Title: Tactical asset allocation with binary regression trees andforests
Authors: Juan Laborda - Carlos III (Spain) [presenting]
Ricardo Laborda - University of Zaragoza (Spain)
Abstract: Given the statistical characteristics of financial series, we propose a methodology that allows us to accurately forecast the relative returns of financial assets and to implement tactical asset allocation (TAA) strategies, as part of a basic portfolio management model (equities, bonds and cash). We will focus on that research topic which, with the implication of predictability in series of returns, allows us, on the basis of fundamental variables, to exploit the set of information that these offer in order to segment and classify homogeneous areas, on the basis of which we can predict higher returns from one asset relative to another. The initially proposed technique is single binary regression trees (classification and regression trees) that naturally extends to ensembles trees, applying methods like random forest. Once the forecast relative returns are calculated, a TAA system is developed from which we derive the structure of the optimal aggressiveness factors of the various tactical strategies, which allows us, using a benchmark portfolio, to calculate the weightings to hold in each of the assets.