Available online at www.sciencedirect.com ScienceDirect Energy Procedia 104 (2016) 305 – 310 CUE2016-Applied Energy Symposium and Forum 2016: Low carbon cities & urban energy systems Measuring the direct rebound effect of China’s residential electricity consumption Yue-Jun Zhang a,b*, Hua-Rong Peng a,b a b Business School, Hunan University, Changsha 410082, PR China Center for Resource and Environmental Management, Hunan University, Changsha 410082, PR China Abstract Due to the rebound effect, the electricity savings by improving the electricity efficiency of China’s households may be not as much as expected For this reason, based on the data from 29 provinces during 2000-2013, this work develops the panel threshold model to investigate the direct rebound effect of China’s residential electricity consumption under different scenarios The results show that the direct rebound effect of China’s residential electricity consumption is 71.53% on average, and the increase in GDP per capita and the decrease in cooling degree days and rainfall will help to reduce the direct rebound effect © 2016 The Authors Published by Elsevier Ltd This is an open access article under the CC BY-NC-ND license © 2016 The Authors Published by Elsevier Ltd (http://creativecommons.org/licenses/by-nc-nd/4.0/) Selection peer-review of under responsibility of CUE Peer-reviewand/or under responsibility the scientific committee of the Applied Energy Symposium and Forum, CUE2016: Low carbon cities and urban energy systems Keywords: Residential electricity consumption; energy rebound effect; panel threshold model Introduction In theory, improving electricity utilisation efficiency has important positive influence on reducing electricity consumption, the share of coal consumption, and greenhouse gas emissions, but why did the residential electricity consumption increase rather than decrease with the improvement of electricity utilisation efficiency in the past decade in China? There are two candidate aspects at play here On the one hand, there may be an energy rebound effect The effectiveness of improving electricity efficiency for energy conservation is not as great as expected, whereas there will be some rebound energy consumption [1] Residential electricity consumption arises mainly from household appliances, such as air conditioners for controlling the temperature, refrigerators, rice cookers, washing machines, and home lighting for daily life, and computers and TVs for work or entertainment When electricity utilisation efficiency improves, * Corresponding author Tel./Fax: 86-731-88822899 E-mail address: zyjmis@126.com (Yue-Jun Zhang) 1876-6102 © 2016 The Authors Published by Elsevier Ltd This is an open access article under the CC BY-NC-ND license (http://creativecommons.org/licenses/by-nc-nd/4.0/) Peer-review under responsibility of the scientific committee of the Applied Energy Symposium and Forum, CUE2016: Low carbon cities and urban energy systems doi:10.1016/j.egypro.2016.12.052 306 Yue-Jun Zhang and Hua-Rong Peng / Energy Procedia 104 (2016) 305 – 310 the power consumed doing the same work (lighting, heating, cooling, etc.) decreases Therefore, the cost for equal energy services may decrease, which in turn leads to a change in behaviour, and residents increase the demand for buying or using the household appliances, thus electricity consumption increases On the other hand, electricity utilisation improves with technologic progress, which promotes economic growth to some degree This will raise the buying power of residents, and they may increase their demand for using or buying household appliances, which results in increased electricity consumption Obviously, residential electricity consumption may vary in different external environments Meanwhile, the rebound effect of residential electricity consumption may also be different in various external environments Therefore, it is imperative to estimate the rebound effect of residential electricity consumption to avoid overestimating the effectiveness of energy efficiency policy There have been a number studies on the energy rebound effect of households, and we can find that previous research about the factors influencing residential electricity consumption mainly employs econometric approaches to detect the linear relationship between residential electricity consumption and its influencing factors and the rebound effect in the linear