How does energy technology innovation affect total factor ecological efficiency: 1 Evidence from China 2

: Recently, with the dual constraints of resources and environment, to accelerate the 6 transformation of low-carbon energy driven by energy technology innovation has become a global 7 development trend. On account of the provincial data during the period of 2000 to 2017, we creatively 8 incorporate the ecological footprint into the measurement of total factor ecological efficiency so as to 9 infer the coordinated development level of 3E system more precisely. In this paper, the dynamic spatial 10 impact of energy technological innovation on regional total factor eco-efficiency is explored through 11 the spatial Durbin model, and the complex nonlinear relationship between the two is further probed by 12 constructing the panel threshold model. The following conclusions are obtained ultimately. First of all, 13 both China's provincial ecological efficiency and energy technology innovation activities possess 14 significant spatial positive correlation, which manifests as the spatial geographical distribution 15 agglomerated by the similar characteristics; Secondly, the regional energy technology innovation has a 16 remarkable spatial effect on ecological efficiency, which displays a U-shaped trend. And compared 17 with the direct effect, the spatial spillover effect is more intense, along with more stronger long-term 18 influence; Finally, taking the level of regional economic development as the moderating variable, the 19 impact of energy technology innovation on eco-efficiency emerges a conspicuous threshold effect with 20 two threshold values. Only when the level of economic development crosses the double threshold, can 21 energy technology innovation activities significantly improve the regional total factor ecological 22 efficiency. After the robustness test and discussion of the empirical model, relevant policy suggestions 23 are put forward based on the conclusions of the paper.


Introduction 28
The deterioration of the ecological environment brought by the massive mining and 29 utilization of fossil fuels, as well as the pollution emissions caused by energy consumption, 30 have increasingly exceeded the environmental carrying capacity, thus posing a potential threat 31 to both the economic security and social stability (Xu et  To be specific, the basic path of the energy transition can be attributed to two aspects of energy 40 structure adjustment and energy technology progress (Zhou 2010;Shafiei 2014). In the course 41 of energy transition, compared with the ultimate goal of energy structure transformation, 42 energy technology innovation, as the main driving force of energy transformation, plays a more 43 decisive role in realizing energy system transformation and achieving sustainable development 44 (Wu 2017). 45 According to China's Action Plan on Energy Technology Revolution and Innovation 46 (2016-2030), a sound energy technology innovation system suited to China's national 47 accelerated transformation of these innovation achievements through the imitation, learning, 246 investment and consumption of external regions, and thus exerts positive externalities, showing 247 that the social benefits are greater than the benefits of individual enterprises. In addition, energy 248 technology reform will also urge the transformation of the green industrial structure in 249 neighboring areas to a certain extent, forming a model for regional energy technology to lead 250 and drive industrial development. 251 Nevertheless, in the process of energy technology spillover, it does not necessarily lead to 252 the favorable growth of total factor ecological efficiency in different regions. The effect of such 253 influence depends on many factors (Fu 2009), among which the geographical distance, the 254 absorptive capacity of receiving region and the technological diffusion capacity of sending 255 region are the three elements that do really matter (Shangguan 2016). To be specific, the 256 geographical proximity of different regions makes for technical overflow and knowledge 257 diffusion no matter from the perspective of economic development level, or the perspective of 258 the level of transportation and information technology. The closer the geographical location is, 259 the more conducive it is to transform the hidden technology spillover into the explicit 260 technology spillover; As a key influencing factor, the absorptive capacity of technology 261 undertaking region is an abstract concept integrating many factors such as regional social 262 culture, management policy, industrial structure and development level. The degree of 263 economic growth often implies the extent of the region's ability to absorb spillover technology; 264 The diffusion of regional technology is mediated by the accumulation and circulation of human 265 capital, and it has various diffusion effects on the external regions, thus giving play to different 266 technology spillover effects. 267 On such a basis, this article comes to formulate Hypothesis 1: There is a spatial spillover 268 effect of energy technology innovation on total factor ecological efficiency, and the spatial 269 effect is uncertain. 270 According to effect of crowding out and factor substitution, the initial energy technology 271 innovation is mostly characterized by high cost, low return and long product innovation cycle, 272 and such immature energy technology innovation cannot bring into play good economies of 273 scale and environmental benefits (Fan 2020). In addition, the innovation input of industrial 274 enterprises in energy utilization and development will crowd out the original productive 275 investment of enterprises to some extent, and produce the crowding out effect on other types of 276 technological innovation, which leads to the low efficiency in distributing enterprises resources 277 and the destruction of overall economic and environmental benefits. According to the 278 infrastructure lock-in effect and the "valley of death" hypothesis, the application and 279 popularization of energy technology in regions with backward economic development is 280 restricted by many institutions, such as technological system, social system and political system 281 (Geels 2007). Due to the imperfect infrastructure construction of energy supply and 282 consumption, the energy technology is easy to fall into the "chicken and egg" paradox, making 283 it hard to develop on a large scale. In regions with different levels of economic development, 284 their market stability and investment environment are widely divergent. Hence, the promotion 285 of energy technology innovation products will face different prospects and risks, and even the 286 application of some energy technology innovation products will fall into the "valley of death" 287 where the capital chain is broken (Ehlers 1999 European countries. Alola (2020) found that among the four types of economies (high, medium 291 high, medium low, low income), energy technology innovation only played a conspicuously 292 inhibiting role on CO2 discharge in the countries with high and medium income. 293 In consideration of previous analysis, this study puts forward Hypothesis 2: There is a 294 complex link between the innovation in energy technology and total factor ecological efficiency. 295 Meanwhile, under the adjustment of the level of regional economic development, the influence 296 of energy technology innovation on total factor ecological efficiency appears as a nonlinear 297 shock. 298

