Predicting poverty through time with publicly available data

Fighting poverty remains challenging due to laborious and expensive tracking and targeting methods, especially over time. Our work presents an accurate, scalable, inexpensive method to estimate consumption expenditure from publicly available data using surveys, satellite images and OpenStreetMap features from four African countries, Nigeria, Tanzania, Ethiopia and Malawi. Our approach is capable of predicting consumption through time. The features explain up to 75% of the variation in local-level economic outcomes, and for the temporal prediction, up to 60%. Our method presents a novel way to predict poverty over time. It could transform efforts to understand the development of poverty in developing countries and the tracking and targeting of poverty.