Prediction Analysis of Maximum Sharpe Ratio Investment Portfolio Allocation Based on NLP Model
Year: 2026
Type: article
Abstract: This article takes major events of A-share listed companies as the research object, uses the FinBERT model to extract text emotions and risk attention features, constructs emotional momentum factors and risk exposure factors, and integrates them with traditional quantitative factors such as valuation and momentum to form multidimensional feature inputs. On this basis, it is combined with the Markowitz mean-variance framework to construct an investment portfolio optimization model with the goal of maximizing the Sharpe ratio. This article selects the China Securities Index (CSI) 300 constituent stocks from 2018 to 2023 as samples, conducts out-of-sample tests using a rolling window, and performs Integrated Circuit (IC) sequence analysis and the group return monotonicity test on text factors. The research results indicate that the emotional momentum factor has stable predictive ability and the Sharpe ratio of the investment portfolio after integrating text factors is significantly better than traditional quantitative models and benchmark portfolios. This article constructs an asset allocation process driven by both "text" and "quantitative" factors, providing empirical evidence for using alternative data to improve portfolio performance.
Author: Kexin Zhang
Institution: Southwestern University of Finance and Economics
Language: English
Subfield: Finance
Domain: Social Sciences
Open Access status: hybrid
OpenAlex ID: W7154040731
