Electricity Prices Impede Industry and Residential Electrification by a Small (But Significant) Amount

1. Introduction

Sustaining declining carbon emissions, i.e., decarbonizing the energy system, requires increasing both the share of energy services that are delivered via electricity and the share of nonfossil fuels used to generate that electricity [1]. Yet, there is evidence that high electricity prices (combined with fossil fuel prices that ignore carbon-based costs) discouraged households (in California) from switching to electricity for energy services like space heating, water heating, and personal mobility [2].

The current paper considers how electricity prices and the price of the direct use of fossil fuels contribute to electrification (share of energy consumption from electricity) in the residential and industrial sectors. It does so by analyzing OECD country macro panel data spanning 1978-2017 using a mean group-based estimator. For both sectors, we find a negative, significant, but relatively small (ranging from -0.1 to -0.07) response for electricity prices and a positive, significant, and small (ranging from 0.1 to 0.05) response for the direct use of fossil fuel price index.

While there is a growing literature on determining how energy prices impact the adoption of electricity services, most of this work has focused on electric vehicles (e.g., [3,4,5]). Indeed, the only paper we know of like the current paper is [6]. Huntington [6] estimated the extent to which electricity prices deterred economy-wide electrification (i.e., the share of total energy use accounted for by electricity) by analyzing OECD country macro data spanning 1980-2019. That paper calculated an electricity price coefficient that ranged from -0.3 to -0.1.

The present paper goes beyond [6] in two important ways: (i) we focus on electrification in the residential and industry sectors; and (ii) we develop a real price index of the direct use of fossil fuels in those sectors, i.e., we estimate a cross-price elasticity of electrification. Huntington [6] considered economy-wide energy prices in addition to economy-wide electricity prices; but the extent to which economy-wide energy prices are comprised of economy-wide electricity prices is determined by the electrification rate (i.e., as electrification approaches 100%, the two prices converge).

2. Materials and Methods

2.1. Dataset

We define electrification as the final electricity consumption in a sector (residential or industry) divided by the total energy consumption in that sector. These data are from the IEA’s World Energy Balances. Real GDP per capita data are from the World Bank Development Indicators. Real electricity prices are from Enerdata’s Global Energy & CO₂ database. The real price index of direct use of fossil fuels (coal, peat, and oil shale; natural gas; and oil products) for each sector (residential, industry) are constructed—as standard—by weighing Enerdata’s real price data for a specific fossil fuel by that fuel’s consumption share from the IEA’s Energy Balances. Table 1 lists each variable with details on its definition and source.

The final dataset consists of two balanced panels that span 1978-2017. The availability of the price data for the direct use of fossil fuels is why the panel ends in 2017 (the other series extend to 2019). The industry panel has 25 countries, and the residential panel has 24 (because Portugal does not have the residential fossil fuel price index, it is dropped from that panel). Table 2 presents descriptive statistics. The cross-sectional dependence test (CD test) from [7] indicates a strong degree of cross-sectional correlation for all variables, and most variables have high mean correlation coefficients.

Table 3 and Table 4 display electrification data at different points in time for the residential and industry sectors, respectively, for each country in the analysis. Even among OECD countries, residential electrification can have dramatic differences. For example, current rates are particularly high for Norway and New Zealand (85 and 72%, respectively) but is only 12% for Poland. Nearly all countries have experienced growth in electrification (Poland, too, but not Germany). Yet, for several countries there has been little change either recently (since 2010), like Australia, or for some time (since 1990), like Belgium, Denmark, and New Zealand. Other countries have grown in fits like Norway, Spain, and Switzerland.

Industry electrification shows less divergence (Table 4). Norway is still particularly high at 64%, but most countries are in the 30-40% range most recently. Finland has experienced little change throughout the time period, while Austria, Japan, Norway, Sweden, and Switzerland have had little change since 1990. In the US, industry electrification peaked in 2000 and has declined since then.

Figure 1 a&b present a last look at the dataset. These figures display the individual country traces for the industry real electricity price (Figure 1a) and the industry direct use of fossil fuel real price index (Figure 1b). For most (but not all) countries the residential and industry electricity price series are highly correlated as are the residential and industry fossil fuel price indices. Thus, to economize space, only the industry price series are shown. The fossil fuel price index (Figure 1b) clearly indicates the two major international oil price spikes: first in the early-to-mid 1980s and then again in the early 2010s. For electricity prices (Figure 1a), the traces are more diverse but most/many countries experienced declining prices.

2.2. Model and Method

Following [6], we model electrification (Elec) as a function of GDP per capita (GDP) and electricity prices (elec pr):

$${Elec}_{it}={\beta }_i^1{GDP}_{it}+{\beta }_i^2{elecpr}_{it-\mathrm{1,2},3}+{{\beta }_i^3{ffpr}_{it-\mathrm{1,2},3}+{\beta }_i^4{trend}_i+\epsilon }_{it}$$

(1)

where t represents the time dimension, i the country dimension, and ε_it is the error term. Additionally, we include a price index of direct fossil fuel use (ff pr). We expect electricity price to have a negative coefficient and direct fossil fuel use price—as a cross-price term—to have a positive coefficient. However, we do not expect prices to have a contemporaneous impact on electrification; so, we lag those terms. Yet, we do not have a priori knowledge on what the lag structure should be; so, for robustness we consider up to three lags (a fourth lag or greater always produced insignificant coefficients). Again, because of a lack of a priori intuition and for robustness, we consider pairs of lagged price terms sequentially (i.e., leading to three regressions).

