Evaluating the Impact of Preferential Trade Agreements on Agricultural and Food Trade: New Insights from Natural Language Processing and Machine Learning

EVALUATING THE IMPACT OF PREFERENTIAL TRADE AGREEMENTS ON AGRICULTURAL AND FOOD TRADE: NEW INSIGHTS FROM NATURAL LANGUAGE PROCESSING AND MACHINE LEARNING

Sponsoring Institution

National Institute of Food and Agriculture

Project Status

ACTIVE

Funding Source

AFRI COMPETITIVE GRANT

Reporting Frequency

Annual

Accession No.

1028020

Grant No.

2022-67023-36399

Cumulative Award Amt.

$650,000.00

Proposal No.

2021-10827

Multistate No.

(N/A)

Project Start Date

Jan 1, 2022

Project End Date

Dec 31, 2025

Grant Year

2022

Program Code

[A1641]- Agriculture Economics and Rural Communities: Markets and Trade

Recipient Organization
UNIV OF CONNECTICUT
438 WHITNEY RD EXTENSION UNIT 1133
STORRS,CT 06269

Performing Department
Agriculture & Resource Econ

Non Technical Summary
This research will generate new knowledge regarding the formation of preferential trade agreements (PTAs), their impact on global trade, and the consequences for U.S. agricultural and food businesses and employment. To accomplish this goal, we will rely on modern statistical modeling techniques to thoroughly investigate the factors that influence the formation of PTAs. This analysis builds on newly collected PTA data captured with the help of Neural Machine Translation and Natural Language Processing systems. To determine factors that influence the formation of PTAs, we will adopt the Random Forest algorithm. This statistical analysis will provide new insights regarding the role of economic, social, and political factors in forming PTAs with agricultural and food provisions. We will use the newly created dataset to investigate the impact of PTA provisions on agricultural and food trade in the sectoral three-way gravity model context relying on an adaptation of the Prior least absolute shrinkage and selection operator to the Poisson pseudo-maximum likelihood estimator. This innovative machine learning approach will enable us to incorporate prior information, reduce over-fitting, and facilitate feature selection in a high-dimensional context. We will also assess the impact of PTA provisions on the structure and conduct of the U.S. agricultural and food sector and evaluate employment effects. A better understanding of these trade policy consequences will shed light on a critical driver of structural change. Such knowledge is essential for the functioning of global supply chains. The project will help to inform federal policies that aim to foster the competitiveness of U.S. farmers and ranchers and increase their participation and success in international markets.

Animal Health Component

20%

Research Effort Categories

Basic

80%

Applied

20%

Developmental

(N/A)

Classification

Knowledge Area (KA)	Subject of Investigation (SOI)	Field of Science (FOS)	Percent
606	5010	3010	100%

Knowledge Area
606 - International Trade and Development;

Subject Of Investigation
5010 - Food;

Field Of Science
3010 - Economics;

Keywords

agricultural trade

machine learning

natural language processing

preferential trade agreements

Goals / Objectives
This project will developnew insights regarding the formation of PTAs, their impact on global trade flows, and theconsequences for U.S. agricultural and food businesses. We rely on modern statistical modeling techniques to thoroughly investigate the factors that influence the formation of PTAs with agricultural and food provisions. This analysis builds on newly collected PTA data, which are structured and analyzed with the help of Neural Machine Translation (NMT) and Natural Language Processing (NLP) systems. To estimate factors that influence the formation of PTAs with agricultural and food provisions, we will adopt the Random Forest (RF) algorithm. This state-of-the-art supervised machine learning technique behaves well in high-dimensional settings allowing us to relax critical assumptions on data required by conventional regression methods. The statistical analysis will provide new insights regarding the role of economic, social, and political factors in forming PTAs with agricultural and food provisions. We will use this newly created dataset to investigate the impact of PTA provisions on agricultural and food trade in the sectoral three-way gravity model context relying on the adaptation of the Prior least absolute shrinkage and selection operator (Prior Lasso) to the Poisson pseudo-maximum likelihood (Poisson PML) estimator. This machine learning (ML) approach enables us to incorporate prior information, reduce over-fitting, and facilitate feature selection. We will instrumentalize PTA policy changes to assess their impact on the structure and conduct of the agricultural and food industry relying on newly collected data on U.S. business activities. A better understanding of these trade policy consequences will shed light on a critical driver of structural change in this vital sector of the economy. The research will help to inform federal policies that attempt to foster the competitiveness of U.S. farmers and ranchers and increase their participation and success in global agricultural and food markets. Consequently, this project will provide essential knowledge on the functioning of markets in light of trade policy changes and, thereby, enhance market efficiency and performance.The proposed research will accomplish the following research objectives:Objective 1: Structure and classify agricultural and food provisions in PTA treaties.Objective 2: Measure the economic, social, and political determinants of PTAs with agricultural and food provisions.Objective 3: Assess the trade creation and diversion effects of PTA provisions on agricultural and food trade.Objective 4: Evaluate the impact of PTA provisions on U.S. agricultural and food business activities and labor markets.

