Accepted Manuscript
Title: ENVIRONMENTAL TAXATION AND THE DOUBLE DIVIDEND HYPOTHESIS IN CGE MODELLING LITERATURE: A CRITICAL REVIEW
Author: Jaume Freire-Gonzalez´
PII: S0161-8938(17)30120-5 DOI: https://doi.org/10.1016/j.jpolmod.2017.11.002 Reference: JPO 6397
To appear in: Journal of Policy Modeling
Received date: 12-3-2017 Accepted date: 10-11-2017
Please cite this article as: & Freire-Gonzalez,´ Jaume., ENVIRONMENTAL TAXATION AND THE DOUBLE DIVIDEND HYPOTHESIS IN CGE MODELLING LITERATURE: A CRITICAL REVIEW.Journal of Policy Modeling https://doi.org/10.1016/j.jpolmod.2017.11.002
This is a PDF file of an unedited manuscript that has been accepted for publication. As a service to our customers we are providing this early version of the manuscript. The manuscript will undergo copyediting, typesetting, and review of the resulting proof before it is published in its final form. Please note that during the production process errors may be discovered which could affect the content, and all legal disclaimers that apply to the journal pertain. ENVIRONMENTAL TAXATION AND THE DOUBLE DIVIDEND HYPOTHESIS IN CGE MODELLING LITERATURE: A CRITICAL REVIEW
Jaume Freire-González1,2
1 Department of Economics, Harvard University, Cambridge, Massachusetts 02138, United States. 2 ENT Environment and Management, Sant Joan, 39, 1, 08800 Vilanova i la Geltrú, Barcelona, Spain.
Abstract
Computable general equilibrium (CGE) modelling is a flexible and open way to model the economic systems that allow practitioners to assess the impacts of different policies or external shocks over an economic system. There is some empirical literature dedicated to test the double dividend hypothesis of an environmental tax reform using CGE models. This hypothesis claims that is possible to obtain an improvement of both environmental and economic conditions by imposing an environmental tax and recycling revenues obtained to reduce other pre-existing taxes. This research provides a comprehensive review of this literature including a statistical and a meta-regression analysis. 69 different simulations from 40 studies have been analyzed. 55% of simulations have achieved a double dividend, concluding that although the environmental dividend is almost always achieved, the economic dividend still remains an ambiguous question that needs further research.
Keywords: Taxation; CGE; environmental policy; modelling; double dividend.
JEL: C6, H2, Q5
1. CGE modelling and the double dividend hypothesis
1.1. Introduction
Environmental taxation has been growingly seen as an effective economic instrument to create incentives in favor of cleaner production and consumption habits. Concerns about environmental problems like climate change, health or use of resources have fostered its useACCEPTED across different countries and regions onMANUSCRIPT several areas. Some authors highlight the advantages it represents over other economic instruments like payments for environmental services, tradable permit systems.
There is an important discussion around the economic and environmental effects of this economic instrument. Conventional economic theory suggests that increasing the price
1 of normal goods and services through the use of a tax reduces its own demand. Pigou (1920) was the first to propose the use of taxes to internalize negative externalities. He showed that a direct tax on carbon emissions is the most efficient tax instrument for
CO2 abatement. Empirical research has analyzed and corroborated this direct effect, however it is not clear the extension and dimension of the economy-wide effects. The affectation of an environmental tax on the overall economic system will ultimately depend on many factors, like its design and implementation, the economic structure on which it is implemented, consumer preferences or other socio-political factors.
The price paid by a buyer and the one received by the seller is different if there is a tax. Given that, under some assumptions, prices coordinate efficient resource allocation, such difference reduces economic efficiency and consumer welfare by an amount greater than the revenue collected by the government. This is known in literature as the "deadweight loss" or "excess burden" of the tax system. Beyond the isolated effects of an environmental tax, there is a research and policy line that claims that an environmental tax reform (ETR) that increases the environmental taxes and, at the same time, reduces other taxes, keeping the same government revenue, generates a double benefit, by reducing negative environmental impacts and improving the economy.1 This is known as the ‘double dividend’ (DD) hypothesis, and it has been of growing interest amongst CGE modelers during the last years. However, there are different variations and interpretations of this hypothesis (Mooij, 1999). The main problem in the determination of the economy-wide effects of environmental taxation or other policies is the high complexity involved in modern economic systems and societies. There is a discussion among economists around the different existing theories and derivate methods used to assess these global effects. From more traditional economic perspectives, some modelling applied methods have been developed in order for account for overall economy-wide effects of policies or other external shocks. The most important ones are: macroeconometric models, input-output methods and applied general equilibrium methods. However, they are often interrelated in more complex assessments.
Computable general equilibrium (CGE) methods and models are the most comprehensive way to take into account the most relevant aspects of an economic system (Dixon and Jorgenson, 2012). The great development of computing and IT technologies during the last half century have fostered their general use and an increasing its complexity. This complexity sometimes makes them difficult to understandACCEPTED for non-specialized audiences orMANUSCRIPT policy-makers, but also allow them to explicitly take into account more aspects of the economic reality than other methodologies. On the other side, these methods sometimes rely on controversial
1 Some countries that have already implemented different versions of explicit environmental tax reforms are: Sweden, Denmark, the Netherlands, the United Kingdom, Finland, Norway, Germany and Italy.
2 assumptions on the general functioning of economies. CGE modelling is an open standard tool for the analysis of the economy-wide impacts of environmental policies, including environmental tax issues, on resource allocation and the associated implications for incomes of different economic agents. They come from traditional input-output models initially developed by Leontief (1941).
On the other hand, input-output models contain a simple representation of the interchanges between industries and other economic agents in a region during a specific temporal period. They can be easily build for impact analysis nowadays, as data needed (input-output tables) are usually available in official statistic institutes under national accounts. Leontief developed a straightforward method to assess exogenous shocks under certain restrictive assumptions like fixed technology. Later on, Leontief (1953) elaborated a dynamic input-output model providing more elements for the future development of general equilibrium modeling. Isard (1968) proposed a methodology based on input-output tables that related economic and environmental variables in order to offer more policy alternatives. Later, Kneese et al. (1970) used a similar input-output approach for environmental policy. These methods, however, maintain the assumption that environmental impacts or resource use are proportional to output, similarly than the fixed coefficients assumption of the traditional input-output model. So they do not allow technical change after an exogenous shock. These and other limitations encouraged researchers to develop more complex methods. Johansen (1960) was the first known who developed an empirical general equilibrium model without the fixed-coefficients assumption of input-output analysis.
This research provides a comprehensive review on the literature of DD hypothesis of environmental taxation using CGE methods. After some decades of extensive use of these models for the analysis of the DD hypothesis it is time to provide an extensive review. We have analyzed studies from different countries and regions, using a wide variety of CGE models with two main objectives: (1) show the effects of an ETR in different contexts using different approaches of this modelling technique; and (2) investigate in which conditions a DD can be obtained.
This research is structured as follows: section 2 describes the theoretical debate around the DD hypothesis; section 3 contains a comprehensive review on the empirical literature on the DD hypothesis and CGE models; section 4 conducts a statistical and meta-regression analysis of the reviewed studies and section 6 concludes. ACCEPTED MANUSCRIPT
2. Theoretical debate around the double dividend hypothesis
While the theoretical basis of environmental taxes has been well documented (specially in early discussions from Baumol, 1972 and Baumol and Oates, 1971, 1988), this
3 section attempts to approach the main theoretical discussions around the DD question and CGE modelling.
2.1. The efficiency dividend
This has been a topic of growing importance during the last three decades. Tullock (1967) argued that a Pigouvian tax could create a DD by internalizing external costs, as Pigou (1932) did not elaborate specifically an environmental tax. The DD hypothesis was however first proposed by Pearce (1991). It states that it is possible to obtain double benefit by increasing environmental taxation and reducing other current taxes, while keeping the same government revenues (Pearce, 1991; Repetto et al., 1992; Nordhaus, 1993; Grubb, 1993).
The DD arises from: (1) an improvement of the environmental conditions, because of the incentives provided by the environmental tax; and (2) improvement of economic conditions, because of the shift from high distorting taxes to less distorting ones. Goulder (1995a) defined two types of DD: the strong DD occurs when welfare improves after implementing an environmental tax and recycling its revenues in reducing another distorting tax, regardless of the improvement in environmental conditions, so this policy involves zero or negative gross costs. The weak DD states that recycling environmental tax revenues through lowering distortionary taxes leads to cost savings relative to the case where revenues are returned via lump-sum transfers.2
To analyze an environmental tax t, we can simply set it according to (Baumol and Oates, 1988):
푡 = 푀퐷 (1)
where MD are the marginal social damages (in monetary terms) of pollution. This way, the ‘externality’ is ‘internalized’ and the tax is produces a welfare gain, differently from other taxes, that cause a welfare loss. However, Lee and Misiolek (1986) concluded that a pollution tax should be greater than MD. They argued that another component has to be added to the tax design:
ACCEPTED푡 = 푀퐷MANUSCRIPT+ 푅퐸 (2)
Where RE is the revenue effect (following the terminology of Parry, 1995). This is the improvement in welfare that can be gained by using the pollution tax revenue to fund a
2 In next sections we will refer to the strong DD as DD.
4 marginal reduction in other pre-existing taxes (in monetary terms).
