On the Employment Effects of Climate Policy: The Evidence from Carbon Tax in British Columbia∗ Akio Yamazaki† University of Calgary Version: September 2014 Working Paper Please Do Not Cite Without Author’s Permission. Abstract This paper examines how the unilateral imposition of the carbon tax in British Columbia (BC) affected the regional labor market. The unique structure of the tax allows the exploitation of variation in tax exposure across time, industries, and provinces to isolate its causal effect on employment. The estimates suggest that the employment effect is heterogeneous across industries due to the differences in emission intensity. The overall employment effect is significantly positive as more industries experienced an increase in employment. As these findings are robust across many specifications, this paper provides an evidence that a carbon tax does not provide an adverse employment effect in a regulated region. Key Words: Environmental regulation; carbon tax; employment; unilateral climate policy JEL Codes: E24, H23, J2, Q5 ∗ I am very grateful to my advisors, Jared C. Carbone and M. Scott Taylor, for their great supervision and invaluable advice. The paper also benefited from extensive discussions with Pamela Campa, Nicolas Rivers, and Trevor Tombe. I also thank Jevan Cherniwchan, Eugene Choo, Arvind Magesan, Kenneth J. McKenzie , Matthew Webb, and seminar participants at University of Calgary for their helpful comments. All remaining errors are my own. † Email: ayamazak@ucalgary.ca 1. Introduction With growing concern about global warming and climate change, many countries around the world have either implemented or are planning to implement various environmental regulations, one of which is a carbon tax. According to a report by the Climate Commission (SBS, 2013), 33 countries and 18 sub-national jurisdictions would have a carbon tax in place by 2013. The benefits of these regulations, such as an improvement of health outcomes and reductions of various pollutants and emissions, have been investigated extensively by many researchers.1 These estimated benefits seem to be quite large. Yet many policy analysts worry that the environmental benefits may come with unintended consequences. One prominent concern is job loss. The public fears that a regulatory burden imposed on firms would lead to an increase in product price, which lowers the quantity demanded and decreases sales and profits of firms. As a result, layoffs might occur due to a shutdown or relocation of firms to an area where there is less, or even no, environmental regulation in place. This issue is especially sensitive when unemployment rates are high such as in the recent recession, and the environmental regulations are easily blamed. Despite the considerable amount of attention in the media about these concerns, some analysts argue that the overall employment effect of environmental regulation is small, as the job loss in the regulated regions will be offset by the job creation in other parts of the economy. To provide additional evidence towards this active debate, this paper examines the employment effect of the unilateral implementation of the carbon tax in British Columbia (BC). On July 1st, 2008, BC implemented a carbon tax on the consumption of all fossil fuels. Anecdotal evidence suggests the policy is a success — achieving large reductions in pollution at relatively modest cost to its economy (CBC, 2013; Elgie and McClay, 2013). The policy intervention has a number of characteristics that make it an ideal natural experiment with which to study the employment effects of environmental regulation. First, the BC carbon tax 1 There are many studies that look at the effect of environmental regulation on health outcomes and pollution level. To name a few, see Chay and Greenstone (2003) and Currie and Neidell (2005). 1 is a textbook pollution tax, subjecting almost all sources of carbon pollution in the region to a uniform price per tonne of carbon emitted, making it easy to connect the predictions of theory to an empirical test. Second, the speed with which the policy was implemented made it a surprise to most stakeholders, ruling out the possibility that polluters would adjust their behavior in anticipation of the regulation. Third, the relatively high tax rate adopted meant that it provided a strong signal to polluters to change their behavior when the policy was introduced. To take advantages of these unique characteristics of the BC carbon tax, my empirical strategy exploits three sources of variation. As BC unilaterally implemented the carbon tax, the policy placed additional costs on residences of the BC2 . This creates temporal (prevs. post-carbon tax) and regional (BC vs. non-BC province) variation. A third source of variation arises from the heterogeneity in emission intensity across industries. Although the carbon tax is levied at a uniform rate, the actual burden differs across industries as it