Corruption and Energy Efficiency in OECD Countries: Theory and Evidence

Transcription

1 Corrupton and Energy Effcency n OECD Countres: Theory and Evdence Per G. Fredrksson * Southern Methodst Unversty Herman R.J. Vollebergh Erasmus Unversty Rotterdam and OCfEB, Rotterdam Elbert Djkgraaf Erasmus Unversty Rotterdam and OCfEB, Rotterdam Aprl 30, 2003 Abstract We nvestgate the effect of corrupton and ndustry sector sze on energy polcy outcomes. The man predctons of our theory are that: () greater corruptblty of polcy makers reduces energy polcy strngency; () greater lobby group coordnaton costs (ncreased ndustry sector sze) results n more strngent energy polcy; and () workers and captal owners lobbyng efforts on energy polcy are negatvely related. These predctons are tested usng a unque panel data set on the energy ntensty of 11 sectors n 14 OECD countres for years The evdence generally supports the predctons. Keywords: Energy polcy, poltcal economy, corrupton, collectve acton. JEL Code: Q48, D73, D78. * We would lke to thank the partcpants at the The Internatonal Dmenson of Envronmental Polcy conference n Aquafredda, Italy, October 2001, the World Congress of Envronmental and Resource Economsts, Monterey, Calforna, June 2002, the APPAM conference n Dallas, November 2002, and at semnars at Erasmus and Tlburg Unverstes, as well as Leon Bettendorf, Robert Dur, Raymond Florax, Bruno Frey, Danel Mllmet, Daan van Soest, Qn Wang, and especally Peter van Foreest and two referees for helpful comments on earler versons of the paper. Vollebergh acknowledges fnancal support from the research program Envronmental Polcy, Economc Reform and Endogenous Technology, funded by the Netherlands Organsaton for Scentfc Research (NWO). Fredrksson worked on ths paper whle vstng the Bejer Insttute at the Royal Swedsh Academy of Scences, and he thanks the nsttute for ts hosptalty. Peter Mulder and Daan van Soest were helpful n provdng us wth data. The usual dsclamers apply. Correspondence: Per Fredrksson, Department of Economcs, SMU, PO Box , Dallas, TX Fax: ; Emal: pfredrk@mal.smu.edu.

2 Corrupton and Energy Effcency n OECD Countres: Theory and Evdence Abstract We nvestgate the effect of corrupton and ndustry sector sze on energy polcy outcomes. The man predctons of our theory are that: () greater corruptblty of polcy makers reduces energy polcy strngency; () greater lobby group coordnaton costs (ncreased ndustry sector sze) results n more strngent energy polcy; and () workers and captal owners lobbyng efforts on energy polcy are negatvely related. These predctons are tested usng a unque panel data set on the energy ntensty of 11 sectors n 14 OECD countres for years The evdence generally supports the predctons. 1

3 I. INTRODUCTION Ths paper nvestgates how the corruptblty of polcy makers, as well as the ncentves of worker and captalst nterest groups, affect the determnaton of energy polcy n ndustralzed countres. 1 Recent evdence suggests that government corrupton s an mportant determnant of growth, nvestment, and envronmental polcy (see, for example, Mauro (1995), Ades and D Tella (1999), We (2000), and Fredrksson and Svensson (2002)). 2 However, although energy polcy s of great envronmental and geopoltcal mportance, the effect of polcymakers corruptblty on energy polcy has been gnored. In ths paper, we seek to fll ths gap n the lterature. 3 Moreover, we contrbute to the lterature on the role of ndustry sze (lobby group sze) n the polcy formaton process, n partcular on energy polcy. In hs semnal work, Olson (1965) argues that coordnaton costs and free-rdng problems wll reduce the nfluence of large poltcal groups. The effect of lobby group sze on the ablty of specal nterests to nfluence polcy s however stll an unresolved ssue n the emprcal lterature, and the lterature on energy polcy s no excepton. 4 In our 1 Some observers may assocate corrupton manly wth developng countres. However, recent hgh-level corrupton scandals n for example France and Germany (embrolng, for example, Presdent Chrac and former Prme Mnster Kohl) ndcate that the OECD countres are not mmune to ths phenomenon. In the Unted States, the recent collapse of the major energy frm Enron has focused the attenton on (energy) frms open and hdden poltcal actvtes. See Oates and Portney (2001) for a nce survey of the lterature on the poltcal economy of envronmental polcy, and Grachen et al. (2001) for a case study. 2 Mauro (1995) and We (2000) show a negatve effect of corrupton on growth and foregn drect nvestment, respectvely, Ades and D Tella (1999) study the relatonshp between corrupton, ndustral polcy, and competton, López and Mtra (2000) develops a theory of the effects of corrupton on the envronmental Kuznets curve, Damana et al. (2002) fnd a negatve drect effect of corrupton, as well as an nteracton effect between trade openness and corrupton, on the standard for lead content n gasolne, and Fredrksson and Svensson (2002) study the nteracton effect between corrupton and poltcal nstablty on agrcultural sector envronmental polces. 3 Our focus on OECD countres also appears unque n the lterature on corrupton. 4 See Rodrk (1995). Potters and Sloof (1996, pp ) pont out that Most scholars ndeed fnd an ncreased scope for poltcal nfluence wth hgher degrees of concentraton, but there are many that fnd no effect or even a negatve effect. and there seems to be relatvely lttle drect emprcal support for the Olson (1965) nfluental theoretcal study on collectve acton. Marvel and Ray (1983), Walter and Pugel (1985), Gardner (1987), Trefler (1993), Gawande (1997), and Gawande and Bandyopadhyay (2000), found a postve effect of ndustry concentraton on poltcal nfluence, whereas Salamon and Segfred (1977), Fnger et al. (1982), and Baldwn (1985) found negatve, ambguous, or nsgnfcant effects. Grer et al. (1991) document an nverted-u relatonshp between an ndustry s prevalence of Poltcal Acton Commttees and both ndustry concentraton and ndustry sales. 2

