Directed Technological Change in the Electricity Sector Preliminary and Incomplete Please Do Not Cite without Permission Abstract Technological innovations in energy, facilitate the use of a higher share of renewable resources in electricity generation. Furthermore, many governments have introduced incentives to shift energy production from fossil fuels toward renewable sources. In this paper, we investigate the determinants of technological innovations in the electricity sector, and assess environmental policies for promoting renewable energy like wind, solar or biomass. We first derive testable hypotheses from a directed technological change model applied to the electricity sector. Then, we construct a global, firm-level energy patent database from 1978 to 2011 to empirically test our hypotheses. Our preliminary results suggest that higher natural gas prices, a firm’s stock of knowledge and knowledge spillovers increase patenting in clean technologies. These results have strong policy implications. Keywords: electricity sector, directed technological change, climate change 1 Introduction Scientists and policymakers seek options to reconcile concerns about climate change with global economic growth. One such option is to increase the share of renewable sources in electricity generation. To this end, many governments have introduced renewable energy incentives to shift energy production from fossil fuels toward renewables. While these incentives have resulted in a range of technological innovations in renewable energy, their use in global electricity generation remains modest. The main challenge is to increase the share of renewables in the electricity mix. In this paper, we investigate the determinants of technological innovations in the electricity sector, and assess environmental policies aimed at increasing the share of renewable energy in electricity generation. We focus on the electricity sector for two reasons. First, electricity generation is the largest single emitter of carbon dioxide in the U.S. For example, in 2011, burning fossil fuels yielded 38% of total carbon dioxide emissions and 32% of total greenhouse gas emissions (EPA 2011). Second, global electricity subsidies are large. For example, estimates suggest that over the period 2002-2008 subsidies in the U.S. to support R&D in fossil fuels were $72 billion while renewable energy received $29 billion (Adeyeye, 2009). The main goal in this study is to investigate the determinants of technological innovations in the electricity sector. Specifically, we want to answer the following three questions: First, what are the drivers of innovation in renewable energy sources? Second, what factors drive the shift from fossil fuels to renewables? And, third, have R&D subsidies been successful at directing innovation toward renewable energy? To answer these questions, we conduct an empirical analysis in two parts. First, to identify firm-level incentives for innovation in renewable energy, we construct a global, firm-level database of energy patents. We combine patents from the OECD Triadic Patent Database, firm-level information from the OECD HAN database and energy prices 2 from the IEA database. The database we built is unique because it contains information about firm-level patents (clean and dirty) from 73 countries from 1978 to 2011. In addition, we focus on the most valuable patents which we identify as patents registered with the three main patent offices globally.1 Second, we study how to direct R&D efforts toward clean electricity generation, and thereby reduce the gap between fossil fuels and renewables. We do so by evaluating the impact of the most significant R&D subsidies from 1978 to 2011. The model we estimate is based on the theoretical framework of Acemoglu et al. (2012a). We apply the model to the electricity sector where dirty and clean firms create new innovations that make the existing technologies obsolete. From this theoretical model, we construct testable hypotheses that describe firm-level incentives to innovate. In particular, we look at how incentives to innovate in clean and dirty sectors depend on coal and natural gas prices, and the knowledge stock of clean and dirty innovations. Our empirical strategy is closely related to that of Aghion et al. (2012). We estimate a patent count model using a poison estimator and a negative binomial estimator. Our preliminary results suggest that an increase in the price of natural gas is associated with more patenting in the clean sector. Furthermore, we find that firms’ existing knowledge stock is an important determinant of clean patenting. Interestingly, we find that industry spillovers related to clean technologies have a positive impact on dirty patenting. This suggests that clean knowledge spillovers may induce energy efficiency in fossil fuel technologies. This paper relates to three branches of the literature: a theoretical literature studying endogenous economic growth and climate change, an empirical literature with applications to endogenous growth, and, an empirical literature that studies energy prices and induced innovation in environmental technology. We contribute to this literature with an empirical 1 OECD Triadic Patent Database contains patents taken at the European Patent Office (EPO), the Japanese Patent Office (JPO), and the U.S. Patent and Trademark Office (USPTO) to protect the same idea. 3 study of directed technological change in the electricity sector. The literature on how to reconcile economic growth and climate change is large. Particularly, the theoretical literature is extensive. See for example, Bovenberg and Smulders (1995, 1996), Acemoglu et al. (2012a), Goulder and Schneider (1999), Goulder and Mathai (2000) and Zwaan et al. (2002). Our paper is closer related to Acemoglu et al. (2012a), since our work is an application of their theoretical model to the electricity sector. The most relevant empirical literature is the one that applies directed technological change models to study climate change (see e.g. Noailly and Smeets, 2012; Aghion et al., 2012; Acemoglu et al., 2012b, 2013). Noailly and Smeets (2012) study the determinants of directed technical change in electricity generation technologies. They employ European patent data, while we construct a global patent database. Aghion et al. (2012) study directed technical change in the auto industry. While we follow their econometric strategy, we study a different industry. Finally, Acemoglu et al. (2012b) and Acemoglu et al. (2013) study how policies can affect the transition from dirty to clean technologies. Acemoglu et al. (2012b) use micro data on R&D expenditures, patents, sales, employment, and firm entry and exit to calibrate their model, while Acemoglu et al. (2013) investigate policies (subsidies) that may either encourage investment in R&D or discourage innovation and growth by reducing reallocation. Our paper differs from this work since we focus on the effect of historical subsidies instead of running simulations. The remainder of the paper unfolds as follows. In Section 2, we present our database. In Section 3, we describe our empirical strategy while in Section 4 we discuss our results. Finally, Section 5 concludes. 4 2 Data In this section, we describe the construction of our global, firm-level, energy patent database. We combine the OECD’s Triadic Patent Families Database, OECD’s HAN database and energy prices from the IEA. 2.1 Patents and patent families In our study, we utilize patent data to measure technological change. The main advantages of using patent data for our purposes are twofold. First, patents are available at the firm and technology level. In contrast to more aggregate measures such as R&D expenditures, which are generally only available at the industry level and for limited technology types, each individual patent contains detailed information about the inventor(s), applicant(s), and the specific type of technology (Popp, 2005). Information about the applicant is the most useful for our purposes, as it allows us to identify specific firms, while the International Patent Classification (IPC) codes assigned to each patent makes it possible to identify technologies related to electricity generation. Thus, the detailed nature of patent data proves especially useful when examining firm-specific incentives to innovate in select technologies. Second, patents provide a measure of innovation output of firms’ research activities that is close to the actual time of invention. Since patent applications are normally submitted early in the research process, as indicated by the “priority date,” they also provide a good measure of overall innovative activity of a given firm (Popp, 2005). While this is the case, there are drawbacks of patent data that must be addressed. First, patents are not granted for every invention and the propensity to patent may differ considerably across countries and industries (Van Pottelsberghe et al., 2001). Second, a submitted patent application does not guarantee that the technology has actually been adopted (Popp, 2005). Third, individual