Scalable Probabilistic Inference for the Analysis of International Product Trade Profiles

Tamara Broderick, Electrical Engineering and Computer Science and Computer Science and Artificial Intelligence Laboratory (CSAIL)
In Song Kim, Political Science

The central goal of this project was to identify hidden economic and societal forces in international trade by developing scalable machine learning algorithms for the probabilistic inference of massive amounts of trade data. Probabilistic inference can establish the complex models required for these analyses, but in practice is often slow to run. Broderick and Kim used computational-statistical trade-offs to obtain the necessary run-time gains to make these analyses more practical.

“Inter-disciplinary collaboration was critical to developing probabilistic models that are based on substantive and theoretical knowledge of international trade,” reported Broderick and Kim. This research combined Broderick’s work in scalable, practical, and reliable probabilistic analysis, and the political economy and international trade expertise of Kim. Kim provided unique product-level trade data and identified empirical challenges in social sciences, while Broderick’s research group developed the computational models that directly address various methodological challenges that the data present.

Studies often ignore the variation in product-level trade for computational and methodological reasons. In this research, product-level trade data was used to identify latent global production networks, such as “how much country A’s steel and country B’s aluminum are required to produce cars in the U.S., and how do the input-output relations change across space and time.”

Broderick and Kim produced a measure of input-output relations at the product-level that offers improvements on today’s modern approach, which is limited by the use of highly aggregated and noisy input-output tables. In the future, they plan to use this measure to study the spillover effects of trade liberalization, or “trade disputes,” between pairs of other countries across the world.

This research project supported MIT CSAIL PhD candidate William Stephenson, who is in Prof. Broderick’s research group.