However, economists have not agreed on the existence of a double dividend. Babiker et al. (2003) for instance, showed analytically that in an economy with multiple distortions, a weak double dividend need not occur. Fullerton and Metcalf (1997) also claimed that the validity of the DD hypothesis cannot be settled as a general matter. Under certain circumstances, a shift to environmental taxes may increase the burden of the tax system. They also demonstrated (in Fullerton and Metcalf, 1997, 2001) that the focus on the revenues is misplaced, by showing that three different types of policies (with different generation of revenues associated) have equivalent impacts on the environment and the labor supply. For them, the magnitude of the welfare gains (or the loss) due to new environmental policies in the face of pre-existing distortions depends critically on (1) whether the policy generates scarcity rents and (2) whether those rents are captured by the government and used to lower other distorting taxes.
Bohm (1997) provided an analysis of the theoretical literature on the DD hypothesis, concluding that there is always a positive ‘second dividend’ of environmental taxation if revenues are recycled in the most efficient (welfare-maximizing) way, and the outcome is compared to lump-sum redistribution or to an alternative command-and-control policy.
Later, Bento and Jacobsen (2007) extended traditional models by incorporating a fixed- factor in the production of the polluting good and, therefore, allowing Ricardian rents to be generated in the economy. They showed how this way, an environmental tax reform with revenues used to cut pre-existing labor taxes can generate a DD. Fernández et al. (2011) developed a stylized dynamic general equilibrium model with endogenous growth. They included a negative environmental externality in households’ preferences, in the form of CO2 global emissions generated by firms as a by-product of their production process using polluting inputs. Under this framework, they found a wide range of dynamically feasible green tax reforms that satisfy DD hypothesis without any need to assume a complex production structure or tax system, or a variety of externalities in production. As they claim, this results arises because of the consideration of public debt issuing and the management of the government budget balance with an intertemporal perspective.
The introduction of an environmental tax can (also) generate an employment dividend (sometimesACCEPTED known as the third dividend in MANUSCRIPTliterature). Employment can increase even without increasing real incomes or non-environmental welfare. It can be of priority interest in some economies with high unemployment rates, like the European ones. Bovenberg and Goulder (2002) described the circumstances that using of CGE frameworks lead to an employment dividend of environmental taxation. For them, the crucial requirement for an increase in employment is that the reforms shift the tax
5 burden from labor to other primary factors. Specifically, in models with capital and labor, an employment dividend can arise if: (1) Industries on which the environmental tax is levied feature a relatively low labor intensity in comparison with other industries; (2) Revenues from the tax are mainly used to reduce labor taxes (rather than taxes on capital). There are some studies that explore how a green tax reform affects equilibrium unemployment in models with endogenous wage-setting (Koskela and Schöb, 1999; Nielsen et al., 1995; Schneider, 1997; and Bovenberg and van der Ploeg, 1998). As Bovenberg and Goulder (2002) point out, in these models, the impact on employment depends mainly on how an environmental tax reform affects unemployment benefits. If unemployment benefits are a fixed proportion of income in employment, employees are who assume taxes (Layard et al., 1991). Koskela and Schöb (1999) showed that in a context of wage bargaining between unions and employers, the effects on employment of an ETR depend on the taxation of unemployment benefits. If they are neither subject to the income tax nor indexed to the consumer price index, employment may expand.
2.2. Criticisms on the double dividend existence
At the same time the DD hypothesis analyses were gaining importance, some authors argued that it is not possible for different reasons. The positive environmental effects, due to the incentives created by the tax were mostly accepted but the reduction of inefficiency from cutting other distortionary taxes were questioned by some economists.
Early criticisms came from Bovenberg and de Mooij (1994a). They claimed that environmental taxes exacerbate, rather than alleviate, preexisting tax distortions even if revenues are employed to cut preexisting distortionary taxes. The same authors, Bovenberg and de Mooij (1994b), used a static CGE and found that DD is only possible if the labor supply function is backward-bending in a static CGE model. Although empirical literature of labor supply behavior does not support this functional form, a static model is not adequate to deal with investment and capital issues. Bovenberg and Van der Ploeg (1994) introduced wage rigidities into in the labor market of the model, allowing for unemployment. They concluded that in a small open economy with involuntary unemployment a DD may result if: substitution between labor and resources is easy, if the production share of labor is large, and if initial tax rates on resources and profits are small.
ParryACCEPTED (1995) analyzed from a theoretical perspectiveMANUSCRIPT the use of an environmental tax to reduce taxes on labor. He identified not only the ‘marginal-damage reduction effect’ (MD) and the ‘revenue effect’ (RE), but also the ‘interdependency effect’, and pointed out that it is often ignored. For him, the efficiency and revenue raised from labor and capital markets is reduced to the extent that environmental taxes discourage employment and investment. McKitrick (1997) explained his point with an illustrative
6 example: “Consider an economy where the government's budget is entirely financed by a tax of x percent on nominal labor income. A pollution tax will increase the cost of production, thus raising the cost of consumer goods and reducing real wages. As real wages fall, the labor supply falls and the excess burden of the income tax will typically increase. This increase in the distortions in the labor market, brought about by a tax on pollution, is the so-called ‘interdependency effect’ (IE)”.
This effect generates a loss in welfare. A lower optimal tax level need to be set if accounting for this effect than the one suggested by Lee and Misiolek (1986) analysis. Fullerton (1997) obtained a similar conclusion, considering that previous analyses, like those in Lee and Misiolek (1986), implicitly assumed that other tax rates were zero. The environmental tax considering all the effects was now:
푡 = 푀퐷 + 푅퐸 − 퐼퐸 (3)
where IE is the interdependency effect. If RE>IE, then the pollution tax should be greater than marginal damage or externality, if RE In this line, Oates (1995) suggested that most measures that increase the costs of polluting industries have unevaluated major negative impacts, so the DD hypothesis is not possible. Bovenberg and Goulder (1997) did not find a DD using a CGE. They concluded that environmental taxes are implicit factor taxes that not only generate factor market distortions similar to those of income taxes, but also impose additional distortions of their own in other markets. These authors claimed that previous studies offered little scope for the DD because they did not consider pre-existing inefficiencies in the tax treatment of labor and capital. Despite the difficulties in obtaining a DD for these authors, they also recognized that if the initial tax system is suboptimal from a non-environmental point of view, a fiscal reform can reduce pollution and improve economic conditions (Bovenberg, 1994; Carraro and Soubeyran, 1996). Similarly, Fullerton et al. (2008) stated that the possibility of a ‘double dividend’ cannot be ruled in or out; it must depend on the circumstances. They analyzed different cases and consider these three possibilities: (1) if a good associated with uncontrolled pollutionACCEPTED is taxed at less than its optimal rate, MANUSCRIPT an increase in that rate may provide a DD; (2) if this good is taxed at exactly its optimal rate already, an increase in that rate may only provide the environmental dividend; (3) if the pollutant is already controlled in other ways like through command-and-control regulation, an increase in its tax rate may not even provide an environmental dividend; (4) if the good is already taxed at a higher rate than its optimal, an increase in the environmental tax may reduce welfare. For these 7 authors, the number of dividends, is not relevant in itself, but the net effect is positive or negative on overall welfare when tax and environmental policy reforms are integrated properly. The consideration of the interdependency effect (especially the negative one), that could lead to set optimal taxation under the marginal damage on the environment, made economists think about the conditions that would be necessary in an economy to achieve a DD. Bovenberg and Goulder (2002) established the conditions they consider necessary: (1) there should be pre-existing distortionary taxes on production factors and new taxation shifts the burden from a factor with large distortion and onto a factor with small distortion.; (2) primary factors have to be supplied inelastically and be relatively undertaxed; (3) capital needs to be relative immobile internationally; (4) the elasticity of substitution between energy and labor has to be greater than the elasticity of substitution between energy and capital; (5) real wages have to be inelastic to unemployment, so a reduction in labor taxes are not offset by wage increases. Moreover, in models where labor is the only primary factor, a DD can be achieved if: (1) improvements in environmental conditions have effects increasing the marginal value of work time relative to leisure; (2) improvements in environmental conditions increase labor productivity; (3) clean goods are better substitute for leisure than dirty goods (or, dirty goods are complements with leisure); (4) polluting