Forecasting the option implied volatility (IV) surface is difficult with standard time-series models because of its time-varying granularity. We propose a new two-step real-time sequential forecasting framework. The first step fits the daily surface and can accommodate any underlying specification for option prices or IVs, including dynamic option-pricing models, nonparametric methods, and machine-learning techniques. In the second step, we sequentially estimate a dynamic IV model using an updating rule. Our framework can accommodate large datasets and high data frequencies. An empirical application on S&P 500 IV surfaces shows that our approach significantly outperforms random-walk forecasts.
The rapid growth of the global derivatives markets over the past few decades continues unabated. Option markets are especially dynamic, with a record 108.2 billion contracts traded and/or cleared in 2023, a 98.4% increase from 2022, and 90% of this trading volume consists of equity and index options. New and improved techniques for modeling and forecasting equity and index option prices are therefore badly needed. However, the literature on option pricing is increasingly fragmented. On the one hand, researchers have studied dynamic models with stochastic volatility and jumps that are extensions of the seminal Black-Scholes model. While these models offer many valuable insights, they are notoriously difficult to implement and their estimation is time-consuming. It is therefore extremely challenging to implement them recursively in real time. Because of this complexity, an alternative approach relies on parametric and nonparametric techniques that directly model the implied volatility (IV) surface, see for instance the ad-hoc Black-Scholes method of Dumas et al. (1998) and the kernel smoother used by OptionMetrics (2022). Recently, more sophisticated machine learning methods have been proposed to model the IV surface. These techniques use option characteristics as features and differ with respect to the nonlinear functions used for predicting IVs. Meaningful comparisons of these very different models are therefore a high priority, as well as frameworks that allow us to compare existing and new models.
You can now read the full whitepaper at the link below
