Pam21720 - Biology

8/10/2019 Pam21720 - Biology

1/40

The Effect of the Kyoto Protocolon Carbon Emissions

Rahel AicheleGabriel Felbermayr

Abstract

Since 1997, CO2emissions have continued to rise in many countries despite their emis-sion caps under the Kyoto Protocol (Kyoto). Failure to meet promised targets, however,does not imply that Kyoto has been pointless. Whether Kyoto has made a differencerelative to the counterfactual of No Kyoto is an empirical question that requires aninstrumental variables strategy. We argue that countries ratification of the statutes

governing the International Criminal Court is a valid instrument for ratification of Ky-oto commitments. In our panel fixed effects estimations, the instrument easily passesweak identification and overidentification tests. It can be plausibly excluded from oursecond-stage equations and does not cause CO2emissions. Our estimates suggest thatKyoto ratification has a quantitatively large (about 10 percent) and robust, thoughonly moderately statistically significant, negative effect on CO2 emissions. We also

show that higher fuel prices and a different energy mix in Kyoto countries support thisresult. C 2013 by the Association for Public Policy Analysis and Management.

INTRODUCTION

Arguably, climate change is the most important global environmental problem ofour times. Its policy dimensions are explored by a large and insightful theoreticalliterature. Due to the public goods nature of CO2emissions, it is individually rationalfor countries to free ride on others emission reductions. International environmen-tal agreements (IEAs), such as the Kyoto Protocol, exist to solve this dilemma.

In the Kyoto Protocol (Kyoto), 37 industrialized nations and the European Union(EU) have agreed to cap their levels of greenhouse gas (GHG) emissions to anaverage of 94.8 percent of their 1990 emissions by the period 2008 to 2012. Yet since1997, emissions have continued to rise in many countries despite their emissioncaps. In 2010the latest year with GHG data available from the United NationsFramework Convention on Climate Change (UNFCCC)many countries were stillfar from achieving their promised GHG emission reductions. It appears as if theKyoto Protocol has been ineffective.

However, our main argument is that failure to meet a promised target does notimply that Kyoto has been completely unsuccessful in bringing down emissionsrelative to the counterfactual situation of No Kyoto (i.e., a counterfactual worldwhere no Kyoto Protocol exists). One needs to apply program evaluation techniques

to this large-scale policy intervention. Therefore, we ask whether there is empirical

Journal of Policy Analysis and Management, Vol. 32, No. 4, 731757 (2013)C 2013 by the Association for Public Policy Analysis and Management

Published by Wiley Periodicals, Inc. View this article online at wileyonlinelibrary.com/journal/pamDOI:10.1002/pam.21720

8/10/2019 Pam21720 - Biology

2/40

732/ Kyoto and Carbon Emissions

evidence that the Kyoto Protocol induced emission savings or not. Can an interna-tional climate treaty without a strong enforcement mechanism help mitigate climatechange?

To this end, we explore the Kyoto Protocol as an international climate policyquasi-natural experiment, and ex-post evaluate its effect on emissions. The mostimportant econometric problem is that selection into Kyoto is most likely not ran-

dom. The public economics literature argues that GDP per capita, initial emissions,development status, and political freedom are important determinants of IEA mem-bership, Kyoto including. This complicates correct statistical inference as randomemission shocks are likely to correlate with Kyoto commitments. It is straight-forward to control for unobserved time-invariant country-specific correlation bymaking use of the panel dimension of the data, but time-specific technology shocksor changes in environmental preferences could still cause biased estimates. Addi-tionally, emission projections could drive commitment (reverse causation), whichrequires an instrumental variables (IV) strategy. Therefore, we explore determinantsof Kyoto membership and contribute an instrument that could potentially be usedin many applications.

In our IV strategy, we use countries membership in the International Criminal

Court (ICC) based in The Hague, Netherlands, as an instrument for Kyoto com-mitment. Using fixed effects (FEs) estimation, we find robust evidence that Kyotocommitment reduces CO2emissions by some 10 percent on average. To corroboratethis surprisingly high effect, we investigate possible channels through which Kyotomay have affected CO2emissions. We identify effects on countries energy and elec-tricity mix, fuel prices, and energy and electricity use. We believe that these resultsare potentially important for negotiations about future climate deals. They implythat even a highly imperfect international climate deal may be better than no dealat all.

The UNFCCC constitutes the legal framework for negotiating internationaltreaties to reduce GHG emissions. In 1997 such an agreement was signed in the

Japanese city of Kyoto. The Kyoto Protocol entered into force in February 2005,after Russias ratification ensured the critical 55 percent of world emissions and55 percent of countries threshold. Aside from a general emission reduction target,the Kyoto Protocol sets legally binding country-specific targets ranging from an 8percent reduction for EU countries to a 10 percent increase for Iceland.1 Primarily,GHG cutbacks have to be achieved with domestic policy measures, but countriescan also use the so-called flexible market mechanisms (emissions trading, cleandevelopment mechanism, and joint implementation) to meet their targets. As inother IEAs (e.g., the Oslo, Helsinki, or Montreal Protocol), committed countries areobliged to report various GHGs emissions and their policy measure implementationstatus to the UNFCCC. And even though the Kyoto Protocol has an enforcementbody, there is no credible enforcement mechanism.2

A key conceptual question is why a voluntary, nonenforceable agreement suchas Kyoto should solve the prisoners dilemma and matter at all.3 The public eco-nomics literature discusses mechanisms through which voluntary IEAs could beeffective. In the case of the Kyoto Protocol, the most compelling argument is thatparticipating countries have the obligation to monitor and report emissions to the

1 Taking into account the EU burden sharing agreement, the targets range from a 28 percent reductionin Luxembourg to a 27 percent increase in Portugal.2 The WTO is one rare example of an international agreement that provides some enforcement instru-ments (e.g., in the form of countervailing duties).3

This question indeed arises for many international institutions and agreements, not just the ones onglobal environmental goods.

