Energy Economics • AI Infrastructure • Applied Econometrics
Working papers and research in energy economics, AI infrastructure, and applied econometrics. Methodology, code, and data sources are documented for each project.
Between 2019 and 2025, electricity demand in ERCOT, the Texas grid, grew by roughly 27 percent. In PJM and MISO, the two largest electricity markets in the country, demand was essentially flat. This divergence coincides with the largest surge in data center investment in American history and with ERCOT interconnection queue filings that reached 97 gigawatts in 2024 alone.
This paper asks how much of that divergence data center investment can explain, and whether the existing investment pipeline is large enough to stress the grid in ways that current planning models are not built to anticipate. The core obstacle is a data gap: every major U.S. electricity market maintains an internal queue of large load requests from data center developers, and none of them make it public. This paper uses the public generation-side queue as a proxy, documents why that proxy is imperfect, and builds an identification strategy designed to produce credible estimates despite that constraint.
The identification strategy has four layers: panel regression with wild cluster bootstrap inference, synthetic control, difference-in-differences with low-exposure control regions, and narrative validation against known data center energization dates. The headline result is a 34.8 index-point divergence in ERCOT minimum hourly demand above a synthetic counterfactual — the baseload floor signature of infrastructure that maintains high continuous power draw regardless of weather or time of day. A three-model forecasting pipeline (ARIMA, Prophet, XGBoost) extends scenario projections through 2027. The paper documents the absence of public load-side queue data as an explicit policy finding and argues it represents a meaningful gap in public oversight of the grid.
Full paper on SSRN • GitHub • Interactive Dashboard
Posted April 2026 • University of Maryland • 48 pp. • Synthetic Control • DiD • Panel Regression • ARIMA • XGBoost • Python • Stata
JEL: Q41, Q47, L94, C23, C14