4 vew, ths lterature suffers from a defcency: t has, to our knowledge, gnored the nterrelatonshp between lobby groups wth smlar nterests n polcy decson-makng. 5 Fnally, we merge the two lteratures dscussed above by explorng an nteracton effect of government corruptblty and lobby group sze on energy polcy. Is the effect of lobby group sze condtonal on corruptblty? We begn the analyss by developng a smple model of brbery, whch serves as gudance for our emprcal work. The model bulds on the menu aucton model orgnatng wth Bernhem and Whnston (1986), whch has been appled by Grossman and Helpman (1994) to trade polcy and by, e.g., Adt (1998) and Damana (2001) to envronmental polcy. The government s assumed to care about aggregate socal welfare and brbes. It takes brbes from both worker and captal owner lobby groups, n return for energy polcy that permts greater energy use n producton, and thus greater emssons of pollutants. 6 The wllngness of government poltcans to devate from optmal polcy makng here reflects the level of corruptblty (see Schulze and Ursprung (2001) and Fredrksson and Svensson (2002)). As n Laffont and Trole (1991), brbery s assumed costly to coordnate due to transactons costs (see also Eerola (2002)). These transactons costs (coordnaton costs) are assumed to be ncreasng n the sze of the ndustry sector, as argued by Olson (1965). Our theory separates lobbes () ncentves to offer a brbe (the amount at stake) and () ts ablty to coordnate brbery (coordnaton costs), from () the government s wllngness to be brbed (degree of corruptblty). These three factors together contrbute to a specal nterest group s success n the energy polcy process. Three man predctons emerge from the model. Frst, greater corruptblty reduces the strngency of energy polcy. Intutvely, an ncrease n corruptblty shfts the government s relatve 5 Oates and Portney (2001) state that It seems to us that approaches that explctly recognze ths nteracton of dfferent nterest groups are the most promsng for an understandng of envronmental polcy. (2001, p. 5). 3

5 weghtng away from welfare and towards brbes, facltatng the purchase of nfluence by makng t cheaper. Second, ncreasng costs of coordnatng brbery causes energy polcy to become more strngent (n general), consstent wth Olson (1965). Ths s because an ncrease n coordnaton costs reduces the lobby groups brbe offers to the government. Thrd, the dstrbuton of the worker and captal-owner lobbes poltcal pressures depends on how energy polcy affects the lobby group members ncome. In turn, the effects of ncreasng coordnaton costs on lobbyng efforts are dstrbuted n a smlar fashon: when the effect of coordnaton costs on worker lobbyng s hgh (low), the effect on captal owner lobbyng s low (hgh). An addtonal mplcaton of the model s that the effect of lobby group sze depends on the level of government corruptblty. We test these predctons usng dynamc panel data on sector energy ntensty (energy use per unt of value added) n OECD countres for the years As measures of sze of the captal owner and worker lobby groups, we use an ndustry s shares of aggregate () contrbutons to value added and () employment, respectvely. 7 The emprcal fndngs support several of the model s predctons. Frst, hgher corruptblty strongly correlates wth lower energy effcency n OECD countres. To our knowledge, ths s the frst emprcal evdence on the (detrmental) effects of corrupton (corruptblty) on energy polcy outcomes, and more generally on the effects of corrupton usng cross-country data exclusvely from OECD countres. Second, an ncrease n coordnaton costs appears to reduce the nfluence of the captal owner lobby, consstent wth our theory. However, for the worker lobby the same result holds only above a certan threshold ndustry sze. 8 6 Energy polcy determnes the maxmum allowed energy use, and thus restrcts the productvty of labor and captal. However, greater energy ntensty of producton mples that the unorganzed consumers suffer greater envronmental damage. 7 Salamon and Segfred (1977) and Grer et al. (1991) have prevously used ndustry sze as the unt of analyss n studes wth a focus on the U.S. 8 The relatonshp between coordnaton costs and poltcal success s stronger n less corrupt countres. 4

6 Thrd, the effects of coordnaton costs on the captal and worker lobby groups polcy success are ndeed related, as predcted by our theory. Focussng exclusvely on energy ntensve sectors, we fnd that the relatonshp between the captal owner lobby s coordnaton costs and ts polcy success s u-shaped. However, for the worker lobby the same relatonshp exhbts an nverted-u shape. 9 Thus, coordnaton costs have opposte effects on the polcy nfluence of the two lobbes. Fnally, we fnd that these relatonshps are condtonal on the level of corruptblty. To our knowledge, these are novel emprcal fndngs n the lterature, whch rarely separates worker and captal owner lobbyng and never has consdered the nteracton wth polcy maker corruptblty. The results may account for some of the ambgutes reported. Snce energy polcy s an mportant focus of the current debate (for example, n the negotatons on global warmng), we beleve our fndngs may have polcy mplcatons. Reforms amed at reducng corrupton (corruptblty) would have the ndrect beneft of mprovng energy effcency n ndustralzed economes. We also beleve our results may have more general applcablty to other envronmental polces, as well as to for example health and safety standards. The paper s organzed as follows. Secton II sets up the model, derves the equlbrum energy polcy, and generates the predctons of the model. Secton III descrbes the emprcal model and the data. Secton IV reports the emprcal results, and Secton V provdes a bref concluson. II. THE MODEL In ths secton, we develop a smple model of corruptblty and the endogenous formaton of energy polcy n a small open economy. 10 Each country contans frms that produce a prvate good, Q, for a perfectly compettve nternatonal market; prce equals unty. Producton requres nputs of captal, K, labor, L, and energy, θ. The captal stock and the labor supply are mmoble nternatonally, 9 Our fndngs may be related to Grer et al. (1991) who argue that PAC formaton and average ndustry sales exhbt an nverted-u shaped relatonshp. They argue that frms n concentrated ndustres may have a lower need for government nterventon snce they are already enjoyng greater profts, ceters parbus. 10 The model partally bulds on Oates and Schwab (1988) and Fredrksson and Gaston (2000). 5

