Avaiabe onine at www.ijpe-onine.com Vo. 13, No. 4, Juy 017, pp. 418-46 DOI: 10.3940/ijpe.17.04.p9.41846 Thresh Effects and Spatia Spiover of Eectricy Consumption on Economic Growth Liping Guo a, *, Jie Zhou a, Xiaowei Yang b a State Grid Zhejiang Yuyao Power Suppy Company, Yuyao315400, Zhejiang, China b Ningbo Cy Coege of Vocationa Technoogy, Ningbo315100, Zhejiang, China Abstract The importance of the suppy of eectricy to economic growth is sef-evident. However, the existing research mosty focuses on the inear reationship between eectricy consumption and economic growth, whie ignoring the noninear reationship between them. What s more, the existing research aso ignores the spatia correation of power consumption and s spiover effect on economic growth. In order to address the deficiency of the existing research, this study empiricay anayzes the effects of Chinese provincia eectricy of economic growth based on the pane threshod mode and pane spatia mode wh 30 provincia sampes from 1995 to 014. The resuts show that China's power consumption causes threshod effect and spatia spiover on economic growth. When per capa GDP, per capa consumption expendure, and per capa investment exceed a certain threshod, the posive roe in stimuating economic growth of eectricy consumption is significanty different from those of the easticy when the per capa GDP, per capa consumption expendure, and per capa investment do not exceed the corresponding thresh vaue. Not ony does the oca eectricy consumption significanty promote economic growth of the oca region, the eectricy consumption spiover effects of the adjacent provinces can aso promote the economic growth of the oca region. If the spatia spiover effect of eectricy consumption is negected, the supportive effect of eectricy consumption on economic growth wi be underestimated. Keywords: Eectricy consumption; economic growth; noninear reationship; threshod effect, spatia spiover (Submted on January, 017; Revised on Apri 15, 017; Accepted on June 13, 017) 017 Totem Pubisher, Inc. A rights reserved. 1. Introduction China's economic take-off accompanied by China's power industry deveopment and eectricy demand has become another "barometer" of China's economic growth [1]. Statistica data shows that China's gross domestic product (GDP) grew from 57733 biion Yuan to 015 676708 biion Yuan during 1995-015; meanwhie, the eectricy consumption demand of the whoe society has increased from 107.311 biion kiowatts per hour in 1995 to 5588.73 biion kiowatts per hour in 015. In the past 0 years, athough China's eectricy consumption and economic growth rates are not consistent wh each other, their absoute amounts show simiary significant characteristics of growth, refecting their cose reationship. A ot of erature focuses on the reationship between economic growth and eectricy consumption, but these studies ony investigated the inear reationship between them. According to the different types of sampe data, these eratures can be divided into two categories: One category is based on cointegration modeing and Grainger causay anaysis wh the time series. For exampe, researchers have verified the existence of the hypothesis on the reationship between China's eectricy consumption and * Corresponding author. E-mai address: ipingguoyy@163.com.
