Conditional Return Quantiles, Machine Learning, and the Implied Volatility Surface

Abstract

This paper evaluates the potential performance of machine learning algorithms to accurately estimate EUR/USD tail risk, using predictive information from the implied volatility surface. We conduct this analysis by estimating a broad set of machine learning models, alongside well-established quantile regression models, compute daily out-of-sample conditional quantile estimates, and evaluate the models on a set of economic loss functions. We find promising results, in that the machine learning models, in contrast to the traditional econometric models, benefit from increasing the dimensions of the feature space. Overall model rankings, however, are highly dependent on the choice of economic loss function.

Keywords: Finance, Tail risk, Implied volatility surface, Forecasting, Machine learning