7 whle energy can be mported free of mport dutes at prce p. Energy use s assumed pollutng and the damage confned to the country usng energy n producton. The producton technology exhbts constant returns to scale, s concave and ncreasng n all nputs, and twce contnuously dfferentable. It s gven by Q = F( K, L, θ ), whch by lnear homogenety can be rewrtten as Q = Kf ( l, α), (1) where l = L / K s the labor-captal rato (the nverted captal-labor rato) and α = θ / K, α Α, Α R + s the energy-captal rato. The energy-captal rato s the envronmental polcy set by the government. 11 Because the amount of captal s fxed n each country, ths mples that α determnes aggregate energy use. Implctly, t also specfes the energy-ntensty of producton. Suppressng arguments and usng subscrpts to denote partal dervatves, the margnal products of captal, energy, ( l α and labor are gven by f lf αf ), f α, and f l. The margnal products are dmnshng,.e., f αα < 0, f < 0, and we assume that f > 0,.e., an ncrease n α rases the margnal product of ll lα labor. Wth constant returns to scale, the sze of each frm s ndetermnate. The aggregate proft functon of the country s frms equals π = Kf ( l, α) Kr Lw θp, (2) where r s the cost of captal and w denotes the wage rate. Assumng many small frms, such that r and w are taken as gven, the FOC of (2) wth respect to energy use, θ, yelds f α = p. We assume 11 Ths specfcaton resembles for example the Corporate Average Fuel Economy (CAFE) standards that have been a recent focus of the energy polcy debate n the U.S., although frst mandated n 1975 (see Portney (2002) and Sterner (2003, p. 249)). CAFE standards specfy fuel use for all new lght-duty vehcles sold n the U.S. (note that the Transport Sector (ISIC Code 71) s the most energy ntensve sector n our sample (see Table 1 below)). Snce CO 2 emssons are proportonal to energy use, another example s the U.K. cap-and-trade system wth a fxed total of carbon emssons that are tradable, the Emssons Tradng Scheme (Sterner (2003, p. 92)) (snce we have dentcal frms wth constant returns to scale, permt trade s not applcable). Fnally, our specfcaton does not drve our results. Snce we assume that damage from polluton equals total energy use (see below), the specfcaton s equvalent to an envronmental polcy that sets an emssons cap. 6

8 that the government s regulaton of energy use s bndng, such that energy use by frms s restrcted to a quantty lower than mpled by the FOC. There are three types of ndvduals n ths economy: workers, consumers, and captal owners. W S Normalzng the populaton n each country to 1, let β = L, β, and β K represent the proporton of the populaton that are workers, consumers, and captal owners, respectvely. All ndvduals gan utlty from consumng the good, but consumers n addton suffer damage, D, from the polluton stemmng from the energy nput. The damage suffered by each consumer s, for smplcty, drectly proportonal to the energy used (and thus the envronmental polcy set by the government),.e. D = θ. Indvduals are assumed to have addtvely separable utlty functons of the followng form: U j j S = c λ D, (3) S where j = W, S, and K ndex workers, consumers, and captal owners, respectvely, and λ = 1 for consumers (0 otherwse). The ncome S Y of each consumer s exogenously determned, e.g. earned from employment n whte-collar jobs unaffected by envronmental polcy. Workers supply one unt of labor each and the wage equals the gan from employng an addtonal worker. The margnal product of captal equals the sum of the margnal product of captal gven permtted energy use, plus the addtonal output arsng from the ncrease n the permtted energy use (whch s ratoned n the model through the energy-captal rato) assocated wth an ncrease n the captal stock, αf α (see also Oates and Schwab (1988)). Hence, the returns to captal s gven by f lf l. Captal owners and workers are able to overcome free-rder problems and form two separate lobby groups, where all captal owners and workers are members, respectvely. Let both the organzed and unorganzed populaton groups be denoted by a superscrpt, =W,K,S. The organzed worker and captal owner lobby groups are assumed to offer the government a brbe schedule each, C (α ), = W, K. The brbe schedule offered by each lobby relates a prospectve brbe to the equlbrum 7

9 energy polcy chosen by the government. We follow Laffont and Trole (1991) by assumng that corrupt actvtes nvolve coordnaton (organzatonal) costs, such that the total cost to lobby of brbery equals ( 1+ λ ) C ( α), where we assume that the larger the coordnaton problems of lobby group, the greater the coordnaton costs (transactons costs), λ (see also Eerola (2002)). The total coordnaton costs may be vewed as reflectng the varable costs drectly related to lobby group sze. 12 The coordnaton costs only affect the brbery actvty level, not partcpaton, and the greater the sze of the ndustry sector, the greater the coordnaton costs. On the other hand, the consumers face suffcently large such costs that they are unable to overcome the problems assocated wth collectve acton (see Olson (1965)). and The net (ndrect) utlty functons of the worker and captal owner lobby groups are gven by W W W V ( α) Lf (1 + λ ) C ( α), (4) l K K K V ( α) K( f lf ) (1 + λ ) C ( α), (5) l respectvely. Both these groups prefer weaker restrctons on energy use. We express the aggregate utlty of the consumers (although not organzed n a lobby group) as V S S S S S ( α) β ( Y D) = β ( Y Kα). (6) The government derves utlty from a weghed sum of aggregate socal welfare and brbes, and thus ts utlty functon s gven by V G ( α ) av ( α) + C ( α), (7) = W, K, S = W, K where a 0 represents the exogenous weght that the government places on socal welfare relatve to brbes, and = W, K, S V (α ) represents aggregate socal welfare. We nterpret the weght a as the level 12 See, e.g., Mtra (1999) for a dscusson of fxed lobbyng costs. 8

