Analysis of the Evolution of IMF Positions in Argentina: A Large Language Models Based Approach

Abstract

In this paper we examine the evolution of the International Monetary Fund's (IMF) stance towards Argentina through an analysis of their programme documents with the country, spanning from 2000 to the present. The exploration of more than 3000 pages of text was carried out using a large language model (LLM). We analysed changes in the thematic coverage of the organisation's documents, its position on various issues of Argentine economic policy (capital controls, inflation, income distribution and social spending) and the depth of knowledge about the main social programmes and the situation of the most vulnerable groups in the country.
We use text mining to segment and vectorise documents. Using retrieval augmented generation (RAG), we extract and compare thematic elements across different governmental periods. We categorically coded LLM responses to a set of questions to identify changes in levels of thematic content and the organisation's positions on the country's economic and social challenges. The results show an evolution of the Fund towards positions more favourable to fiscal consolidation, with concern for social spending.
This work is a collaboration between specialists in computer science, economics and political science. Its main contribution is a semi-automated, data-driven workflow that allows for the analysis of changes in policy approaches by analysing an organisation's publications over time.