Selected working papers

• Jie Jian, Aaron Schein. Bayesian Poisson-Randomized Gamma Tensor Factorization with Application to International Trade Flows. [arXiv] [poster] 
  
  Overview:
  International trade data provide a rich window into global economic and political relationships, but they are also difficult to analyze: most country-pair-product-year flows are zero, while positive flows are highly skewed and vary greatly across products. We develop a Bayesian tensor factorization approach for this type of sparse, semi-continuous data, motivated by the structure of international trade. The model jointly captures whether trade occurs and how large it is when it does occur, while allowing dispersion to differ across product categories. By imposing a low-rank structure across exporters, importers, products, and years, the method uncovers latent multi-way dependence that cannot be recovered from standard bilateral analyses or lower-dimensional summaries. To make estimation feasible at scale, we combine variational inference with an augmented-data Monte Carlo algorithm. Applied to approximately 60 million trade flows, the approach reveals systematic patterns of dependence in the global trading system and provides a new way to study how trade relationships vary across countries, sectors, and time.