Abstract
This paper investigates how humanitarian aid allocations respond to the Integrated Food Security Phase Classification (IPC) system, which is designed to direct aid to those most in need, using evidence from Afghanistan. A key challenge in evaluating these responses is the lack of systematically geocoded data on aid flows. To address this, I apply Natural Language Processing (NLP) to develop a geocoded, subnational-level aid flow dataset from the UN OCHA Financial Tracking Service (FTS). Using a staggered difference-in-differences approach, I analyze aid responses to IPC Phase 4 escalations, which indicate a food emergency. The results show that while aid distributions increase significantly following Phase 4 designations, per capita aid for affected populations remains insufficient. Furthermore, the analysis reveals that European countries are the primary drivers of these responses, whereas the U.S., despite being the largest contributor, does not specifically respond to IPC escalations. This research provides critical subnational insights into humanitarian responses within the global food insecurity measurement system.