Spatial econometric model 300
Characteristics of technology spillovers is widely accepted by the academic point of view. 301 Therefore, this paper focuses on the transformation of energy technology for space effect of 302 total factor of ecological efficiency. On this basis, space factors in the innovation of energy 303 technology are added in the model. The spatial panel Durbin model is firstly constructed, and 304 the statistical tests are used to determine whether the spatial panel Durbin model can be 305 degenerated into the model of spatial lag model or spatial error, and then the spatial effects of 306 energy technology transformation on the total factor ecological efficiency are further explored. 307 In this paper, the spatial Durbin model (Anselin 1988) is constructed as follows: 308 Where, is the column vector of the explained variable in different regions of 310 each year. Following the STIRPAT framework which is widely used in environmental economics, this paper selects energy technology innovation as a variable to measure 312 technological level and uses population density op and capital affluence to represent 313 population factors and regional affluence, respectively. In addition, due to the increasing 314 number of factors affecting total factor eco-efficiency, environmental regulation and 315 openness to the outside world are also included in this paper. is a matrix composed 316 of core variable nET, quadratic item and control variables such as op、 respectively represent two different interaction effects in spatial metrology，that is 318 endogenous interaction effect and exogenous interaction effect; is the spatial 319 autoregression coefficient, while is the spatial autocorrelation coefficient and represents 320 the error term. Additionally, the model also contains two parameter column vectors and 321 to be estimated. 322

Threshold model 323
This study adopts the non-dynamic panel threshold regression model proposed by 324 Hansen (Hansen 1999) to examine whether there is a threshold effect between energy 325 technology innovation and total factor eco-efficiency. As an econometric model of nonlinear 326 relation test, this method can not only accurately calculate the threshold value, but also verify 327 the significance of endogenous "threshold characteristics". Therefore, a single threshold 328 model is established as follows: 329 In Equation (2), the meanings of dependent variable, core explanatory variable and each 331 control variable are the same as above. The threshold variable in the model is expressed by 332 the level of economic development , is the corresponding coefficient vector, and 333 is the threshold value. The formula also contains an index function (•), whose value is 1 334 when the corresponding condition holds, otherwise is 0. !"~( 0, # ) is the random 335 interference. Moreover, once the model passes the double threshold test, the following 336 equation can be set up. 337 It should be noted that, in the above formula, $ < # and the meanings of other 340 indicators are consistent with that of formula (2). 341