relationship Therefore, this work mainly compensates for the deficiency of the linear framework First, excepting investigating the linear relationship between China’s residential electricity consumption and its main influencing factors, this work also explores the non-linear relationship among them using the panel threshold model Second, excepting the estimation of the direct rebound effect of residential electricity consumption in the linear relationship, this work also chooses a few threshold variables, and measures the direct rebound effect of residential electricity consumption and its changes under different regimes of the threshold variables The contribution in this paper includes three aspects First, this paper investigates the key factors influencing China’s residential electricity consumption, and the linear and non-linear relationships between the influencing factors and residential electricity consumption Second, this paper estimates the direct rebound effect of residential electricity consumption in linear and non-linear relationships in households Finally, using GDP per capita, population, cooling degree days, and rainfall as threshold variables, this paper develops panel threshold model to estimate the direct rebound effect of residential electricity consumption in different regimes The remainder of this paper is organised as follows: Section proposes models and data definitions, Section presents the results and detailed discussions, and Section concludes the paper Methods and data definitions (1)Methods We choose the main factors influencing residential electricity consumption: electricity price, income, polulation, temperature and rainfall [2, 3, 4] In general, increasing the price of electricity will lead to a decrease in electricity consumption Income is closely linked to residential electrcity consumption In developing countries, the increase of GDP per capita often makes people increase their demand for electricity services, for example, using washing machines and air-conditioners Residential electricity consumption is also affected by population Generally, with other conditions unchanged, more people use more electricity Additionally, temperature and rainfall have some influence on residential electricity consumption [5], and some research has proved the non-linear relationship between temperature and residential electricity consumption [6] Based on the analyses above, we build the linear panel model to investiage the direct rebound effect of China’s residential electricity consumption in a linear relationship as Eq (1): (1) ln Ei ,t E E1 ln PE E ln PGDPi ,t E ln POPi ,t E ln CDDi ,t E5 ln RAIN i ,t Pi ,t i ,t where Ei ,t , PE , PGDPi ,t , POPi ,t , CDDi ,t , and RAINi ,t represent residential electricity consumption (unit: 100 million kilowatt hour), residential electricity price (unit: Yuan/kilowatt hour), GDP per capita (unit: Yuan), the urban population at the end of the year (unit: 10,000 persons), cooling degree day (unit: i ,t 307 Yue-Jun Zhang and Hua-Rong Peng / Energy Procedia 104 (2016) 305 – 310 centigrade) and rainfall (unit: millimetres) in province i in year t , respectively E0 is a constant term, and E1, E2 , E3 , E4 , and E5 are coefficients to be estimated, and P i ,t is the random error In this way, E1 denotes the size of the direct rebound effect of China’s residential electricity consumption To examine the non-linear relationship between residential electricity consumption and its main influencing factors, we develop the panel threshold models, and set different threshold variables to estimate the direct rebound effect of China’s residential electricity consumption under different regimes The single threshold model and double threshold model can be expressed as given by Eqs (2) and (3), respectively [7].