Escaping the Poverty Trap: Poverty Alleviation with SCouT

Abstract

Poverty is a widespread epidemic that has plagued millions of people across the world for centuries, yet its incidence is only increasing. Economists Abhijit V. Banerjee and Esther Duflo took an experimental approach to finding the most effective strategies in alleviating poverty by performing many randomized controlled trials (RCTs). Unfortunately, RCTs raise a variety of issues concerning ethics and cost. In this paper, we instead propose a novel method that not only bypasses the problems RCTs present but even outperforms RCTs. In particular, we will use a transformer-based approach to introduce synthetic interventions that forecast the impact of policy interventions in impoverished nations. Our proposed approach differs from traditional synthetic control methods, which use time-agnostic weights to capture similarities across the unit space, because of its flexibility across space and time. We have previously found that this method performs significantly better than existing SC methods, which allows us to more accurately examine the efficacy of policies supported by Banerjee and Duflo’s research.