A0260
Title: Augmented LASSO with textual analysis in the financial market
Authors: Shuyi Ge - University of Nankai (China) [presenting]
Abstract: An augmented lasso (ALasso) framework that integrates textual information is proposed to enhance asset return prediction. Specifically, firm linkages and word-level sentiment are extracted from news reports, and these signals are embedded into the Lasso estimation process. Firm linkage strength is used to impose discriminative penalties across firms, enhancing cross-firm predictability, while word sentiment serves as a directional prior to constrain the signs of textual predictors. The oracle properties of ALasso are theoretically established under varying levels of news information quality. Empirically, ALasso demonstrates superior performance in predicting returns in the A-share market.
