No COVID-19 climate silver lining in the US power sector

The study investigates whether the COVID-19 pandemic led to statistically significant and sustained reductions in U.S. power sector CO2 emissions, challenging the common assumption of a causal link between pandemic-related demand declines and emissions reductions. The U.S. power sector, responsible for roughly one-third of national CO2 emissions, is central to decarbonization policies and utility targets. The authors aim to quantify the significance of emissions changes from March–December 2020 using a counterfactual approach that controls for weather, seasonality, and recent trends, and to assess whether COVID-19 materially increased the likelihood of coal plant retirements. The importance lies in distinguishing temporary fluctuations from durable changes relevant to decarbonization trajectories and policy planning.

Prior work reports COVID-19-related declines in energy demand and CO2 emissions globally and in the U.S., often implying causality between the pandemic and emissions reductions. Additional studies focus on changes in electricity demand and supply mix during the pandemic. Policy context highlights extensive decarbonization commitments by U.S. states and utilities, with coal retirements a key component. The present study departs from much of the literature by explicitly testing the statistical significance of observed emissions reductions against a counterfactual that accounts for typical variability.

The analysis uses three data types: net generation by fuel (EIA Form EIA-923, Jan 2016–Dec 2020), fuel-specific emissions factors (EIA and EPA), and population-weighted heating degree days (HDD) and cooling degree days (CDD) from EIA. Plant-level monthly CO2 emissions are computed by multiplying fuel consumption by fuel-specific emissions factors and aggregating across plants for the contiguous U.S. Gaussian process (GP) regression is used to construct a counterfactual (no-COVID) forecast and quantify uncertainty. For CO2 emissions (C), the GP model (implemented in Python GPy) includes a bias (constant) kernel, a linear (trend) kernel, and a constrained one-year standard periodic kernel to capture seasonality; HDD and CDD effects are modeled with bias and linear kernels. The model is fit on Jan 2016–Feb 2020 data and forecasts Mar–Dec 2020 monthly emissions, yielding Gaussian predictive distributions per month. Statistical significance is evaluated via 95% confidence intervals: if the observed monthly value lies outside the 95% CI, the deviation is deemed statistically significant at the 5% level. Parallel GP regressions are fitted for electricity generation (E) and carbon intensity of electricity supply (C/E) using analogous kernels and the same training/forecast windows; E and C/E are modeled independently (their product need not equal C). Additional fuel-specific GP regressions estimate CO2 contributions from coal, natural gas, and oil, noting wider uncertainty bands. To assess coal retirements risk, the authors estimate monthly profitability for 845 coal units (in seven U.S. organized market regions) from Mar 2020–Dec 2022 under two price scenarios: (1) counterfactual based on EIA’s Jan 2020 wholesale price forecast; (2) current expectations combining actual Mar–Dec 2020 prices with EIA’s Jan 2021 forecast. Historical hourly zonal prices (2018–2020) are from S&P Global Market Intelligence; monthly regional forecasts (EIA) are mapped to hourly zonal profiles via a similarity-and-scaling procedure using historical hourly shapes. Capacity market revenues are included for MISO, New England, New York, and PJM using observed/assumed zonal clearing prices; assumptions include averaging prior periods to extend missing future auction prices. Unit-specific characteristics (location, capacity, variable and fixed O&M) are from S&P (2019 data). A WACC of 4.61% (0.38% monthly) from NREL ATB 2019 is used to discount monthly net cash flows. Operational logic assumes units run in hours when price ≥ variable cost; price duration curves are used to compute monthly revenues and variable costs per unit. Capacity bids are modeled such that non-profitable energy-market units bid to cover variable plus fixed costs net of expected energy revenues; profitable units bid zero. Profitability differences between scenarios identify units that are no longer profitable due to COVID-19-era price effects and are thus at risk of earlier retirement.