activities are subsidized; (5) pre- existing taxes on goods and services are highly distortionary; (6) there is involuntary unemployment because of imperfection in labor markets. In a more recent work, Goulder (2013) also claimed that DD is not possible under relatively simple and ‘neutral’ conditions. This is because the environmental tax is narrower than the tax it replaces. The intuition he exposes is this: if one focuses on the non-environmental dividend, it would be hard to justify replacing a broad tax with a narrow one. For him, the narrower the tax, the greater the opportunities to adjust behavior and escape it. Such adjustments imply efficiency losses. The circumstances conducive to the double-dividend can be summarized for him in: (1) inefficient relative taxation of capital and labor; (2) inefficiently light taxation of resource rents; (3) an informal labor market and associated inefficiently low taxation of informal labor income; and (4) significant positive relationship between environmental quality and labor productivity. ACCEPTED2.3. Responses to the criticisms and newMANUSCRIPT developments McKitrick (1997) pointed out that most of the arguments used in criticisms (specifically referring to Parry’s work) derive from the use of simple partial or general equilibrium models which assume that the economy consists of a single firm with a linear production function and few outputs, no savings, no investment, no fixed capital, no 8 intermediate goods or international trade, only one or two pre-existing ad valorem taxes, and only one factor of production (labor). McKitrick (1997) also criticized Parry (1995) for the values of aggregate elasticities used in a simple graphical model that suggest that the optimal tax on polluting production should be only 63-78% of the MD, by considering that this result is sensitive to the substitution elasticity between the polluting good and labor supply (or, equivalently, leisure demand). He stated that even in a simple model, unambiguous predictions about the possibility of a double dividend are elusive. Schöb (1997) observed that the difference in the results between Bovenberg and de Mooij (1994a) and other arguments and researchers in favor of the existence of DD appears because, implicitly, different definitions of the second-best optimal pollution tax are chosen. Regarding the assumption considering only labor as a productive factor, van Heerden et al. (2006) noticed that (also in other works like Fullerton and Metcalf, 1997; Goulder et al., 1997) makes that the environmental tax is like an implicit tax on labor and then, generates a decrease in labor supply. The recycling of revenues through lower labor taxes only partly compensates for the lower labor supply because the environmental tax not only distorts the labor market, but also the commodity market by reducing the demand for the dirty good (however, this is actually the goal of an environmental tax). Bovenberg and de Mooij (1994a,b) for instance, assumed that the tax on the dirty good was initially at its Pigouvian level, so an extra distortion of the distribution of consumption over the different kind of goods must decrease total welfare. For van Heerden et al. (2006), unlike previous approaches, the consideration of more production factors (like capital or land) increases the potential for a DD, by shifting the tax burden from the over-taxed to the under-taxed factor. Also in response to the criticisms, other researchers have considered other factors and arguments that previous developments had not taken into account. Williams (2002), using a similar analytical model than the one used by Parry, demonstrated that beyond the marginal damage effect (MD), the revenue effect (RE) and the interdependency effect (IE) previously considered in literature, there is an additional interdependency effect that he called ‘benefit-side tax-interaction effect’ (IB) from the pollution reduction, that can yield a DD. To differentiate both interdependency effects, he renames the IE as ‘cost-side tax-interaction effect’. So the new equation defining the optimal environmental tax is: ACCEPTED푡 = 푀퐷 + 푅퐸MANUSCRIPT− 퐼퐸 + 퐼퐵 (4) He showed that when reducing pollution through an environmental tax boosts labor productivity, benefit-side interactions yield a welfare gain. However, when it boosts fixed-factor productivity or reduces medical expenses, these interactions yield a welfare 9 loss. Moreover, when lowering pollution reduces time lost to illness, benefit-side interactions may produce either a welfare gain or a loss, though typical parameter estimates suggest that it will be a loss. In a similar line, Gruver and Zeager (1994) had also considered the effects of environmental quality on productivity some years before. Other researchers, have challenged the size of the interaction effect by analyzing different conditions. This is the case of Bento and Jacobsen (2007), that did it by including a sector-specific factor in their model. They added two more effects that need to be taken into account when designing a tax: the ‘Ricardian rent effect’ (a loss in real income due to the reduction in Ricardian rents earned from the ownership of the specific factor), and the ‘surrogate tax effect’ (the representative consumer's income is reduced from the specific factor, which has a positive income effect on labor supply). For them, the presence of these additional effects due to the existence of a specific factor in the production of the dirty good weakens the (negative) tax interaction effect, as these effects increase the optimal pollution tax. Adding these factors to the optimal tax equation, we could set it as: 푡 = 푀퐷 + 푅퐸 − 퐼퐸 + 퐼퐵 + 푅푅 + 푆퐸 (5) where RR is the Ricardian rents effect and SE is the surrogate tax effect. From a quite different perspective, Fraser and Waschik (2013) paid special attention to the role of specific factors of production in affecting the size of DD. They considered that the circumstances leading to DD are: (1) the role of specific factors; (2) labor supply and the tax interaction effect; and (3) the choice of tax instrument used to recycle the environmental tax revenue and the carbon tax base. 3. Empirical literature on the double dividend This section focuses on the review of the empirical literature on the DD hypothesis with the use of CGE models. First, a more detailed description is done on early studies, as they are key to understand the most recent ones. Then a specific section summarizes the results and main conclusions of the most recent studies classified by geographical area. In total we have analyzed a total of 40 empirical studies from 1993 to 2016. We have consideredACCEPTED these are the most relevant studies MANUSCRIPT that explicitly test the DD hypothesis in literature using CGE models. 3.1. Origins and early studies 10 Although CGE models started in 1960 with Johansen’s model (Johansen, 1960), the explicit analysis of the DD hypothesis of environmental taxation started several years later with Jorgenson and Wilcoxen (1993) work. These authors found a strong DD by recycling the revenues from a carbon tax to reduce taxes on capital, using a dynamic model for the US with solid econometric foundations. They did not found a strong DD if the revenues were used to reduce labor taxes. Goulder (1995b) however, did not find a DD by using an intertemporal CGE model for the US economy, but found that the use of revenues to reduce other distortionary taxes (he analyzed the personal income tax) significantly reduce the overall policy cost. In his model he incorporated non-renewable resources supply dynamics (including the depletion of oil and gas stocks), transitions from conventional to synthetic fuels and the consideration of capital as imperfectly mobile across industries in relation to Jorgenson-Wilcoxen model. Böhringer and Pahlke (1997) also used a dynamic model and they neither found strong DD. Some years later, Bye (2000) also developed an intertemporal general equilibrium framework for Norwegian to test the non-environmental welfare costs of a carbon tax. She found the existence of a DD by using its revenues to reduce tax on labor income (payroll tax rate). One important difference between this study and Jorgenson and Wilcoxen (1993), in Bye’s words is the higher marginal excess burden of capital taxation compared to labor income taxation in the United States. In Norway, however, the marginal excess burden of labor taxation is especially high. Other factors that produce differences are: the small open economy framework that implies that the interest rate and prices on export are given in the world market and the assumption of internationally mobile capital that implies that net savings in financial capital may be uncorrelated with net investment in real capital. Capros et al. (1996) used a CGE model for 12 European countries (GEM-E3) obtaining a DD in employment, provided that the degree of labor supply flexibility is significant. Carraro et al. (1996) used an econometric CGE model and also found an employment DD for 12 European Union countries, but only in the short run. However, Barker (1999) used a CGE Keynesian model for Europe and found a large strong double dividend. Bernard and Vielle (1999) used a CGE model for France and also suggested evidence of a strong double dividend. Also in Europe, Kemfert and Welsch (2000) estimated econometrically elasticities of substitution for capital labor and energy for Germany, and then incorporate them in a dynamic CGE model to assess the effects of a carbon tax. When using the tax revenue to reduce labor costs (social security), they found a DD on employment and GDP. Bach et al. (2002) analyzed the effects of environmental tax reformACCEPTED also in Germany. In order to do it theyMANUSCRIPT used an econometric input-output and a CGE model. They found a DD with an employment increase using both approaches. Although the initial interest raised in the US and European countries, amongst these early studies on the DD hypothesis we can also find studies for other countries like McKitrick (1997), who found a DD for Canada using a CGE model, and recycling 11 revenues or environmental taxes by reducing labor taxes (payroll taxes). However, when revenues were recycled through lump-sum transfers, he did not find a DD. The first study of