Stata 19 introduced the h2oml suite of commands, which bring machine learning into Stata via H2O, a scalable machine learning platform. Users can fit gradient boosting machines and random forests for regression and classification, tune hyperparameters, evaluate model performance, and make predictions for unseen data. The suite also provides explainability tools—variable importance, partial dependence plots, ICE curves, and Shapley values—so predictive power comes without sacrificing interpretability. This talk introduces the workflow and shows how users can apply these methods from within Stata.

Many machine learning methods involve setting hyperparameters, which significantly impact predictive performance. Interactive calibration is a common method for determining optimal hyperparameters, while Bayesian optimization offers an alternative. This presentation introduces the basic framework of Bayesian optimization, its application in determining machine learning hyperparameters, and its Stata implementation.

Online news media, as a supervisory force, play a crucial role in promoting corporate green transformation. Based on big data from online recruitment from 2016 to 2023, this study constructs a corporate green transformation index to explore the impact and mechanism of online media supervision on corporate green transformation. The study finds that online media supervision can significantly promote corporate green transformation. Mechanism analysis shows that online media supervision mainly promotes corporate green transformation by increasing investor attention, strengthening government environmental regulatory constraints, and tightening bank credit supply. Heterogeneity analysis shows that the green transformation effect of online media supervision is more significant in nonheavily polluting industries, regions with lower environmental regulatory intensity, and small-scale enterprises. Further analysis indicates that the demand for new green positions in enterprises is mainly concentrated in digital skills-based positions, social task-based positions, and signaling-based functional positions, and the green transformation effect of online media supervision has significant spillover effects at both the enterprise and industry levels. The research conclusions provide important reference for how to effectively leverage informal institutional forces to promote corporate green transformation under the “dual carbon” target.

Rolling over maturing debt by borrowing new funds (rather than using development resources such as land transfer fees for substantive debt repayment) is a strategic behavior of officials competing for political achievements and shirking debt responsibility under the system of rotating appointments. This leads to the continuous accumulation of local debt and a persistent tension in preventing and resolving local debt risks. This presentation focuses on the unique scenario of “mayors being promoted to municipal party secretaries”, matching the issuance data of local government financing vehicle (LGFV) bonds by prefecture-level cities before and after the promotion to examine the impact of official rotation on the behavior of borrowing new funds to roll over debt. The study finds that because of the zeroing out of previous political achievements, municipal party secretaries promoted from mayors of other prefectures (more inclined to borrow new funds to roll over debt than municipal party secretaries promoted from local mayors) are more inclined to borrow new funds to roll over debt. In the later stages of their tenure, the moral hazard of debt shirking leads to a convergence of borrowing new funds to roll over debt behavior between the two types of municipal party secretaries. “Heavy borrowing and light repayment” drives the continuous accumulation of local debt in China. Unlike existing studies that focus on why local officials “borrow more”, this presentation discusses why they “repay less”, providing a new explanation for the mechanism of local debt accumulation in China.

This talk introduces the Stata ifete command, designed for estimation and inference of treatment effects in panel-data models with interactive fixed effects. As a major extension of two-way fixed effects, interactive fixed effects provide a flexible way to construct counterfactual predictions for causal inference and present point estimates of treatment effects as well as confidence intervals and p-values for inference, while covariates and nonstationary trends are allowed. I illustrate the use of the ifete command by revisiting examples of Hong Kong's economic integration with mainland China (Hsiao et al. 2012) and the California tobacco control program (Abadie et al. 2010).