Variable description 342
The explained variable: Total factor ecological efficiency . In this paper, a super 343 efficiency SBM model considering non-expected outputs is adopted to measure total factor 344 ecological efficiency which can effectively avoid the problem of efficiency overestimation 345 and non-radial adjustment of input and output efficiency. When conditions are relaxed, it is 346 more realistic to assume that returns to scale are variable. At the same time, this paper selects  Table 1 shows the inputs of various biological accounts and energy 352 accounts and the elements of input-output listed in Table 2  technology and the research on the exploitation and application of clean energy technologies 360 (Sagar, 2004). According to the reality in China, the innovation in energy technology of new 361 energy technology research is mainly manifested in the technological innovation of non-fossil 362 energy (such as the energy of wind, ocean, biomass energy, etc.), while the technological 363 innovation in the original energy system is mainly reflected in the improvement and 364 breakthrough of technologies such as energy conservation and pollution reduction (Guo 2013). . Since foreign direct investment can affect the environment and regional economy 382 through technology spillovers or knowledge spillovers and pollution transfer effects (Ma 383 2014), the degree of openness is calculated by dividing foreign direct investment by gross 384 domestic product. 385 Spatial weight matrix: 0-1 adjacent distance weight matrix. Based on Rook's neighbors, 386 this study establishes a 0-1 adjacency matrix. In particular, when two spatial decision-making 387 units have a common boundary, it is 1, otherwise it is 0. The significance of 0-1 spatial weight 388 matrix lies in that only when two regions are adjacent can certain spatial correlation occur. In 389 the matrix construction, it is assumed that Hainan Province and Guangdong Province have the 390 condition of being adjacent to Rook. The matrix is set up as follows: 1，region is adjacent to region 0，region and region are not adjacent (4) 392

Data source 393
Setting the year from 2000 to 2017 as the research period, this paper selects 30 mainland 394 regions in China as the research data. Due to the obvious data missing in Hong Kong, Taiwan, 395 Tibet and Macao, we have eliminated them. The total factor ecological efficiency data 396 processed in this paper are nearly obtained from China Statistical Yearbook and Wind -397 Economic Database; The data of energy technology innovation came from the public patent 398 database retrieved by Shanghai Intellectual Property (Patent) Public Service Platform. In the 399 specific operation, the search scope is positioned at "non-fossil energy" and "energy 400 conservation and emission reduction" technologies. The abstract and keywords are set as 401 "solar energy or wind energy or ocean energy or biomass energy or nuclear energy or 402 hydrogen energy or hydro energy or geothermal energy or chemical energy or renewable 403 energy or new energy" and "energy saving and pollution reduction" respectively. In order to avoid the lack of credibility and comparability of the data caused by price 412 fluctuations, the paper sets the base period as 2000, deflates the prices of all monetary 413 quantities, and adjusts them to comparable prices by means of a basket of price indexes such 414 as fixed asset investment price indexes. Moreover, for fear of the heteroscedasticity and 415 multicollinearity, the logarithm processing is carried out on the related variables. Table 3  416 shows the specific descriptive statistical results of the correlation coefficient matrix of each 417 variable. 418 Before proceeding with the specific selection and application of the spatial measurement 425 model, the spatial correlation analysis of economic activities should be carried out first, which 426 usually adopts Moran index, Lagrange multiplier form LMLAG, LMERR and its robust form 427 Robust-LMLAG, Robust-LMERR test, etc. In this study, Moran's index is firstly adopted to 428 examine whether the target data is spatially dependent, and then Lagrange multiplier form and 429 spatial effect decomposition are applied to make a more comprehensive judgment. 430 Specifically, the Moran index is defined as follows: 431 Secondly, in order to make up for the shortcomings of the global Moran index 433 measurement, this paper introduces the Moran scatter diagram and Lisa cluster diagram, 434 which are the local spatial correlation test indexes, so as to concretely analyze the spatial 435 distribution characteristics within 30 provinces. The following is the definition of the local 436   Note: The statistical values at 10%, 5% and 1% levels are indicated by * , ** and *** respectively. 