Abstract: We study the critical issue of handling missing entries in hedge fund data. We introduce a deep learning approach, the BRITS, for recovering data for fund returns and 23 fund predictors. We compare its performance with popular imputation methods, such as the cross-sectional mean and singular value thresholding. BRITS' ability to capture information from past and future values in time series and the whole cross-section of observations yields the highest imputation fidelity in our simulations. The recovered information improves predictions of nonlinear and linear methods. At the same time, it helps to select top-performing funds that earn significant out-of-sample annual alphas of 13.4% net of all costs. 

Presentations: MFA 2026 (Chicago, scheduled), Paris December Finance Meeting 2025 (Paris), FMA 2025 (Vancouver), SFA 2025 (Orlando), Alpine Finance Summit 2025 (Grenoble), 32 Finance Forum (Spanish Finance Association, AEFIN) (Pamplona), University of Edinburgh 2025.

Abstract: We select equity pricing factors using second-order stochastic dominance (SSD) rather than the Sharpe ratio. Drawing from 177 candidate factors spanning prominent models and the factor universe of Jensen et al. (2023), we apply the Sparse Spanning methodology of Arvanitis et al. (2025) to identify the smallest set such that no risk-averse investor would benefit from adding another factor. The procedure converges to a ten-factor model whose utility loss is indistinguishable from zero across all subperiods. The selected factors shift across regimes: momentum and growth dominate before 2000, while quality, value, and investment emerge after the dot-com bust and persist through the post-crisis period. Out-of-sample, the model achieves the highest Sharpe ratios and the lowest pricing errors in the most recent subperiods, and optimal portfolios formed on its abnormal returns mimic the risk-free rate. The Harvey and Liu (2021) lucky factors test confirms that the SD-Sparse factors provide incremental explanatory power beyond every benchmark, including adaptive LASSO and RP-PCA.

Abstract:  We propose an information-based multiple test, namely the family-wise error rate plus (fwer+) method, and apply it to evaluate hedge fund performance. This method ensures low risk of selecting any non-performing funds, yet allows significance thresholds to vary by informative covariates to accommodate investors’ conditional expectation. Using a large set of hedge funds and informative covariates, we show that fwer+ is sufficiently powerful to identify a substantial number of outperforming funds each year. Our portfolios consisting of in-sample outperforming funds beat passive benchmarks and portfolios formed by unconditional multiple tests, and generate significant alphas based on various factor models (up to 7% per annum). We further show that hedge funds’ exposures to macroeconomic variables play a dominant role as informative covariates and may explain fund managers’ skill.

Presentations: International Finance and Banking Society 2025, 31 Finance Forum 2024 (Spanish Finance Association, AEFIN) (Tenerife), EFMA 2024 (Lisbon)