Journal of Policy Analysis and Management DOI: 10.1002/pamPublished on behalf of the Association for Public Policy Analysis and Management

8/10/2019 Pam21720 - Biology

3/40

Kyoto and Carbon Emissions/ 733

UNFCCC, which summarizes emission reduction achievements in annual reports.The resulting league tables are usually the matter of substantial debate and maywell matter for the political process in countries. So, the lack of formal sanctionshas certainly hampered Kyoto, but it does not automatically imply that Kyoto hasnot added incentives to engage in mitigation policies with the objective to saveemissions. Whether Kyoto has affected policy and thus emissions is an empirical

question. In this paper, we attempt to provide evidence on this issue.The rest of the paper proceeds as follows. The next section describes our empir-ical strategy. First, CO2 emissions are explained by variables capturing economicdevelopment, population growth, political preferences, and trade openness. To ac-count for relevant time-invariant country features such as endowments with nat-ural resources, industrial structure, geographical position, and climate, as well asfor unobserved heterogeneity, we use a FEs estimation strategy. Second, we arguethat countries ICC membership is a valid and relevant instrument. The ICC is apermanent, international tribunal that prosecutes war crimes and crimes againsthumanity. The Rome Statute, which governs the ICC, was adopted in 1998 andratified by the necessary quorum of 60 countries at the end of 2002. The Kyoto Pro-tocol was negotiated one year earlier, and has been ratified by countries since 2001.

The timing of the two multilateral initiatives coincided nicely. The two treaties alsoposed similar domestic policy issues. For example, commentators such as Groves(2009) describe both the Kyoto Protocol and the Rome Statute as threats to thesovereignty of the United States, which has ratified neither. It is thus not surpris-ing that ICC membership correlates robustly to countries commitments under theKyoto Protocol. In terms of content, in contrast, the two treaties have nothing incommon. ICC membership does not relate to environmental outcome variables suchas the level of CO2and GHG emissions or fuel prices, nor is it likely to directly causethose variables. These features make ICC membership and its spatial lag (i.e., othercountries membership dummies, weighted by their distance and size) candidateinstruments for Kyoto commitment.

The following section presents results from first-stage regressions. We find a ro-bust relationship between ICC and Kyoto membership, both in a linear probabilityas well as in a Probit model. The subsequent section presents our core results; it in-

vestigates how Kyoto has affected CO2emissions and presents a host of robustnesschecks. Amongst others, we run a placebo check, exclude economies in transition(EIT) from the sample, use total GHG emissions as an alternative dependent vari-able, and explore the timing of effects by including lags of Kyoto ratification. Theresults show a negative effect of Kyoto commitment on CO2 emissions of around10 percent. To support this result, the penultimate section turns to the channelsthrough which Kyoto may have affected emissions. The results suggest that Kyotoworks through changes in the energy mix, higher fuel taxes, and more efficientenergy and electricity use. The last section contains concluding remarks.

RELATED LITERATURE

Our work is related to two important strands of literature. The first group of papersdeals with the economic rationale for IEAs and with their effectiveness. IEAs areusually understood as ways to solve the free-riding problem inherent in climatepolicy. The paper by Congleton (2001) offers a comprehensive survey. Andreoni andMcGuire (1993), Welsch (1995), Hoel (1997), Carraro and Siniscalco (1998), andLange and Vogt (2003) describe how international IEAs can solve this dilemma byproviding commitment devices and policy forums for coordination.

However, IEAs (like all international treaties) are based on the voluntary coop-eration of countries. This raises the question of whether such agreements can be


8/10/2019 Pam21720 - Biology

4/40


effective and how emission targets compliance can be achieved by means of appro-priate institutional designs. Gerber and Wichardt (2009) propose a specific exampleof an explicit mechanism that induces compliance even in the absence of a centralinstitution empowered to enforce the agreement. Ringquist and Kostadinova (2005)and Aakvik and Tjtta (2011) survey the literature and provide more general theoret-ical arguments on why international agreements can be useful even in the absence

of an enforcement mechanisms. The central reasons are that, first, IEAs could in-duce scientific research, raise environmental awareness, and change preferences,thereby affecting technological options and the regulatory environment. Second,noncompliance with a voluntary IEA could entrain a loss of trust in other interna-tional policy arenas like international lending so that countries find it optimal tocomply (Rose & Spiegel, 2009). There is also a growing empirical body of literatureon the efficacy of international institutions. IEAs such as the Montreal and HelsinkiProtocol have been scrutinized, with inconclusive results on their effectiveness tobring emissions down (see Aakvik & Tjtta, 2011; Finus & Tjtta, 2003; Murdoch,Sandler, & Vijverberg, 2003; and Ringquist & Kostadinova, 2005).

While focusing on a very specific IEA, the Kyoto Protocol, our work is moregenerally related to the literature on international treaties, such as, in the area of

international trade. For example, there is a prominent discussion on whether theWorld Trade Organization (WTO) had trade-enhancing effects or not. The initial andsurprisingly inconclusive results by Rose (2004) have triggered important method-ological advances; see, for example, Tomz, Goldstein, and Rivers (2007). Whenapplicable, we make use of the insights of that literature. And the IMF (2009) findsthat voluntary or unenforceable fiscal rules have had significant effects on countriesdebt levels. Countries may have dramatically fallen short from proclaimed targets,but this does not imply that the rules have not had any effect. We contribute to thisstrand of literature by investigating the Kyoto Protocol.

