Quantile Regression Forest for Value-at-Risk Forecasting Via Mixed-Frequency Data

In this paper we introduce the use of mixed-frequency variables in a quantile regression framework to compute high-frequency conditional quantiles by means of low-frequency variables. We merge the well-known Quantile Regression Forest algorithm and the recently proposed Mixed-Data-Sampling model to build a comprehensive methodology to jointly model complexity, non-linearity and mixed-frequencies. Due to the link between quantile and the Value-at-Risk (VaR) measure, we compare our novel methodology with the most popular ones in VaR forecasting.