Quantitative Tactical Asset Allocation Using Ensemble Machine Learning Methods

Abstract

Beating the SP500 Index benchmark is a do-or-die among active portfolio managers. We propose a new method to add a 2-layer augmentation to relative strength and momentum based active portfolio management methods; first layer is to add a filtering mechanism to add a momentum filter in the recommendation engine and second is to include a multi level- multi layer machine learning method to integrate an ensemble model to decision making process. The ensemble model consists of gradient boosted decision trees and neural network models. Our initial results show that it is possible to beat the SP500 benchmark index by 600 basis points (in the calculations industry standard trading costs are included) as it is demonstrated by comparing the overall performance of the proposed method.

Keywords: Tactical Asset Allocation, Stocks, Bonds, Real Estate, Quantitative, Momentum, ETFs, GTAA, Machine Learning, Ensemble, alpha, neural network, CAGR

JEL Classification: C10, C50, E00, G10, G11