A second stream of research deals with the econometric issues that arise dueto the fact that IEA membership or the stringency of climate policy can hardly

be taken as exogenous. Murdoch and Sandler (1997) and Beron, Murdoch, andVijverberg (2003) discuss how countries select into such treaties as a function oftheir economic development, their demographic situation (see, in particular, York,2005), or political determinants that simultaneously affect countries emissions andtheir willingness to engage in international policy efforts. This is an importantchallenge for statistical inference which the literature is only starting to take up.

This challenge looms particularly large when it comes to the quantification ofKyotos effect on countries CO2 emissions, the most pervasive GHG. Determinantsof emissions are predominantly discussed in the carbon Kuznets curve literature,which stresses a dynamic relationship between development (measured by GDPper capita) and CO2 emissions per capita, and started as a purely empirical exer-cise, see, for example, Grossman and Krueger (1995) and Holtz-Eakin and Selden

(1995) for early contributions. Subsequently, theoretical explanations have been putforward; Andreoni and Levinson (2001) and Brock and Taylor (2010) provide excel-lent examples. Dinda (2004) and Galeotti, Lanza, and Pauli (2006) have presentedcomprehensive surveys. Much of the Kuznets curve papers attempt estimating theturning point beyond which further GDP per capita growth lowers emissions percapita. The evidence for the existence on such a turning point is mixed, and we donot find conclusive results. Azomahou, Laisney, and Nguyen-Van (2006) present arecent skeptical view.

More closely related to our work are recent studies that are based on theKuznets curve framework, but investigate the role of the Kyoto Protocol. Grunewaldand Martnez-Zarzoso (2009) include a dummy for Kyoto ratification in the car-

bon Kuznets curve framework. In a panel of 123 countries over the period1974 to 2004, the authors find that Kyoto commitment reduces CO2 emissions.


8/10/2019 Pam21720 - Biology

5/40


Iwata and Okada (in press) investigate different types of GHG emissions, and findKyoto commitment leads to emission reducing effects for CO2 and CH4, but notfor NO2, and positive and significant effects for other GHGs (HFCs, PFCs, SF6).However, both Grunewald and Martnez-Zarzoso (2009) and Iwata and Okada (inpress) do not instrument Kyoto ratification. Aichele and Felbermayr (2012) use aninstrument similar to the one employed in this paper to study the carbon footprints

for a sample of about 40 countries.

4

EMPIRICAL STRATEGY AND DATA

A Model of CO2 Emissions: The Second Stage

We are interested in understanding the effect of Kyoto commitments on countriesCO2emissions. In a later section on transmission channels, we will also investigateother dependent variables such as the energy mix, fuel prices, and per capita energyuse.

Denoting country i s outcome variable of interest at time t by Yit , we want toestimate the parameter, 1, in the following relationship

Yit = 0 + 1Kyotoit + 2Xit + t + i + it . (1)

In our core exercise,Yit is the log of CO2emissions. We also work with a broadermeasure of all GHG emissions, but due to data availability, we present results as arobustness check. Xit is a vector of controls that influence emissions. These controlscan be divided into three different categories. The first set of controls is comprisedof economic and demographic determinants of emissions (the log of GDP and itssquare, the log of population, the log of the share of agriculture, manufacturing,and services in GDP, and economic openness). We expect that an economicallylarge country has higher levels of emissions, all else equal; population growth in-

creases emissions; and countries with a high share of manufacturing experiencehigher levels of emissions, whereas a large share of services and agriculture in GDPshould be associated with fewer emissions. The second category includes measuresfor preferences and policy (the stock of other IEAs, a countrys political orienta-tion measured by the chief executives party affiliation, a dummy variable for WTOmembership, and the Polity index).5 The stock of other IEAs is a proxy for envi-ronmental awareness. So we expect a country that has signed up for more IEAs tohave lower emissions. Finally, Xit contains the spatial lag of Kyoto commitment,that is, the Kyoto status of other countries weighted with their respective size overdistance squared. This measure reflects carbon leakage. We expect that countrieswith large Kyoto countries nearby (i.e., with a larger value of the spatial Kyoto lag)are less prone to competitiveness effects and thus have higher own emissions. To

account for time-invariant country-specific determinants such as endowments offossil fuels, patterns of comparative advantage, climatic and geographic conditions,

4 Aichele and Felbermayr (2012) differ from this study in several dimensions: first, the country coverageis limited to only 40 countries; second, it is more narrowly focused on the proper calculation of carbonfootprints of countries and on their behavior relative to territorial emissions under Kyoto.5 We do not include an EU dummy. In our fixed effects approach, only the change in EU status matters. Inour time period, with the exception of Malta and Cyprus, new EU members are EIT that have experiencedmore changes than just a change in EU status. So, including an EU dummy would not be informativeabout the effect of becoming an EU member country. However, in a robustness check we exclude

transition countries from the sample.


8/10/2019 Pam21720 - Biology

6/40


and historical features (such as historical emission levels), we add a full set of coun-try dummies, i . We also include a full set of year dummies, t, to control for theworld business cycle or the world market price of oil.

