A Toolkit for Exploiting Contemporaneous Stock Correlations with Kazuhiro Hiraki
Journal of Empirical Finance, Volume 65, January 2022, Pages 99-124.
Stock correlations are important in portfolio optimization problems (see DeMiguel etal., 2014). We exploit stock correlations using a newly developed machine learning tool, OWL, which: 1) encompasses a LASSO setup that encourages sparsity in stock positions; 2) identifies correlated stocks contemporaneously by assigning similar positions to stocks; 3) enables bespoke constraints. We develop algorithm to allow lower and/or upper bounds for stocks positions if investors have prior beliefs. We compare OWL-(un)constrained strategies with some popular existing strategies in five different asset classes. We find strong evidence that OWL related strategies outperform other candidates when stocks are correlated. We show that OWL component exploits stock correlations while constraints set a threshold to prevent stocks from receiving extreme positions. In particular, while OWL regularised portfolio receives very similar positions to (but not the same as) the 1/N strategy (DeMiguel etal., 2009), it outperforms 1/N in terms of both out-of-sample Sharpe ratio and turnovers, significantly.
Journal of Empirical Finance, 77, March 2024
Abstract: Factor correlation is an important consideration when selecting factors for explaining the cross-sectional asset returns, and ignoring it often compromises the robustness and validity in standard methods for such an exercise. Yet, despite its importance, there is a lack of vigorous discussion about it in the related literature: what the implications are and what we can do about it. This paper investigates high-dimensional factor models for cross-sectional asset returns with a focus on robust estimation when factors are (highly) correlated. We utilize the stochastic discount factor (SDF) and combine it with a newly developed Machine Learning method (Figueiredo and Nowak, 2016) to select factors and to disentangle correlated factors without imposing structural assumptions. This method can identify highly correlated factors by assigning them with similar coefficients while simultaneously shrinking off useless/redundant ones. We develop asymptotic properties for this estimator with relaxed assumptions and show that it is consistent under mild conditions. Empirically, we illustrate that our method is robust with correlated factors and consistently identifies that the ‘market’ factor is the most important factor for cross-sectional asset returns, while other benchmarks (such as the LASSO, the Elastic-Net and the Fama-MacBeth regression) are adversely affected by factor correlations, rendering the ‘market’ factor redundant. In addition, we demonstrate that our method can be used for factor investing and we find ample evidence that hedged portfolios using our method outperform portfolio strategies based on other benchmarks in an out-of-sample framework.
A Sorted Penalty Estimator: Inference for a Correlation-Robust Shrinkage Method (submitted) with Marcelo Medeiros [paper] [slides]
Presentation: ASSA 2021, 33rd EC^2 2022, AMES2023, NASM2024
Variable correlations present significant challenges for a wide range of LASSO-type shrinkage methods in big data modeling. This paper introduces a correlation-robust shrinkage estimator, advancing both theoretical and practical aspects of high-dimensional estimation. We establish the (non-)asymptotic properties of this estimator under relaxed assumptions, including a mixing condition and allowance for heavier tails beyond the typical sub-Gaussian setting. Additionally, we demonstrate model selection consistency under mild conditions. We further propose a de-biased version of the estimator, proving its asymptotic normality. Simulated data reveal that the de-biased estimator outperforms traditional benchmarks. In an empirical application, we employ this de-biased estimator to identify key Economic Policy Uncertainty (EPU) factors that explain inflation levels. Our findings suggest that news-based EPU factors play a crucial role in explaining CPI dynamics.
A New Perspective on Cross-Sectional Factors in Asset Pricing with Xiao Xiao
Presentation: Erasmus University
Option-Implied Expected Returns and the Construction of Mean-Variance Portfolios with Kazuhiro Hiraki
Presentation: OptionMetrics 2019
Learning the High-freqency Forex Rate: a Machine Learning Approach with Ian Marsh
Online Learning with Factor Selection
Leveraging Large Language Models for Inflation Prediction with Marcelo Medeiros and Samuel Efraim
Time-varying Parameter Models: A New Approach with Marcelo Medeiros