WP-047: Peter Christensen, Paul Francisco, Erica Myers, Hansen Shao, and Mateus Souza Energy Efficiency Can Deliver for Climate Policy: Evidence from Machine Learning-Based Targeting 

(February 2021 (Updated: November 2021))

Energy efficiency is a key component of climate and energy policy around the world. While engineering projections suggest that energy efficiency is one of the most cost- effective strategies to reduce greenhouse gas emissions, impact evaluations have called this into question by demonstrating that realized benefits typically fall short of projections. This study shows that machine learning models, in conjunction with pre-retrofit billing data, can be used to more accurately predict the causal impact of retrofits in an energy efficiency program and improve projections of economic ben- efits. The ML models are trained with comprehensive data on characteristics and monthly energy use from over thirteen thousand homes in a large energy efficiency program in the United States. Comparison with a widely-used engineering model indicates that the ML approach outperforms predictions that currently guide deci- sions in energy efficiency programs. Results further demonstrate that more accurate machine-learning predictions can be used to target funds to high-return interven- tions and increase the cost-effectiveness of energy efficiency investments by 21%. Advances in modeling and expanded access to home-specific billing data could pave the way for maximizing energy efficiency impacts at low cost.