The key independent variable is a dummy for Kyoto commitment, (Kyotoit ). Ittakes the value of 1 if a countryi has ratified the Protocol at timet and has a cap ondomestic CO2 emissions. It takes the value of 0 otherwise:

Kyotoit =

1 ratification of emission cap andt ratification year,

0 else.

It implies that Kyoto starts to matter for committed countries once ratificationthrough the parliament has occurred. Studies evaluating the treatment effects ofIEAs such as the Montreal, Helsinki, and Oslo Protocol take a similar stance andalso use ratification as the decisive treatment date (see Ringquist & Kostadinova,2005).

Econometric Issues

Our default strategy to eliminate the country-specific unobserved heterogeneity, i ,from equation (1) is FE estimation on yearly data. It is fairly standard and used inmuch of the empirical Kuznets curve and IEA treatment effects literature. Yet, thisstrategy is subject to some criticism. Bertrand, Duflo, and Mullainathan (2004) ar-gue that standard errors of treatment effects in FE estimation are inconsistent, andthe estimators standard deviation is underestimated if the outcome and treatment

variable are both serially correlated over time. This might lead to an over-rejectionof the Null of no effect. The authors suggest applying a FE estimator to the pre-treatment and posttreatment averages (long FEs estimator) to cure the spuriouscorrelation problem. Although there has been some heterogeneity in the timing ofKyotos ratification across countries, most countries ratified it between 2001 and2003. So we assume treatment takes place in this period, but conduct robustnesschecks pertaining to this choice. The pretreatment and posttreatment windows of1997 to 2000 and 2004 to 2007 are chosen such that they constitute symmetricintervals around this treatment window.6 Whereas the within transformation onyearly data are subject to the Bertrand, Duflo, and Mullainathans (2004) critique, ithas the advantage that it does not require assumptions about a treatment window.Moreover, the number of useable observations is about 10 times larger than in thelong FE model. For these reasons we show results for both methods.

FE estimation eliminates any correlation between the second-stage equation er-rors and the Kyoto dummy that would be due to time-invariant unobserved hetero-geneity at the country level. However, such a correlation may arise for two further

reasons. First, if our Kyoto dummy is a noisy measure for a countrys true climatepolicy stance, estimates based on a simple within-transformed model would be bi-ased towards zero (attenuation bias). Second, a bias also arises in the presence ofreverse causation, that is, if a transitory shock on the outcome variable makes itmore or less likely that a country has a binding Kyoto commitment. Instrumentingthe Kyoto status cures these biases. Therefore, the key empirical challenge consistsin finding a valid instrument for Kyoto commitment.

6 Note that Russia and Ukraine ratified Kyoto in 2004 and Belarus in 2005 and are assigned to the

treatment group. Australia and Croatia ratified the Kyoto Protocol in 2007 and are assigned to thecontrol group.


8/10/2019 Pam21720 - Biology

7/40


IV Strategy

In order to identify the causal effect of Kyoto onYit (emissions and other outcomes),we require instruments, Zit , that (i) correlate with Kyotoit and that (ii) are uncor-related with the error term in equation (1). Condition (ii) implies that Zit must nothave an effect onYit except throughKyotoit , and thatZit has explanatory power forK yotoit conditional on the vector of controls, Xit , included into equation (1). An IVapproach then exploits the exogenous variation in instruments for causal inference.Thus, the first-stage model is

Kyotoit = + X

it + Z

it + i + t + it , (2)

wherei is a country-specific FE and t is a year dummy.In this paper, we propose ICC membership as an instrument for Kyoto commit-

ment. The Rome Statute, which governs the ICC, was finally signed in 1998, enteredinto force in 2002, and by December 2010 ratified by 114 countries. Thirty-fourcountries, including the United States, India, and China, decided not to ratify theRome Statute. Since the ICC statutes were ratified around the same time as the

Kyoto Protocol, time variation in the ICC membership dummy has the potential tocorrelate with time variation in the Kyoto status variable.

The idea behind our instrumentation strategy is that both treaties reflect a coun-trys preferences for international policy initiatives. Some countries are more willingthan others to give up national sovereignty and subject themselves to an interna-tional organization. And indeed, Groves (2009) argues that both the Kyoto Protocoland the Rome Statute constitute a threat to U.S. sovereignty.7 Further on we test thislinkthat is, condition (i)and estimate the first-stage equation (2) with a linearprobability model.8 If a countrys ICC status is a valid instrument for Kyoto com-mitment, then its spatial lag (all other countries membership dummies weightedby a strictly exogenous proxy for trade links) is a valid instrument as well.

The exclusion restriction (ii) cannot be tested formally. It requires that a coun-trys ICC involvement is not caused by carbon emissions and that it does not directlyaffect the outcome variables, either. Meeting these requirements appears plausi-ble enough, but the exclusion restriction also requires that the instrument is notcorrelated with the error term. One concern may be that altruistic or cooperativecountries have lower emissions and a higher likelihood of ICC ratification. How-ever, our FE estimation strategy deals with country-specific heterogeneity, whichmakes the exclusion restriction more likely to hold. Moreover, we include the stockof other IEAs that captures how a countrys environmental preferences evolve overtime. We also add variables related to the political orientation of governments, thepolity index, and a WTO dummy to capture any political preferences that may berelated to the ratification of the ICC and Kyoto and also influence emission levels.

