Forecasting Inflection Points: Hybrid Methods with Multiscale Machine Learning Algorithms

This paper investigates hybrid time series forecasting models, which are based on combinations of ensemble empirical mode decomposition and least squares support vector machines. Several algorithms are considered: the genetic algorithm, the grid search, and particle swarm optimization. Theoretical guarantees of prediction accuracy are tested with sine curves. From a numerical testing perspective, we are interested in showing the superiority of one approach to another based on theoretical prediction and time series applications in finance (S&P 500), commodities (WTI oil price), or cryptocurrencies (Bitcoin). The superiority of hybrid models to soft- and hard-computed models is further assessed through a ‘horse race’ and trading performance, as well as through fine-tuning of the algorithms.