"Causal Machine Learning for Cost-Effective Allocation of Electricity Aid" thesis for my Masters in Management and Digital Technologies at Ludwig-Maximillian Univeristy, Munich.
TOPIC: Causal Machine Learning for Cost-Effective Electricity Aid Allocation
ABSTRACT: The United Nations Sustainable Development Goals (SDG 7) aim to ensure access to a!ordable, reliable, sustainable, and modern energy by 2030. Achieving this goal by 2030, particularly in developing countries, requires substantial volumes of development aid. In this thesis, we develop a causal machine learning framework to predict heterogeneous treatment effects of electricity development aid disbursements to make informed decisions for more e!ective aid allocation. Specifically, our framework employs a causal forest to analyze continuous treatment effects while addressing high-dimensional country characteristics, treatment selection bias, and small sample-size settings. Our framework uses data on electricity development aid, electricity access, and country characteristics and achieves a Root Mean Integrated Squared Error (RMISE) of 0.1685 and Root Mean Squared Error (RMSE) of 0.1346 on factual data, indicating reliable predictive performance. Analysis of country-level variations shows that countries with lower access rates exhibit higher treatment e!ect heterogeneity, while countries with 100% access show minimal variation in treatment response. Our framework successfully generates country specific aid-response curves with uncertainty quantification using bootstrap ensemble methods, demonstrating the potential for more targeted electricity aid allocation strategies.