10 of corruptblty (corrupton) of the regme. The greater the level of corruptblty (the smaller s a), the greater the nfluence of lobbyng actvtes relatve to socal welfare. Clearly, n some countres, even democratc OECD countres, the transfer of funds to poltcans s legal. In our vew ths represents a form of corrupton, and clearly other authors share ths perspectve. Shlefer and Vshny (1993) and Bardhan (1997) defne governmental corrupton as the propensty to sell polces for personal gans n the form of monetary transfers. Moreover, Schulze and Ursprung (2001) and Fredrksson and Svensson (2002) vew the nteracton between lobby groups and the government n the model by Grossman and Helpman (1994) as closely descrbng corrupton. The monetary transfers (brbes) are amed at nfluencng government polcy and not electons. The level of corruptblty n our model s, n essence, reflected by the government s wllngness to allow lobby groups to dstort energy polcy. 13 The equlbrum energy polcy s determned as the outcome of a two-stage, non-cooperatve game. In stage one the lobby groups offer the government a brbe schedule C (α ), =W,K. The strategy of each lobby s a dfferentable functon C : Α R+ ; and each lobby offers the government a monetary reward for selectng polcy α. In stage two, the government selects an energy polcy and collects the assocated brbe from each lobby. The lobbes are assumed not to renege on ther promses n the second stage. The lobbes receve payoffs descrbed by V : Α R+. The equlbrum n the common agency model by Bernhem and Whnston (1986) maxmzes the jont surplus of all partes and s formally equvalent to the Nash barganng soluton (see also Grossman and Helpman (1994) and Dxt et al. (1997)). 14 The characterzaton of the equlbrum 13 Coate and Morrs (1999) also vew the monetary transfer n ths type of model as a brbe. See also López and Mtra (2000) and Damana et al. (2002) for smlar formulatons. 14 * * The subgame perfect Nash equlbrum s defned as a feasble brbe schedule C and an energy polcy α such that the G energy polcy maxmzes the government's welfare V (α ), takng the brbe schedule as gven; and gven the government's and lobby s strateges, lobby j does not have a feasble strategy that results n a net payoff greater than the equlbrum net payoff. 9

11 * energy-captal standard, α, can be derved from the followng two necessary condtons (see Eerola (2002) for a complete proof): * G α arg maxv ( α); (C1) α * G α arg max{[ V ( α) (1 + λ ) C ( α)] + V ( α)},. α (C2) The dervaton of the equlbrum characterstcs s standard n the lterature, and we therefore omt t. The FOC of condton (C2) mples V ( α) = (1 + λ ) C ( α), whch can be substtuted nto the FOC of (C1) to yeld the characterzaton of the equlbrum energy polcy: α α 1 avα ( α) + V ( ) = 0. α α (8) 1+ λ = W, K, S = W, K The falure of the consumers to partcpate n poltcal brbery gves rse to a poltcal dstorton. From (8) t follows that ther nterests are not represented to the same degree as the workers and captal owners nterests. Whereas these latter two groups receve a weght of [ 1 (1 + λ ) + a], the consumers receve a weght of only a. However, as λ, the weght of group converges to a. Moreover, note that the greater the organzatonal costs nvolved wth brbery, the lower the nfluence of the organzed specal nterest groups. To fnd an explct expresson for the equlbrum energy polcy (correspondng to (8)), we need to fnd the effects of a change n the energy polcy on the gross welfare of the groups. The effect of energy polcy on the workers, captal owners, and consumers welfare s gven by and V W V ) = Lf, (9) V α ( α lα ) = K( f lf ), (10) K α ( α α lα S α S ( α) = β K, (11) 10

12 respectvely. Notce that the effect of energy polcy on the welfare of the worker and captal owner lobby groups are closely related. If the wage effect n (9) s relatvely large (small), the effect on the returns to captal n (10) s relatvely small (large). 15 Below, we wll see that the effect on ncome determnes the equlbrum lobbyng efforts of the two lobbes. Substtutng (9), (10) and (11) nto (8) yelds an equlbrum condton for the energy polcy equal to ( ) S + a lf + + { = W lα a f lf K α lα aβ 144 λ λ C A B (12) Equaton (12) shows the forces on energy polcy, n equlbrum. Term A captures the nfluence of the worker lobby group, term B mrrors the nfluence of the captal owner lobby group, and term C reflects the government s consderaton of consumers welfare (due to changes n emssons). All terms multpled by a are ncluded n socal welfare, and terms contanng λ reflect the brbery effort of lobby. Note that smple rearrangements of (12) mply that n equlbrum, β S > f α,.e. the margnal dsutlty from energy related polluton s greater than the margnal productvty of energy. Ths s due to the lack of partcpaton by the consumers n brbery actvtes. The prevous lterature has made a dstncton between nterest group actvty and condtons for poltcal success (see the survey by Potters and Sloof (1996)). Let us dscuss ths n terms of our framework, usng the worker lobby as an example. In our theory the aggregate nfluence of the worker lobby depends on three factors, n equlbrum. Frst, the worker lobby s ncentve to offer a brbe depends on the amount at stake (reflected by lf lα ). 16 Second, the ablty to offer a brbe s W affected by coordnaton costs (reflected by 1/(1 + λ )). The lobby s ncentve and ablty combnes nto an actvty level. Thrd, the degree of government corruptblty (reflected by a) s the 15 Expresson (10) s postve, gven that a weaker energy polcy rases the productvty of captal. The equaton reflects the share of the ncome ncrease (due to weaker energy polcy) not clamed by labor (see (9)). 11

13 wllngness of the government to sell polcy favors (.e. to dstort polcy). The level of a lobby s success n nfluence-seekng s consequently the combnaton of all three factors, as seen n (12). Predctons To examne the effect of corruptblty on energy polcy, we assume for smplcty that all thrd-order condtons are approxmately zero (as do Laffont and Trole (1998, p. 66)). Dfferentaton of (12) yelds S α β fα = a 1 + a f K 1+ λ αα, (13) whch s negatve, snce from (12), β S > f α. An ncrease n corruptblty (a lower a) rases the dstorton n energy polcy. Ths s smply because the government s wllngness to sell polcy favors rses. We summarze our frst testable predcton as: Predcton 1: Greater corruptblty reduces the strngency of energy polcy. Next, we fnd that the effects of captal owners and workers coordnaton costs equals: ( f lf ) α 1 α lα = < 0, (14) K 2 λ K ( 1+ λ ) 1 + a f K αα 1+ λ and α 1 lf lα = < 0. (15) W 2 λ W ( 1+ λ ) 1 + a f W αα 1+ λ Rsng lobby group coordnaton costs cause energy polcy to become more strngent, because the brbe offers are scaled back. Each lobby group member gans less by contrbutng to the lobby, on 16 Energy sector restrctons hurt energy-ntensve sectors more, and therefore more s at stake for both the worker and the 12

