Modelling global trade with optimal transport.

Area of Science:

• Economics
• Econometrics
• Machine Learning

Background:

• Global trade is influenced by complex factors beyond supply and demand, including transport costs, tariffs, and political-economic relations.
• Traditional gravity models struggle to capture these subtler drivers of trade due to reliance on explicit covariates.

Purpose of the Study:

• To develop a novel framework for modeling time-dependent trade costs using optimal transport and deep neural networks.
• To overcome limitations of traditional gravity models in capturing complex trade determinants.

Main Methods:

• Employed optimal transport and deep neural networks to learn a time-dependent cost function from data without predefined functional forms.
• Utilized a data-driven approach to model trade costs, allowing for natural uncertainty quantification.

Main Results:

• The proposed approach consistently outperformed traditional gravity models in accuracy.
• Demonstrated disproportionately higher increases in trade costs for low-income countries following the war in Ukraine's impact on wheat markets.
• Uncovered hidden patterns in global food and agricultural trade, free-trade agreements, trade disputes with China, and Brexit's impact on UK-Europe trade.

Conclusions:

• The novel framework offers a more accurate and nuanced understanding of global trade dynamics.
• Highlights the significant impact of geopolitical events on trade costs, particularly for vulnerable economies.
• Provides insights into the effects of trade policies and disputes, revealing patterns not evident from trade volumes alone.