ABSTRACT
Improving energy efficiency is essential for sustainable development; therefore, an accurate assessment of energy efficiency is needed to inform policymakers on how to set energy efficiency goals. Using a stochastic frontier model, we estimate energy efficiencies of 44 countries in Europe from 1990 to 2015. Comparing average energy efficiencies across countries among three time periods, 1990 to 1998, 1999 to 2007, and 2008 to 2015, we find the following trends: (1) countries with highest energy efficiency cluster geographically suggesting spillover effects; (2) from the first period to the second period, 67.44% of European countries’ average energy efficiency increased, while only 59.09% of European countries’ average energy efficiencies increased between the first and the third period; (3) gains in energy efficiency may be levelling off or decreasing over time; and (4) when countries face economic downturn, they experience a decrease in energy efficiency signalling possible future decreases in energy efficiency in light of the recent recessions caused by the Covid-19 pandemic. Finally, we compute potential energy savings from a counterfactual experiment in which countries realize full efficiency. The results show substantial energy savings can be obtained by Russia in particular, and less savings can be obtained by countries with smaller populations.
Disclosure statement
No potential conflict of interest was reported by the authors.
Notes
1 See Kuosmanen and Kortelainen (Citation2012) for a combination of DEA and SFA that incorporates statistical noise.
2 Our review of the literature above focuses on works that examine European countries, but we refer the reader to a more focused review of papers using DEA to estimate energy efficiency across sectors and industries in Apergis et al. (Citation2015).
3 In line with standard stochastic frontier models, we assume that our model does not suffer from non-stationarity issues such as spurious regression results. In our case, the number of time periods is relatively small. For other potential time series related problems, see Wagner (Citation2008). For further discussion of non-stationarity in the inefficiency terms, see Kneip, Sickles, and Song (Citation2012), Duygun, Kutlu, and Sickles (Citation2016), and Almanidis, Karakaplan, and Kutlu (Citation2019). These studies address some of the non-stationarity issues in the context of stochastic frontier analysis. Also, Kutlu (Citation2018, Citation2021), Kutlu at al. (Citation2020) and Glass et al. (Citation2014, Citation2016) discuss spatial spillover issues in the context of stochastic frontier analysis.
4 Note that L/E and K/E normalizations in the output are based on the theoretical restrictions that we described in EquationEquations 2(2) (2) –Equation4(4) (4) .
5 Following a referee’s suggestion, in line with Fillipini and Hunt (Citation2011), as a robustness check, we also included monthly average temperature, value added from services as a percent of GDP, and the value added from the industrial sector as a percent of GDP. The number of observations for this model is 960 due to missing observations. The correlation of efficiency estimates based on this model and our benchmark model (Model 2) is 0.94, which is reasonably high. We picked our benchmark model based on AIC and BIC model selection criteria. Particularly, AIC values for Model 2 and this extended model are −3737.9 and −3424.9, respectively; and the corresponding BIC values are −3437.3 and −3123.2. Thus, giving AIC and BIC values, our benchmark model is preferred. We also estimated our benchmark model using only formerly communist countries. The correlation of efficiency estimates with our benchmark estimates is 0.73.