Thresh Effects and Spatia Spiover of Eectricy Consumption on Economic Growth 419 economic growth using cointegration anaysis and error correction mode wh the time series data during 195-001. Shiu and Lam appied the error-correction mode to examine the causay between eectricy consumption and rea GDP in China during 1971-000 and the estimation resuts indicate that rea GDP and China s eectricy consumption are cointegrated and there is unidirectiona Granger causay from eectricy consumption to rea GDP but not vice versa. Yuan et a. appied the cointegration theory to examine the causa reationship between eectricy consumption and rea GDP in China during 1978-004 and the estimation resuts indicate that rea GDP and eectricy consumption for China are cointegrated and there is ony unidirectiona Granger causay from eectricy consumption to rea GDP but not the vice versa. Researchers have confirmed that there existed a ong-term equiibrium reationship between China's eectricy consumption and the rea GDP. Sun and Zhang aso confirmed that Granger causay ony existed from eectricy consumption to economic growth using quartery data since 004 []. Another category is based on cointegration modeing and error correction anaysis wh pane data. For exampe, using the data of 30 provinces and cies in China. Researchers used the pane cointegration method to confirm that the eectricy consumption payed a significant roe in promoting the economic growth of provinces and cies. Using simiar methods, Herrerias et a., using inter-provincia data of China from 1995 to 009, verified the mutua dependence between eectricy consumption and economic growth in China [3]. Li et a. [4] confirmed the difference of the reationship in the East and West between eectricy consumption and economic growth wh inter-provincia data of China from 1990 to 011; based on the pane error correction mode, Liu and Ma not ony confirmed the cointegration reationship between China's eectricy consumption and economic growth, but aso proved the Granger causay from the power consumption to economic growth [5]. Some schoars doubt whether there is a symmetric reationship between eectricy consumption and economic growth. Zhang regarded that woud get the miseading resuts using the vector error correction mode if we coud confirm the existence of asymmetric adjustment between eectricy consumption and economic growth. But using the two-mechanism threshod cointegration test wh China's time series data during the years 1955-009, Zhang didn t find the existence of the noninear cointegration reationship between China's eectricy consumption and economic growth. This difference in resuts stems from mutipe factors: Li [4] used the threshod cointegration mode integrated wh the time series data from 1980 to 009. He confirmed the existence of the ong-term noninear reationship between economic growth and China's eectricy consumption, the usage efficiency of eectricy, eectricy prices and other factors; Liang seected the time series during 1953-00 and appied the Markov regime method. The Granger test method aso confirmed the existence of the noninear reationship between eectricy consumption and economic growth in China [6]. Athough this noninear reationship between eectricy consumption and economic growth may be objective, the pubic may doubt the reiabiy of the empirica resuts ony wh the time series but not the pane data. In addion, a major faw in the existing studies on the reationship between eectricy consumption and economic growth is that these studies ignore the spatia correation of different provincia power consumption. According to Ansein, if eectricy consumption is in the presence of significant spatia correation, negecting the effect of spatia correation woud ead to errors or inconsistencies of estimation. In fact, the erature on the reationship between eectricy consumption in other countries has noticed the space reated issues of eectricy consumption. To our knowedge, ony Leticia et a. used the spatia autoregressive mode wh autoregressive disturbances to anayze the presence of spatia spiover effects in eectricy consumption using an aggregate pane data set on the 46 Spanish provinces for the period of 001 to 010 [7]. But in their paper they ony estimated the price and income easticy of eectricy demand. In order to provide more convincing evidence for the existence of the noninear reationship between eectricy consumption and economic growth, this study firsty constructed the pane threshod mode to empiricay anayze the threshod effect of China power consumption on economic growth. In addion, this study further constructs a pane spatia econometric mode to test the spatia spiover effects of eectricy consumption to economic growth, which is negected in the existing erature. The empirica resuts show that there not ony exists significant threshod effects of the provincia eectricy consumption on economic growth in China, but aso causes spatia spiover effects. When the provincia per capa GDP is more than 4361.703 Yuan, the expendure of per capa consumption is more than 1308.098 Yuan, and per capa investment is more than 1896.945 Yuan, the effects of eectricy consumption on economic growth are strikingy different from the suations when the per capa GDP, per capa consumption and per capa investment are ess than the corresponding threshod vaue. Eectricy consumption of the provinces not ony pays a promoting effect on the economic growth of oca provinces, but aso has the spatia spiover effects on the economic growth of neighbouring provinces. Not considering the spatia spiover effects may underestimate the roe of eectricy consumption in economic growth. The setup of this artice is arranged as foows: the second part contains the research design on modes, methodoogy and data. The third part contains the empirica resuts and interpretation, and the fina part contains the concusion and poicy impication.