14 the margn (the lobby group sze s gven). Ths yelds a form of free-rdng by the lobby group members, causng lower brbe offers. The predcton that emerges s: Predcton 2: Greater lobby group coordnaton costs cause energy polcy to become more strngent. Note also that the effect of coordnaton costs: () depends on the level of corruptblty; () s nonlnear snce λ appears on the rght hand sde and thus depends on ts own value, and λ 0 α as λ. Coordnaton costs have lttle margnal effect where these costs are very hgh. These observatons wll be explored further n our emprcal analyss. Fnally, (14) and (15) also reveal that the effects of the two groups coordnaton costs are nter-related. They are proportonal to the lobbyng ncentves, whch are reflected n the numerators of (14) and (15). Ceters parbus, when the wage effect of energy polcy, lf lα, s large (n the numerator of (14)), the effect on energy polcy strngency of ncreased worker coordnaton costs s hgh, and the effect of captal owner coordnaton costs s low. The crucal determnant of the dstrbuton of the burden s thus the allocaton of the effect on factor rewards of energy polcy. If a partcular lobby carres a relatvely large burden from strcter energy polcy, coordnaton costs have a relatvely greater margnal effect on ts lobbyng effort. There s consequently a negatve nterdependence between the lobby groups n ther response to greater coordnaton costs. In sum, we have the followng predcton: Predcton 3: In equlbrum, when the wage effect of energy polcy s large (small), the effect on energy polcy strngency of worker coordnaton costs s large (small), and the effect on energy polcy strngency of captal owner coordnaton costs s small (large), ceters parbus. captal owner lobby n these sectors, ceters parbus. 13

15 III. EMPIRICAL STRATEGY Our theoretcal framework yelds three man predctons, whch we test emprcally n ths secton: () Greater corruptblty reduces the strngency of energy polcy; () Lobby group coordnaton costs cause energy polcy to become more strngent; () Worker and captal owner brbery efforts are negatvely related. Econometrc Model To test the man predctons of the model we analyze varatons n energy polcy strngency at the sector level wthn and between countres. We ft the followng model: Y jt = c j + β t + γz jt + δx jt + ε jt, (16) where Y jt s the sector specfc energy polcy strngency measure for sector n country j at tme t, c j are tme-nvarant country-specfc sector fxed effects, β t are sector specfc tme trends, Z jt represents the matrx of explanatory varables, X jt s the matrx of sector or country specfc control varables, and ε jt s the normally dstrbuted error term corrected for cross-sectonal heteroskedastcty (wth ε jt N(0, σ j )). Vector γ s the set of parameters of nterest and δ the vector of control parameters. In order to test whether the explanatory varables dffer sgnfcantly from zero, we start from the presumpton that sectors are non-homogeneous. The varaton n varances across dfferent sectors across countres s consderable. Therefore, we employ a Generalzed Least Square (GLS) specfcaton to correct for the presence of cross-secton heteroskedastcty. 17 The sample varancecovarance matrx s obtaned from a frst-stage OLS regresson and s terated untl convergence. 17 The Breusch-Pagan test statstc s 262 and clearly rejects the homoskedastcty assumpton. Furthermore, there s lttle reason to employ an error component or random effects model as the number of statstcal unts s large and sector data are not drawn from a random set as the sample s closed (see Matyas and Sevestre (1996)). Testng for the equalty of varances across the dfferent sectors clearly rejects the assumpton of homogenety employed by the ordnary OLS estmator. Due to the relatve short tme dmenson we estmated the model usng a determnstc trend. Senstveness for estmaton n frst dfferences, however, s very lmted. 14

16 Fnally, we perform several robustness tests. Below, we dscuss our data, ts sources and lmtatons, and next we address the lnks between theory and our estmaton equaton. Data Sources and Measurement Issues We have panel data set on the energy use n 11 sectors n 14 OECD countres for years Energy consumpton, n partcular the consumpton of fossl fuels, has been a major focus n envronmental polcy over the last decades. As our dependent varable, we use sector level energy effcency defned as physcal energy unts per unt of value added (ENERGYINTENSITY). Ths measure has the advantage that: () t reflects the combned regulatory strateges used by OECD countres (physcal energy constrants, effcency standards, and energy taxes); () t s potentally nfluenced by corrupton; 19 () t s avalable as a large panel data set; (v) t exhbts suffcent ndustry, country, and tme varaton to help us control for mportant unobservable factors that may nfluence our man regressors. 20 By restrctng energy use (per unt of output), frms nternalze several ar polluton externaltes that are drectly related to the combuston of fossl fuels, ncludng smog, acd ran (SO 2 and NO x ), and clmate change (CO 2 ). Energy use per unt of output s the ultmate outcome of polces that seek to constran energy use for envronmental reasons. Our unbalanced panel combnes OECD sectoral energy use data and economc data from the Intersectoral Data Base (ISDB) (see the Appendx for detals). Compared to the earler lterature, our data set s partcularly large (and focused). Table 1 specfes the sectors and ther levels of energy use per unt of output. 21 [TABLE 1 HERE] 18 Ths focus reduces the number of unobservables related to nsttutonal dfferences between hgh- and low-ncome countres. 19 Accordng to Leveque et al. (1996), the evaluaton of avalable technologes for energy effcency mprovements on the bass of Best Avalable Technologes Not Entalng Excessve Costs (BATNEEC) at the plant level leaves much room for regulatory capture. 20 Moreover, by lookng at the varaton n energy ntensty measured by the physcal amount of energy use (tons of ol equvalent) per unt of value added for each sector, our measure s not obstructed by changes n the sector s energy prce composte. 21 The ten countres are: Belgum, Canada, Denmark, Fnland, France, Italy, Japan, Netherlands, Norway, Sweden, the UK and the US. As seen n Table 1, the level of energy effcency s partcularly hgh n the transport sector, the basc 15

