Measuring Firm Quality Using Machine Learning
Log in to collectAcademic paper
Measuring Firm Quality Using Machine Learning
Changyi Chen; Bin Ke; Qi Zhao
- Shanghai University of Finance and Economics
- SGNational University of Singapore
- ?National University of Singapore (NUS)
- ?Shanghai Univeristy of Finance and Economics
- ?Independent
- South China University of Technology
- Shenzhen University
- ?College of Economics, Shenzhen University
Strategy in a nutshell
This strategy uses machine learning, specifically XGBoost, to predict firm quality based on historical financial and accounting data. The predicted quality measure is then used to build stock portfolios. Data is carefully cleaned, standardized, and winsorized to reduce noise, while model hyperparameters are optimized using time-series cross-validation. Portfolios are formed annually, rebalanced monthly, and evaluated using standard error metrics. Risk is managed through diversification, position sizing, and stop-loss rules, ensuring the approach can be realistically applied in practice.
Economic rationale
The idea builds on value investing principles: buying quality companies at fair prices. While cheapness is easy to measure, firm quality is harder because it reflects future performance. Machine learning offers a better way to capture this forward-looking measure by analyzing detailed financial data and handling missing values. By improving how firm quality is estimated, XGBoost models can explain stock returns more accurately and help investors, regulators, and firms make better decisions. Ultimately, this method provides a stronger foundation for value investing and can generate higher returns.