Improved inference for fund alphas using high-dimensional cross-sectional tests

Abstract

Abstract The traditional fund-by-fund alpha inference suffers from various econometric problems (e.g., cross-sectional independence assumption, lack of power, time-invariant coefficient assumption, multiple-hypothesis-testing). Recognizing the panel nature of fund industries, we tailor four high-dimensional cross-sectional tests to shed light into both the zero-alpha hypothesis and ratio of non-zero alphas. Particularly, we augment Gagliardini et al. (2016) with a time-varying alpha estimator. Our results reject the zero-alpha joint hypothesis as the statistical significance of alphas is too high to be explained by luck. After controlling for luck, our empirical studies show that the power enhancement helps to identify a large portion of significant fund alphas, which cannot be achieved using the usual Wald test. Meanwhile, the time-varying approach shows that fund alphas diverge during the late 2000s Global Financial Crisis, which cannot be observed using the time-invariant model. Overall, relative to the literature, we draw a more accurate and complete picture, and provide several powerful tools for future research.