17 Explanatory varables Corruptblty s dffcult to measure. The wdely used Transparency Index (CORRUPT) from Transparency Internatonal (2000) s however a good and wdely used proxy for corruptblty (corrupton) (see, for example, Persson et al. (2000)). The ndex shows enough varablty across rch countres (mportant for our purposes) and over tme (to some degree). The CORRUPT data show that Denmark, Fnland, Norway and Sweden have become less corrupt over the tme perod , whereas Belgum, France, Japan and the USA have become more corrupt. In order to construct a smooth estmate of the long-term trend n corruptblty we nterpolated the avalable data ponts usng the Hodrck-Prescott flter. 22 In order to measure the coordnaton costs of the worker and captal owner lobbes we focus on sector sze, followng Olson (1965) and our theory. As a measure of the captal owner lobby s coordnaton costs, we use an ndustry sector s contrbuton to total value added (VALUEADDED%). Larger sectors represent a greater number of frms and greater coordnaton problems, gven frm sze. 23 As a measure of worker coordnaton costs, we use the share of all workers employed n a partcular sector (EMPLOYMENT%). 24 Moreover, n order to capture non-lneartes (dscussed by the theory), we nclude the squared terms VALUEADDED% 2 and EMPLOYMENT% 2. Our theory predcts that the effect of sector sze on brbery efforts depends on the margnal effect of energy polcy on lobby group ncome,.e. the amount at stake. A small change n energy polcy has a greater mpact on the returns to labor and captal when more energy s used, thus the metal ndustry, the non-metallc mneral products ndustry, and the paper ndustry, whereas the commercal servce sector and constructon have the lowest levels. The ntra-sectoral varaton s consderable. 22 The Hodrck-Prescott flter s normally used to smooth macro-economc tme seres to obtan a smooth estmate of the underlyng long-term trend. The smoothng process mnmzes the varance of the seres around the smoothed seres usng a penalty parameter on the second dfference of the smoothed seres. In contrast, our objectve s to maxmze the varance n the corrupton seres gven the observed data ponts. Therefore, we choose very low values for ths parameter to create a trend as close as possble to the gven data ponts. 23 To verfy that VALUEADDED% s postvely correlated wth the number of frms n an ndustry we used the 4-dgt data avalable from OECD (2003) whch covers a subset of our data (2121 observatons). The correlatons (t-statstcs wthn brackets) were 0.82 [30.4] for the Netherlands (69 sectors; years ), 0.84 [29.4] for Norway (166; ), 0.63 [18.9] for Portugal (179; ), 0.21 [6.0] for Span (132; ), and 0.62 [36.4] for the full sample. 16

18 economc stakes n the polcy outcome are greater for sectors that use relatvely more energy. Interest groups wll lkely engage more ntensvely n brbery f they have hgher stakes n the energy polcy outcome. 25 To capture dfferences n stakes between sectors, we create the nteracton varables VALUEADDED%*ENERGY% and EMPLOYMENT%*ENERGY%, where ENERGY% s the share of total energy used by a sector. Control varables Assumng that envronmental qualty s a normal good, demand wll ncrease wth ncome. Our measure s ncome per capta (GDPPC). Followng Antweler et al. (2001), we assume that energy polcy s lkely to respond slowly to polluton problems, and we therefore use the average GDP per capta of the prevous 3 years. The ntensty of energy use at the sector level s affected by (adaptve) frm behavor. Indeed, dfferences n the structure of energy producton across countres, as well as shfts n producton structure through prce or technologcal effects, are lkely to affect energy ntensty. We address ths problem by explct and mplct controls. In order to capture structural changes n energy prces n the observed perod, and the assocaton between domestc energy resources and lower local energy prces (OECD (1999)), we nclude domestc energy prces for heavy fuel ol (OILPRICE) and electrcty (ELECTRICITYPRICE). OILPRICE should capture common trends n both the ol and natural gas markets, whle electrcty prces usually reflect country specfc crcumstances. 26 Furthermore, we add factor prces such as mean hourly wage of producton workers (WAGES) and cost of captal (CAPITALPRICE). Changng (relatve) nput prces mght nduce substtuton between producton factors. Antweler et al. (2001) fnd that captal s complementary to energy, and we 24 Workers n larger sectors may have greater nfluence at the votng booth (see Stgler (1971)). Whle workers votng power lkely plays a role, we need to keep n mnd that our theory gnores votng ssues. 25 Lobby group sze s often used to measure both a lobby s stake n the polcy outcome and the lobby s poltcal strength, and therefore ambguous results are common n the lterature. Gven that energy and captal are usually complements (see Antweler et al. (2001)), restrctons on energy use are lkely to ht captal owners harder n captal-ntensve ndustres. 17

19 therefore expect substtuton between a captal/energy composte and labor. Whereas we expect OILPRICE and ELECTRICITYPRICE to take negatve sgns (hgher energy prces should lead to substtuton away from energy as a factor of producton), we expect WAGES and CAPITALPRICE to have the opposte effects on ENERGYINTENSITY. Fnally, we use sector specfc fxed effects (c j ) to allow for uncontrolled varables such as excluded economc determnants or technologcal dfferences between sectors. 27 For example, the producton process n a sector may dffer between countres. 28 We take these parameters as fxed over tme. In order to account for tme related unobservables, we add sector specfc tme trends (β t). The pattern of energy ntensty shows a clear trend for most sectors. Ths may reflect a gradual shft over tme towards new technques n response to the ol prce shock n the 1970s. Gven our focus on sectors as our basc estmaton unt, allowng for tme related varaton between sectors appears preferable, compared to ntroducng heterogeneous tme dummes that are homogenous for all sectors. Table 2 provdes descrptve statstcs. [TABLE 2 HERE]. IV. RESULTS Basc fndngs Table 3 presents the man results from our estmatons. [TABLE 3 HERE] In Models 1-3, we nclude the varables of nterest, but not all nteractons are ncluded n each model. Ths enables us to better understand the effect of the nteractons dscussed n our theory. CORRUPT has the expected postve sgn n all models and s sgnfcant at the 1% level, mplyng that greater corruptblty ncreases energy ntensty. Moreover, consstent wth our theory, VALUEADDED% has a sgnfcant 26 We also expermented wth ndces reflectng ndgenous energy producton varables, measured as the share of total energy use comng from domestc fossl fuels, nuclear power, and green sources (bomass, hydro, wnd, and sun). None of these are mportant. Results are avalable upon request. 27 We extensvely tested our model wth and wthout country-specfc, as well as sector-specfc fxed effects, n partcular because varaton of CORRUPT s farly lmted. These results ndcate that country-specfc sector fxed effects are to be preferred (F-test). Moreover, our man results do not depend on ths specfcaton. Results are avalable upon request. 18