452 Actually, Moran index test is a preliminary test of spatial dependence and heterogeneity 457 of total factor ecological efficiency. Before the formal analysis of spatial measurement 458 models, we also need to estimate the non-spatial panel models and examine their statistics, 459 that is, the existence of spatial correlation should be further judged by LM test. In this paper,  The statistical values at 10%, 5% and 1% levels are indicated by * , ** and *** respectively.  we finally choose to establish the dual fixed-effect spatial Durbin model. 480 positive, which further proves the positive spatial correlation of the regional total factor 484 ecological efficiency. As far as the internal regions are concerned, the influence of energy 485 technology patent on total factor ecological efficiency presents a U-shaped 486 pattern, showing a change from negative to positive. Capital affluence exerts an 487 effectively positive force on total factor eco-efficiency, while population density has an 488 adverse impact to some extent; From a spatial perspective, by integrating * and 489 * ( ) # , it is easy to find an apparent spatial effect between the patents of energy 490 technology innovation and regional green development, which also displays a U-shaped 491 change. Notably, the spatial influence coefficients are greater than those within the region, 492 which are -0.276 and 0.021, respectively. In terms of the four control variables, except for the 493 level of environmental regulation , other variables all demonstrate significant spatial 494 influence. 495 items can not fully reflect their spatial correlation. More critically, we focus on the spatial 500 decomposition effect of explanatory variables on explained variables after treating the spatial 501 econometric model with partial differentiation, including direct effect, spillover effect and the 502 total effect of both. Among them, the direct effect includes the spatial feedback cumulative 503 effect of the spillover effect of the province to its neighboring provinces, that is, it includes 504 the feedback effect of the spillover effect of its own province and the spillover effect of the 505 neighboring provinces (Yuan et al., 2020); The indirect effect refers to the spillover effect, 506 that is, the spatial diffusion of the influence of the province on the neighborhood. And the 507 total spatial effect covers the previous two types of effects in a given province, but it is not 508 simply a summation. 509 Table 8 demonstrates the spatial effect decomposition results of both short term and 510 dynamic long term. The following is the concrete analysis. ①In terms of direct effect, the 511 level of technological innovation within the region has a non-linear U-shaped relationship 512 with the provincial total factor eco-efficiency, that is, the number of energy technology 513 patents has different influences on the green productivity in various areas. In view of the 514 coefficient, every 1% change in the weighted number of the energy technical innovation in the 515 early stage will reduce the regional total factor ecological efficiency by 0.073%, while it will 516 increase the economic level by 0.006% in the later stage. Meanwhile, compared with the 517 short-term direct effect, the significance level of the long-term direct effect has no obvious change, but the influence coefficient is larger, manifesting that the long-term influence is 519 stronger;②In terms of indirect effect, the overflow influence of energy technology innovation 520 on ecological efficiency in external regions also has a U-shaped relationship, which is 521 consistent with the above analysis results and provides empirical support for Hypothesis 1. In 522 the long run, the spatial effects are greater than the short-term spillover effects, which are 523 -0.392 and 0.030 respectively. In addition, the inter-regional impact coefficients of energy 524 technology innovation are all greater than its direct effect coefficients, indicating that the 525 spatial indirect effect of energy technical patents cannot be ignored;③In terms of the total 526 effect, Since energy technology innovation has the same influence on total factor ecological 527 efficiency in direct and indirect effects, its cumulative total effect is a larger with a more 528 significant level. Similarly, the spatial total effect shows a significant U-shaped effect, which 529 verifies the first half of hypothesis 2 in this paper. In general, the spatial impact of energy 530 technology innovation level on total factor ecological efficiency reflects a significant 531 U-shaped relationship with a stronger spatial spillover effect, and emerges as a stable 532 long-term shock. 533 Note: The statistical values at 10%, 5% and 1% levels are indicated by * , ** and *** respectively. 535