40 Liping Guo, Jie Zhou, and Xiaowei Yang. Modes, Methodoogy and Data.1. The construction of the basic mode In order to empiricay test the reationship between eectricy consumption and economic growth, we need to buid the corresponding econometric mode. We introduced the power consumption (E), capa (K), and abor (L) input as input eements into the production function named Cobb and Dougas, and reaxed the assumption of constant returns to scae. The foowing is the equation: Y A K L E (1) Whereas i indicates the provincia domain, t represents the period, A is the technoogy progress, Y is a measure of economic growth. Greek etters,, are parameters to be estimated. The formua () is obtained by taking the natura ogarhm on both sides of the equation (1): og( Y ) og( A ) og( K ) og( L ) og( E ) () If we do not consider the infuence of other factors, according to formua (), we can anayze the impact of capa, abor and eectricy consumption on the economic growth in a cosed economy. However, considering the infuence of other factors on the adjustment of China's industria structure, some contro variabes need to be added to the formua () to buid the basic econometric mode (3): og( Y ) og( A ) og( K ) og( L ) og( E ) j Contro j1 (3) i.. The anaytica method of the threshod effect We introduce the pane threshod regression method to anayze the noninear effect of power consumption on economic growth and the method needs to find the crica point of the mutation. According to [8], assumption ony a singe threshod, the mode can be constructed as foows: og( Y ) c og( K ) og( L ) og( E ) I ( Thresh ) 1 og( E ) I ( Thresh ) j + i j1 Contro (4) Among them, i is the representative function, Thresh is the threshod variabe, indicates the corresponding threshod vaue, and the threshod vaue is estimated by the grid search method, ˆω= argmin S (ω). 1 In mode (4), when the crica vaue exceeds, means that the infuencing coefficient of the eectric power consumption on economic growth can be changed. It usuay needs to construct the F statistic ( F ( S - S (ˆ )) / ˆ ) to test the significance of the threshod. In the formua of F, S 0 0 1 is the sum of squares of residuas whout a threshod effect, and S ( ˆ ) 1 is the residua sum of squares wh a threshod effect, and is the variance of the perturbation term. For the singe threshod regression mode, the origina hypothesis indicates that there is no threshod effect, and the aternative hypothesis represents that there is ony one threshod vaue. Using the bootstrap method, the asymptotic distribution of the corresponding F statistic and the corresponding probabiy vaue can be obtained based on the ikeihood ratio ( LR( ) ( S ˆ ˆ 1( ) - S1( )) / ). When the probabiy vaue is sufficienty sma, the origina hypothesis H0 is rejected. Otherwise, the aternative hypothesis H1 is accepted. H 0 ˆ If there exists mutipe threshods, the formua (4) needs to be abe to test the existence of the threshods. Taking the doube threshod mode as an exampe, the threshod mode can be constructed as the foowing form: H 1
Thresh Effects and Spatia Spiover of Eectricy Consumption on Economic Growth 41 og( Y ) c og( K ) og( L ) og( E ) I ( Thresh ) 1 1 og( E ) I ( Thresh ) 1 og( E ) I ( Thresh ) 3 3 j Contro j 1 i (5) The methods for judging the doube threshod or singe threshod are not the same. Usuay the threshod vaue determined first, and the second threshod vaue S ( ˆ ) ˆ is defined ater foowed by the minimized sum of squared residua. When the second threshod vaue is fixed, we go back to adjust the vaue of the first threshod vaue. Next, minimize the sum of squared residuas is updated and so on. S ( ˆ ) 1 1.3. The anaytica method of spatia spiover effect. The test method about tripe threshod is simiar to the doube threshod test, According to the different approach on spatia pane mode setting, the spatia pane mode can be divided into two kinds: one is the pane spatia autoregressive mode (SAR). The expression is as foows: ˆ 1 is N og( Y ) w og( Y ) og( K ) j1 ij jt 1 og( L ) og( E ) 3 Contro j j1 i (6) The pane spatia error mode (SEM). The expression is as foows: og( Y ) og( K ) og( L ) og( E ) 1 3 N Contro ( j) j1 i w j1 ij jt (7) In the (6) and (7) mode, i denotes the provinces and t denotes the year; the expained variabe y is N T vector and expanatory variabes is N T vector; and represents the k-th and the j-th parameter to be estimated respectivey; is the xk ( ) ( k ) eement of the spatia weight matrix; i ( j) is the error term wh an independent