A Network and Machine Learning Approach to Factor, Asset, and Blended Allocation

Abstract

The main idea of this article is to approach and compare factor and asset allocation portfolios using both traditional and alternative allocation techniques: inverse variance optimization, minimum-variance optimization, and centrality-based techniques from network science. Analysis of the interconnectedness between assets and factors shows that their relationship is strong. The authors compare the allocation techniques, considering centrality and hierarchal-based networks. They demonstrate the advantages of graph theory to explain the advantages to portfolio management and the dynamic nature of assets and factors with their “importance score.” They find that asset allocation can be efficiently derived using directed networks, dynamically driven by both US Treasuries and currency returns with significant centrality scores. Alternatively, the inverse variance weight estimation and correlation-based networks generate factor allocation with favorable risk–return parameters. Furthermore, factor allocation is driven mostly by the importance scores of the Fama–French–Carhart factors: SMB, HML, CMA, RMW, and MOM. The authors confirm previous results and argue that both factors and assets are interconnected with different value and momentum factors. Therefore, a blended strategy comprising factors and assets can be defensible for investors. As argued in previous research, factors are much more overcrowded than assets. Therefore, the centrality scores help to identify the crowded exposure and build diversified allocation. The authors run LASSO regressions and show how the network-based allocation can be implemented using machine learning. TOPICS:Factor-based models, portfolio theory, portfolio construction Key Findings • The authors compare network-based asset and factor allocations and blended strategies and argue that network analysis can be used to derive allocation, monitor interactions, and provide additional layers of risk control mechanisms. • The authors find that factors and assets are strongly interconnected. Investors should pay close attention to currencies and the Fama–French–Carhart factors (RMW, HML, CMA) because they have large centrality scores. • Using machine learning and predictive models, investors can find asset and factor allocation solutions. The authors argue that factor exposure is desired within asset allocation.