20 negatve sgn, ndcatng that captal owner lobby groups encounter ncreasng coordnaton costs as sector sze rses. VALUEADDED% 2 s mostly nsgnfcant suggestng a lnear effect. However, EMPLOYMENT% has a postve and sgnfcant sgn, but the margnal effect of worker lobby coordnaton costs s rsng as EMPLOYMENT% 2 has a (small) negatve coeffcent. Ths mples that workers n large sectors receve less strct energy polces, but the effect declnes as ndustry employment rses. In Model 2, CORRUPT*VALUEADDED% s sgnfcant and postve as expected, suggestng that the captal owners coordnaton costs have less mpact where polcy makers are more corruptble. Second, CORRUPT*EMPLOYMENT% has a sgnfcant and negatve coeffcent, ndcatng the opposte relatonshp for workers (compared to captal owners). Note that the estmated coeffcent on EMPLOYMENT% roughly doubles from Model 1 to 2. The effect s lower n countres wth more corruptble polcy makers, however. Thrd, the mportance of ncludng the nteracton terms s also reflected by the fact that the sgnfcance of the CORRUPT coeffcent rses strongly as we move from Model 1 to 2. In Model 3, the nteractons wth ENERGY% capture the stakes nvolved n energy polcy. ENERGY%*VALUEADDED% s sgnfcant, ndcatng that captal owners suffer less from coordnaton costs f a sector uses a larger share of total energy consumpton and thus has more at stake. In Model 4 all nteractons are ncluded. Ths s our preferred model, as t captures the drect and ndrect effects dscussed by our theoretcal model. Most coeffcents are sgnfcant at the 1% level. CORRUPT lowers the strngency of energy polcy, consstent wth our theory. Ths s partcularly the case n sectors wth hgh VALUEADDED% and low EMPLOYMENT% The paper ndustry n Fnland uses raw pulp, whch s more energy ntensve than the recycled fbers used by the Dutch paper ndustry, for example. 29 Model 4 suggest that a unt ncrease n CORRUPT (evaluated at the mean) ncreases the value of ENERGYINTENSITY by 1.87 (= * *4.6). 19

21 In Fg. 1 we plot the relatonshp between CORRUPT and ENERGYINTENSITY at the sample means of all sgnfcant varables. [FIG. 1 HERE] Hgher corruptblty clearly correlates wth greater energy ntensty, whch lends support for our frst hypothess. A unt ncrease n CORRUPT ncreases ENERGYINTENSITY (at the mean) by 0.86 unts (= * *4.6). The effect of government corruptblty, however, also depends on coordnaton costs of the lobby groups (the ndrect effect). Ths senstvty s llustrated for a rse of both VALUEADDED% and EMPLOYMENT% by 1 std. dev. above the means. At ths hgher value for VALUEADDED%, the level of ENERGYINTENSITY s below average for all values of corruptblty, confrmng the hypothess that captal owners n larger sectors face greater coordnaton costs. However, the margnal effect of CORRUPT s stronger n ths case (the slope beng greater). In contrast, the evdence for sectors wth large employment shares (1 std. dev. above the mean) s mxed. Large worker lobbes face lower-than-average coordnaton problems at low levels of corruptblty, but hgher-thanaverage problems at hgh corruptblty values. Greater corruptblty thus rases coordnaton costs for worker lobbes. The results suggest that the effects of energy polcy brbery actvtes are senstve to lobby groups coordnaton costs. Wth respect to the captal owner lobby, the (sgnfcant) coeffcents mply that a unt ncrease n VALUEADDED% (evaluated at the sample means) causes a declne n energy use per unt of output (a reducton of the value of ENERGYINTENSITY) by 2.53 unts (= * *4.7). Ths lends support to both Olson s and our theory. Fg. 2 plots the relatonshp between VALUEADDED% and ENERGYINTENSITY, condtonal on three dfferent levels of CORRUPT. [FIG. 2 HERE] The negatve (lnear) relatonshp suggests that captal owners n larger sectors are less successful n nfluence actvtes. However, ther organzatonal problems are less problematc n more corrupt countres For captal owners, ncreased corruptblty ncreases the ablty to gan nfluence on energy polcy. At a CORRUPT level two std. dev. above the mean, the effect of an ncrease n VALUEADDED% falls somewhat (n absolute value), to 20