Project Methods
We will achieve the research objectivesby deploying modern statistical modeling techniques to thoroughly investigate the factors that influence the formation of PTAs with agricultural and food provisions. The proposed analysis builds on newly collected PTA data, which we will analyze with the help of innovative NMT and NLP systems. This analysis provides the foundation to thoroughly assess the factors that determine the formation of PTAs with agricultural and food provisions. We will adopt modern ML techniques that behave well in high-dimensional settings allowing us to relax critical assumptions on data required by conventional regression methods. The data-driven RF regression analysis will provide new insights regarding the role of economic, social, and political factors in forming PTAs with agricultural and food provisions. We will use this newly created dataset to investigate the impact of PTA provisions on agricultural and food trade in the sectoral three-way gravity model context, relying on an innovative adaptation of Prior Lasso to the Poisson PML estimator, which will allow us to incorporate prior information, reduce over-fitting, and facilitate feature selection. We will instrumentalize county-level PTA impact estimates to assess their effect on the structure and conduct of the agricultural and food industry, relying on establishment-level data on U.S. business activities. Our research project will enhance the understanding of a highly relevant foreign trade issue that has been largely neglected in the empirical literature while being of vital importance for the future of U.S. agriculture. Because trade policies can affect foreign trade significantly, they play a critical role in the functioning of agricultural and food supply chains. The research will shed light on a crucial driver of structural change in the agricultural and food sector. Such knowledge is essential because it will help to inform federal policies that attempt to foster the competitiveness of U.S. farmers and ranchers and increase their participation and success in global agricultural and food markets. The research will provide essential knowledge on the functioning of markets in light of trade policy changes and enhance market efficiency and performance.We will disseminate the research findings to the public and academic community. This research is highly relevant to the public because it will generate new knowledge regarding the formation of PTAs with agricultural and food provisions, their impact on trade flows, and the domestic consequences of trade liberalization. Furthermore, this research will provide highly relevant insights for the academic community by developing modern ML techniques paired with solid economic modeling and high-frequency trade data for policy impact evaluation. The project will have a dedicated webpage that presents the research findings and conclusions relevant to policymakers. We will work closely with the USDA to publish relevant reports for policymakers and share our research results widely in the executive and legislative branches. It is essential to improve our understanding of the relationship between trade policies and global trade flows. Such insights will help to inform federal policies that attempt to foster the competitiveness of U.S. farmers and ranchers and increase their participation and success in global agricultural and food markets. Because this question is of national scope, we will explore ways to disseminate our research findings through the media actively.We also plan on interacting actively with the academic community. In addition to presentations at academic conferences and industry workshops, we plan on publishing our research in well-recognized academic journals.We will make all non-proprietary data and research methods available to the academic community by providing web access to our datasets and codes via the dedicated project webpage. The dataset on PTAs with agricultural and food provisions, the methods for processing and analyzing this data, and the newly developed ML algorithms, have the potential to be particularly useful for future research activities.