same distribution; zero mean and expectations named ; refects the effect of a specific space; is the space autoregressive coefficient; autoregressive error condions, and is the spatia autocorreation coefficient. w ij is the spatia The maximum ikeihood estimation method (ML) is usuay adopted to estimate the parameters of the spatia econometric mode. Because the estimated resuts are different between the SAR mode and the SEM mode, is necessary to determine the mers of the different resuts. Maximum ikeihood vaue and the Lagrange mutipier method usuay are adopted to determine which way is better, the former usuay judges according to the ogarhm of the maximum ikeihood vaues (og L), the atter usuay judges according to the Lagrange mutipier (LM) statistic vaue. It is generay considered that the estimated resut of the mode wh arger statistics is better for both of the two methods. Tabe 1. Variabes description Variabes Code Index Construction Un Name Economic growth Y GDP 100 miion Yuan Capa input K newy fixed assets investment 100 miion Yuan Labor input Eectricy consumption L E number of empoyees at the end of the year eectricy consumption of the whoe society Ten thousand peope kw/h
4 Liping Guo, Jie Zhou, and Xiaowei Yang Industria structure stru secondary and tertiary industry/primary industry Trade openness open import and export voume open/gdp % Per capa GDP avgdp Per capa GDP Yuan/ peope Per capa Consumption avcon Per capa consumption Yuan/ peope Per capa investment avinv K/ popuation at the end of the year Weight matrix W If region i is adjacent to region w ij=1; Otherwise, w ij=0 % Yuan/ Ten thousand peope Tabe. Summary statistics of the variabes Variabe Obs Mean Std.Dev. Min Max og(y) 600 8.464 1.180 5.13 11.14 og(k) 600 7.691 1.355 4.018 10.657 og(l) 600 7.49 0.87 5.497 8.796 og(e) 600 6.48 0.909 3.466 8.563 og(stru) 600 1.970 0.893 0.556 5.35 og(open) 600-1.761 1.013-3.44 0.718 og(avgdp) 600 9.554 0.90 7.510 11.564 og(avcon) 600 8.598 0.766 7.098 10.669 og(avinv) 600 8.777 1.169 6.05 11.159.4. Data source and variabes description In this paper, 30 provinces (except for Tibet) are seected as the research objects. The sampe period begins in 1995 and ends in 014. Basic data comes from the Chinese Nationa Bureau of Statistics webse coumn and Chinese Economic and Socia Deveopment Statistics Database. The variabes in this paper are described in Tabe 1. It s necessary to point out that setting the vaue of eementary eements to 0-1 adjacency spatia weight matrix W accords to geographic adjacent or not. If province i is geographicay adjacent to region w ij=1; Otherwise, w ij=0. The geographic adjacent information of 30 provinces comes from The Map of Peope's Repubic of China. Tabe shows the summary statistics of the variabes. 3. Empirica Resuts and Anaysis 3.1. The resuts of pane threshod regression According to [9], we firsty tested whether there exists a singe threshod, doube threshod, tripe threshod condion or not when the power consumption effects on economic growth. Tabe 3 ists the corresponding F statistics vaues, probabiy vaues and the corresponding crica vaue at 1%, 5%, 10% significant eve when the threshod variabe is per capa GDP og(avgdp), per capa consumption og(avcon) and investment per capa og(avinv) wh singe threshod, doube threshod or tripe threshod, respectivey. As Tabe 3 shows, in the process of threshod mode test wh threshod variabes named GDP per capa, we firsty tested whether there is a singe threshod mode or not. Because the probabiy of accepting the nu hypothesis H 0 "inear mode" is 0.000, therefore the aternative hypothesis H 1 "singe threshod mode" is accepted; we further tested doube threshod mode whether there exists a doube threshod mode or singe threshod mode. Because of the accepting probabiy, the nu hypothesis H 1 is 0.070; therefore, shoud accept the nu hypothesis at the 5% significance eve. Finay, the resuts show that shoud accept the concusion of "singe threshod". Simiary, when per capa consumption is the threshod variabe, shoud refuse the nu hypothesis "inear mode" at 1% significance eve, whie accepting the aternative hypothesis of "singe threshod mode"; For the doube threshod test, the accepting probabiy of singe threshod hypothesis is 0.150, F statistic vaue beow the crica vaue at the 1%, 5% or 10% of significance eve. So, the nu hypothesis shoud be accepted. That is to say, when per capa consumption is acted as the threshod variabe, the singe threshod mode maybe the best choice.
