Boosting the Anatomy of Volatility

Risk and, thus, the volatility of financial asset prices plays a major role in financial decision making and financial regulation. Therefore, understanding and predicting the volatility of financial instruments, asset classes or financial markets in general is of utmost importance for individual and institutional investors as well as for central bankers and financial regulators. In this paper we investigate new strategies for understanding and predicting financial risk. Specifically, we use componentwise, gradient boosting techniques to identify factors that drive financial-market risk and to assess the specific nature with which these factors affect future volatility. Componentwise boosting is a sequential learning method, which has the advantages that it can handle a large number of predictors and that it-in contrast to other machine-learning techniques-preserves interpretation. Adopting an EGARCH framework and employing a wide range of potential risk drivers, we derive monthly volatility predictions for stock, bond, commodity, and foreign exchange markets. Comparisons with alternative benchmark models show that boosting techniques improve out-of-sample volatility forecasts, especially for medium- and long-run horizons. Another finding is that a number of risk drivers affect volatility in a nonlinear fashion