Topic: Can Machines Learn Weak Signals?

Speaker: XIU Dacheng, Joseph Sondheimer Chair in Econometrics and Statistics, University of Chicago Booth School of Business

Time: April 28, 2025, 09:30

Venue: EMS 231

Abstract:

ln high-dimensional regressions with low signal-to-nolise ratilos，we assess the predictive performance ofseveral prevalent machine learning methods.Theoretical insights show Ridge regression's superioriy in exploitingweak signals,surpassing a zero benchmark. In contrast, Lasso fails to exceed this baseline, indicating its learninglimitations. Simulations reveal that Random Forest generally outperforms Gradiant Boosted Regression Trees whensignals are weak.Moreover，Neural Networks with l2-regularization excel in capturing nonlinear functions of weaksignals. Our empirical analysis across six economic datasets suggests that the weakness of signals, not necessarilfythe absance of sparsity, may be Lassd's major limitation in economic predictians.

Guest Bio:

Xiu Dacheng is the Joseph L. Sandheimer Distinguished Service Professor of Econometrics and Statistics at the University of Chicago Booth School of Business and a Research Associate at the National Bureau of Economic Research (NBER).

Professor Xiu serves as an editorial board member or editor for leading economics, finance, and statistics journals, including the Journal of Business & Economic Statistics, Journal of Financial Econometrics, Journal of Finance, Review of Financial Studies, Journal of the American Statistical Association, Management Science, and Journal of Econometrics. His research has been published in top-tier academic journals such as Econometrica, Journal of Political Economy, Journal of Finance, Review of Financial Studies, Journal of the American Statistical Association, and Annals of Statistics.