these characteristics in Asia, was done for China by Garbaccio et al. (1999). They found a DD using a dynamic CGE, by recycling revenues via reducing all other taxes proportionately. They differentiated between both plan and market institutions in the Chinese economy. Bosquet (2000) analyzed the empirical literature of effects of an environmental tax reform from 56 studies using a great variety of modeling techniques and assumptions (including partial equilibrium, CGE, macroeconomic, input–output, etc.). He concluded that when revenues from an environmental tax are recycled to reduce labor taxes, significant reductions in pollution, small gains in employment, and marginal gains or losses in production are likely in the short to medium term, while investments fall back and prices increase. Results however, are more uncertain in the long term. The first analysis for Australia was done by Wedner (2001). He found a DD by using a CGE model and recycling part of the carbon tax revenues for financing social security retirement benefits and the remaining portion employed to subsidize labor costs. Boyd and Ibarrarán (2002) investigated the potential for a DD in Mexico through a CGE model but find little support for its existence except under unusual circumstances. They found that only under significantly high rates of technological change in the Mexican economy (about 5 or 6%), a reduction in the rate of growth of carbon emissions and an increase in welfare can be attained for all income groups simultaneously. A different implementation of these models to test DD hypothesis was conducted by Böhringer et al. (2003). They analyzed from a joint implementation perspective of two countries (Germany and India) if an ETR provides employment and overall efficiency gains as compared to an environmental tax reform in one country. With the use of a large-scale computable general equilibrium (CGE) model for both countries, they found that joint implementation offsets the adverse effects of carbon emission constraints on the German economy providing employment gains. Labandeira et al. (2004) found a DD for a carbon tax using a CGE model for Spain, and recycling revenues through social security contributions. Also using a CGE model for Spain, Labandeira et al. (2009), estimated the effects of an increase of the consumption taxes of energy goods (electricity, refined oil products, natural gas and coal) by 20% and recycling them through a proportional reduction of indirect tax rates of all the remainingACCEPTED commodities. They found a DD, MANUSCRIPTwith a GDP increase by approximately one percent after the simulated reform, with a null impact on employment and prices for labor and capital in real terms. André et al. (2005) simulated the effects of an Environmental Tax Reform in the region of Andalusia (Spain). The reform assessed consisted in imposing a tax on CO2 and SO2 emissions while reducing at the same time either the income tax or the payroll tax of employers to Social Security, keeping public 12 deficit unchanged. They found a DD (improvement of environment and the economy) when the payroll tax was reduced and when CO2 emissions were selected. However, they did not found DD when income tax was reduced to compensate the environmental tax. Also for Spain, Manresa and Sancho (2005) developed a CGE model and found a DD by implementing a tax on energy goods and reducing payroll taxes. However, when they only considered a tax on a specific polluting energy use (i.e., a petrol tax) the distortions were greater than the previous policy, and they did not find DD. Conrad and Löschel (2005) pointed out the fact that recycling tax revenues from an environmental tax to lower employers’ non-wage labor cost depends on how the costs of labor are modeled. They performed two main simulations using a CGE model for Germany. In their simulations carbon taxes were levied to meet a 21% reduction of domestic CO2 emissions as compared to 1990 levels, following the reduction target committed by the German government. They found a DD found under the user cost price of labor scenario, but not under the market price of labor scenario. After reviewing some DD and CGE studies from the 90s decade, Bergman (2005) concluded that the existence of a strong DD can be taken for neither granted nor entirely ruled out. 3.2. Double dividend in recent empirical literature on CGE3 This section describes by geographical areas (continents) and chronologically, the main aspects and results of the most recent studies found in literature on the DD and CGE modelling. 3.2.1. Europe This is, along with Asia, the most analyzed continent in recent literature of CGE and the DD hypothesis. This may reflect the raising interest on the use of environmental taxation to reduce carbon emissions and other environmental impacts in this continent over the last years. Saveyn et al. (2011) used a world CGE model (GEM-E3) for the EU and found a DD using auctioning (tradable permits) for energy intensive sectors and taxation for the othersACCEPTED to generate additional government revenuesMANUSCRIPT and use them to reduce social security contributions of employees. They found that higher private consumption was the main driver in explaining the DD. Ciaschini et al. (2012) used a CGE model for a bi- 3 We consider recent literature that one starting in 2006 (approximately the second half of the period 1993-2016). 13 regional economy for Italy, for regions South-Islands and North-Centre. They assumed two scenarios. A first one where the environmental tax revenue is recycled to reduce the income tax, and a second where it is used to reduce regional tax on activities. They identified an employment DD only in the North–Centre region. Orlov and Grethe (2012) found a DD by using revenues of an environmental tax to reduce labor taxes by modelling the Russian economy with a CGE model. Unlike most of previous studies, they developed a CGE with imperfect competition, and suggested that this could lead to higher welfare costs of environmental taxation than in the case of perfect competition. They specifically modelled the case of a Cournot oligopoly with homogenous products and symmetric firms in the markets for natural gas, petroleum and chemical products, metals, and minerals. Orlov et al. (2013) found in a similar research that the maximum rate of CO2 emission reductions consistent with the double dividend hypothesis is approximately 28%. Pereira et al. (2016) analyzed the implementation of a carbon tax with a dynamic CGE model for Portugal. They simulated the tax assuming that it only affects the non-ETS (European Union's Emissions Trading System) sectors, but their model only considered one industry. The Commission for Environmental Tax Reform in Portugal proposed 50% of revenues to be are recycled on investment tax credits (ITC), 25% to be used to alleviate Social Security contributions (SSC) paid by employers, and 25% to go to finance lower personal-income taxes (PIT). Under this recycling scenario they found a strong DD (triple adding a reduction of public debt/GDP) by 2050. Finally, Allan et al. (2014) investigated the economic and environmental impact of a Scottish specific carbon tax under three scenarios about the use of the revenue raised by the tax: not recycled within Scotland; used to increase general government expenditure or to reduce Scottish income tax. They only found a DD in this last scenario. 3.2.2. Asia There has been a growing interest during the last years on environmental taxes in Asia, reflected in the number of studies found there compared to the first half of the period analyzed in this continent (no studies have been found). This is especially visible in China, where the concerns on pollution have been growing at the same time the economy has. ACCEPTED MANUSCRIPT Takeda (2007) developed a CGE model for Japan and found a DD by using carbon tax revenues in reducing capital taxes. However, DD did not arise if revenues were recycled to reduce labor and consumption taxes. He attributed this result because capital taxes are more distortionary than labor and consumption taxes. 14 Lu et al. (2010) assessed the impacts of a carbon tax on the Chinese economy using a dynamic recursive CGE model. They found that reducing indirect tax in the meantime of imposing carbon tax helped to reduce the negative impact of the tax on production and competitiveness. Instead of reducing tax rate they recycled revenues via lump-sum transfers to enterprises as a subsidy. They did not find a DD but found limited economic impacts and large carbon emissions reductions. Bor and Huang (2010) used an ORANI-recursive dynamic CGE model (Monash University) to assess the economic impacts of the energy tax in Taiwan, and found a DD by reducing both personal income tax and business income tax, however, the first one is stronger. Li and Lin (2013) used a CGE for China to test a carbon tax and an energy tax with revenue recycled to reduce VAT (indirect tax). In the simulations they conducted, revenue recycling was not helpful in mitigating welfare losses from both taxes, not achieving a DD. However, they used a static small model with only three industries (agriculture, industry and services) one representative consumer, who owns all revenues from factors of production, tax and goods, with no explicit representation of the public sector. In this sense, Cao et al. (2013) used a more elaborated dynamic CGE model for China taking onto account not only the market part of the economy but also the planned part, similarly to Garbaccio et al. (1999). They assessed the impacts of a carbon tax using the revenues to reduce all pre-existing distortionary taxes (capital income, value added and sales tax) and using them through a lump-sum transfer to households. They did not found a DD but economic impacts were very limited by this recycling scenario. From a regional perspective, Zhang et al. (2016) analyzed the impact of the carbon tax in three Chinese regions: Henan, Fujian and Chongqing, using a static CGE model. They did not find a DD by recycling the revenues through indirect tax on products. 