Thresh Effects and Spatia Spiover of Eectricy Consumption on Economic Growth 43 Threshod variabe og(avgdp) og(avcon) og(avinv) Tabe 3. Test for threshod effects Crica vaues Test assumption F-statistics P-vaue 1%,5%,10% H 0: None threshod; H 1: Singe threshod 11.530 0.000 47.793, 34.730, 9.975 H 0: Singe threshod; H 1: Doube threshod 31.960 0.070 55.178, 36.901, 5.481 H 0: Doube threshod; H 1: Tripe threshod 16.450 0.647 6.1, 48.780, 40.987 H 0: None threshod; H1: Singe threshod 114.780 0.000 51.678, 39.14, 3.058 H 0: Singe threshod; H 1: Doube threshod 8.870 0.150 48.134, 39.307, 30.74 H 0: Doube threshod; H 1: Tripe threshod 1.550 0.837 61.186, 47.10, 40.933 H 0: None threshod; H1: Singe threshod 34.560 0.053 50.44, 34.809, 9.61 H 0: Singe threshod; H 1: Doube threshod 17.190 0.47 39.006, 8.659, 3.315 H 0: Doube threshod; H 1: Tripe threshod 16.10 0.73 40.137, 6.965,.545 When per capa investment is taken as the threshod variabe, seems that we shoud accept the "inear mode" nu hypothesis at 1% and 5% significance eve, but at the 10% eve of significance, the resut refuses the nu hypothesis H 0 and accepts the aternative hypothesis H 1 of "singe threshod mode"; whie for the two threshod test, the probabiy of the singe threshod hypothesis is 0.47 and the F statistics is ess than the crica vaue at 1%, 5% or 10% significance eve. So, shoud accept the nu hypothesis. That is to say, using the singe threshod mode to estimate the coefficient is the best choice when per capa investment is the threshod variabe. Tabe 4 ists the threshod vaue of og(avgdp), og(avcon) and og(avinv) obtained by the grid search method in the singe threshod mode. From Tabe 4, we can draw the concusion that the threshod vaue of per capa GDP, per capa consumption and per capa investment is 10.654 (avgdp=4361.703), 9.479 (avcon =1308.098) and 7.548 (avinv =1896.945), respectivey. Tabe 4. The estimated threshod vaue Threshod variabe Test resuts Threshod vaue Confidence Interva (95%) og(avgdp) singe threshod 10.654 [10.619, 10.665] og(avcon) Singe threshod 9.479 [9.471, 9.483] og(avinv) Singe threshod 7.548 [7.489, 7.574] The pane data mode can be divided into the fixed effect mode and random effect mode according to the difference of the residua component. According to [10], the fixed effect mode is a better choice when the regression anaysis of the sampe is imed to some particuar individuas [9]. The regression resuts in Tabe 5 are based on the pane fixed effect. The first coumn in Tabe 5 is the resuts of ordinary inear regression, the second coumn in Tabe 5 is the threshod regression resuts based on the threshod variabe named per capa GDP, the third coumn in Tabe 5 is the threshod regression resuts based on the threshod variabe named per capa consumption, the fourth coumn in Tabe 5 is the threshod regression resuts based on the threshod variabe named per capa investment. The resut of inear regression mode shows that the variabes named og(k), og(l) and og(e) are in a significanty posive reationship wh og(y), and the sum of the easticy is 1.046. Because the sum of easticy of capa China, abor and eectricy consumption to the output easticy is greater than 1, indicating in returns to scae stage of increasing. Further anaysis of three noninear singe threshod regression mode, we find that the effects of eectricy consumption on economic growth are different when the per capa GDP, per capa consumption and per capa investment are more than or ess than the corresponding threshod vaues. When the per capa GDP is not more than 4361.703 Yuan, the easticy of eectricy consumption to economic growth is 0.31; and when the per capa GDP is more than 4361.703 Yuan, the easticy of the power consumption to economic growth is 0.350. However, when the per capa consumption is not greater than 1308.098 Yuan, the easticy of eectricy consumption to economic growth