patents differ considerably in their worth, with 5 many patents having low values (Aghion et al., 2012). These limitations motivate the extraction of our patents from the OECD’s Triadic Patent Database which contains patent family data from 1978 to 2011. Triadic patent families are useful because they are collections of patents that protect the same idea in different countries. For example, a particular patent application must have an equivalent application at the European Patent Office (EPO), Japanese Patent Office (JPO) and the United States Patent Office (USPTO) in order to qualify as a patent family member. Because triadic patents are applied for in three separate offices, they include only the highest valued patents and allow for a common worldwide measure of innovation that avoids the heterogeneity of individual patent office administrations (Aghion et al., 2012). Furthermore, the OECD utilizes “extended families,” which are designed to identify any possible links between patent documents (Martinez, 2010). This is advantageous, as it provides the most comprehensive method of consolidating patents into distinct families, allowing us to to include an extensive number of patented ideas. One disadvantage of triadic patent families includes the lag times associated with the USPTO. Legal delays for publishing applications is 18 months after the priority date and up to 5 years between the priority date and publication date (Dernis and Khan, 2004). As a consequence, U.S. patent grants may delay the completion of data on triadic patent families. In order to mitigate this limitation, the OECD utilizes forecasts called “nowcasting” in order to improve the timeliness of triadic patents (Dernis and Khan, 2004). Despite this difficulty, triadic patents still provide the most inclusive measure of high-value, firm-level, innovative performance. 2.2 Electricity generation technologies Electricity generation is chosen as a specific case study for several reasons. Although renewable energy can provide a clean source of electricity production, the majority of the 6 world’s electricity generation, 67.8% in 2011, still comes from burning fossil fuel (IEA, 2013). Recent predictions indicate that the demand for energy is expected to grow by 84% from 2007 to 2035 in the developing world alone, likely leading to serious consequences for greenhouse gas emissions as these countries continue to gain greater access to electricity (Wolfram et al., 2012). Given the signifiance of these statistics in terms of climate change and environmental quality, investigating the incentives to innovate in renewable sources of electricity can shed light on policies that are designed to direct innovation from fossil fuel to renewable generation sources. In order to accomplish this objective, we select patents that are specific to electricity generation based on IPC codes. We then categorize them into two groups: renewable energy and fossil fuel based technologies. Renewable energy technologies are identified from the World Intellectual Property Office’s (WIPO) IPC Green Inventory.2 Specifically, we select patents whose technology classes are related to alternative energy production. This includes integrated gasification combined cycle (ICGG), fuel cells, pyrolysis, harnessing energy from manmade waste, hydro energy, wind, solar, geothermal energy, other production or use of heat, using waste heat, and devices for producing mechanical power from muscle energy. Specific descriptions of the IPC codes used in this paper are presented in Tables 1 - 4. Fossil fuel technologies are selected from the general fossil fuel technology IPC codes reported in Lanzi et al. (2011). The selection of technology classes through IPC codes requires careful consideration. For example, recent work by Noailly and Smeets (2012) classifies firm-level patents into renewable and fossil fuel technologies to study the main drivers of innovation and the direction of technological change. They select patents related to renewable technologies from IPC codes reported in Johnstone et al. (2010). When comparing patent families selected by these IPC 2 The IPC codes listed in the IPC Green Inventory have been compiled by the IPC Committee of Experts in concordance with the United Nations Framework Convention on Climate Change (UNFCCC). For more information see http://www.wipo.int/classifications/ipc/en/est/. 