3.2.3. America Most of recent studies found in America have been done for the US economy. The number of empirical studies in this continent have decreased compared to the first half of period analyzed. Glomm et al. (2008) examined the DD hypothesis using a dynamic CGE model for the USA economy. They introduced non-market valuation techniques into CGE in order to evaluateACCEPTED the environmental and market effects MANUSCRIPT of green tax policy reform in a unified framework. Given that a lower tax rate on capital encourages capital accumulation (as pointed out by Jorgenson and Wilcoxen, 1993), the new steady state levels of capital and consumption of the clean good are higher than their pre-reform levels and, as a result, the quality of the environment may worsen in the new steady state. For them, following this analysis, the environment dividend, or higher discounted utility from a 15 cleaner environment, is much smaller than the efficiency dividend, or higher discounted utility from the consumption of market goods, one of the reasons is because of the low willingness to pay for improvements in air quality found in households and used in this study. They did not find a DD by increasing gasoline taxes and using the generated revenues to reduce capital income taxes. Also for the US economy, Carbone et al. (2013) developed a dynamic CGE overlapping-generations model, combined with a disaggregated model of near-term distributional impacts and found a DD by recycling revenues through reductions in capital taxes. Using this model, recycling the revenues via reductions in payroll taxes or personal income taxes is slightly less economically efficient, and via lump-sum rebates is worse than the other options in terms of economic efficiency. Jorgenson et al. (2013) developed an updated version of their intertemporal CGE model for the US economy elaborated in Jorgenson and Wilcoxen (1993), using parameters and exogenous variables econometrically estimated. They assessed the impacts of a carbon tax under different recycling alternatives: lump sum distribution, reduction in capital tax rates, reduction in labor tax rates and reduction in all rates. They found that reductions in capital tax is ambiguously the best option for recycling carbon tax revenues, finding a DD for this option. On the other hand, they found that lump sum distributions are not the worst option. They only study found for South American countries is Chisari and Miller (2015). They developed a small recursive dynamic CGE model to analyze the impact of carbon taxes under alternative close rules and hypothesis for several countries: Argentina, Brazil, Chile, El Salvador, Jamaica and Peru. They assumed a first case where the additional revenue, obtained from carbon taxes is compensated with a reduction of labor taxes. In this case they found benefits (DD) in all cases except for Brazil, which had very significant reduction in costs by recycling revenues. They also found changes in welfare distribution. 3.2.4. Africa Only two studies that explicitly tests the DD hypothesis in CGE models have been found in recent literature for South Africa. There is clearly a lack of studies for other African countries. More evidence is needed on how economic structures in developing countries respond to environmental taxes and recycling options. Van Heerden et al. (2006)ACCEPTED used a CGE model for South Africa MANUSCRIPTto explore the DD hypothesis. They analyzed the impact of different taxes: a CO2 tax, a fuel tax, an electricity tax, an energy tax (this one is a combination of the fuel tax and the electricity tax) and a water tax. Three recycling schemes were analyzed: (1) a decrease in direct taxes; (2) a general decrease in indirect taxes; and (3) a decrease in taxes on food. The simulations showed that for all the taxes and recycling options, there was a fall in CO2 emissions and water 16 use. Two recycling schemes, direct tax and indirect tax, did not yield a second dividend for the three first taxes, only the food tax gave the second dividend for them. However, for the water tax, a DD was found for all recycling options. Devarajan et al. (2011) also developed a CGE for the South Africa economy to assess the potential of a DD of a CO2 tax. They did it by recycling revenues in reducing all indirect taxes (production taxes, sales taxes, and import tariffs). As they pointed out, factor taxes are generally less important than indirect taxes as revenue sources in developing countries, and actually, South Africa does not impose factor taxes. They found that partial reforms of labor market distortions can lead to a strong DD from a tax on carbon or a sales tax on energy inputs. 3.2.5. Oceania We find two studies for this continent in recent literature on this matter. Fraser and Waschik (2013) used a CGE model for Australia to analyze the DD hypothesis. They examined three different environmental taxes on the production of energy goods: production of energy goods; production of carbon; and usage of carbon. They found a strong DD when revenue was recycled through reductions in consumption taxes. A novel approach was used in this research to modelling specific factors, finding that DD was stronger when the share of specific factors was higher and when carbon taxes where charged on the production rather than the usage of carbon. Sajeewani et al. (2015) analyzed the potential effects of the carbon tax in Australia using a CGE model. Results indicated that revenue recycling through income tax reductions improved post tax income levels and welfare. Although GDP was reduced under this scenario these authors consider that there is DD as households have more capacity to alter their decisions toward the consumption patterns that have initially been affected by the carbon price policy. 4. Statistical and meta-regression analysis This section contains a statistical analysis of studies, simulations and their results, as well as a description of the meta-regression analysis conducted in order to obtain a more detailed overview of the literature. As shown in the previous section, there is a high heterogeneity that makes them difficult to compare, as there are different countries, a variety of taxes and recycling instruments, modelling schemes, assumptions, etc. The heterogeneityACCEPTED problem along with the lack ofMANUSCRIPT enough empirical evidence make difficult to set comparisons and general conclusions. However, we present in this section some statistics and a meta-regression analysis that provide a global perspective on the state- of-the-art and helps to find the gaps and elements to improve in further research. Table 1 summarizes the main results of the analyzed studies. 17 4.1.Statistical analysis We analyzed a total of 69 simulations included in 40 studies on the DD hypothesis of environmental taxation using CGE methods from 1993 to 2016. Studies analyzed are summarized in table 1. Fig. 1 shows that 55.1% of simulations achieved a DD by recycling revenues from environmental taxes to reduce other taxes or via lump-sum transfers. This represents 38 simulations. On the other side, 44.9% of simulations analyzed (a total of 31) assured that the DD was not achieved. However, if we analyze these results by the recycling instrument used in simulations, we find a variety of results, as shown in Fig. 2. We can see that most of studies that recycled revenues to reduce social security contributions (labor costs) achieved a higher percentage of DD (9 simulations out of 10). On the other side, simulations that recycled revenues towards lump-sum transfers are the less successful in achieving a DD (only 1 out of 10). Simulations that recycled via labor incomes, capital taxes and other taxes achieved mostly a DD. In the analyzed sample 61.8% (24 simulations) used dynamic CGE models while 38.2% (15 simulations) used static CGE models to obtain their results. Regarding the tax implemented, 74.4% of these studies have simulated the effects of a carbon tax (29 out of 39) and 18% have assessed energy taxes (7 out of 39). If we focus on the countries object of the simulations, most of them have been done on Europe (20) and North America (18). Fig. 3 shows the geographical distribution of the sample. We can see how most of simulations conducted in Europe have achieved a DD, while most of them conducted in North America and Asia have not achieved it. Africa, South America and Oceania simulations also lead to confirm the DD hypothesis. Finally, table 2 shows a summary of some quantitative data collected from the studies. Given that not all simulations have provided the same results in terms of macroeconomic variables, this results need to be carefully considered before deriving general conclusions, as in some cases the sample is small. For this reason, we have added an average value with its standard deviation and a column that shows the number ofACCEPTED simulations that include each figure. MANUSCRIPT We can see that on average, simulations lead to a reduction of carbon emissions by 11.76%. Those studies that showed GDP variation show in average an increase of 0.11% and those who focused on a welfare measure were in average unaffected by taxation. It is interesting to see that employment, consumption, exports and prices tend 18 to increase, while investment and imports tend to decrease. 