is 0.319; and when the per capa consumption is more than 1308.098 Yuan, the easticy of eectricy consumption to economic growth is 0.346. When the per capa investment is not more than 1896.945 Yuan, the easticy of eectricy consumption to economic growth is 0.339; whie the per capa investment is more than 1896.945 Yuan, the easticy of eectricy consumption to economic growth is 0.319. The estimation resuts show that there exist the "threshod" characteristics when eectricy consumption affects economic growth. That is to say, when the per capa GDP, per capa consumption and per capa investment exceed a specific vaue, the margina output of eectrica consumption wi
44 Liping Guo, Jie Zhou, and Xiaowei Yang change. But further comparing the parameter size before and after threshod, we find that per capa GDP and per capa consumption for threshod estimation of eectricy consumption of the margina output easticy is increasing, whie per capa investment for the easticy of margina output of eectricy consumption for threshod estimation is reversed, which indicates that the margina output rate decreased when a arge number of new investments drive the growth of the eectricy consumption. Tabe 5: Regression resuts of threshod mode c og(k) og(l) og(e) γ 1 γ og(e) og(e) og(stru) og(open) Threshod(None) Threshod (og(avgdp)) Threshod(og(avcon)) Threshod (og(avinv)) 0.96 *** 0.340 0.343-0.868 [1.500] [0. 900] [0.910] [-.330] 0.634 *** 0.48 *** 0.481 *** 0.515 *** [56.760] [36. 40] [36.050] [37.160] 0.377 *** 0.3 *** 0.34 *** 0.453 *** [3.50] [5.910] [5.960] [8.140] 0.035 *** [3.050] R 0.978 adj R 0.977 0.31 *** 0.319 *** 0.339 *** [14.500] [14.380] [14.50] 0.350 *** 0.346 *** 0.319 *** [15.650] [15.50] [13.560] 0.64 *** 0.44 *** 0.459 *** 0.546 *** [5.460] [.730] [.930] [.980] 0.163 *** 0.008 0.003 0.001 [19.550] [0.510] [0.170] [0.040] F(6,564) 5668.74 *** 5667.90 *** 500.61 *** R : whin 0.9837 0.9837 0.9816 between 0.8995 0.8987 0.8731 overa 0.9371 0.9366 0.99 Note: *** indicates significance at the 1% eve, and the data in the square brackets is the t statistics vaue. In addion, due to independenty expained variabes such as og(k), og(l), og(stru) and og(open), the regression coefficients are sti posive, and is simiar to the concusion made from the inear regression mode. It confirms that input of capa and abor and industry structure pay promoting roes in economic growth and the roe of foreign trade promoting economic growth needs to be further verified. 3.. The resuts of spatia pane regression The resuts in Tabe 6 are obtained by using the maximum ikeihood method (ML) combined wh the 0-1 adjacency matrix to estimate modes (6) and (7). The cacuation resuts of Log L statistics, LM statistics and Robust LM statistics are aso isted in Tabe 6. The Log L statistic vaue shows that the SAR mode, under the condion of spatia fixed effects, is better than the estimation resuts based on the SEM mode wh the fixed time. However, the SEM mode wh space and time fixed effect are better than the corresponding estimation resuts based on the SAR mode. Further testing of the estimated resuts according to the LM SAR and LM SEM, Robust LM SAR and Robust LM SEM statistics gives resuts that are competey consistent wh the Log L statistics resuts. Therefore, this paper seects mode SAR (1), SEM () and SEM (3) to expain the spiover effects of eectricy consumption and other factors on economic growth. In the SAR mode (1), the estimated parameters of W*og(Y) refecting the spatia correation is 0.193 at the 1% eve of significance. In the SEM mode () and SEM mode (3), the estimated parameters of the variabe named spat.err refecting the of the spatia correation aso meets 1% eve of the significance, which indicates China's economic growth in the presence of strikingy spatia correation, and aso indicates the how previous studies negected the spatia effects.
