Forecasting the Uncertainty about VIX Futures Prices: VVIX Predictions with exogenous Drivers

Abstract

This study investigates the predictability of the VVIX, representing the volatility-of-volatility in the U.S. stock market, by leveraging both endogenous and exogenous variables within an LSTM network framework. We contrast the LSTM’s forecasting performance against the HAR, LASSO, and ARMA models from 2009 to 2022, focusing on 1-day and 21-days forecast horizons. Our findings reveal that the LSTM not only significantly surpasses the other models in terms of lower MSE and enhanced directional accuracy but also effectively incorporates exogenous factors to refine its predictions. Despite challenges in capturing extreme volatility events, the robust performance of the LSTM network underscores its potential in financial market volatility forecasting and risk management, highlighting its superior capability in detecting underlying patterns and non-linear relationships within market data.

Keywords: VVIX forecasting, volatility-of-volatility, LSTM

JEL Classification: C5, C22, C45, G17