How can innovative data-driven approaches like reinforcement learning revolutionize risk management for financial institutions?
Financial institutions constantly deal with the challenge of managing risks tied to factors like interest rates, stock prices, and more. These risks, often unpredictable in nature, add complexity to the financial landscape. Managing risk is simpler when dealing with straightforward assets like stocks, where risks are typically linked to the asset’s price. However, complexities arise when dealing with financial derivatives like options, where risks are shaped by intricate non-linear relationships and unpredictable market changes.
Historically, the financial industry has relied on parametric models to understand financial variable dynamics. The Black-Scholes model, introduced in 1973 by Black, Scholes, and Merton, became renowned for its constant-volatility assumption.
Data Sciences Institute (DSI) members and University of Toronto professors Sebastian Jaimungal, (Department of Statistical Sciences, Faculty of Arts & Science) and John Hull (Joseph L. Rotman School of Management), propose a new frontier in financial risk management. Their aim is to develop alternative methods for quantifying and managing risk within financial institutions, utilizing reinforcement learning — a data-driven approach. Their strategies prioritize robustness to model misspecification and dynamic time consistency.
Jaimungal explains, “Thanks to the invaluable support from DSI, our team has achieved a significant milestone with the development of ‘FuNVol: A Multi-Asset Implied Volatility Market Simulator using Functional Principal Components and Neural SDEs. This research employs Legendre polynomials to represent the surface and employs neural stochastic differential equations, a form of stochastic evolution driven by neural networks, to capture its intricate dynamics. With DSI’s support, we’ve been able to delve deeper into understanding volatility surface dynamics and its implications for risk management.”
With the support of a DSI Catalyst Grant, this collaborative research team is working to better understand how volatility surfaces change using generative models. Their research has significant implications for risk assessment, risk management, and portfolio valuation, primarily benefiting financial institutions. Understanding the various ways volatility surfaces can evolve promises to enhance portfolio hedging strategies for financial institutions.
“The DSI Catalyst Grant program underscores our commitment to advancing the frontiers of data-driven research, and we are delighted to witness the significant progress it has facilitated,” says Gary Bader, associate director, research and software, at DSI.
Hull’s group is looking at the same challenge using variational autoencoders (VAEs), a model where latent factors determine option price location and spread.
What sets this research apart is the goal of incorporating these generative models as inputs into reinforcement learning algorithms. Their aim is to develop sophisticated strategies for mitigating risks within portfolios of financial options and toward more robust and effective risk mitigation strategies in an ever-evolving financial landscape.