MACRO FACTOR-MIMICKING PORTFOLIOS

Abstract

The estimation of risk factors and their replication through mimicking portfolios are of critical importance for academics and practitioners in finance. In this paper, we propose a general optimization framework to construct macro factor-mimicking portfolios (FMP) that encompasses existing portfolio mimicking approaches such as two-pass cross-sectional regression models (Fama and MacBeth, 1973) and maximal correlation approaches (Huberman et al., 1987, Lamont, 2001). We also incorporate potential empirical estimation improvements through machine learning methodologies. In our empirical application, we compare the ability of various FMP methodologies to replicate the characteristics of the two most well-known macro factors (growth and inflation). Our sample covers the period from January 1974 to April 2022, spanning different economic regimes. We use a set of investible assets representative of (developed market) cross-asset portfolios that can be invested through liquid and cost-efficient vehicles, such as futures/swaps derivatives or ETFs. Overall, the results prove the superiority of our machine learning macro-mimicking approach over the traditional FMP approaches, both in-sample and out-of-sample. We conclude by suggesting some potential future developments on the methodology.
JEL Classification: G11, D81, C60.