Penetrating the fog of US stocks and predicting stock returns

TSAI

Stock return prediction is a core issue in asset pricing research. Based on the Fama-French five-factor model framework and combined with machine learning algorithms, the TSAI platform has built a multi-factor prediction system suitable for the US stock market, providing investors with scientific stock selection decision support.

1. Factor system construction

On the basis of traditional valuation, scale, and profitability factors, 20 new quantitative factors are added, covering alternative data such as institutional position changes, option implied volatility, and social media sentiment index. Through factor validity tests (t statistics, IC_IR values), 12 core factors are screened out, among which the analyst expectation revision factor has an information coefficient (IC) of 0.32 in predicting the probability of exceeding expectations in quarterly financial reports.

2. Prediction model development

The gradient boosting tree (XGBoost) is used to build a nonlinear prediction model, and the parameter settings are optimized through cross-validation. The model input contains factor data for the past 12 months, and the output is the expected return ranking of stocks in the next 3 months. Backtesting shows that the model has an annualized excess return of 11.8% for the top 20% high expected return portfolios in the S&P 500, with a Sharpe ratio of 1.57.

3. Practical application scenarios

The platform embeds the prediction results into the intelligent stock selection system, allowing users to set screening conditions based on risk preferences. In the 2025 AI sector investment, the high-growth stock portfolio selected by the model has a quarterly increase of 42%, significantly better than the industry benchmark. At the same time, the system provides factor exposure analysis to help investors understand the risk sources and return drivers of the portfolio.