Return Predictability from Industry Network Effects: Evidence from Rolling Window Adaptive Lasso

Abstract

The literature argues that industry network effects imply the predictability of industry returns, while the return predictability can estimate the industry network in reverse. The paper employs rolling window adaptive lasso regressions to test the robustness of return predictions, showing the inherent model instability of the industry network and raising caveats for using the adaptive lasso method to estimate the industry network with a long time series. The out-of-sample predictability is examined by creating long-short trading portfolios based on the predictions. With non-parametric tests, the paper discovers that the out-of-sample predictability exists but is primarily due to the momentum effect. The adaptive lasso portfolios perform significantly worse than the momentum portfolio before the adjustment of risks. Based on the multifactor models, the trading portfolios also fail to continuously generate significant alphas for risk-averse investors.

Keywords: return predictability, network effects, market efficiency, adaptive lasso

JEL Classification: D85, G12, G14, G17