Andreas Dür (Salzburg) and Manfred Elsig (Bern), FWF/SNF Weave Project, 2025-2029
Preferential trade agreements (PTAs) reduce trade barriers between two or more countries without extending this liberalisation to all countries worldwide. Since the Second World War, countries from across the world have signed more than 800 such agreements. Well-known examples include the EU, the USA, Mexico and Canada trade agreement, and the free trade agreement between China and Australia. These agreements contain provisions to reduce tariffs on trade in goods. A growing number of PTAs also regulate trade in services, intellectual property rights, and foreign direct investment. Many even contain provisions on environmental protection, labour rights or national security.
PTAs differ greatly from one another. Some are narrower (i.e. they cover fewer topics) than others and some contain more obligations for signatories than others. Capturing these differences is important to understand why states sign PTAs, how they evolve, and what consequences PTAs have for trade, the environment or labour rights.
So far, the design of PTAs has mainly been measured through manual coding. The Design of Trade Agreements (DESTA) dataset, founded by the lead applicants of this project, is the most ambitious attempt to do this. However, the manual coding of large numbers of PTAs is both labour-intensive and potentially error-prone. Human coders assigning values to the agreements for a large number of items (e.g. what types of environmental provisions are included in an agreement) also requires a lot of expertise. This makes it difficult to continuously update the dataset as new agreements are signed.
With this project, we therefore want to develop a new approach that relies on computational text analysis instead of human coders. We strive to develop an algorithm that provides valid and reliable measures of the contents and design of PTAs largely without human intervention. To this end, we will work with large language models, using previous manual coding to train the models for the task at hand.
We will use the resulting data to gain new insights into PTAs. In particular, we are interested in how states design PTAs in response to global value chains. A global value chain exists when the production of a good takes place in several countries. In the automotive industry, for example, a car is often designed in one country, the parts and components are manufactured in another country, and the vehicle is then assembled in a third country. We also want to better understand the extent to which PTAs contribute to the opening of markets and how and when provisions travel from one agreement to another.
Overall, we will develop a new method to obtain novel data on the design of PTAs that will allow us to better understand the creation, evolution, and impact of PTAs.
Call for applications for a postdoc position in this project