7 Table 1: Definitions of IPC patent classes for clean patents Description Fuel cells Electrodes Inert electrodes with catalytic activity Non-active parts Within hybrid cells IPC code H01M 4/86-4/98, 8/00-8/24, 12/00-12/08 H01M 4/86-4/98 H01M 4/86-4/98 H01M 2/00-2/04 , 8/00-8/24 H01M 12/00-12/08 Pyrolysis or gasification of biomass C10B 53/00 C10J Harnessing energy from manmade waste Agricultural waste Fuel from animal waste and crop residues Incinerators for field, garden or wood waste Gasification C10L 5/00 C10L 5/42, 5/44 F23G 7/00, 7/10 C10J 3/02, 3/46 F23B 90/00 F23G 5/027 B09B 3/00 F23G 7/00 C10L 5/48 F23G 5/00, 7/00 C21B 5/06 Chemical waste Industrial waste Using top gas in blast furnaces to power pigiron production Pulp liquors Anaerobic digestion of industrial waste D21C 11/00 A62D 3/02 C02F 11/04, 11/14 F23G 7/00, 7/10 B09B 3/00 F23G 5/00 B09B B01D 53/02, 53/04, 53/047, 53/14, 53/22, 53/24 C10L 5/46 F23G 5/00 Industrial wood waste Hospital waste Landfill gas Separation of components Municipal waste Hydro energy Water-power plants Tide or wave power plants Machines or engines for liquids E02B 9/00-9/06 E02B 9/08 F03B F03C F03B 13/12-13/26 F03B 15/00-15/22 Using wave or tide energy Regulating, controlling or safety means of machines or engines Propulsion of marine vessels using energy derived from water movement B63H 19/02, 19/04 F03G 7/05 Ocean thermal energy conversion (OTEC) 8 Table 2: Definitions of IPC patent classes for clean patents Description Wind energy Structural association of electric generator with mechanical driving motor Structural aspects of wind turbines IPC code F03D H02K 7/18 B63B 35/00 E04H 12/00 F03D 11/04 B60K 16/00 B60L 8/00 Propulsion of vehicles using wind power Electric propulsion of vehicles using wind Power Propulsion of marine vessels by windpowered motors Solar energy Photovoltaics (PV) Devices adapted for the conversion of radiation energy into electrical energy B63H 13/00 H01L 27/142, 31/00-31/078 H01G 9/20 H02N 6/00 H01L 27/30, 51/42-51/48 H01L 25/00, 25/03, 25/16, 25/18, 31/042 C01B 33/02 C23C 14/14, 16/24 C30B 29/06 G05F 1/67 Using organic materials as the active part Assemblies of a plurality of solar cells Silicon; single-crystal growth Regulating to the maximum power available from solar cells Electric lighting devices with, or rechargeable with, solar cells Charging batteries Dye-sensitised solar cells (DSSC) F21L 4/00 F21S 9/03 H02J 7/35 H01G 9/20 H01M 14/00 F24J 2/00-2/54 F24D 17/00 F24D 3/00, 5/00, 11/00, 19/00 F24J 2/42 F03D 1/04, 9/00, 11/04 F03G 6/00 C02F 1/14 F02C 1/05 Use of solar heat For domestic hot water systems For space heating For swimming pools Solar updraft towers For treatment of water, waste water or sludge Gas turbine power plants using solar heat source Hybrid solar thermal-PV systems Propulsion of vehicles using solar power Electric propulsion of vehicles using solar power Producing mechanical power from solar energy Roof covering aspects of energy collecting devices Steam generation using solar heat H01L 31/058 B60K 16/00 B60L 8/00 F03G 6/00-6/06 E04D 13/00, 13/18 F22B 1/00 F24J 1/00 F25B 27/00 Refrigeration or heat pump systems using solar energy Use of solar energy for drying materials or objects Solar concentrators F26B 3/00, 3/28 F24J 2/06 G02B 7/183 Solar ponds F24J 2/04 9 Table 3: Definitions of IPC patent classes for clean patents Description Geothermal energy Use of geothermal heat Production of mechanical power from geothermal energy Other production or use of heat, not derived from combustion, e.g. natural heat Heat pumps in central heating systems using heat accumulated in storage masses Heat pumps in other domestic- or spaceheating systems Heat pumps in domestic hot-water supply systems Air or water heaters using heat pumps Heat pumps Using waste heat To produce mechanical energy Of combustion engines Of steam engine plants Of gas-turbine plants As source of energy for refrigeration plants For treatment of water, waste water or sewage Recovery of waste heat in paper production For steam generation by exploitation of the heat content of hot heat carriers Recuperation of heat energy from waste incineration Energy recovery in air conditioning Arrangements for using waste heat from furnaces, kilns, ovens or retorts Regenerative heat-exchange apparatus Of gasification plants Devices for producing mechanical power from muscle energy IPC code F01K F24F 5/00 F24J 3/08 H02N 10/00 F25B 30/06 F03G 4/00-4/06, 7/04 F24J 1/00, 3/00, 3/06 F24D 11/02 F24D 15/04 F24D 17/02 F24H 4/00 F25B 30/00 F01K 27/00 F01K 23/06-23/10 F01N 5/00 F02G 5/00-5/04 F25B 27/02 F01K 17/00, 23/04 F02C 6/18 F25B 27/02 C02F 1/16 D21F 5/20 F22B 1/02 F23G 5/46 F24F 12/00 F27D 17/00 F28D 17/00-20/00 C10J 