Electricity Demand Stayed Price-Inelastic Across Ninety Years

Electricity policy often assumes that consumers will react more to prices as smart meters, time-varying tariffs, automation, and storage spread. That assumption matters because many decarbonization plans rely on price signals to flatten peaks, absorb renewable supply, and avoid expensive grid reinforcement. This paper tests the premise against a long empirical record: 4,720 own-price elasticity estimates from 462 studies, with underlying data spanning 1934 to 2024.

The result is less a new estimate than an audit of a policy belief. The authors find small electricity-demand responses, larger elasticities only when consumers have more time to adjust, and no evidence that responsiveness has risen over calendar time.

Figure 1 summarizes the paper’s central time-series claim: across nine decades of estimates, the fitted elasticity path does not show the upward drift that the technology story predicts.

Core Contribution

The paper’s main contribution is to put heterogeneous electricity-demand studies on a common empirical ladder. Rather than treating all estimates as interchangeable, the authors classify price identification from randomized or experimentally assigned tariff variation through natural experiments, difference-in-differences, instrumental variables, panel fixed effects or structural systems, and naive regressions. That ranking is then used alongside publication-bias correction and study-level controls.

This changes the reading of the literature. The more credible designs do not reveal hidden large responses. They tend to produce smaller elasticities than less credible regressions, which is awkward for the view that weak identification has been masking a more flexible demand side.

Technical Approach

The meta-analysis uses funnel-asymmetry and precision-effect methods to correct for selective reporting, then layers in horizon, identification tier, data vintage, publication timing, sector, tariff type, geography, and study characteristics. The paper distinguishes a “data clock,” which asks whether consumers in later underlying data years behave differently, from a “publication clock,” which asks whether later papers report different estimates.

Figure 2 captures the horizon result. Bias-corrected responsiveness grows as consumers have more time to change appliances, capital stock, contracts, or production processes: the short-run estimate is about -0.16, while the long-run estimate is about -0.38. That is a real adjustment gradient, but it is not the same as a trend toward high flexibility. The long run is already represented in the historical data.

The identification ladder is the paper’s most useful methodological device. It lets the authors ask whether cleaner sources of price variation imply larger or smaller responses. Their answer is smaller: the best-identified studies yield an adjusted short-run elasticity around -0.09, statistically indistinguishable from zero in the reported comparison, while naive specifications are more elastic.

Results and Analysis

The headline estimate is economically modest. A short-run elasticity of -0.16 means a 10% electricity-price increase is associated with less than a 2% reduction in total consumption. The long-run estimate of -0.38 is roughly twice as large, but still far from unit elasticity. Table 3 reports that this horizon gradient is broadly stable across eras: long-run corrected estimates sit between about -0.29 and -0.47 across the reported vintage bins, with imprecision in some cells.

The calendar-time tests are the stronger challenge to the policy premise. In Table 4, the short-run data-clock coefficient is -0.018 per decade when controlling for precision, -0.033 when separating data and publication clocks, and +0.017 after adding composition controls. None supports a reliable drift toward more negative elasticities. The paper also reports equivalence-style checks: a scenario in which short-run responsiveness doubles over three decades would require a rise of 0.05428 in absolute elasticity per decade, and the data reject a rise that large in the main specifications.

Technology-rich settings do not rescue the premise. Figure 4 shows the least elastic estimates in cells with the technologies or tariff regimes most often associated with demand flexibility, including time-of-use pricing. That should be read carefully: the paper studies total electricity consumption, not necessarily load shifting across hours. A household may respond to a time-varying tariff by moving use from peak to off-peak while leaving total kWh nearly unchanged.

Caveats in Practice

The evidence is broad and unusually well organized for an energy-economics meta-analysis, but it is still a synthesis of heterogeneous studies. The authors’ ladder depends on coded descriptions of primary designs, many instrumental-variable studies lack reported first-stage diagnostics, and some long-run design-based cells are thin. The paper also does not prove that future automation will fail; it shows that the historical record through 2024 does not yet contain the rising total-consumption elasticity assumed in many planning narratives. The practical implication is narrow but important: if planners want demand flexibility, price signals alone are a weak instrument unless paired with enabling technology, contracts, and program design that pay for the response they need.