Changes in the production structure of a country are taken into account in thesecond stage. Components left in the error term of the second-stage equation areunobservable changes in technological change or changes in comparative advantagethat influence emissions as well as Kyoto commitment. However, this is unrelated to

7 Similarly, Mike Huckabee, former Governor of Arkansas, argues that the Kyoto Protocol would havegiven foreign nations the power to impose standards on us (Huckabee, 2007, p. 70). China expressedsimilar concerns in the Copenhagenclimate change negotiations. Other developed or newly industrializedcountries that are neither a Kyoto nor an ICC member include Israel, Korea, Singapore, Chile, and Turkey.8

This appears to be the natural choice since we are working with a linear fixed effects approach in thesecond stage. However, the correlation also holds in a Probit framework.


8/10/2019 Pam21720 - Biology

8/40


ICC membership. With these considerations in mind, we argue that we can excludeIC Cit and its spatial lag from the second-stage equation (1).

9

Data

We briefly describe the data used in our empirical exercise. CO2 emissions are

from the World Banks World Development Indicators (WDI) 2010. They compriseemissions due to the burning of fossil fuels and the manufacture of cement, andinclude carbon dioxide emissions produced during consumption of solid, liquid,and gas fuels, and flared gas. Alternatively, we work with a broader measure of GHGemissions that we also take from WDI. These data are available for most countriesin our sample; however, GHG emissions are not reported on a yearly basis, so wecannot apply the same econometric strategy when using them.10

Unfortunately, for a sample comprising Kyoto and non-Kyoto countries for asufficiently long time window, there is no comprehensive data base on fuel taxesand, even more desirably, on implicit or explicit carbon prices. However, due to thenature of our econometric exercise, one can attribute the time variation in fuel pricesto time variation in taxes.11 Data on diesel and gasoline pump prices,12 electricity

and energy use per capita, and the shares of different energy sources in energy andelectricity production were compiled from WDI 2010.

Data on ICC membership stem from the UN Treaty Series database. The ICCdummy takes a value of 1 if a country has ratified the Rome Statute governingthe ICC and 0 otherwise. The spatial lag of ICC membership is the average ICCmembership of other countries (all other countries membership dummies weightedby population over distance, and added up). The Kyoto dummy is constructed fromthe UNFCCC homepage. Kyotos spatial lag is constructed exactly as the ICC lag.13

GDP, population, and openness data stem from the Penn World Table 6.3. Open-ness is the usual ratio of exports plus imports over GDP (measured in current prices).The shares of manufacturing, agriculture, and services (value added) in GDP are ob-

tained from the WDI 2010. The stock of other IEAs was calculated using the IEAsDatabase Project.14 It gives the number of IEAs other than Kyoto a country hasratified or accepted up to a given year. The chief executives party orientation isfrom the World Banks Database on Political Institutions (DPI) 2010, and codes thegovernments orientation with respect to economic policy as 1 (right wing, conserva-tive, Christian democratic), 2 (centrist), or 3 (left wing, socialist, social democratic,or communist). A 0 indicates cases where none of the previous categories fit orthe party does not focus on economic issues. The WTO dummy takes a value of1 if a country is member to the WTO and 0 otherwise, and was compiled fromthe WTO homepage. The Polity Index was obtained from the Center of SystemicPeaces Polity IV Project Database. The index classifies countries according to their

9 Other international treaties, such as those governing the WTO or international environmental ques-tions, cannot be easily excluded since they will affect emissions directly through green preferences of

voters and consumers, or through trade policy.10 GHG data are available for the years 1990, 2000, 2005, and 2008. The country coverage ranges from109 nations in 1990 to 134 for the latest data point.11 The nonparametric time trend takes care of the global fuel price; country-level fixed effects take careof endowment-related or geographic determinants of fuel price differences.12 Pump prices of diesel and gasoline are only available every other year from 1998 to 2006.13 The exact calculation of the spatial weighting matrix (e.g., with distance squared instead of simpledistance or with all others trade shares, i.e., exports plus imports, in GDP as a more direct measureof economic competition) does not make a significant difference. Distance data come from the CEPII

distance database.14 http://iea.uoregon.edu/.


8/10/2019 Pam21720 - Biology

9/40


Table 1.Summary statistics.

Variable Obs. Mean Std. dev. Min Max Source

Dependent variables, second stageln CO2 emissions (kt) 1,456 9.78 2.28 4.19 15.69 (a)ln GHG emissions (kt of CO2 eq.) 335 11.20 1.56 8.39 16.04 (a)

Diesel pump price (USD/L) 622 0.58 0.33 0.01 1.73 (a)Gasoline pump price (USD/L) 622 0.73 0.35 0.02 1.90 (a)Renewables, share in energy use 1,209 22.64 27.45 0.00 93.77 (a)Alternative energy, share in 1,209 2.52 5.11 0.00 29.26 (a)

electricity productionFossil fuels, share in energy use 1,209 67.82 26.81 5.32 102.43 (a)Coal, share in electricity prod. 1,209 18.04 27.11 0.00 99.46 (a)ln electricity use (kWh/capita) 1,198 7.29 1.49 3.02 10.15 (a)ln energy use (kg oil eq./capita) 1,209 7.17 0.99 4.83 9.38 (a)

Kyoto variablesKyoto ratification (0,1) 1,456 0.12 0.33 0.00 1.00 (b)Kyoto, spatial lag 1,456 0.28 0.97 0.00 13.29 (b)

Kyoto stringency (0,1,2) 1,456 0.20 0.57 0.00 2.00 (b)InstrumentsICC ratification (0,1) 1,456 0.33 0.47 0.00 1.00 (c)ICC, spatial lag 1,456 0.12 0.38 0.00 3.46 (c)

