Combining Satellite Imagery and Machine Learning to Predict Economic Impact of Land Registration in Georgia

with Irakli Barbakadze

Property rights are a key factor in economic development. In order to identify the causal effect of land ownership, one should exploit a natural experiment; otherwise, it is difficult to exogenously identify the effect, as typically, registration decisions are not random and there is a potential positive selection bias among registered households. To overcome the identification problem, we study the Systematic Land Registration Pilot Reform (2016-2019) in Georgia. We contribute to the literature with a novel way to evaluate such an experiment based on high-resolution data and machine learning methods. Using remotely sensed daytime satellite images and cadastral maps, we find positive changes in household welfare, which we measure in terms of the quality of rooftops and land use, in a recent free land registration program in rural Georgia.