A1308
Title: Online forecasting of unbalanced implied volatility surfaces
Authors: Arnaud Dufays - EDHEC Business school (France) [presenting]
Jeroen Rombouts - ESSEC Business School (France)
Kris Jacobs - University of Houston (United States)
Abstract: The daily option implied volatility surface is difficult to forecast with standard time series models because of its time-varying granularity. Approaches using option pricing models face a time-consuming estimation problem because realistic models require multiple latent factors. To address these challenges, a sequential surface forecasting approach is proposed that involves daily fitting combined with a dynamic model based on the parameter estimates as summary statistics of the previously implied volatility surfaces. The method works with any surface fitting method, such as option pricing processes, nonparametric methods, and machine learning models. In the empirical application to forecasting S\&P 500 implied volatility surfaces, it is found that nonparametric and machine learning approaches typically outperform advanced option pricing models. The dynamic model with a particular case called the Surface HAR model, is shown to generally lead to significant forecast improvements for all models.
