ML improves volatility forecasts by capturing option heterogeneity

Despite documented heterogeneity in volatility dynamics across the option surface, standard implied volatility forecasting models apply homogeneous parameters.

This Federal Reserve working paper introduces a machine-learning framework that uses regression trees to partition the surface along moneyness and maturity dimensions.

This identifies data-driven regions where distinct forecasting models perform best.

Extending the Surface Heterogeneous Autoregressive (SHAR) framework, the authors develop tree-based SHAR specifications that preserve interpretable structure while allowing model parameters to vary across the surface.

This approach substantially improves forecasting accuracy by adapting the model to regional heterogeneity across the surface and leveraging these localized dynamics.

The empirical analysis uses S&P 500 options.

Market segmentation and variation in investor clienteles across the option surface are well-documented.

Different clienteles, such as volatility sellers for short-dated options or institutional hedgers for near-the-money contracts, create distinct supply and demand patterns.

The empirical approach employs regression trees, using moneyness and maturity as splitting variables to recursively partition the option panel into more homogeneous subregions.

This tree-based local estimation scheme preserves the economic interpretation of underlying models while allowing parameters to adapt flexibly.

The Boosted Tree-SHAR model achieves the lowest out-of-sample forecast errors, reducing one-month-ahead RMSE by 13% versus the benchmark SHAR model.