Thresh Effects and Spatia Spiover of Eectricy Consumption on Economic Growth 45 og(k) og(l) og(e) og(stru) og(open) W*og(Y) spat.err. Tabe 6. Regression resuts of spatia modes SAR(1) SEM(1) SAR() SEM() SAR(3) SEM(3) 0.355 *** 0.47 *** 0.598 *** 0.61 *** 0.368 *** 0.455 *** [7.663] [40.351] [64.980] [73.399] [41.54] [43.79] 0.45 *** 0.68 *** 0.417 *** 0.408 *** 0.81 *** 0.316 *** [6.574] [6.564] [37.949] [38.441] [7.79] [7.074] 0.18 *** 0.5 *** 0.034 *** 0.0 *** 0.179 *** 0.59 *** [9.871] [14.448] [3.155] [.0] [10.707] 13.543] 0.37 *** 0.407 *** 0.159 *** 0.153 *** 0.39 *** 0.359 *** [19.95] [17.993] [1.491] [1.406] [18.853] 16.447] -0.045 *** -0.06 * 0.11 *** 0.104 *** 0.115 *** 0.108 *** [-4.05] [-1.904] [1.101] [10.455] [13.156] [11.361] 0.193 *** 0.01 ** 0.194 *** [8.99] [.186] [33.55] 0.439 *** 0.375 *** 0.3 *** [10.46] [8.4] [6.950] spatia fixed effects yes yes no no yes yes time period fixed effects no no yes yes yes yes R 0.996 0.995 0.980 0.98 0.995 0.994 sigma^ 0.006 0.007 0.08 0.05 0.006 0.009 og-l 703.15 65.74 0.931 46.56-358.88 570.373 LM sar 81.687 *** 0.130 5.007 *** Robust LM sar 14.591 *** 0.73 1.9734 LM sem 67.144 *** 37.166 *** 5.373 *** Robust LM sem 0.048 37.759 *** 9.339 *** Note: *, **, *** indicate significance at the 10%, 5% and 1% eve respectivey, and the data in the square brackets are the t statistics vaue. In the SAR mode (1), the direct effect of the provincia eectricy consumption on economic growth is 0.18, and the spiover effect of the neighborhood eectricy consumption on the oca economic growth is 0.035 (=0.18*0.193). Two effects is 0.17, which means for every 1% growth in eectricy consumption of the oca province and the neighboring provinces, the economic growth of oca province wi grow by 0.17%. In the SEM mode (), a of the expanatory variabes meet the t-test. The infuencing coefficient of eectricy consumption on economic growth is 0.0. The estimated parameter of spat.err. is 0.375, which indicates that there exists spiover of the eectricy consumption in the adjacent region on economic growth, and s size is about 0.008 (=0.0*0.375). So, the direct effects and spatia spiover promote roe is 0.030. Simiary, in the SEM mode (3), the direct effect of the oca province eectricy consumption on economic growth is 0.59, and adding the corresponding spatia spiover effect 0.083 (=0.59*0.3), the tota effects of eectricy consumption on economic growth is 0.34. The above anaysis shows that if the spatia spiover effect is negected, the promotiona roe of eectricy consumption on economic growth woud be underestimated. Further investigation of the estimated parameters of the other variabes, the regression coefficients of og(k) on economic growth is at intervas between 0.355 to 0.61 and the regression coefficients of og(l) on economic growth is at intervas between 0.45 to 0.417, which indicate that capa and abor both promote economic growth. The industria structure optimization index as a contro variabe, the regression coefficients of og(stru) are at intervas between 0.153 to 0.37, which indicate that the industria structure of the adjustment and optimization is conducive to economic growth. The effect of foreign trade on economic growth in the Mode SAR (1) is significanty negative, whie the estimated coefficient of og(open) on the economic growth in the mode SEM () and SEM (3) are both posive, which shows that the reationship between foreign trade and economic growth may be uncertain. That is to say that the choice of the econometrics mode payed a very important roe in the empirica resuts.
