Machine Learning Projects
Machine Learning Applications in Trade and Industrial Policy
This section presents a series of machine learning applications designed to analyze U.S. automotive production and evaluate the economic effects of Section 301 tariffs. These projects complement the econometric analysis in the research papers by incorporating predictive modeling, counterfactual estimation, and policy simulation.
The objective is to demonstrate how modern machine learning techniques can enhance empirical trade analysis, particularly in high-frequency macroeconomic environments.
Forecasting Motor Vehicle Production
This project applies machine learning methods to forecast U.S. motor vehicle production using macroeconomic indicators from FRED. The model captures nonlinear relationships between industrial production, labor markets, prices, and interest rates.
The results illustrate how machine learning can improve short-term forecasting performance relative to traditional linear models. The model achieves strong out-of-sample performance, demonstrating the ability of machine learning methods to capture short-run fluctuations in industrial production.
Counterfactual Analysis of Section 301 Tariffs
This project constructs a machine learning-based counterfactual to estimate how U.S. motor vehicle production would have evolved in the absence of Section 301 tariffs. The approach leverages predictive models trained on pre-policy data to generate a synthetic no-tariff trajectory.
The divergence between observed and predicted production provides an estimate of the aggregate impact of trade policy. The results indicate a measurable divergence between observed production and the no-policy counterfactual following the implementation of Section 301 tariffs.
Policy Simulation under Section 301 Tariffs
This project extends the counterfactual framework by simulating alternative policy scenarios. Instead of estimating a single treatment effect, the model evaluates how different tariff intensities would affect production outcomes.
This approach allows for a more flexible and forward-looking analysis of trade policy, highlighting the potential range of economic impacts under different policy regimes. The simulation highlights the sensitivity of production outcomes to tariff intensity, providing a flexible framework for policy evaluation.
Summary
Together, these machine learning applications complement the econometric framework developed in the main research papers. While traditional causal inference methods provide identification of policy effects, machine learning enables:
- flexible functional forms
- improved predictive accuracy
- counterfactual construction
- forward-looking policy simulations
This integration of econometrics and machine learning reflects a modern approach to applied economic research in trade and industrial policy.