On the pricing of capped volatility swaps using machine learning techniques

Abstract

A capped volatility swap is a forward contract on an asset's capped, annualized, realized volatility, over a predetermined period of time. This paper presents data-driven machine learning techniques for pricing such capped volatility swaps, using unique data sets comprising both the strike price of contracts at initiation and the daily observed prices of running contracts. Additionally, the developed model can serve as a validation tool for external volatility swap prices, flagging prices that deviate significantly from the estimated value. In order to predict the capped, future, realized volatility, we explore distributional information on the underlying asset, specifically by extracting information from the implied volatilities and market-implied moments of the asset. The pricing performance of tree-based machine learning techniques and a Gaussian process regression model is evaluated in a validation setting tailored to the use of financial data.