Superstar Firms Identification Using Machine Learning : Evidence from Chinese Manufacturing Firms

Superstar firms, the few firms that outperform the rest significantly in profitability and/or productivity, are increasingly important for driving industry growth or export patterns. It is thus critical to understanding the vital factors contributing to their success. However, it is also challenging because, arguably, a myriad of factors could be at play. Recent innovations in machine learning approaches and their increasing use in the economics literature offer a potential solution. Focusing on the Chinese textile manufacturing industry and leveraging machine learning approaches, we provide the first systematic evaluation of the critical factors that help explain the top manufacturing firms’ export success. In particular, we use two machine learning algorithms—random forest and extreme gradient boosting with feature selection—and compare them against two baseline models: (a) naïve; and (b) multiple linear regression. We find that machine learning algorithms could be used to help explain factors behind these superstar exporter firms, and random forest has the best performance among all models, with an average accuracy of 78.8% for predicting the top 200 firms and 77.5% for the top 10. We find the most significant features in the textile industry are the previous year’s market share, main business cost, intermediate input, and main business revenue