3/86 F03G 5/00-5/08 10 Table 4: Definitions of IPC patent classes for dirty patents Description Production of fuel gases by carburetting air or other gases without pyrolysis Steam engine plants; steam accumulators; engine plants not otherwise provided for; engines using special working fluids or cycles Gas-turbine plants; air intakes for jet-propulsion plants; controlling fuel supply in air-breathing jet-propulsion plants Hot-gas or combustion-product positive-displacement engine; use of waste heat of combustion engines,not otherwise provided for Steam generation Combustion apparatus; combustion processes Furnaces; kilns; ovens; retorts IPC code C10J F01K F02C F02G F22 F23 F27 codes to the WIPO IPC classifications, significantly more patent families can be extracted using the latter. Table 5 provides a comparison of the number of patent families that are selected using codes solely from Johnstone et al. (2010) and those included in both Johnstone et al. (2010) and the WIPO IPC Green Inventory. In total, 38,370 patent families are gained by using the more inclusive codes, with the majority of additional families stemming from biomass and waste. The IPC codes in Johnstone et al. (2010) also omit certain categories, such as hydro energy, that are important sources of renewable electricity generation. Table 5: Comparison of patent family counts Technology class All Renewables Wind Solar Geothermal Ocean Biomass and Waste 2.3 (Johnstone et al., 2010) 8,245 662 2,996 170 189 4,306 (Johnstone et al., 2010) + WIPO 46,615 1,118 9,730 1,443 192 35,577 Families gained 38,370 456 6,734 1,273 3 31,271 Firms We aggregate individual patent counts at the firm-level. By utilizing the OECD Harmonized Applicants Names (HAN) Database, a register that contains clean applicant names which are matched against company names from business register data, we are able to link patents to 11 firms and individuals. Unfortunately, the HAN database does not contain firm information for every patent application in our sample. Names that cannot be matched using the HAN are synchronized using applicant information contained in the Triadic Patent Families Database. Although this allows us to match every patent to an applicant, it poses two difficulties. First, applicant names in the Triadic Patent Database contain a number of spelling, character, and name variations. For example, “3M INNOVATIVE PROPERTIES” and “3M INNOVATIVE PROPERTIES CO” would be incorrectly treated as separate firms in the absence of name harmonization. Second, the Triadic Patent Families Database does not directly link patent applications to applicant names. Instead, applicant names are linked to family identifiers. Thus, if a given family contains more than one firm name, it is impossible to tell which firm to associate with each patent. In order to minimize complications that may result from these challenges, we restrict our sample to those patents applications that can i.) be matched fully from the HAN register or ii.) have a single applicant and are the sole member of patent family. We conduct further harmonization using algorithms, although some name variation still remains. In contrast to Noailly and Smeets (2012), who select only on European firms, we consider 15,081 total firms that claim residence in 73 countries. Table 6 illustrates the countries that contain the highest concentrations of innovating firms. Since the majority of innovative activities originate in the United States, Japan, and Germany, it is not surprising that these countries contain the highest numbers of firms in the sample. 2.4 Descriptive statistics In total, we identify 99,739 unique triadic patent applications across all firms during the sample time period of 1978 to 2011. Of this total, 61,709 are designated as clean technologies, while 38,014 are classified as dirty. In contrast to prior empirical works,3 our the number 3 See Aghion et al. (2012) and Noailly and Smeets (2012) 12 Table 6: Distribution of firms across countries United States Japan Germany France United Kingdom Switzerland Sweden Canada Number of firms 5,210 2,359 2,011 1,131 669 418 384 367 of clean patents is greater than the number of dirty patents. This is likely due to our usage of more comprehensive IPC codes for clean technologies. It may also be the case that clean technologies, though newer, have more general applications that are more likely to be patented Dechezleprêtre et al. (2013). Figure 1 illustrates the time series of clean and dirty patenting over the sample time period. While clean and