† ln Eit D D1 ln PEit I (qit d J ) D ln PEit I (qit ! J ) D ln PGDPit D ln POPit D5 ln CDDit D6 ln RAINit Xit ln Eit G G1 ln PEit I (qit d J ) G ln PEit I (J qit J ) G ln PEit I (qit t J ) (2) (3) G ln PGDPit G3 ln POPit G ln CDDit G5 ln RAINit Q it where qit represents a threshold variable, and we here take ln PGDP , ln POP , ln CDD , and ln RAIN as the threshold variable, respectively I ( x ) is the indicator function; J is the threshold of the single threshold model, and J and J are the thresholds in the double threshold model Before the estimation of panel threshold models, it is necessary to test whether, or not, the threshold effect is statistically significant Testing for the null hypothesis of no threshold effect ( H 01 : D1 D ) is done through Eq (4): F1 ( S0 S1 (Jˆ )) / Vˆ (4) where S0 is the sum of the squared residual of the linear model, S1 is the sum of the squared residual of the single threshold model, Jˆ is the OLS estimate of J , and Vˆ is the variance estimate of the error term of the single threshold model If the null hypothesis of no threshold is rejected, it is necessary to test whether, or not, there is a single threshold effect Testing for the null hypothesis of the single threshold effect ( H 02 : G1 G or G G ) is through the LR statistic in Eq (5): LR1 ( S1 (J ) S1 (Jˆ )) / Vˆ (5) Using the bootstrap approach suggested by Hansen [7], we can find the asymptotic distribution, so pvalues constructed from the bootstrap are asymptotically valid If the null hypothesis of the single threshold effect is rejected, the LR statistic for testing double threshold effects should be adapted [7] (2)Data definitions Based on data availability, we collect the panel data from 29 provinces of China except Tibet, Ningxia, Taiwan, Hong Kong, and Macao for 2000-2013, including residential electricity consumption, residential electricity price, GDP per capita, population at the end of the year, cooling degree days, and rainfall It should be mentioned that China has implemented three-tier-tariffs for household electricity in 2012, and in every province, more than 80% of residential electricity demand is covered in the first-tier-price Hence, we use the first-tier-price as the residential electricity price in 2013 In addition, we use the cooling degree days and rainfall in the 29 capital cities to represent the values for their respective provinces Meanwihle, the temperatures in all capital cities are represented by monthly data Besides, the data of residential electricity consumption come from the China Energy Statistical Yearbook 2001-2013 and the Provincial Statistical Yearbook 2014; the residential electricity price data are provided by the Provincial Development and Reform Commission; data for GDP per capita, population at the end of the year, mean monthly temperature, and rainfall are from the China Statistical Yearbook 2001-2014 GDP per capita and residential electricity price data are all at the 2005 constant RMB price † The empirical results not involve multiple thresholds, we, therefore, just take single and double threshold models as examples 308 Yue-Jun Zhang and Hua-Rong Peng / Energy Procedia 104 (2016) 305 – 310 Empirical results and discussion 3.1 Linear regression results The panel unit root test is widely used to examine the stationarity of each variable [8], and the results indicate that all the variables are integrated at order one.‡Then we develop the linear panel regression to estimate the coefficients according to Eq (1), and the F-test results lead us to adopting the individual fixed effect variable intercept model, as shown in Table We can find that the model fits well, and all coefficients are significant at the 5% significance level, and in the linear relationship, there is a direct rebound effect in China’s residential electricity consumption, with the average size of the direct rebound effect for China’s residential electricity consumption arriving at 71.53% during 2000-2013 This indicates that when electricity utilisation efficiency improves, there are 71.53% of the expected electricity savings being offset by the extra electricity consumption due to efficiency improvements and the cheaper electricity service cost, and only 28.47% of the expected electricity savings can be attained The result is approximate to the direct rebound effect of China’s residential electricity consumption estimated by Wang et al [9] and larger than the direct rebound effect in some developed countries For example, Chitnis and Sorrell [10] estimate that the rebound effect of UK household electricity consumption is 48% using the almost ideal demand system model and input-output approach This is likely that the energy rebound effect in developing countries appears more significant [11] Table Estimation of linear panel a Regressor Coefficient estimate Standard error constant -9.5843 (0.0000) 0.7725 ln PE -0.7153 (0.0325) 0.1077 0.8104 (0.0000) 0.0404 ln PGDP ln POP 0.7001 (0.0000) 0.0733 ln CDD 0.0229 (0.0456) 0.0114 -0.0346 (0.0246) 0.0153 ln RAIN R 