Price elasticity of electricity demand: Using instrumental variable regressions to address endogeneity and autocorrelation of high-frequency time series

This paper examines empirical methods for estimating the response of aggregated electricity demand to high-frequency price signals, the short-term elasticity of electricity demand. We investigate how the endogeneity of prices and the autocorrelation of the time series, which are particularly pronounced at hourly granularity, affect and distort common estimators. After developing a controlled test environment with synthetic data that replicate key statistical properties of electricity demand, we show that not only the ordinary least square (OLS) estimator is inconsistent (due to simultaneity), but so is a regular instrumental variable (IV) regression (due to autocorrelation). Using wind as an instrument, as it is commonly done, may result in an estimate of the demand elasticity that is inflated by an order of magnitude. We visualize the reason for the Thams bias using causal graphs and show that its magnitude depends on the autocorrelation of both the instrument, and the dependent variable. We further incorporate and adapt two extensions of the IV estimation, conditional IV and nuisance IV, which have recently been proposed by Thams et al. (2022). We show that these extensions can identify the true short-term elasticity in a synthetic setting and are thus particularly promising for future empirical research in this field.