• CO2 emissions (C): Observed monthly CO2 emissions were significantly below the counterfactual only in April and May 2020 (outside 95% CIs). Deviations in other months (Mar, Jun–Dec 2020) were within 95% CIs. Table 1 deviations: April -14.0% (±10.5% CI), May -13.7% (±9.9% CI).
• Electricity generation (E): Observed E fell below the 95% CI in six months—March, April, May, June, August, and October 2020; within typical ranges by November and December (Table 1, Fig. 2). Average deviation Mar–Dec 2020: -2.9%.
• Carbon intensity (C/E): Significant reductions only in April and May 2020; otherwise within 95% CIs. Average deviation Mar–Dec 2020: -2.8%.
• Fuel-specific emissions (Table 2): Average Mar–Dec 2020 deviations vs counterfactual: coal -8.6%, natural gas -2.0%, oil +12.7% (annualized MMT averages: coal counterfactual 853.5 vs observed 780.1; gas 674.7 vs 664.6; oil 9.8 vs 10.7).
• Reversion toward pre-COVID patterns: Returns to pre-COVID expected levels observed for CO2 emissions and C/E by June 2020, and for E by November 2020.
• Coal plant retirements risk: Of 845 units, 90 units (2.8 GW; 1.9% of coal capacity) were profitable pre-COVID in the counterfactual but become unprofitable under current expectations due to COVID-19-era price effects—thus at risk of early retirement. Mean age 59.1 years; mean capacity 31.1 MW; distribution: 1 in SPP, remainder in MISO (79 in Zone 6, 9 in Zone 4, 1 in Zone 9).
• Profitability impact: Aggregate coal unit profits are $6.5 billion lower (present value) in current expectations vs counterfactual over Mar 2020–Dec 2022; $4.5 billion of this reduction occurs during Mar–Dec 2020.

The findings indicate that COVID-19-related reductions in U.S. power sector CO2 emissions were short-lived and statistically significant only in April and May 2020. Subsequent months fell within ranges consistent with historical variability when controlling for weather, trends, and seasonality, suggesting limited evidence for persistent emissions reductions attributable to the pandemic. Electricity generation showed temporary but non-persistent declines, aligning with macroeconomic data indicating a rebound in GDP after Q2 2020 and expectations for recovery. Carbon intensity reductions were also confined to April–May 2020, with average fuel-specific effects showing larger relative reductions from coal than gas and a relative increase from oil. The coal profitability analysis suggests that COVID-19-era price effects modestly increase retirement risk for a small fraction of capacity, implying that broader coal retirements are likely driven by longer-term structural factors rather than the pandemic. Overall, the results do not support a durable COVID-19-driven decarbonization in the U.S. power sector.

Using GP regression to construct a weather-, trend-, and seasonality-adjusted counterfactual, the study finds statistically significant reductions in U.S. power sector CO2 emissions only in April and May 2020, with a rapid return to expected levels thereafter. Electricity generation and carbon intensity similarly reverted to typical ranges by late 2020. A coal profitability assessment indicates that less than 2% of coal capacity is newly at risk of early retirement due to COVID-19-related market conditions through 2022. These results suggest no sustained COVID-19 climate “silver lining” in the U.S. power sector and highlight the importance of long-term policies and market dynamics in driving decarbonization.

• The GP models are trained on Jan 2016–Feb 2020 and assume stationarity captured via linear and annual periodic kernels; unmodeled nonlinear or structural shifts beyond these may reside in the Gaussian noise term.
• E and C/E are modeled independently; their product need not equal C in any given month.
• Fuel-specific GP regressions exhibit wider uncertainty (e.g., coal-related CO2 April 2020 ±24.9% of mean), indicating lower predictability at the fuel level.
• Profitability modeling relies on assumptions about unit operation (dispatch when price ≥ variable cost), capacity market bidding behavior, WACC (4.61%), and extrapolation of future capacity prices via historical averages where data are unavailable.
• Cogeneration units supplying electricity directly to customers are excluded, limiting generalizability to all coal-fired facilities.
• Data availability for prices and costs includes proprietary S&P datasets; some inputs are forecasts (EIA) that may differ from realized outcomes.
• Analysis covers the contiguous U.S. and the period through Dec 2022 for profitability; impacts beyond this scope are not assessed.