22 Model 4 also mples that the predctons of our theory regardng worker coordnaton costs hold only for sectors employng a suffcently large share of workers: we detect an nverted-u shaped relatonshp between employment numbers and the effects of coordnaton costs. Evaluated at the mean, a unt ncrease n EMPLOYMENT% causes an ncrease n ENERGYINTENSITY by 1.23 unts (=3.42 2*0.02* * *4.7). The effect of EMPLOYMENT% s less pronounced n countres wth greater corruptblty, however. 31 As llustrated by Fg. 3, for suffcently hgh values of EMPLOYMENT% a unt ncrease ndeed reduces energy ntensty. 32 [FIG. 3 HERE] Moreover, at hgh values of EMPLOYMENT%, the mpact of a further ncrease n employment share s greater the lower the level of corruptblty. 33 The generally opposte effects of ncreases n VALUEADDED% and EMPLOYMENT% lend a measure of support to our Predcton 3. In stuatons where captal owner coordnaton costs have a relatvely large (small) effect on the polcy outcome, the effect of the worker lobby s coordnaton costs are small (large). Model 4 suggests that corruptblty, as well as energy share dfferences between sectors, have the opposte effects on the two lobby groups coordnaton problems (both varables reduce the captal owner lobby s coordnaton problems, but ncrease the worker lobby s). In Table 3 we also present some alternatve specfcatons that brng out the dfferences between relatvely energy ntensve (Model 5) and energy extensve (Model 6) sectors. 34 Snce the economc stakes may dffer consderably between these types of sectors, brbery behavor on energy Whle our theoretcal model correctly predcts the nteracton between captal owner coordnaton costs and corruptblty, t fals to predct the drecton of ths condtonalty. 31 Evaluated at one std. devaton above the sample means for CORRUPT, the margnal effect declnes to 0.54 (= *4.6*2 0.53* *4.7) and becomes negatve at two std. devatons. 32 At one std. dev. above the mean for EMPLOYMENT%, the margnal effect s 0.88 (= *13.4*2-0.53* *4.7), whereas at the maxmum level t equals The peak occurs at EMPLOYMENT%=24.9, wth 7% of all observatons to the rght of the peak. 33 The nteractons wth ENERGY% (agan) ndcate that workers and captal owners have opposte polcy effects as sector energy use rses, although some sgn shfts occur. One explanaton may be that the correlaton between ENERGY%*VALUEADDED% and ENERGY% *EMPLOYMENT% s qute strong for some sectors. 34 See Table 1 for these categores. Note that an energy-ntensve sector does not necessarly also have a hgh share of energy consumpton wthn a country. For nstance, the relatvely energy ntensve BMI ndustry n Denmark has a value of 1590 for ENERGYINTENSITY, but uses only 1.3% of total energy. 21

23 polcy may potentally also dffer notceably. The results for the energy ntensve sectors (Model 5) yeld coeffcent szes substantally larger than n Models 1-4 and 6. Moreover, the varables of mportance reman sgnfcant wth dentcal sgns as n Model 4, and VALUEADDED% 2 becomes sgnfcant. Captal owner coordnaton costs and energy ntensty dsplay a u-shaped relatonshp for energy ntensve sectors. Fg. 4 shows a smlar pattern as n Fg. 1 between CORRUPT and ENERGYINTENSITY, but the effects are now much stronger (notce the dfferences n scale on the axs). [FIG. 4 HERE] An ncrease n CORRUPT by one std. dev. mples a 4% ncrease n ENERGYINTENSITY n energy ntensve sectors, whereas ths effect s 0.4% for the full sample. The result can be expected, snce the stakes n energy polcy outcomes are sgnfcantly greater n energy ntensve ndustres. In general, energy restrctons tend to hurt ncomes more n energy ntensve sectors, gven the larger amount of energy (per unt of output) used n these sectors. Therefore the lobbes stakes are greater (see (12)). Fg. 4 also ndcates that lobby group coordnaton costs n large energy ntensve sectors have smlar effects as n the full sample (Fg. 1). Ths agan suggests that the effect of corruptblty depends on lobby group coordnaton costs. Fg. 5 and 6 explctly plot the separate effects of the two lobby groups n energy ntensve sectors. The relatonshp between ENERGYINTENSITY and VALUEADDED% for the energy ntensve sectors (condtonal on CORRUPT) s non-lnear. 35 [FIG. 5 HERE] It appears that when the stakes are hgh, captal owners are able to overcome coordnaton problems once the sector s sze s suffcently large. 36 Fg. 6 shows the relatonshp between ENERGYINTENSITY and EMPLOYMENT% for energy ntensve sectors (condtonal on CORRUPT). 37 [FIG. 6 HERE] In energy-ntensve sectors, the peak of the nverted-u occurs substantally earler than n the full sample 35 Ths fndng s consstent wth Guttman (1978) who fnds that the effect of the number of producers on poltcal nfluence on agrcultural polcy s more lkely postve when a state has a relatvely large number of farmers. 36 An alternatve explanaton could be that large sectors beneft from varous subsdes that reduce the net energy prce. Ths s a topc for future research that s outsde the scope of ths paper. 37 The peak for ENERGYINTENSITY and EMPLOYMENT% occurs at 3.7% wth 22% of the observatons to the rght of the peak. These fgures for ENERGYINTENSITY and VALUEADDED% are 5.1% and 9%, respectvely. 22

24 (Model 4). The non-lnear effects are strong n these sectors, suggestng that the worker lobby s coordnaton problems emerge at substantally lower levels of sector employment. Ths may be explaned by the fact that accordng to Antweler et al. (2001), energy ntensve sectors are relatvely captal-ntensve. Thus, the workers stakes are relatvely lower n energy ntensve sectors. For energy extensve sectors (Model 6), most of the (relevant) coeffcents are smlar to Model 4. Thus, the energy extensve sectors appear to drve the results n the full sample (Model 4), but sector energy ntensty s mportant for the effect of corruptblty and coordnaton costs. Turnng to the control varables n Table 3, GDPPC has a sgnfcant postve sgn n some models, mplyng that rcher countres may permt greater energy use per unt of output. ELECTRICITYPRICE s sgnfcant wth a postve sgn n most models. CAPITALPRICE s also sgnfcant n most models wth the expected (postve) sgn, whle WAGES s never sgnfcant. The explct controls reflectng factor prces thus appear less robust than our man varables, ndcatng that poltcal varables are relatvely mportant for energy effcency. Dscusson of Results Why does EMPLOYMENT% tend to rase ENERGYINTENSITY for part of the sample range (especally n energy ntensve sectors)? We can provde several explanatons that together contrbute to ths emprcal result, relyng both on our theoretcal framework and on the prevous lterature: () Predcton 3 suggests that when the effect of coordnaton costs on captal owner lobbyng s great, the effect s small on worker lobbyng. In partcular, when lf lα n the numerator of (15) s small, the W numerator of (14) s large (thus, for small values of lf lα, α λ 0 ). Fg. 2 and 3 (partally) reflect such a relatonshp. Whereas captal owners suffer severely from coordnaton costs, the effect on workers s mnor for a large part of the sample range. () Moreover, expressons (14) and (15) also mply that coordnaton costs have lttle mpact when these costs are hgh. If coordnaton costs ( λ W ) are hgh n the average sector even for small worker groups (possbly due to fxed costs), a further 23