Progress 01/01/24 to 12/31/24

Outputs
Target Audience:The targeted audience includes academic researchers and policymakers. The research informs the policy debate regarding free trade through applied economic research, supported by insights drawn from data science and advanced machine learning. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?We provided training opportunities for four graduate students and one postdoctoral scholar. How have the results been disseminated to communities of interest?We presented the research at an internal seminar, the Midwest Trade Conference, IATRC, AAAI, and the AAEA Annual Meeting. Additional presentations are planned. What do you plan to do during the next reporting period to accomplish the goals?We expect to complete the final work on Objective 4 by Fall 2025.

Impacts
What was accomplished under these goals? We completed the analysis for Objectives 1 through 3 and submitted the corresponding papers to academic journals. Objective 1: Structure and classify agricultural and food provisions in PTA treaties. Objective 2: Measure the economic, social, and political determinants of PTAs with agricultural and food provisions. Objective 3: Assess the trade creation and diversion effects of PTA provisions on agricultural and food trade. We have also initiated the analysis for Objective 4: Objective 4: Evaluate the impact of PTA provisions on U.S. agricultural and food business activities and labor markets.

Publications

Type: Other Status: Awaiting Publication Year Published: 2025 Citation: Steinbach, S., & Zurita, C. (2025). Assessing the Impact of the Trade Facilitation Agreement on Agri-Food Global Value Chain Integration. The World Economy
Type: Peer Reviewed Journal Articles Status: Under Review Year Published: 2025 Citation: Kim, Dongin, 2022. "Preferential Trading in Agriculture: New Insights from a Structural Gravity Analysis and Machine Learning," 2022: Transforming Global Value Chains, December 11-13, Clearwater Beach, FL 339469, International Agricultural Trade Research Consortium. Available at SSRN: https://ssrn.com/abstract=4412389 or http://dx.doi.org/10.2139/ssrn.4412389
Type: Peer Reviewed Journal Articles Status: Published Year Published: 2024 Citation: Gordeev, S., & Steinbach, S. (2024). Determinants of PTA Design: Insights from Machine Learning. International Economics, 178, 100504. https://doi.org/10.1016/j.inteco.2024.100504
Type: Other Status: Published Year Published: 2024 Citation: Jiahui Zhao, Ziyi Meng, Stepan Gordeev, Zijie Pan, Dongjin Song, Sandro Steinbach, and Caiwen Ding. "Key Information Retrieval to Classify the Unstructured Data Content of Preferential Trade Agreements." The 38th Annual AAAI Conference on Artificial Intelligence (AAAI) AI for Time Series Workshop (AI4TS).
Type: Other Status: Published Year Published: 2024 Citation: Zijie Pan, Stepan Gordeev, Jiahui Zhao, Ziyi Meng, Caiwen Ding, Sandro Steinbach, Dongjin Song. "International Trade Flow Prediction with Bilateral Trade Provisions ." The 38th Annual AAAI Conference on Artificial Intelligence (AAAI) AI for Time Series Workshop (AI4TS).
Type: Peer Reviewed Journal Articles Status: Published Year Published: 2025 Citation: Gordeev, S., Jelliffe, J., Kim, D., & Steinbach, S. (2025). What Matters for Agricultural Trade? Assessing the Role of Trade Deal Provisions using Machine Learning. Applied Economic Perspectives and Policy. First published online. https://doi.org/10.1002/aepp.13525

Progress 01/01/23 to 12/31/23

Outputs
Target Audience:The targeted audience isacademic researchers and policymakers. The researchintends toinform the policy debate regarding freetrade throughapplied economic research supportedby insights drawnfrom data science and advanced machine learning. Changes/Problems: Nothing Reported What opportunities for training and professional development has the project provided?We provided disciplinary and interdisciplinary training opportunities for four graduatestudents and one postdoctoral scholar. How have the results been disseminated to communities of interest?We presented the research at an internal seminar, the Midwest Trade Conference, IATRC, AAAI, and the AAEA annual meeting. Additional presentations are planned. What do you plan to do during the next reporting period to accomplish the goals?We expect to complete the final work on Objectives 1 to 3 by Winter 2024. Objective 4 shouldbe closed by Winter 2025.