Robustness test 536
In the spatial panel, an appropriate spatial weight matrix is the key factor for the success 537 of model building. The results may differ significantly depending on the type of matrix. In 538 consequence, this paper selects two spatial weight matrices concerning the geographic 539 distance and information distance as a robustness test of the model, so as to provide evidence 540 for the credibility and stability of the above empirical results of spatial Durbin model and its 541 decomposition effects. Table 9 collects the models of two types of robustness tests, which are 542 conducted on the basis of double fixed SDM models. The results show that the number of 543 significant variables and the influence direction of variable coefficient are the same as the 544 results in this paper. Moreover, there is no contradiction between the three kinds of effects and 545 the above conclusions, and the spatial effect coefficient is even larger, manifesting that the 546 model establishment is more rational. 547 Note: The statistical values at 10%, 5% and 1% levels are indicated by * , ** and *** respectively. 549

Empirical results of threshold panel model 551
On account of theoretical analysis and statistical analysis, it can be seen that the core for 552 the non-linear relationship between energy technology innovation and total factor ecological 553 efficiency lies in the intervention of intermediate mechanism. In light of the extremely uneven 554 development of various provinces in China, this study empirically explores the complex 555 mechanism among the innovation in energy technology and regional total factor ecological 556 efficiency under the heterogeneous level of economic development in different areas. In the 557 threshold model, the F value and the corresponding self-sampling P value are obtained after 558 400 repeated sampling, as demonstrated in Table 10. Based on the value of P in the Table 10, 559 it can be judged that the model not only passes a single threshold, but also has a second 560 threshold. In other words, it is highly possible to have a double threshold effect of economic 561 development level, with two thresholds 9.0933 and 9.5651. Consequently, this paper will 562 analyze the double threshold effect in detail. 563 Note: The statistical values at 10%, 5% and 1% levels are indicated by * , ** and *** respectively. 565 To acquire the threshold and the confidence interval in a more intuitive way, we further 566 identify the threshold value by feat of the least square likelihood ratio statistic LR. The 567 threshold estimate is the statistic when LR is zero. Figure 3 presents the likelihood ratio 568 function graphs covering the two threshold values respectively. 569  Table 11 presents the two existing thresholds and their confidence intervals of the 571 threshold model which are obtained through software analysis. In combination with Figure 3, 572 it is easy to find that the threshold values at the 95% confidence level are respectively [9.0626, 573 9.1199] and [9.4934, 9.5788], and all the LR values are less than the critical value of 7.35 574 which is at the significance level of 5% (as shown by the dotted line in the figure). 575 In accordance of the threshold regression, it is concluded that the driving impact of 577 energy technology patents on total factor eco-efficiency is not monotonically incremental (or 578 degressive). The effect coefficient of energy technology innovation varies evidently in 579 different provinces, that is, as the economic development level continues to increase, it will 580 first inhibit the regional total factor ecological efficiency and then have a completely opposite 581 effect. To a certain extent, it is consistent with the "U" shaped curve in spatial Durbin model 582 with the addition of spatial lag term and spatial direct effect. Specifically, when the level of 583 economic development is lower than 9.0933, every 1% optimization of energy technology 584 innovation will lead to a 0.056% decrease in the level of green economy; When the value of 585 per capita income crosses the first threshold, that is, when the is between 9.0933 586 and 9.5651, the direction of the influence of energy technology patents on the regional total 587 factor ecological efficiency changes structurally. The effect coefficient changed from negative 588 to positive, while the parameter estimates does not pass the significance test; As the level of 589 economic development continues to rise, its inhibitory effect is weakened, whereas not 590 significant; Once the adjustment variable is greater than 9.5651, the elasticity coefficient of 591 energy technology innovation activities turns to 0.017， passing the significance level test of 592 5%, which further validates the hypothesis 2 of this article. The above results illustrate that 593 the optimal interval is the high value interval of the economic development level, at which 594 point the energy technology innovation can raise the regional total factor ecological efficiency 595 in a more productive way. 596 The statistical values at 10%, 5% and 1% levels are indicated by * , ** and *** respectively. 598

Robustness test 599
In order to avoid instability of the estimation, a robustness test is inevitably performed to 600 examine the threshold effect of different types of energy technology innovation on total factor 601 ecological efficiency. For this purpose, the energy technology innovation is divided into 602 technology innovation for energy conservation and emission reduction of traditional energy 603 1 and technology innovation for comprehensive utilization of renewable energy 2. 604 This paper conducts threshold regression for the two variables respectively, and the estimation 605 are summarized in Table 13. It is not hard to find that for each type of energy technology 606 innovations, no significant fluctuations have occurred in the value of the impact coefficient or 607 the level of significance. More specifically，both the threshold effect and threshold value are 608 similar to the above, and there is no apparent fluctuation in the measurement results of the 609 control variables. On this basis, it can be considered that the threshold model constructed in 610 this paper has good robustness. 611 Note: The statistical values at 10%, 5% and 1% levels are indicated by * , ** and *** respectively. 613

614
Spatial econometric models indicate a significant U-shaped spatial impact between 615 energy technology innovation and regional total factor ecological efficiency, among which the 616 spillover effect between regions is more evident. The reason may lie in that although 617 technology patent is a kind of intangible asset, the positive externality of knowledge and the 618 mobility of human capital will facilitate the circulation and imitation of technology elements. 619 In the short term, due to the immaturity energy technology, the region's own technology 620 diffusion and technology reception capacity are very limited. At this time, the spillover 621 technology elements can not bring positive learning and imitation between regions. On the 622 contrary, due to the absorption of immature energy technologies which are not suitable for the 623 region itself, the comprehensive economic benefits are not ideal as expected. In the long run, 624 the energy technology matures into a viable technology, during which time the absorption 625 capacity of the technology undertaking region is relatively strong. Under the premise of 626 economic stability, the integration and imitation of different types of energy technology 627 elements can further optimize industrial structure and improve the level of productivity, thus 628 increasing the green ecological efficiency.