4.2. Meta-regression analysis To synthesize the effect on different variables in results obtained in different simulations we use meta-regression analysis method proposed by Stanley and Jarrell (1989): 퐾 푏푗 = 훽 + ∑푘=1 훼푘푍푗푘 + 푒푗 (6) where 푏푗 is the effect size of the study j, 훽 is the independent term, Z is the set of independent variables that measure relevant characteristics of the empirical study that influenced the estimated effects, 훼푘 are the meta-regression coefficients and 푒푗 is the error term. Basically, the idea behind meta-regression analysis is to explore and identify factors that drive results obtained in literature. Analyzing the quantitative data obtained from simulations, we observe that it is difficult to obtain general conclusions because of the lack of sufficient studies and the enormous heterogeneity we find in them in CGE specifications, data used, underlying assumptions of models, economic structures, regions and even subjective considerations on when a DD is achieved in each study. In other words, all studies are too different and we have few of them. Given this heterogeneity and the lack of studies, there is no possible to conduct a sound regression analysis that controls for all factors that would affect the achievement of a DD. Equation (6) is a general specification that can include several characteristics of studies and simulations. From collected data in simulations we have conducted some estimates using Ordinary Least Squares method, with different specifications, and no one of them have provided significant results. This is normal given the characteristics of the sample of studies described above. Seven different dependent variables have been analyzed: a dummy variable that includes all simulations that authors claim reach a DD; different measures of welfare variation; GDP variation; consumer price index variation; investmentACCEPTED variation, Imports variation and MANUSCRIPT exports variation. Regarding the independent variables: CO2 emissions variation; a dummy variable for studies that recycle revenues to income taxes and a dummy variable that recycles via lump-sum transfers. Table 3 summarizes nine different models estimated. Other options have been tested but for most of them there is not enough data to conduct estimates. Although estimates are not enough robust to state general conclusions, we observe that except for the model (1), in general, a reduction in CO2 emissions is followed by a 19 (low) reduction in welfare/GDP, however, it is close to zero. This would suppose that researchers follow other criteria than only considering effects on GDP or on welfare when considering simulations achieve DD. Lump-sum recycling would be worse than income tax, in GDP terms (models 4 and 5). On the other side, models (6), (7), (8) and (9) show that investment and imports tend to decrease when a reduction in emissions is tested in CGE models, and that the consumer index price and exports tend to increase, especially the latter. Figure 4 shows the relationship between the variation in GDP and the variation in CO2 emissions obtained from some simulations that showed these variables. We can see how there is no a clear relationship between them. Gago et al. (2013) used a similar methodology with other variables and considering a wider set of methods, not only CGE modelling. They found that the effects of recycling tax revenues in different ways, on average were positive on GDP, employment and consumer prices (CPI), and negative on energy demand, energy prices and welfare. Patuelli et al. (2005) used a similar methodology but also including different methods, finding that CGE models tended to produce positive results regarding employment and GDP in relation to other methods. 5. Conclusions The comprehensive review of literature done in this research shows that there is still long way to be done regarding the use of CGE modelling to test the DD hypothesis of an ETR. However, some interesting insights, lessons and suggestions for further research arise from it. Main criticisms of the DD hypothesis focus on the non-environmental dividend, as there is a high consensus amongst economists that environmental taxes reduce environmental impacts. So focusing on the economic dividend, one of the main research lines against the existence of the DD –coming from Goulder, Bovenberg, Parry, etc.– argues, in simple terms, that replacing a broad tax with a narrow one limit the opportunities to adjust behavior and escape from it. However, it is important to point out that even several criticisms on the hypothesis recognize that the environmental benefits of these taxesACCEPTED might exceed the non-environmental costs,MANUSCRIPT and so, the use of environmental taxes with recycled revenues would be better than the initial situation Goulder (2013). Moreover, most of these criticisms could be solved by reconsidering the object and the design of the tax system. Regarding the object, most of the applied and analytical research is focused on carbon or energy taxes. This implies a special focus of taxation over specific economic agents. However, there are a lot of different environmental 20 impacts and problems with resources depletion that can be dealt with a comprehensive environmental tax reform, and at the same time improve the non-environmental side of the DD. Two main conclusions derive from this review: (1) an ETR improves the environment: almost all studies recognize that an environmental tax improves environmental conditions, and this is the main objective of imposing such a tax; and (2) the occurrence of the second dividend (improvement of economic efficiency) is still ambiguous: we have seen that most of simulations analyzed have achieved a DD, but only by a narrow margin (55% versus 45%). Despite this ambiguity, it is quite accepted than under certain conditions, some related to the design of the ETR, the DD is possible. However, as noted in several parts of this paper, even amongst researchers against the existence of the DD, as Glomm et al. (2008) noted: there is a general consensus that can be summarized with this sentence in de Mooij (2000): “Whereas the second dividend may be in doubt, the first dividend (i.e. a cleaner environment) remains a powerful reason for the introduction of pollution taxes.” There are several factors that makes that evidence is not conclusive. They can be summarized in: still few studies and heterogeneity of models, assumptions, data, scenarios, etc. Moreover, beyond these general factors, deeply analyzed in this paper, the achievement of a DD also depends on many other factors like the tax structure considered, the individual and collective preferences, the factor mobility, the factor substitution, etc. It is more likely to achieve a DD by recycling revenues on reducing social security contributions, capital taxes and income taxes. On the other side, recycling of revenues via lump-sum transfers is the worst option. Comparing the two continents with more evidence, European economies are more likely to achieve a DD than American ones. From literature we can also conclude that a correct design and implementation of an ETR would reduce environmental impacts and improve economic conditions, including employment. This design need to be based on the economic structure and the pre- existing taxation system to lead to a DD. So the question is not whether a DD is possible, but what design of an ETR we need for that specific economic system to yield a DD. There are a lot of potential possibilities in designing a tax system and recycling revenues and most of the empirical literature focuses on few limited scenarios. ACCEPTED MANUSCRIPT Beyond all these issues there is an underlying question on the effects of environmental taxes and dynamics that need further research. Environmental taxes are expected to reduce pollution or resources use, so the tax base is expected to reduce in the future, However, this effect, at the same time would reduce revenues and then, recycling possibilities and the potential DD. To avoid this effect, environmental taxes need to be 21 updated and set every few years and/or shift to other pollutants or resources to keep the same revenues. This and other aspects highlighted in this research is what further research will have to address the next years. Acknowledgements This project has received funding from the European Union’s Horizon 2020 research and innovation programme under the Marie Sklodowska-Curie grant agreement No 654189. ACCEPTED MANUSCRIPT 22 References Allan, G., Lecca, P., McGregor, P., & Swales, K. (2014). The economic and environmental impact of a carbon tax for Scotland: a computable general equilibrium analysis. Ecological Economics 100, 40–50. André, F.J., Cardenete, M.A., & Velázquez, E. (2005). Performing an environmental tax reform in a regional economy. A computable general equilibrium approach. The annals of Regional science, 39(2), 375–392. Babiker, M.H., Metcalf, G.E., & Reilly, J. (2003). Tax distortions and global climate policy. Journal of Environmental Economics and Management 46(2), 269–287. Barker, T. (1999). Achieving a 10% cut in Europe's carbon dioxide emissions using additional excise duties: coordinated, uncoordinated and unilateral action using the econometric model E3ME. Economic Systems Research 11(4), 401–422. Baumol, W. (1972). On taxation and the control of externalities. American Economic Review 62(3), 307–321. Baumol, W., & Oates, W. (1971). The use of standards and prices for the protection of the environment. Swedish Journal of Economics 73, 42–54. Baumol, W., & Oates, W. (1988). The Theory of Environmental Policy, Cambridge, University Press, Cambridge. Bento, A. M., & Jacobsen, M. (2007). Ricardian rents, environmental policy and the ‘double-dividend’ hypothesis. Journal of Environmental Economics and management 53(1), 17–31. Bergman, L. (2005). CGE modeling of environmental policy and resource management. Handbook of environmental economics 3, 1273–1306. Bernard, A.L., & Vielle, M. (1999). Report to the MIES on the Greenhouse Effect: An Evaluation of the Kyoto Protocol with the GEMINI-E3 Model. Ministere de L'Equipment des Transports et du Logement, Commissariat a L'Energie Atomique, Paris. Bohm, P. (1997). Environmental taxation and the double dividend: fact or fallacy, in: Ecotaxation, ed. T. O’Riordan, London: Earthscan. Böhringer, C., & Pahlke, A. (1997). Environmental tax reforms and the prospects for a double dividend: An intertemporal general equilibrium analysis for Germany. Working Paper,ACCEPTED Boulder: University of Colorado. MANUSCRIPT Böhringer, C., Conrad, K., & Löschel, A. (2003). Carbon taxes and joint implementation. an applied general equilibrium analysis for germany and india. Environmental and Resource Economics 24(1), 49–76. Bor, Y.J., & Huang, Y. (2010). Energy taxation and the double dividend effect in 23 Taiwan's energy conservation policy—an empirical study using a computable general equilibrium model. Energy policy 38(5), 2086–2100. Bovenberg, A.L. (1994). Environmental policy, distortionary labor taxation, and employment: Pollution taxes and the double dividend. In: New directions in the environment (Carraro, C., Siniscalco. D., Eds.). Cambridge University Press. Cambridge. Bovenberg, A.L., & de Mooij, R.A. (1994a). Environmental policy in a small open economy with distortionary taxes. In: International Environmental Economics (Ierland, Ekko C. Ban., Ed.). Elsevier, New York. Bovenberg, A.L., & de Mooij, R.A. (1994b). Environmental levies and distortionary taxation