dirty patent counts tend to be very close in the late 1970s and early 1980s, clean patenting surpasses dirty patenting by a wide margin by 2000. It may be the case that dirty technologies, which have existed for a longer time period and are built upon a larger knowledge stock, are harder to innovate upon. Thus, dirty innovations may only be incremental and harder to patent. Both series exhibit a sharp decline around 2000.4 It may also be of interest to study the main patent holders of each innovation category over the time period considered. Table 7 illustrates that the main clean patent holders, with the exception of Air Products Chemicals, are located outside of the United States. Specifically, Japanese firms tend to dominate this list. In contrast, Table 8 illustrates that the top two dirty patenting firms, General Electric and United Tech Corp, are both U.S. firms. Furthermore, these two firms have a higher total number of patents than the other firms which are ranked within the tables. This indicates that although the total number of clean patents exceeds the total number of dirty patents in the entire sample, a higher 4 Though often noted, no concrete explanation for this has been cited in the literature. 13 0 Number of patents 1000 2000 3000 4000 Figure 1: Patenting over time 1980 1990 2000 2010 Year Clean patent counts Dirty patent counts concentration of dirty patenting occurs for the top patent holding firms. Although most of the firms in these rankings tend to specialize in either dirty or clean patenting, Germany’s Siemens AG and Japan’s Matsushita Electric has a substantial share of both types. 2.5 Energy prices Country level energy price data, quoted in U.S. dollars per unit, comes from the International Energy Agency (IEA). We consider the following fuel types based on availability: steam coal, coking coal, auto diesel fuel, high sulpher fuel oil, light fuel oil, and natural gas. There are two limitations of this specific dataset that we now address. First, price data exists for only 33 countries. This is significantly less than our firm coverage of 73 countries. Second, there are many missing values within the dataset. Although energy prices do differ across countries due to taxes and regulations, they tend to move together over time. Figure 2 14 Table 7: Main clean patent holders Firm name CANON KK MATSUSHITA ELECT IND NISSAN MOTOR AIR PRODUCTS CHEMICALS TOYOTA JIDOSHA KK NGK INSULATORS SHIN ETSU HANDOTAI PANASONIC CORP SIEMENS AG WOBBEN ALOYS firm country JP JP JP US JP JP JP JP DE DE clean patents 1353 1078 665 566 477 467 457 442 428 427 dirty patents 1 857 27 129 74 92 5 17 998 0 total patents 1354 1935 692 695 551 559 462 459 1426 427 Table 8: Main dirty patent holders Firm name GENERAL ELECTRIC UNITED TECH CORP SIEMENS AG MATSUSHITA ELECT IND SNECMA HITACHI WESTINGHOUSE ELECT CORP ASEA BROWN BOVERI AG PRATT WHITNEY CANADA CORP NGK SPARK PLUG firm country US US DE JP FR JP US CH CA JP 15 dirty patents 3752 1923 998 857 755 369 362 348 317 217 clean patents 476 175 428 1078 36 172 172 38 15 29 total patents 4228 2098 1426 1935 791 541 534 386 332 246 illustrates prices for coal, one of the leading sources of electric power generation, for the top innovating countries in our sample (?). The series are similar in their movement, illustrating greater volatility in the 1980s and rising prices in the last decade. 50 Price ($ per tonne) 100 150 200 Figure 2: Coking coal prices for top innovating countries 1980 1990 2000 2010 Year United States Japan France Sweden Germany Switzerland United Kingdom In order to mitigate the aforementioned problems we use U.S. prices when faced with zero or missing values. Figure 3 illustrates the selection of energy prices for the United States over the time period considered. Prices for oil, including light fuel oil and high sulpher fuel oil, are considerably higher and more volatile than the other energy prices, which remain relatively stable from 1978 to 2011. 16 0 Price ($ per unit) 400 200 600 800 Figure 3: U.S. fuel prices over time 1980 1990 2000 2010 Year Auto Diesel Fuel Steam Coall Light Fuel Oil 3 Coking Coal Natural Gas High Sulphur Fuel Oil Econometrics and identification strategy In this section we consider the following Poisson specification for the determinants of clean and dirty patent counts related to electricity generation. P ATz,it = exp(βz,p lnF Pjt−1 + Az,it−1 + Tz,t )ηzi + µz,it (1) where z indicates separate equations for clean (c) and dirty (d) technologies, i indicates firms, j represents countries, and t illustrates time. P ATz,it is the number of triadic patents applied for, either clean or dirty, by firm i in year t. F Pjt represents the country-level fuel prices described in the data section.5 Az,it indicates the firm’s existing stock of knowledge which includes its cumulative history of clean and dirty patenting activity as well as knowledge 5 We take the natural log of fuel prices, however, we do not do so for knowledge stocks as these can contain zero values. 