We also allow for an individual country time trend (trend). Time trends have been suggested to approximate technology/energy efficiency improvements in energy demand models (e.g., [8,9]). Huntington [6] does not motivate the inclusion of GDP per capita other than to acknowledge that electrification is considered part of the development/economic growth process. One might hypothesize that GDP represents the resources to afford infrastructure investments that make the electrification of energy services more attractive. Such investments would be done on the individual country level. Whereas, time trends represent technological improvements that encourage/facilitate electrification that all OECD countries are likely to find accessible (e.g., via trade/transfer) even when an individual country’s GDP growth stalls.

We analyze the residential and industry sectors in separate equations, but to avoid clutter do not include any more sub-/super-scripts in Equation 1. Lastly, we take natural logs of all the variables so that their coefficients can be interpreted as elasticities.

Although we are interested in panel coefficients, we use the mean group estimator from [10] that makes the most general assumption that the individual country coefficients are different and then averages those estimations to arrive at the panel coefficients (standard errors are constructed nonparametrically). Indeed, the Pasaran and Yamagata [11] test for slope homogeneity (results not shown) confirms that mean group panel average coefficients are different from fixed effects coefficients, i.e., fixed effects would be biased.

As the CD test results reported in Table 2 suggested, residual cross-sectional dependence (CSD) could be an issue. Untreated CSD in regression residuals suggests omitted variable bias and endogeneity, and thus, biased and inconsistent estimates. This problem is typically addressed in mean group regressions by adding cross-sectional average terms. Since, international energy prices are likely to be the main source of such dependence in our model (e.g., see Figure 1b), we include one (international energy prices) cross-sectional average. Data from the IEA Real Price Index for industry and households was used to construct the term. As will be discussed/shown in the following section, adding this one cross-sectional average variable removed evidence of stronger CSD; so, to preserve degrees of freedom, we add only one cross-sectional average term.

3. Results and Discussion

Table 5 displays the regression results; the bottom rows contain diagnostic statistics. There is a concern that the CD test from [7] over-rejects weak dependence when applied to residuals; so, as a residual diagnostic, we use the bias corrected CD* test from [12]. Since weak cross-sectional dependence could not be rejected (second to last row), we employ the Maddala and Wu [13] panel unit root test to demonstrate that the residuals are stationary (last row). (The CIPS test [14] was used to demonstrate stationarity for the one case that marginally rejected weak dependence, i.e., residential electrification with three-year lags).

The electricity price coefficients are negative, always statistically significant, but relatively small—always around -0.1 for residential and always less than -0.1 (ranging from -0.08 to -0.07) for industry. The fossil fuel price coefficients are positive, usually significant, and mostly of similar (absolute) magnitude as the electricity price coefficients. These fossil fuel price coefficients ranged from 0.1 to 0.05 for residential and, when significant, from 0.09 to 0.06 for industry (depending on the lag). The time trend is positive, always significant, but even smaller in absolute magnitude. The GDP coefficient is always significant and large for industry, but it is significant for only one regression for residential. Huntington’s [6] most similar regression produced a larger electricity price coefficient of -0.3 and a smaller GDP coefficient of 0.2 (both highly statistically significant).

Huntington [6] raised the issue of endogeneity between electrification and economic growth. This endogeneity is unlikely for the residential sector but could exist for the industry sector since electricity is considered a more/the most productive form of energy. However, the GDP term in our model is aggregate output per capita and not industry output. Furthermore, for the developed OECD countries in our panel, industry’s share of GDP is rather small; e.g., it is mostly below 30%. Further still, industry’s share of GDP is negatively correlated with GDP per capita for nearly every country (Ireland and Norway are the exceptions). Huntington [6] tried to instrument for GDP using the output shares of agriculture, construction, industry, and services. Not only does this instrument seem inappropriate for our purposes (e.g., the dependent variable is electrification for industry only), it would lead to several years of lost data. Ultimately, [6] argued that for richer (i.e., OECD) countries, economic growth should cause electrification (in other words, endogeneity was not a concern). Nevertheless, for robustness, we consider two alternative specifications for the industry panel: (i) drop GDP (as did [6]); and (ii) replace contemporaneous GDP with its one-year lag to further mitigate any possible endogeneity. Those results—shown in Appendix Table A1—are mostly in concert with the results from Table 5.

5. Conclusions

Liddle and Parker [1] determined that, for the 15 countries that have sustained reductions in their per capita CO₂ emissions and decoupled those emissions from GDP (i.e., their emissions and GDP are negatively associated/correlated), they decarbonized their energy systems by increasing both the share of energy services that are delivered via electricity and the share of nonfossil fuels used to generate that electricity. Yet, there is a concern that high electricity prices deter the electrification process (e.g., [2,6]). This analysis determined that electricity prices do impede the electrification process in the residential and industrial sectors by a small amount. However, the analysis also found that direct use of fossil fuel prices encourage electrification by a mostly similar small amount. Hence, there is a policy opportunity to harness these dual (albeit modest) price incentives to simultaneously encourage electrification by coupling carbon taxes with renewable inducements/subsidies (i.e., raising the price of the direct use of fossil fuels relative to the price of electricity).