Geometric Deep Learning of Volatility Surfaces

Options are financial instruments that are used to manage risk, hedge investments, and speculate. The value of these options depends on the price of the underlying asset and a multitude of different variables. As a result, pricing models can become complex, requiring infeasibly expensive routines or simulations to be run to price a single option. One reason this procedure can be slow is that the model’s parameters need to be tuned to the market’s current conditions, reflected by an implied volatility surface (IVS), which gives the value of options with different parameters. While the IVS has been researched extensively, it is still not understood well. We propose the use of deep learning to better understand the IVS, and plan to release our models to the public for future research. Riskfuel creates options pricing tools using deep learning, and will be using the model for accelerating pricing.

Faculty Supervisor:

Andreas Veneris

Student:

Nicholas Fung

Partner:

Riskfuel

Discipline:

Engineering - computer / electrical

Sector:

Professional, scientific and technical services

University:

University of Toronto