17 spillovers, proxied by aggregate counts of clean and dirty technologies of all firms within the same country. More specifically, a firm’s total knowledge stock is given by Az,it = βz,1 Kc,it + βz,2 Kd,it + βz,3 SP ILLc,it + βz,4 SP ILLd,it (2) where Kz,it is firm i’s patent stock of the designated technology type and SP ILLz,it is the spillover for firm i, calculated as the sum of all other firm’s patent stocks that are located P domestically. This is defined as Kz,ijt . i We lag these and knowledge stocks and prices by one period, as in Aghion et al. (2012), in order to adequately reflect the delayed response of firms. We also include a full set of time dummies, represented by Tz,t , and firm fixed effects indicated by ηzi . 4 Estimation results Our baseline regression results are shown in Tables 9 and 10. All estimates include firm fixed effects and year dummies. Table 9 illustrates six separate regressions, one for each fuel type considered, where the number of clean patents in a firm is the dependent variable. There are several observations based on these results that we now discuss. First, only four of the fuel price coefficients are significant in the clean patent regressions: coking coal, auto diesel fuel, high sulpher fuel oil, and natural gas. These coefficients are negative, with the exception of natural gas, indicating that an increase in fuel prices is associated with a decline in clean patenting. Second, firms’ existing knowledge stocks are important determinants of clean patenting. Firms who have innovated in clean technology in the past and those who are exposed to larger clean knowledge stocks from other geographically local firms are significantly less likely to produce clean patents. While this is largely inconsistent 18 Table 9: Regression of clean patents with various fuel types (1) Steam coal price (2) Clean patent counts (3) (4) (5) (6) -0.128 (0.116) Coking coal price -0.447** (0.210) Auto diesel fuel price -0.632*** (0.244) High sulphur fuel oil price -0.559*** (0.143) Light fuel oil price 0.0669 (0.179) Natural gas price 0.418* (0.245) -0.000671** -0.000627** -0.000671** (0.000294) (0.000295) (0.000285) 0.000967*** 0.000960*** 0.000979*** (0.000259) (0.000263) (0.000266) -6.67e-05** -9.07e-05*** -0.000147*** (2.97e-05) (3.06e-05) (4.45e-05) 1.72e-05 2.60e-05 6.13e-05** (2.01e-05) (2.11e-05) (2.65e-05) Own clean stock Own dirty stock Clean spillover Dirty spillover -0.000632** (0.000294) 0.000941*** (0.000269) -8.67e-05*** (3.00e-05) 2.85e-05 (1.96e-05) Observations 11,184 Number of firms 1,874 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 -0.000622** (0.000303) 0.000933*** (0.000257) -8.12e-05*** (2.89e-05) 2.20e-05 (2.02e-05) -0.000627* (0.000334) 0.000850** (0.000359) -1.28e-05 (3.51e-05) -2.45e-06 (2.13e-05) 11,184 1,874 9,578 1,693 19 11,184 1,874 11,184 1,874 11,184 1,874 with the idea of path dependency, a firm’s own history of dirty patenting is associated with more clean innovation. This result is consistent with Aghion et al. (2012). Exposer to larger stocks of dirty patents from other firms does not have a significant effect on clean patenting, with the exception of regression (6) which includes natural gas prices. The magnitudes of these coefficients are similar regardless of the fuel price included in the regression. Table 10: Regression of dirty patents with various fuel types (7) Steam coal price (8) Dirty patent counts (9) (10) (11) -0.214* (0.125) Coking coal price -0.375* (0.226) Auto diesel fuel price -1.121** (0.474) High sulphur fuel oil price -0.653*** (0.249) Light fuel oil price -0.397 (0.339) Natural gas price Own stock clean Own stock dirty Clean spillover Dirty spillover (12) -0.00107 (0.00114) 0.000961*** (0.000207) 8.61e-05 (5.33e-05) -5.26e-05* (2.95e-05) Observations 9,039 Number of firms 1,365 Robust standard errors in parentheses *** p<0.01, ** p<0.05, * p<0.1 -0.000900 (0.00116) 0.000938*** (0.000207) 9.32e-05* (5.41e-05) -6.38e-05** (3.14e-05) 3.86e-05 (0.00116) 0.000678*** (0.000237) 0.000135* (7.69e-05) -0.000100*** (3.84e-05) 9,039 1,365 7,533 1,200 -0.000868 -0.000767 (0.00114) (0.00113) 0.000953*** 0.000956*** (0.000214) (0.000215) 0.000113** 9.46e-05* (5.56e-05) (5.45e-05) -7.00e-05** -6.74e-05** (3.27e-05) (3.40e-05) 9,039 1,365 9,039 1,365 0.161 (0.294) -0.000838 (0.00116) 0.000952*** (0.000213) 5.98e-05 (6.78e-05) -4.56e-05 (4.67e-05) 9,039 1,365 Table 10 illustrates the same set of regressions with dirty patents as the dependent variable. Coefficients for steam coal, coking coal, auto diesel fuel prices, and high sulpher fuel oil are significant and also negative. A firm’s own history of clean innovation does not appear to have any significant impact on dirty patenting, while its history of dirty patenting 20 has a positive and significant impact on future dirty innovations. The magnitudes of the “own stock dirty” coefficients are very similar to those in Table 9. Interestingly, spillovers related to clean technologies have a positive impact on dirty patenting in regressions (8) (11), indicating the possibility that clean knowledge spillovers may induce energy efficiency or energy saving fossil fuel based technologies. Interestingly, dirty knowledge spillovers have a significant negative impact on dirty patenting in regressions (7) - (11). 5 Conclusions To be written 21 References Acemoglu, Daron, Philippe Aghion, Leonardo Bursztyn, and David Hemous (2012a) “The Environment and Directed Technical Change,” American Economic Review, Vol. 102, pp. 131–66. Acemoglu, Daron, Ufuk Akcigit, Nicholas Bloom, and William R. Kerr (2013) “Innovation, Reallocation and Growth,” Working Paper 18993, National Bureau of Economic Research. Acemoglu, Daron, Ufuk Akcigit, Douglas Hanley, and William Kerr (2012b) “The Transition to Clean Technology,” University of Pennsylvania mimeo. Adeyeye, Adenike (2009) Estimating US Government Subsidies to Energy Sources, 2002-2008: Environmental Law Institute. Aghion, Philippe, Antoine Dechezleprêtre, David Hemous, Ralf Martin, and John Van Reenen (2012) “Carbon taxes, path dependency and directed technical change: evidence from the auto industry,”Technical report, National Bureau of Economic Research. Bovenberg, A Lans and Sjak Smulders (1995) “Environmental quality and pollution-augmenting technological change in a two-sector endogenous growth model,” Journal of Public Economics, Vol. 57, pp. 369–391. Bovenberg, A Lans and Sjak A Smulders (1996) “Transitional impacts of environmental policy in an endogenous growth model,” International Economic Review, pp. 861–893. Dechezleprêtre, Antoine, Ralf Martin, and Myra Mohnen (2013) “Knowledge spillovers from clean and dirty technologies: A patent citation analysis.” Dernis, Hélène and Mosahid Khan (2004) “Triadic patent families methodology,”Technical report, OECD Publishing. 22 Goulder, Lawrence H and Koshy Mathai (2000) “Optimal CO¡ sub¿ 2¡/sub¿ Abatement in the Presence of Induced Technological Change,” Journal of Environmental Economics and Management, Vol. 39, pp. 1–38. Goulder, Lawrence H and Stephen H Schneider (1999) “Induced technological change and the attractiveness of CO2 abatement policies,” Resource and Energy Economics, Vol. 21, pp. 211–253. IEA (2013) “IEA Electricity Information 2013,”Technical report, OECD Publishing. Johnstone, Nick, Ivan Haščič, and David Popp (2010) “Renewable energy policies and technological innovation: Evidence based on patent counts,” Environmental and Resource Economics, Vol. 45, pp. 133–155. Lanzi, Elisa, Elena Verdolini, and Ivan Haščič (2011) “Efficiency-improving fossil fuel technologies for electricity generation: Data selection and trends,” Energy Policy, Vol. 39, pp. 7000–7014. Martinez, Catalina (2010) “Insight into different types of patent families,”Technical report, OECD Publishing. Noailly, Joëlle and Roger Smeets (2012) “Directing technical change from fossil-fuel to renewable energy innovation: An empirical application using firm-level patent data.” Popp, David (2005) “Lessons from patents: using patents to measure technological change in environmental models,” Ecological Economics, Vol. 54, pp. 209–226. Van Pottelsberghe, Bruno, H Dernis, and Dominique Guellec (2001) “Using patent counts for cross-country comparisons of technology output,”Technical report, ULB–Universite Libre de Bruxelles. 23 Wolfram, Catherine, Orie Shelef, and Paul Gertler (2012) “How Will Energy Demand Develop in the Developing World?” Journal of Economic Perspectives, Vol. 26, pp. 119–38. van der Zwaan, Bob CC, Reyer Gerlagh, Leo Schrattenholzer et al. (2002) “Endogenous technological change in climate change modelling,” Energy economics, Vol. 24, pp. 1–19. 24