Additional controlsln GDP 1,456 17.95 1.91 13.23 23.28 (d)ln GDP, squared 1,456 325.94 69.64 175.05 542.06 (d)ln population 1,456 9.28 1.51 6.04 14.09 (d)ln manufacturing (in percent of GDP) 1 ,421 2.62 0.56 0.35 3.79 (a)ln agriculture (in percent of GDP) 1,440 2.30 1.11 2.63 4.36 (a)ln service (in percent of GDP) 1,437 3.93 0.35 1.01 4.40 (a)ln stock of other IEA 1,456 3.28 0.59 1.79 4.84 (e)Chief executive party orientation 1,456 0.17 0.09 0.10 0.30 (f)

Openness (current price) 1,456 0.86 0.48 0.05 4.57 (d)WTO dummy (0,1) 1,456 0.82 0.39 0.00 1.00 (g)Polity index 1,456 3.66 6.41 10.00 10.00 (h)

Note:The table shows summary statistics for variables over the period 1997 to 2007.

Sources: (a) World Bank WDI 2010, (b) http://www.unfccc.int, (c) UN Treaty Series database, (d) PWT6.3, (e) http://iea.uoregon.edu, (f) World Bank DPI 2010 (series: execrlc), (g) http://www.wto.org, (h)http://www.systemicpeace.org.

political authority characteristics and ranges from 10 to 10, where 10 is a per-fectly autocratic regime and 10 is a full democracy. Table 1 lists summary statisticsand sources.15

The WDI database has information on emissions for almost 200 geographicalentities. However, we drop entries that do not constitute independent countries(such as, e.g., Puerto Rico, Greenland, Monaco, and Hong Kong). Moreover, for anumber of countries, we do not have the full list of covariates as defined above. Atthis stage we lose mainly small (island) states like Grenada, Barbados, but also some

very recently created countries (such as Kosovo). In our analysis, we drop countries

15 Additional information on Kyoto status, average emission growth from the pretreatment to posttreat-ment period, and sample info (rich, large, Organization of the Petroleum Exporting Countries (OPEC),transition country) is given in Appendix Table A1. All appendices are available at the end of this article as

it appears in JPAM online. Go to the publishers Web site and use the search engine to locate the articleat http://www3.interscience.wiley.com/cgi-bin/jhome/34787.


8/10/2019 Pam21720 - Biology

10/40


with fewer than five observations and countries that only appear in either the pre-or the postpolicy period to get a more balanced sample and have comparability withthe long FE sample. This leaves us with a sample of 133 countries. In 2005, samplecountries account for about 85 percent of global CO2 emissions, which totaled 2.92billion tons of CO2. The period of observation covers the years 1997 to 2007. Wechose this time frame because it is symmetric around the treatment time window.

Data availability would allow us to extend the analysis up to 2009; however, theinclusion of the crisis years 2008 and 2009 does not seem recommendable to us.

SELECTION INTO KYOTO: THE ROLE OF ICC MEMBERSHIP

In our sample, 32 of 133 countries have commitments under the Kyoto Protocol.16

Within the group of countries that had commitments as of 2007, there is somevariation as to the timing of national ratification. The first countries to ratify acommitment were Romania and the Czech Republic (in 2001), 27 countries ratifiedtheir commitment in 2002, followed by Lithuania (2003), Ukraine (2004), Belarus(2005), and finally Australia and Croatia (2007). The tetrachoric correlation betweenthe Kyoto and ICC ratification dummy is 0.78. A two-sided test that Kyoto and ICC

are independent is rejected at all conventional significance levels (P-value 0.00). Toexamine what drives Kyoto ratification and verify condition (i), we now estimateequation (2) with a linear probability model for Kyoto commitments for the timespan 1997 to 2007, applying the same methods (FE and long FE estimation) as forequation (1).

Table 2 presents results on the first-stage regressions. Columns 1 to 3 report FEestimations on levels of yearly data. We adjust the variancecovariance matrix forheteroskedasticity and for clustering of standard errors within countries (Stock &Watson, 2008). Column 1 shows that ratification of the Rome Statute governing theICC correlates strongly with ratification of the Kyoto Protocol. The estimated coef-ficient of 0.19 implies that ICC ratification increases the odds of Kyoto ratification

by 19 percentage points. The estimate is different from zero at the 1 percent levelof statistical significance and, together with country and time dummies, explainsabout 25 percent of the variance in the Kyoto dummy. In column 2, adding thespatial lag of ICC ratification, that is, ICC ratification by spatially close countries,further increases the share of variance explained to 52 percent in the FE model.17

Column 3 adds the vector of controls,Xit , which also features in the second-stageregressions. The spatial lag of Kyoto ratification adds no explanatory power giventhe spatial lag of ICC and the other covariates.18 The log of GDP enters negatively,its square positively (though the latter is not statistically distinguishable from zero).This signals that economic growth deters countries from ratifying their Kyoto com-mitment, but the effect eventually levels off. Population size has a large negativeeffect on the odds of Kyoto ratification. Since we identify all effects in Table 2 by

time variation at the country level, our results suggest that countries with higherpopulation growth are less likely to have commitments. This finding is in line withYork (2005), who underlines the importance of demographic factors for the ratifica-tion of the Kyoto treaty. Countries industrial structure matters to a certain degree

16 Five committed Kyoto countries (Iceland, Liechtenstein, Luxembourg, Russia, and Switzerland) arenot included due to data availability.17 We have also experimented with other international agreements such as the Comprehensive Nuclear-Test-Ban Treaty and the Anti-Personnel Land Mines Convention. Ratification of these texts also tends tomake Kyoto commitments more likely; however, the effects are weaker and less statistically significant.18

Note that a positive and statistically significant effect is obtained when the spatial lag of ICC is notincluded in the equation.


