Boosting Model-free predictions for econometric datasets

In this paper we propose some novel non-parametric prediction methods to perform short- and long-term aggregated forecasting for the log-returns of econometric datasets. The previous works in the regime of NoVaS model-free prediction are restricted to short-term forecasting only. Often practitioners and traders want to understand the future trend for a longer time into the future. This article serves two purposes. First it explores robustness of existing model-free methods for long-term predictions. Then it introduces, with systematic justification, some new methods that improve the existing ones for both short- and long-term predictions. We provide detailed discussions of the existing and new methods and challenge the new ones with extensive simulations and real-life data. Interesting features of our methods are that these entail significant improvements compared to existing methods for a longer horizon, strong volatile movements and shorter sample size.