and labor market distortions. European Journal of Political Economy 10 (4), 655–683. Bovenberg, A.L., & Goulder, L.H. (1997). “Costs of environmentally motivated taxes in the presence of other taxes: General equilibrium analysis”. National Tax Journal 50, 59–88. Bovenberg, A.L., & Goulder, L.H. (2002). Environmental taxation and regulation. In: Handbook of Public Economics (Auerbach, A., Feldstein, M., Eds.). Elsevier, Amsterdam, North Holland, pp. 1471–1545. Bovenberg, A.L., & van der Ploeg, F. (1994). Consequences of environmental tax reform for involuntary unemployment and welfare. Tilburg University Center Discussion Paper 9408. Bovenberg, A.L., & van der Ploeg, F. (1998). Tax reform, structural unemployment and the environment. Scandinavian Journal of Economics 100(3), 593–610. Boyd, R., & Ibarrarán, M.E. (2002). Costs of compliance with the Kyoto Protocol: a developing country perspective. Energy Economics 24(1), 21–39. Bye, B. (2000). Environmental tax reform and producer foresight: An intertemporal computable general equilibrium analysis. Journal of Policy Modeling 22 (6), 719–752. Cao, J., Ho, S., & Jorgenson, D.W. (2013). The economics of environmental policies in China. In: Clearer skies over China (Nielsen, C.P., Ho, M.S., Eds.). Cambridge. MA: MIT press. Carbone, J., Morgenstern, R.D., Williams III, & R.C., Burtraw, D. (2013). Deficit reduction and carbon taxes: Budgetary, economic, and distributional impacts. ResourcesACCEPTED for the Future. MANUSCRIPT Carraro, C., Galeotti, M., & Gallo, M. (1996). Environmental taxation and unemployment: some evidence on the ‘double dividend hypothesis’ in Europe. Journal of Public Economics 62(1), 141–181. Carraro, C., & Soubeyran, A. (1996). Environmental taxation and employment in a 24 multi-sector general equilibrium model. In: Environmental fiscal reform and unemployment (Carraro, C., Siniscalco, D., Eds.). Kluwer. Dordrecht. Ciaschini, M., Pretaroli, R., Severini, F., & Socci, C. (2012). Regional double dividend from environmental tax reform: An application for the Italian economy. Research in Economics 66 (3), 273–283. Chisari, O.O., & Miller, S.J. (2015). CGE Modeling: The Relevance of Alternative Structural Specifications for the Evaluation of Carbon Taxes' Impact and for the Integrated Assessment of Climate Change Effects: Simulations for Economies of Latin America and the Caribbean. Inter-American Development Bank. Conrad, K., & Löschel, A. (2005). Recycling of eco-taxes, labor market Effects and the true cost of labor-A CGE analysis. Journal of Applied Economics 8 (2), 259–278. Devarajan, S., Go, D.S., Robinson, S., & Thierfelder, K. (2011). Tax policy to reduce carbon emissions in a distorted economy: Illustrations from a South Africa CGE model. The BE Journal of Economic Analysis & Policy 11(1). Dissou, Y., & Sun, Q (2013). GHG mitigation policies and employment: A CGE analysis with wage rigidity and application to Canada. Canadian Public Policy, 39 (Supplement 2), S53-S65. Dixon, P.B., & Jorgenson, D.W. (2012). Handbook of Computable General Equilibrium Modeling. Newnes. Fernández, E., Pérez, R., & Ruiz, J. (2011). Optimal green tax reforms yielding double dividend. Energy policy, 39(7), 4253–4263. Fraser, I., & Waschik, R. (2013). The double dividend hypothesis in a CGE model: Specific factors and the carbon base. Energy Economics 39, 283–295. Fullerton, D., Leicester, A., & Smith, S. (2008). Environmental taxes (No. w14197). National Bureau of Economic Research. Fullerton, D., & Metcalf, G.E. (1997). Environmental taxes and the double-dividend hypothesis: Did you really expect something for nothing? (No. w6199). National bureau of economic research. Fullerton, D., & Metcalf, G.E. (2001). Environmental controls, scarcity rents, and pre- existing distortions. Journal of public economics 80(2), 249–267. Gago, A., Labandeira, X., & López-Otero, X. (2013). A Panorama on Energy Taxes and GreenACCEPTED Tax Reforms. Economics for Energy MANUSCRIPT. WP 08/2013. Garbaccio, R.F., Ho, M.S., & Jorgenson, D.W. (1999). Controlling carbon emissions in China. Environment and development economics 4(04), 493–518. Glomm, G., Kawaguchi, D., & Sepulveda, F. (2008). Green taxes and double dividends in a dynamic economy. Journal of policy modeling 30(1), 19–32. Gómez-Plana, A., Kverndokk, S., & Faehn T. (2003). Can carbon taxation reduce 25 Spanish unemployment? Second World Congress of Environmental and Resource Economists. Monterey, California. Goulder, L.H. (1995a). “Environmental taxation and the ‘double dividend’: A reader’s guide”. International Tax and Public Finance 2 (2), 157–183. Goulder, L.H. (1995b). “Effects of carbon taxes in an economy with prior tax distortions: An intertemporal general equilibrium analysis”. Journal of Environmental Economics and Management 29 (Part 1) (3), 271–297. Goulder, L.H. (2013). Climate change policy's interactions with the tax system. Energy Economics, 40, S3-S11. Goulder, L.H., Parry, I.W.H., & Burtraw, D. (1997). Revenue-raising versus other approaches to environmental protection: The critical significance of pre-existing tax distortions. The RAND Journal of Economics 28(4), 708-731. Grubb, M., Edmonds, J., ten Brink P., & Morrison M. (1993). The cost of limiting fossil-fuel CO2 emissions. Annual Review of Energy and Environment 18, 397–478. Gruver, G., & Zeager, L. (1994). Distributional Implications of Taxing Pollution Emissions: A Stylized CGE Analysis. Paper presented at the Fifth International CGE Modeling Conference, October 27–29, University of Waterloo, Ontario, Canada. Holmlund, B., & Kolm, A.S. (2000). Environmental tax reform in a small open economy with structural unemployment. International Tax and Public Finance 7(3), 315–333. Isard, W., Bassett, K., Choguill, C., Furtado, J., Izumita, R., Kissin, J., Romanoff, E.; Seyfarth, R., & Tatlock, R. (1968). On the linkage of socio‐economic and ecologic systems. Papers in Regional Science 21(1), 79–99. Johansen, L. (1960). A Multisectoral Study of Economic Growth. Contributions to Economic Analysis 21. North-Holland, Amsterdam. Jorgenson, D.W., & Wilcoxen, P.J. (1993). Reducing U.S. carbon emissions: An econometric general equilibrium assessment. Resource and Energy Economics 14, 243– 268. Jorgenson, D.W., Goettle, R.J., Ho, M.S., & Wilcoxen, P.J. (2013). Double dividend: environmental taxes and fiscal reform in the United States. Cambridge, MA: MIT press. Kemfert, C., & Welsch, H. (2000). Energy-capital-labor substitution and the economic effectsACCEPTED of CO 2 abatement: evidence for Germany. MANUSCRIPT Journal of Policy Modeling 22(6), 641–660. Koskela, E., & Schöb, R. (1999). Alleviating unemployment: the case for green tax reforms. European Economic Review 43, 1723–1746. Labandeira, X., Labeaga, J.M., & Rodríguez, M. (2003). A new micro-macro approach to analyse tax changes: the effects of green tax reforms in Spain. Universidade de Vigo. 26 Workshop on Climate Policy after 2012, Ghent University. Labandeira, X., Labeaga, J.M., & Rodríguez, M. (2004). Green tax reforms in Spain. European Environment 14(5), 290–299. Labandeira, X., Labeaga, J.M., & Rodríguez, M. (2009). An integrated economic and distributional analysis of energy policies. Energy Policy 37(12), 5776–5786. Layard, R., Nickell, S., & Jackman, R. (1991). Unemployment. Macroeconomic Performance and the Labor Market (Oxford University Press, Oxford). Lee, D.R., & Misiolek, W.S. (1986). Substituting pollution taxation for general taxation: Some implications for efficiency in pollutions taxation. Journal of Environmental Economics and Management 13(4), 338–347. Leontief, W. (1941). The structure of American Economy, 1919-1939. Nova York: Oxford University Press. Li, A., Lin, B. (2013). Comparing climate policies to reduce carbon emissions in China. Energy Policy 60, 667–674. Manresa, A., & Sancho, F. (2005). Implementing a double dividend: Recycling ecotaxes toward lower labor taxes. Energy Policy 33, 1577–1585. McKitrick, R. (1997). Double dividend environmental taxation and Canadian carbon emissions control. Canadian Public Policy/Analyse de Politiques, 417–434. Meng, S., Siriwardana, M., & McNeill, J. (2013). The environmental and economic impact of the carbon tax in Australia. Environmental and Resource Economics 54(3), 313–332. Mooij, R.A. (1999). The double dividend of an environmental tax reform. In: Handbook of environmental and resource economics (van der Bergh J.C.J.M., Ed.). Edward Elgar. Nielsen, S.B., Pedersen, L.H., & Sørensen, P.B. (1995). Environmental policy, pollution, unemployment and endogenous growth. International Tax and Public Finance 2, 185–205. Nordhaus, W. (1993). Optimal greenhouse gas reductions and tax policy in the ’DICE’ model. American Economic Review 83: 313–317. Oates, W.E. (1995). Green taxes: can we protect the environment and improve the tax system at the same time? Southern Economic Journal 61 (4), 915–922. Orlov, A., & Grethe, H. (2012). Carbon taxation and market structure: A CGE analysis forACCEPTED Russia. Energy Policy 51, 696–707. MANUSCRIPT Orlov, A., Grethe, H., & McDonald, S. (2013). Carbon taxation in Russia: prospects for a double dividend and improved energy efficiency. Energy Economics 37, 128–140. Parry, I.W.H. (1995). Pollution taxes and revenue recycling. Journal of Environmental Economics and Management 29(3), s64–s77. 