0.99 F-statistic 1564.27(0.0000) Direct rebound effect 71.53% a The p-value of corresponding statistics are reported in parentheses 3.2 Results of panel threshold model According to Eqs (2) and (3), we employ GDP per capita, population, cooling degree days, and rainfall as the threshold variable of the panel threshold model, respectively Moreover, we use 300 bootstrap replications to test whether, or not, the model has threshold effects and estimates the direct rebound effect of China’s residential electricity consumption for different regimes of the threshold variables The results indicate that the models with GDP per capita, cooling degree days, and rainfall as threshold variables have only one threshold effect; however, the model with population as its threshold variable has no threshold effect at the 10% significance level Then, according to Eqs (4) and (5), we can attain the threshold estimates and their confidence intervals of corresponding models Taking the threshold estimates of GDP per capita, cooling degree days, and rainfall as the dividing line, there are two different regimes, respectively § Next, we get the regression results of the models as shown in Table 2, which use GDP per capita, cooling degree days and rainfall as the threshold variables, respectively All regression coefficients of the models are statistically significant at the 10% level Based on the results above, we can obtain the following arguments ‡ § Detailed results can be obtained upon request Detailed results can be obtained upon request 309 Yue-Jun Zhang and Hua-Rong Peng / Energy Procedia 104 (2016) 305 – 310 Table Threshold effect results Regressor constant ln PGDP ln POP ln CDD ln RAIN ln P * I (ln PGDP d 10.1151) ln P * I (ln PGDP ! 10.1151) ln P * I (ln CDD d 7.2749) ln P * I (ln CDD ! 7.2749) ln P * I (ln RAIN d 7.4851) ln P * I (ln RAIN ! 7.4851) F-Statistic a Model A Model B Model C -11.8205 (0.0000) 0.8596 (0.0000) 0.9415 (0.0000) 0.0302 (0.0060) -0.0683 (0.0110) -0.6786 (0.0000) -0.5474 (0.0000) -9.3403 (0.0000) 0.7773 (0.0000) 0.7429 (0.0000) 0180 (0.1100) -0.0789 (0.0040) -9.6429 (0.0000) 0.8056 (0.0000) 0.7671 (0.0000) 0.0264 (0.0190) -0.1043 (0.0000) -0.7452(0.0000) -0.8923(0.0000) 1009.17 (0.0000) 1031.72 (0.0000) -0.6830 (0.0000) -0.8631 (0.0000) 1011.03 (0.0000) The p-value of corresponding statistics are reported in parentheses Model A refers to the model whose threshold variable is ln PGDP ; Model B refers to the model whose threshold variable is ln CDD ; and Model C refers to the model whose threshold variable is ln RAIN First, GDP per capita and population contribute greatly to promoting residential electricity consumption, while cooling degree days contribute slightly to its promotion On the contrary, rainfall is helpful at restraining residential electricity consumption, but the effect is small From Tables 2, we can find the coefficients of GDP per capita and population are positive and close to one, and the coefficient of cooling degree days is positive but small Second, the increase of GDP per capita will contribute to declining the direct rebound effect of China’s residential electricity consumption During 2000-2013, in the provinces where GDP per capita is lower than 24713.38 Yuan (most provinces, accounting for 73.40%), i.e., under the low income regime, the direct rebound effect of residential electricity consumption is 67.86% In the provinces with GDP per capita higher than 24713.38 Yuan (such as Beijing, Shanghai, Guangdong, Tianjin, Jiangsu, and Zhejiang), i.e., under the high income regime, the direct rebound effect is 54.74% Third, the decrease in cooling degree days will contribute to declining the direct rebound effect of residential electricity consumption During 2000-2013, in the provinces where cooling degree days are less than 1443.61 centigrade (most provinces, accounting for 83.99%), i.e., under the low cooling degree day regime, the direct rebound effect is 74.52% In provinces with cooling degree days larger than 1443.61 centigrade (such as Fujian, Guangdong, Guangxi, Hainan, Chongqing, and Sichuan), i.e., under the high cooling degree day regime, the direct rebound effect arrives at 89.23% Finally, the decrease in rainfall will contribute to diminishing the direct rebound effect of residential electricity consumption During 2000-2013, in the provinces where rainfall is less than 1781.3 millimetres (most provinces, accounting