Robust Inference in Large Panels and Markowitz Portfolios

Abstract

We propose a general framework for testing the significance of parameters in large panels of multiple linear regression models, focusing on mean-variance spanning (MVS) tests. The proposed methodology is versatile and applicable even when the number of equations is large, requiring only stationary data, and allows the number of regressors to grow asymptotically toward the sample size. Monte Carlo simulations demonstrate that the testing procedure maintains correct size and power, even when residuals exhibit asymmetry, fat-tails, serial correlation, and GARCH effects, outperforming existing methods. We apply the methodology to assess whether including blue-chip stocks from the U.S., Europe, and Switzerland enhances each country's domestic meanvariance efficient frontier. The findings suggest that the benefits of international diversification depend on economic conditions and vary by country, with the rejection of the MVS hypothesis linked to variance reduction within the domestic global minimum-variance portfolios.

Keywords: Mean-variance spanning tests, Robust Student-t tests, Large scale inference JEL codes: B23, C12, C52

JEL Classification: B23, C52, C12