GRI and Joseph L. Rotman, School of Management, University of Toronto
  • Jay Cao, Joseph L. Rotman, School of Management, University of Toronto
  • Jackie ChenJoseph L. Rotman, School of Management, University of Toronto
  • Zissis PoulosJoseph L. Rotman, School of Management, University of Toronto
Analyst working with Business Analytics and Data Management System on computer to make report with KPI and metrics connected to database. Corporate strategy for finance, operations, sales, marketing

OVERVIEW:

One of the areas in finance where machine learning is having an impact is the hedging of derivative portfolios. If there were no transaction costs, one strategy would be to zero out all the Greek letters (delta, gamma, vega, etc.). But in practice, a trader must follow a partial hedging strategy that reflects trade-offs between risk reduction and transaction costs.

In its first foray into this area, Rotman’s research center, FinHub, considered only delta hedging. It showed that, in both a Black-Scholes environment and a stochastic volatility environment, it is possible to use reinforcement learning to develop strategies reflecting a specified trade-off between the mean and standard deviation of profits. An optimal delta-hedging strategy which included transaction costs is derived using a machine learning approach. Insights into a number of key framework decisions are discussed in the following paper.

Overview provided by:
John Hull
Senior Research Fellow, Global Risk Institute
Maple Financial Group Professor in Derivatives and Risk Management, University of Toronto
Academic Director, Rotman Financial Innovation Hub
February 2022