27 Patuelli, R., Nijkamp, P., & Pels, E. (2005). Environmental tax reform and the double dividend: A meta-analytical performance assessment. Ecological Economics 55(4), 564–583. Pearce D (1991) The role of carbon taxes in adjusting to global warming. Economic Journal 101: 938–948. Pereira, A.M., Pereira, R.M., & Rodrigues, P.G. (2016). A new carbon tax in Portugal: A missed opportunity to achieve the triple dividend? Energy Policy 93, 110–118. Pigou, A.C. (1920). The Economics of Welfare. Macmillan and Co., Ltd, London. Repetto, R., Dower, R., Jenkins, R., & Geoghegan, J. (1992). Green fees: How a tax shift can work for the environment and the economy. World Resource Institute, New York. Sajeewani, D., Siriwardana, M., & Mcneill, J. (2015). Household distributional and revenue recycling effects of the carbon price in Australia. Climate Change Economics 6(03), 1550012. Saveyn, B., Van Regemorter, D., & Ciscar, J. C. (2011). Economic analysis of the climate pledges of the Copenhagen Accord for the EU and other major countries. Energy Economics 33, S34–S40. Schneider, K. (1997). Involuntary unemployment and environmental policy: the double dividend hypothesis. Scandinavian Journal of Economics 99, 45-59. Schöb, R. (1997). Environmental taxes and pre-existing distortions: The normalization trap. International tax and public finance, 4(2), 167-176. Shackleton, R., Shelby, M., Cristofaro, A., Brinner, R., Yanchar, J., Goulder, L.H., Jorgenson, D.W., Wilcoxen, P.J., & Pauly, P. (1996). The efficiency value of carbon tax revenues. In: Reducing Global Carbon Dioxide Emissions: Costs and Policy Options (Gaskins D., Weyant, J., Eds.). Energy Modeling Forum, Stanford, CA). Stanley, T., & Jarrell, S. (1989). Meta-regression analysis: a quantitative method of literature surveys. Journal of Economic Surveys 3, 54-67. Tullock, G. (1967). Excess benefit. Water Resources Research 3, 643–644. Van Heerden, J.H., Blignaut, J.N., Mabugu, M., Gerlagh, R., Hess, S., Tol, R.S., Horridge, M., Mabugu, R., de Wit, M., & Letsoalo, T. (2006). Redistributing environmental tax revenue to reduce poverty in South Africa: the cases of energy and water.ACCEPTED South African Journal of Economic andMANUSCRIPT Management Sciences, 9(4), 537–553. Wendner, R. (2001). An applied dynamic general equilibrium model of environmental tax reforms and pension policy. Journal of Policy Modeling 23(1), 25–50. Williams, R.C. (2002). Environmental tax interactions when pollution affects health or productivity. Journal of Environmental Economics and Management 44(2), 261–270. Yahoo, M., & Othman, J. (2015). Carbon and energy taxation for CO2 mitigation: a 28 CGE model of the Malaysia. Environment, Development and Sustainability 1–24. Zhang, X., Guo, Z., Zheng, Y., Zhu, J., & Yang, J. (2016). A CGE Analysis of the Impacts of a Carbon Tax on Provincial Economy in China. Emerging Markets Finance and Trade 52(6), 1372–1384. ACCEPTED MANUSCRIPT 29 44.9% 55.1% Yes DD No DD Fig. 1. Percentage of simulations that obtained a strong DD. 100% 10.0% 90% 36.4% 80% 46.4% 40.0% 70% 60% 90,0% 50% 90.0% 40% 63.6% 30% 53.6% 60.0% 20% 10% 10.0% 0% HH income SS Capital Other Lump-sum Yes DD No DD Fig. 2. Percentage of simulations that obtained a strong DD by recycling instrument. Note: HH income: labor incomes; SS: social security contributions; Capital: capital taxes; Other:ACCEPTED other taxes; Lump-sum: lump-sum transfers. MANUSCRIPT 30 16 14 12 10 8 6 4 2 0 Europe North Asia Africa South Oceania America America Yes DD No DD Fig. 3. Number of simulations by distribution area of the sample. 6% 5% 4% 3% P D 2% G n o i 1% t a i r 0% a v -1% % -2% -3% -4% -60% -50% -40% -30% -20% -10% 0% 10% 20% % variation CO2 emissions Fig.ACCEPTED 4. Relationship between GDP and CO2 emissions MANUSCRIPT in simulations analyzed. 31 Table 1. Summary of studies on the double dividend hypothesis using CGE models Strong double dividend? Region Labor tax Lu Static/ Study Tax Capit Oth Other features / comments (Country) mp- Dynamic Inco Soc. al tax er me Seg. sum Jorgenson and - Capital mobility across industries. USA CO No - Yes - - Dynamic Wilcoxen (1993) 2 - Econometrics to estimate parameters. - No capital mobility across industries. Goulder (1995b) USA CO No No No - - Dynamic 2 12 European Capros et al. (1996) Countries (EU- Energy - Yes - - - Dynamic - 12) 12 European - Econometric CGE model. Carraro et al. (1996) Countries (EU- Energy - Yes - - - Dynamic - Employment DD only in the short run. 12) Shackleton et al. USA CO - - Yes - - Dynamic - Jorgenson-Wilcoxen model. (1996) 2 Bovenberg and Dynamic USA Gasoline No - - - - - No capital mobility across sectors. Goulder (1997) - Reduction of CO2 emissions by 12.5% off a year McKitrick (1997) Canada CO Yes - - No - Static 2 2000 base case (approx. Rio targets). - They contemplate market and planned economy. Garbaccio et al. China CO - - - - Yes Dynamic - Same reduction in tax rates of some distortionary (1999) 2 taxes (capital income, value added and sales tax). Bye (2000) Norwegian CO2 Yes - - - - Dynamic - Recycling by reducing payroll tax rate. Kemfert and Welsch - Substitutions elasticities between capital, labor and Germany CO - Yes - No - Dynamic (2000) 2 energy econometrically estimated. - Part of the revenues used for financing social Wedner (2001) Australia CO2 - Yes - - - Dynamic security retirement benefits and the remaining portion employed to subsidize labor costs. Fuel oil, gasoline, diesel oil, Bach et al. (2002) Germany - Yes - - - Dynamic - Distributive effects. electricity and natural gas Böhringer et al. Germany and - Joint implementation of ETR in both countries. CO - Yes - - - Static (2003) India 2 - Focused in employment gains. - Tax rates on capital and on labor are both reduced Boyd and Ibarrarán Mexico CO No - No - - Dynamic in the same proportion. (2002) 2 Labandeira et al. - Use of a CGE model with a microeconomic Spain CO - - Yes - - Static (2004) 2 household energy demand model. Andalusia André et al. (2005) CO and SO No Yes - - - Static (Spain) 2 2 - DD for the user cost price of labor scenario but not Conrad and Löschel Yes/ for the market price of labor scenario. Germany CO - - - - Static (2005) 2 No - Carbon taxes levied to meet a 21% reduction of domestic CO2 emissions as compared to 1990 levels. Manresa and Sancho - A 10% energy tax increases utility and employment Spain Energy goods - Yes - - - Static (2005) and a 2.8% reduction in CO2 emissions. - Based on the structure of the ORANI-G. - They also obtained a ‘poverty divided’. ACCEPTEDCO2, fuel, MANUSCRIPTYes Van Heerden et al. No / No / - DD found for CO , fuel, electricity, energy by South Africa electricity, energy - - / Static 2 (2006) Yes Yes recycling revenues on food taxes. / water use Yes - DD found for water tax by recycling revenues on direct taxes, indirect taxes and food taxes. - Recycling by reducing capital, labor and Takeda (2007) Japan CO2 No - Yes - No Dynamic consumption taxes. - Most elasticity values taken from GTAP. 32 Glomm et al. (2008) USA Gasoline - - No - - Dynamic - Non-market valuation into the model. Electricity, Labandeira et al. refined oil Spain - - - - Yes Static - Recycling by reducing VAT rates of other goods. (2009) products, natural gas and coal Bor and Huang - Use of ORANI-recursive dynamic CGE model Taiwan Energy Yes - - - Dynamic (2010) (Monash University). - Recursive-dynamic model. Lu et al. (2010) China CO2 - - - No - Dynamic - Recycling via lump-sum transfers to enterprises as a subsidy. Devarajan et al. - Revenue recycling by reducing all indirect taxes South Africa CO - - - - Yes Static (2011) 2 (production taxes, sales taxes, and import tariffs). - World GEM-E3 CGE model. Saveyn et al. (2011) EU GHG - Yes - - - Dynamic - Recursive-dynamic model. - Two regions: South-Islands and North-Centre. - Labor market equilibrium admits a positive rate of Ciaschini et al. Yes/ Italy CO - - - - Static unemployment. (2012) 2 No - Second dividend defined as the reduction in unemployment rate. - DD under perfect competition but no DD under Orlov and Grethe Yes/ imperfect competition (Cournot oligopoly). Russia CO - - - Static (2012) 2 No - GTAP database. - Effects on equality. - They contemplate market and planned economy. - Two recycling scenarios: lum-sum and the same Cao et al. (2013) China CO - - - No No Dynamic 2 reduction in tax rates of some distortionary taxes (capital income, value added and sales tax). - Overlapping-generations model. Carbone et al. (2013) US CO - - Yes No - Dynamic 2 - Focus on distributional impacts. Energy goods, Fraser and Waschik - Data from GTAP. Australia generation CO , - - - - Yes Static (2013) 2 - Revenue recycling by reducing consumption taxes. usage of CO2 - Econometrically estimated. Jorgenson et al. US CO No - Yes No - Dynamic - Updated and amplified version of Jorgenson and (2013) 2 Wilcoxen (1993). - Model with three industries one representative Li and Lin (2013) China Energy and CO - - - No Static 2 consumer Meng et al. (2013) Australia CO2 - - - No - Static - ORANI-G model. - Use of AMOSEVI model. Allan et al. (2014) Scotland (UK) CO Yes - - - - Dynamic 2 - Labor market with imperfect competition. Argentina, Brazil, Chile, Chisari and Miller - Not DD in Brazil, but it has a very significant El Salvador, CO Yes - - - - Dynamic (2015) 2 reduction in costs by recycling. Jamaica and Peru - ORANI-G model. Sajeewani et al. Australia CO Yes - - No - Dynamic - Only 50% of revenues are recycled. (2015) 2 - Focused on distributional impacts. Yahoo and Othman Malaysia CO - - - Yes - Static - (2015) 2 ACCEPTED MANUSCRIPT- Carbon tax to non-ETS sectors. - Recycling revenues: 50% on investment tax credits Pereira et al. (2016) Portugal CO Yes - - - Dynamic 2 (ITC), 25% Social Security contributions, and 25% personal-income taxes (PIT). Henan, Fujian Zhang et al. (2016) and Chongqing CO2 - - - No No Static - Recycling by reducing indirect taxes on goods. (China) Source: own elaboration. 33 Table 2. Average values of some quantitative results in simulations. Number of simulations Variable Average value in % change that include this result 20.38 CO2 tax ($/Tn)* 37 (19.94) -11.76 CO2 emissions 46 (0.13) 0.11 GDP/GNP 43 (0.01) 0.00 Welfare 14 (0.01) 0.56 Employment 15 (0.01) 0.01 Consumption 11 (0.01) -0.50 Investment 9 (0.02) -0.69 Import 11 (0.19) 3.70 Export 19 (0.01) 0.66 Consumer price index 5 (0.01) * Several currencies used in simulations translated into $ at October 2016 exchange rates. Note: standard deviations in parenthesis. ACCEPTED MANUSCRIPT 34 Table 3. Meta-regressions of DD hypothesis. Dependent variable (% variation) Independent Consumer variables DD* Welfare GDP GDP GDP Price Investment Import Export (% (1) (2) (3) (4) (5) Index (7) (8) (9) variation) (6) 0.54 -0.13 0.12 -0.18 -0.77 -0.067 -0.25 -0.55 -1.58 훽 (5.54) (0.11) (0.26 (0.49) (0.54) (0.46) (0.69) (0.44) (4.91) -0,004 0.008 0.003 0.016 0.013 -0.103 0.03 0.014 -0.52 Emissions (-0.66) (0.01) (0.01) (0.02) (0.03) (0.05) (0.03) (0.02) (0.27) Income 1.23 ------tax** (0.58) 1.24 Lump-sum** ------(1.14) Observations 46 10 41 19 10 5 8 11 19 included R2 0.009 0.09 0.001 0.22 0.16 0.57 0.14 0.03 0.17 Note: standard deviations in parenthesis. * Dummy variable: 1 if authors claim that simulation achieve DD; 0 otherwise. ** Dummy variable: 1 if recycling instrument is income tax/lump-sum; 0 otherwise. ACCEPTED MANUSCRIPT 35