Impacts
What was accomplished under these goals? We completed the analysis for objectives 2 and 3 and submitted the papers to academic journals. Objective 2: Measure the economic, social, and political determinants of Preferential Trade Agreements (PTAs) with agricultural and food provisions. Objective 3: Assess the trade creation and diversion effects of PTA provisions on agricultural and food trade. We have made progress on objective 1. Objective 1: Structure and classify agricultural and food provisions in PTA treaties. We also started data collection for objective 4. Objective 4: Evaluate the impact of PTA provisions on U.S. agricultural and food business activities and labor markets.

Publications

Type: Other Status: Published Year Published: 2024 Citation: Jiahui Zhao, Ziyi Meng, Stepan Gordeev, Zijie Pan, Dongjin Song, Sandro Steinbach, and Caiwen Ding. "Key Information Retrieval to Classify the Unstructured Data Content of Preferential Trade Agreements." The 38th Annual AAAI Conference on Artificial Intelligence (AAAI) AI for Time Series Workshop (AI4TS).
Type: Other Status: Published Year Published: 2024 Citation: Zijie Pan, Stepan Gordeev, Jiahui Zhao, Ziyi Meng, Caiwen Ding, Sandro Steinbach, Dongjin Song. "International Trade Flow Prediction with Bilateral Trade Provisions ." The 38th Annual AAAI Conference on Artificial Intelligence (AAAI) AI for Time Series Workshop (AI4TS).
Type: Conference Papers and Presentations Status: Under Review Year Published: 2024 Citation: Dongin Kim and Sandro Steinbach (2023). Preferential Trading in Agriculture: New Insights from a Structural Gravity Analysis and Machine Learning, CAPTS Working Paper 2022-12, under review
Type: Journal Articles Status: Under Review Year Published: 2024 Citation: Stepan Gordeev and Sandro Steinbach (2023). Determinants of PTA Design: Insights from Machine Learning, under review.
Type: Journal Articles Status: Under Review Year Published: 2024 Citation: Stepan Gordeev, Jeremy Jelliffe, Dongin Kim and Sandro Steinbach (2023). What Matters for Agricultural Trade? Assessing the Role of Trade Deal Provisions using Machine Learning, under review.

Progress 01/01/22 to 12/31/22

Outputs
Target Audience:The targeted audience of this project is academic researchers and policymakers. The research's intent is to inform the policy debate regarding free trade through sound applied economic research supported by insights drawn from data science and advanced machine learning techniques. Changes/Problems:There was a change in the leadership structure for this project. Dr. Steinbach moved to NDSU. Dr. Ding took over as the PI. Work on all project objectives is progressing according to plan. What opportunities for training and professional development has the project provided?We provided disciplinary and interdisciplinary training opportunities for three graduatestudents and one postdoctoral scholar. How have the results been disseminated to communities of interest?We presented the research at an internal seminar. Weare scheduled to present the work at the AAEA annual meeting. Additionalpresentations are planned. What do you plan to do during the next reporting period to accomplish the goals?We expect to complete work on Objectives 1 to 3 by Winter 2023. We will then proceed towork on objective 4.

Impacts
What was accomplished under these goals? We developed code and run initial analysis for objectives 2 and 3. Objective 2: Measure the economic, social, and political determinants of PTAs with agriculturaland food provisions. Objective 3: Assess the trade creation and diversion effects of PTA provisions on agricultural andfood trade. We also advance on objective 1. Objective 1: Structure and classify agricultural and food provisions in PTA treaties.

Publications

Type: Conference Papers and Presentations Status: Published Year Published: 2022 Citation: Sahidul Islam*, Shanglin Zhou*, Ran Ran, Yu-Fang Jin, Wujie Wen, Caiwen Ding, Mimi Xie. EVE: Environmental Adaptive Neural Network Models for Low-power Energy Harvesting System. In the 41st International Conference On Computer-Aided Design (ICCAD), 2022.
Type: Conference Papers and Presentations Status: Accepted Year Published: 2023 Citation: Bingbing Li, Zigeng Wang, Shaoyi Huang, Mikhail Bragin, Ji Li, Caiwen Ding. Towards Lossless Head Pruning through Automatic Peer Distillation for Large Language Models. In Proceedings of the 30th International Joint Conference on Artificial Intelligence (IJCAI), 2023