Predicting Institutional Investment Trends from SEC Form 13F Filings Using Machine Learning

The Securities and Exchange Commission (SEC) Form 13F filings provide insights into the investment strategies of large institutional investors, offering a quarterly snapshot of their holdings. This research explores the predictive modelling of institutional asset values based on historical Form 13F filings from 2013 to 2024. Machine learning techniques, including Linear Regression, Decision Trees, Random Forests, and XGBoost, are applied to predict the total market value of institutional holdings using engineered financial metrics and firm-level attributes. Extensive feature selection and transformation techniques, including log transformations and scaling, are utilized to optimize the model performance. Experimental results demonstrate that tree-based models outperform linear regression, with Random Forest and XGBoost achieving the highest predictive accuracy. Feature importance analysis reveals that an institution's maximum- and minimum-security values are the most influential predictors. The findings contribute to quantitative finance research by providing an automated approach for analyzing institutional investment patterns and estimating portfolio valuations. The proposed methodology offers practical implications for financial analysts, hedge funds, and market researchers aiming to enhance investment decision-making.