8/10/2019 Pam21720 - Biology

11/40


Table 2.First-stage regressions: explaining Kyoto commitment.

Dependent variable: Kyoto commitment (0,1)

Linear probability Probit longFEModel Long FE diff.a

Method (1) (2) (3) (4) (5)

Excluded instrumentsICC (0,1) 0.19*** 0.11*** 0.10*** 0.25*** 0.16***

(0.05) (0.04) (0.03) (0.07) (0.03)ICC, spatial lag 0.51*** 0.41*** 0.37*** 0.04***

(0.10) (0.08) (0.12) (0.01)Other controls

Kyoto, spatial lag 0.01 0.01(0.01) (0.01)

ln GDP 0.94* 2.46** 2.04***

(0.52) (1.02) (0.69)ln GDP, squared 0.02 0.07** 0.05***

(0.02) (0.03) (0.02)ln population 1.72*** 1.31*** 3.42***

(0.38) (0.45) (0.59)ln manufacturing (percent of GDP) 0.01 0.02 0.16

(0.04) (0.10) (0.12)ln agriculture (percent of GDP) 0.14** 0.29** 0.05

(0.06) (0.13) (0.08)ln services (percent of GDP) 0.01 0.15 0.25**

(0.08) (0.17) (0.13)ln stock of other IEA 0.02 0.27 0.07

(0.09) (0.17) (0.11)Government orientation (0.1, 0.2, 0.3) 0.03 0.07 0.11

(0.19) (0.59) (0.20)Openness, (Exp + Imp)/GDP 0.22*** 0.27 0.63***

(0.07) (0.18) (0.14)WTO (0,1) 0.15** 0.28* 0.16**

(0.06) (0.14) (0.07)Polity (1 to 1) 0.00 0.02 0.02***

(0.00) (0.01) (0.01)

No. of observations 1,456 1,456 1,418 266 133Adj.R2 0.25 0.52 0.61 0.69F-stat 4.32 6.69 7.36 5.69Log-likelihood 13.922 39.60

Note:Linear probability and Probit models. Sample: 133 countries. Probit regression shows marginaleffects. Heteroskedasticity robust standard errors (clustered at country level) in parentheses. *P < 0.1;**P < 0.05; ***P < 0.01. LPM: Year dummies and constant included (not shown).aEstimation on cross-section of differences between pretreatment and posttreatment averages.

for Kyoto commitment: The higher the share of agriculture in a countrys GDP, thelower the probability it will ratify the Kyoto Protocol. The logs of the manufacturingand services shares are not relevant for Kyoto commitments.

We include two variables to proxy for green preferences. The first variable, thelog stock of other (than Kyoto) IEAs ratified by a country is expected to affect

the likelihood of commitment positively, but does not show up significantly inthe analysis. The second variable is the countrys chief executive partys political


8/10/2019 Pam21720 - Biology

12/40


orientation. One would think that left-leaning governments are more likely to acceptcommitments, but this does not show up in our regression.

The next two variables measure openness to international trade: exports plusimports over GDP and the WTO dummy. Both variables correlate negatively withKyoto ratification. On average, doubling openness makes ratification less likely by22 percentage points. Note that 15 countries became WTO members in the sample

period;

19

with the exception of the three Baltic states, none of these countries havecommitments under Kyoto.As a robustness check, column 4 applies the long FE model, where the change

of Kyoto status (i.e., ratification) is regressed on changes in instruments and othercontrols. Comfortingly, the explanatory power of that model is comparable to theFE model while parameter estimates typically are slightly larger. Column 5 reportsmarginal effects pertaining to a Probit model on long-differenced data.20 Resultsare again comparable to the long FE linear probability model.

KYOTOS EFFECT ON EMISSIONS

In this section we present our core results about the effect of Kyoto commitment on

CO2emissions. First, we present the evidence in a graph, and then we employ moreformal econometric FE models. Finally, we turn to several robustness checks.

Graphical Inspection of Kyotos Effects

Figure 1 plots the density of changes in the log of CO2emissions over two groups ofcountries: countries that end up with emission caps, and countries that do not. Thechanges are computed over period averages 1997 to 2000 (before the first countryhas ratified the Protocol) and 2004 to 2007 (following the ratification of Russiaand Ukraine in 2004). Ratification occurred later only in Belarus (2005), Australiaand Croatia (2007).21 In both groups, emissions have fallen in some countries, but

increased in others. In the sample of non-Kyoto countries, the distribution is moreto the right and the dispersion of changes is much more pronounced. Emissionshave fallen substantially in some developing countries affected by civil war (such asBurundi), and increased strongly in countries recovering from crises (such as Chador Angola). In the group of committed countries, Norway and Spain have increasedemissions by more than 20 percent, while they went down by roughly 8 percentin Belgium and Germany. Between the two periods, emissions have increased onaverage by 27 percent in the group of noncommitted countries, while they haveincreased on average by 3.4 percent in the group of committed countries (see the

vertical lines). The difference of the means is statistically significant at the 1 percentlevel of significance.