46 Liping Guo, Jie Zhou, and Xiaowei Yang 4. Concusions This study empiricay anayzes the effects of China s power consumption to economic growth based on the pane threshod mode and pane spatia mode wh 30 provincia sampes during 1995-014, and two concusions are made: (1) The promotiona roe of Chinese provincia power consumption to economic growth wh noninear conversion mechanism, and the promotiona effects are effected and restricted by the economic basis of the province, per capa consumption and per capa investment eve. When a provincia per capa GDP and per capa consumption are more than 4361.703 Yuan and 1308.098 Yuan respectivey, the supporting roe of eectricy consumption on economic growth is strikingy greater than that the per capa GDP and per capa consumption are ower than the corresponding threshod vaue; whie per capa investment is more than $1896.945, the supporting roe of eectricy consumption in economic growth is ower than the effect of that when the per capa investment is ess than the corresponding threshod vaue. () There exists a significant spatia correation between the economic growth of different provinces. Eectricy consumption of different provinces has a direct impact on oca economic growth, but aso has a spiover effect on the economic growth of neighboring provinces. What s more, the more neighboring provinces, the more the growth spiover may be from the eectricy consumption of neighboring provinces. If we ignore the spatia spiover effect of eectricy consumption on economic growth coming from the adjacent provinces, is, to some extent underestimating the effects of China's eectricy consumption on oca economic growth. There is no doubt that this paper broadens peope's understanding of the mechanism of eectricy consumption to economic growth. The above concusions have important impication for the corresponding poicy adjustments. Because the eectricy consumption has significant growth effects, is necessary to fuy meet the demand for eectricy production and ife in order to maintain steady economic growth; because the eectricy consumption of different provinces has significant spatia spiover effects, each province shoud pay fu attention to the roe of spatia association of eectricy consumption in different provinces, and get rid of the artificia segmentation of the eectricy consumption market. The cooperative distribution of cross regiona transport may secure the eectricy suppy for the eectricy consumption of production and ife in the oca and adjacent region; taking into account the imed amount of eectricy suppy, the contro strategy of eectricy consumption shoud take into account the differences and dynamics of regiona economic condions. If the eectricy demand sharpy increases too much for investment, is necessary to contro the eectricy consumption and priorize regions wh high margina output. References 1. Sun, X. D and Zhang, L.L., Eectric power consumption and economic growth: a causa anaysis based on quartery data, Journa of North China Eectric Power Universy, no.5, pp. 34-38, 015.. Feng J J, Chen B P, Ma D C, Meng F L., Comparative study of reationship between eectricy consumption and economic growth of China and US based on cointegration anaysis, Resource Deveopment & Market, vo. 3, no. 3, pp. 88-9, 016. 3. Herrerias M. J., Joyeux, R. and Girardin E., Short-and ong-run causay between energy consumption and economic growth: evidence across regions in China, Appied Energy, vo. 11, no. 4, pp.1483-149, 013. 4. Li, Q., Wang, H.C. and Hu, A.G., China's eectricy consumption and economic growth: a causa anaysis of China's provincia pane data, China Industria Economy, no.9, pp. 19-30, 013. 5. Liu. L. and Ma X.Q., The correation between regiona economic growth and eectricy consumption, Systems Engineering, no. 6, pp. 84-90, 015. 6. Bernard Njindan Iyke., Eectricy consumption and economic growth in Nigeria: A revis of the energy-growth debate, Energy Economics, vo. 51, pp. 166-176, 015. 7. Liang, J.W., Reationship between energy consumption and economic growth from the perspective of noninear, Doctor Dissertation of Tianjin Universy, 013. 8. Leticia, M., Bázquez, G., Massimo, F. and Fabian, H., Regiona impact of changes in disposabe income on Spanish eectricy demand: A spatia econometric anaysis, Energy Economics, vo.40, no. 6, S58-S66, 013. 9. Wang, Q.Y., Fixed-effect pane threshod mode using Stata, The Stata Journa, vo. 15, no.1, pp. 11-134, 015. 10. Batagi, B.H., Econometric Anaysis of Pane Data (Fifth Edion), Chichester, Uned Kingdom: John Wiy & Sons, 013.