for 95.32%), i.e., under the light rainfall regime, the direct rebound effect is 68.30% However, in provinces with rainfall more than 1781.3 millimetres (such as Hainan and Guangdong), i.e., under the heavy rainfall regime, the direct rebound effect reaches 86.31% Conclusions Based on the results above, we may safely come to several important conclusions First, in both linear and non-linear relationships, GDP per capita and population have great positive impact on residential electricity consumption; and cooling degree days have small positive impact on residential electricity consumption; while rainfall has small negative influence on residential electricity consumption Second, 310 Yue-Jun Zhang and Hua-Rong Peng / Energy Procedia 104 (2016) 305 – 310 in the linear relationship, the direct rebound effect of China’s residential electricity consumption during 2000-2013 reached 71.53% on average Finally, the increase of GDP per capita and the decrease in cooling degree days and rainfall may help to reduce the direct rebound effect of China’s residential electricity consumption The direct rebound effect of China’s residential electricity consumption is 67.86% (54.74%) in the low (high) income regime, 74.52% (89.23%) in the low (high) cooling degree regime, and 68.30% (86.31%) in the light (heavy) rainfall regime These conclusions may help policy makers to better understand the impact of rebound effect so as to not to overestimate energy savings achieved by implementing energy efficiency policies in China’s residential sector Acknowledgements This work was supported by the National Natural Science Foundation of China (nos 71273028, 71322103), National Special Support Program for High-Level Personnel from the central government of China and Hunan Youth Talent Plan References [1] Sorrell S, Dimitropoulos J, Sommerville M Empirical estimates of the direct rebound effect: A review Energy Policy 2009; 37: 1356-1371 [2] Ang BW, Goh TN, Liu XQ Residential electricity demand in Singapore Energy 1992;17:37-46 [3] Lam JC Climatic and economic influences on residential electricity consumption Energy Conversion and Management 1998;39: 623-629 [4] Zhang M, Song Y, Li P, Li H Study on affecting factors of residential energy consumption in urban and rural Jiangsu Renewable and Sustainable Energy Reviews 2016;53:330-337 [5] Wangpattarapong K, Maneewan S, Ketjoy N, Rakwichian W The impacts of climatic and economic factors on residential electricity consumption of Bangkok Metropolis Energy Buildings 2008;40:1419-1425 [6] Bessec M, Fouquau J The non-linear link between electricity consumption and temperature in Europe: a threshold panel approach Energy Econmics 2008;30: 2705-2721 [7] Hansen BE Threshold effects in non-dynamic panels: Estimation, testing, and inference Journal of Econometrics 1999;93: 345368 [8] Ozturk I, Aslan A, Kalyoncu H Energy consumption and economic growth relationship: evidence from panel data for low and middle income countries Energy Policy 2010;38: 4422-4428 [9] Wang Z, Lu M, Wang JC Direct rebound effect on urban residential electricity use: an empirical study in China Renewable and Sustainable Energy Reviews 2014;30,124-132 [10] Chitnis M, Sorrell S Living up to expectations: Estimating direct and indirect rebound effects for UK households Energy Econmics 2015;52:S100-S116 [11] Sorrell S, Dimitropoulos J UKERC review of evidence for the rebound effect: Technical Report 5: Energy, productivity and economic growth studies.2007 Biography Dr Yue-Jun Zhang is a Professor at Business School, Hunan University, China, and the Director of Center for Resource and Environmental Management, Hunan University He got his PhD degree in Management Sciences from Chinese Academy of Sciences (CAS) in 2009, and his research focuses on energy economics and policy modeling ... of the linear model, S1 is the sum of the squared residual of the single threshold model, Jˆ is the OLS estimate of J , and Vˆ is the variance estimate of the error term of the single threshold... model If the null hypothesis of no threshold is rejected, it is necessary to test whether, or not, there is a single threshold effect Testing for the null hypothesis of the single threshold effect. .. factors, this work also explores the non-linear relationship among them using the panel threshold model Second, excepting the estimation of the direct rebound effect of residential electricity consumption