Regression Results

Figure 1 does not control for the effects of time-varying controls and self-selection into the Kyoto Protocol. Therefore, we turn to more elaborate estimationtechniques. Table 3 reports our benchmark results. Column 1 uses ratification of

19 These countries are Albania, Armenia, Cambodia, China, Croatia, Estonia, Georgia, Jordan, Lithuania,Latvia, Moldova, Macedonia, Nepal, Saudi Arabia, and Vietnam.20 Time-differencing pre- and postpolicy averages, the selection equation yields a model in which thedependent variable (Kyoto commitment) and the instrument (ICC) are again binary variables. This

procedure eliminates unobserved heterogeneity; yet, the Probit model remains applicable.21 Note that Belarus is considered as treated, while Australia and Croatia are put in the control group.


8/10/2019 Pam21720 - Biology

13/40


0

.01

.02

.03

.04

.05

Densit

y

50 25 0 25 50 75 100Emission growth (in %)

Kyoto NonKyoto

Note: The figure shows a kernel density plot (Epanechnikov, optimal bandwidth) of CO 2emission growthrates (i.e., lnEMit ) of non-Kyoto and Kyoto countries, where t = 0 is the pretreatment (1997 to 2000)average andt = 1 is the posttreatment (2004 to 2007) average. The graph also shows the mean emissiongrowth rate in each subpopulation.

Figure 1. Differences in Emission Growth Rates in Kyoto and Non-Kyoto Countries.

Kyoto commitment to explain variation in yearly emissions.22 The Kyoto dummy ishighly statistically significant and suggests that Kyoto commitment reduces emis-sions by about 17 percent. The simple model explains about 27 percent of the

variation in emission growth across countries.Column 2 turns to a more comprehensive model. Adding the control vector, Xit ,

more than halves the Kyoto effect and also weakens statistical significance to the5 percent level.23 The explanatory power of the model rises to about 48 percent.We find that a 1 standard deviation increase in the spatial lag of Kyoto ratification(more countries that are close ratify) increases emissions by 1.4 percent.24 Thisfinding is consistent with the carbon leakage hypothesis: Noncommitted countriesincrease emissions as they step up exports of carbon-intensive goods to committedcountries. The coefficients on log GDP and its square are positive and negative, re-spectively, and not statistically different from zero. Jointly, however, they are highlysignificant (theP-value of theF-test statistic of joint significance is 0.00), signalinga problem of collinearity. Dropping the GDP squared term yields a coefficient onthe log of GDP of 0.56. So, a 1 percent increase of GDP is associated with 0.56 per-

cent higher CO2 emissions. Population increases emissions; the elasticity is highlysignificant and statistically not distinguishable from unity.25 This result is in linewith the literature (see, e.g., Cole & Neumayer, 2004). The share of manufactur-ing in GDP has a positive though not statistically significant effect on emissions,while a higher share of agriculture or services in GDP reduces CO2 emissions. Thechief executives party (government) orientation has a positive effect on emissions:

22 In principle, this replicates Figure 1, but with yearly data.23 In the discussion paper version of this paper, we used a larger set of controls and obtained very similarresults.24

0.965 0.014 = 0.014.25 TheP-value of theF-test statistic is 0.68.


8/10/2019 Pam21720 - Biology

14/40


Table 3.Second-stage regressions: the effect of Kyoto on CO2 emissions.

Dependent variable: ln CO2 emissions

Long LongFE-OLS FE-OLS FE-IV FE-OLS FE-IV

Method (1) (2) (3) (4) (5)

Kyoto (0,1) 0.17*** 0.06** 0.10** 0.09** 0.12*

(0.03) (0.02) (0.05) (0.04) (0.07)Kyoto, spatial lag 0.01** 0.02*** 0.03*** 0.03***

(0.01) (0.00) (0.01) (0.01)ln GDP 0.65 0.62 1.13** 1.02*

(0.55) (0.50) (0.54) (0.56)ln GDP, squared 0.00 0.00 0.02 0.01

(0.01) (0.01) (0.01) (0.02)ln population 1.10*** 0.95*** 1.12*** 1.05***

(0.24) (0.26) (0.26) (0.29)ln manufacturing (percent of GDP) 0.12 0.12 0.25*** 0.25***

(0.10) (0.09) (0.09) (0.09)

ln agriculture (percent of GDP) 0.09* 0.10** 0.26*** 0.27***(0.05) (0.04) (0.08) (0.09)

ln services (percent of GDP) 0.10 0.10 0.26* 0.26*

(0.10) (0.09) (0.15) (0.15)ln stock of other IEA 0.18** 0.18*** 0.23** 0.21*

(0.07) (0.07) (0.10) (0.11)Government orientation (0.1,0.2,0.3) 0.15** 0.15** 0.29** 0.28*

(0.06) (0.06) (0.15) (0.15)Openness, (Exp + Imp)/GDP 0.01 0.02 0.16 0.16

(0.07) (0.07) (0.12) (0.12)WTO (0,1) 0.01 0.00 0.04 0.03

(0.04) (0.04) (0.05) (0.06)

Polity (1 to 1) 0.00 0.00 0.01 0.01(0.00) (0.00) (0.01) (0.01)

No. of observations 1,456 1,418 1,418 266 266No. of countries 133 133 133 133 133First-stage diagnostics

Sheas partialR2 0.28 0.43HansenSarganJ-stat (P-value) 0.44 0.71Weak-ID test (F-stat) 19.09 37.70

Second-stage diagnosticsAdj.R2 0.27 0.48 0.49F-stat 11.99 13.85 17.44 21.40 23.09

Note:Standard errors in parentheses adjusted for within-group clustering and heteroskedasticity. *P

Pam21720 - Biology

Documents

Transcript of Pam21720 - Biology