depends on how much emission each industry emits. The heterogeneity in tax exposure across industries creates a cross-industry variation. Having these three types of variation allow me to employ an empirical approach similar in spirit to triple differences that control for any unobserved time-invariant industry and province heterogeneity, time-varying industry shocks, and aggregate trends that would otherwise confound identification of the causal effect. Using this approach and after controlling for various shocks at province and industry level, I find that the implementation of a carbon tax in BC led to decreases in employment for industries whose emission intensity is greater than 0.57 tonnes per thousand dollars of production, and employment increases for industries whose emission intensity falls below this threshold. For example, at $10 per tonne of carbon dioxide equivalent (CO2 e), the cement and concrete product manufacturing sector, one of the most emission intensive sectors, ex2 Quebec introduced a carbon policy in 2007 as well (CBC, 2007); however, given that Quebec’s carbon tax rate is set at much lower rate, and its treatment of tax revenue differs from the BC carbon tax, I exclude it from the analysis. Alberta also implemented a carbon levy in 2007 (Alberta Environment and Sustainable Resource Development, 2014). But only facilities that emit more than 100,000 tonnes of GHG emission a year is subject to pay $15 a tonne. As the levy is only targeted to the small fraction of industries, I included Alberta in the analysis as non-carbon tax province. 2 perienced a 22% decline in employment relative to the pre-carbon tax periods. In contrast, the insurance carrier sector, one of the least emission intensive sectors, experienced a 7% increase in employment. As there are many more industries whose emission intensity falls below the threshold, the overall employment effect of the carbon tax is significantly positive. The results suggest that, on average, the implementation of the carbon tax generated 33,000 net job gains annually during the first 5 years following the implementation of the carbon tax, which is approximately a 2.5% increase in aggregate employment in BC during the period. This means that job gains in relatively clean industries exceed job losses in relatively dirty industries. In the context of the literature, these results are in line with studies that found a positive employment effect of the environmental regulations in the United States (EPA, 1981; Berman and Bui, 2001; Bezdek et al., 1989; Bezdek, 1993; Morgenstern et al., 2002; Wendling and Bezdek, 1989). Among these studies, Berman and Bui (2001) and Morgenstern et al. (2002) suggested an important channel through which the positive employment effects arise. They both point out that as regulated firms or industries engage in production, their labor demand could increase if either abatement activity requires additional labors or the production process becomes more labor intensive. This is particularly interesting as it relates to the degrees of substitutability between fossil fuels and labor in the case with a carbon tax. Other studies suggested that environmental regulations will create a new environmental protection (EP) industry. EP industry would provide services to abate pollution and maintain these services. As such, it may create more jobs in EP industry than jobs lost in regulated industries. Alternatively, stringent environmental standards lead firms to invest in technology to pollute less but also lower costs and improve quality. This would lead to more production, sales, and profits, which encourage increases in labor demand. These arguments are also quite relevant to a carbon tax. To avoid facing higher fuel prices, firms may invest in technology to obtain higher fuel efficient operation and labor demand may increase in EP industries that provide the technology. 3 Many earlier studies in this literature, including the studies just mentioned, estimated the effect based on simulation of macroeconomic and general equilibrium models. As the results of these studies are quite sensitive to the model assumptions, also other studies found a negative effect (Hollenbeck, 1979; Hazilla and Kopp, 1990; Greenstone, 2002). Hollenbeck (1979) and Hazilla and Kopp (1990) both showed that employment declined due to a negative effect of the Clean Air Act (CAA) of the United States on output. Hollenbeck’s results are based on the assumption that employment is proportional to physical output, and the results of Hazilla and Kopp are driven from the reduction of quantity supplied due to an increase in output prices. These assumptions about the output effect on employment are also recognized by the studies that found positive employment effects. However, these studies point out that the negative output effect on employment is rather small and that other positive effects on employment outweighs the negative effects. From the perspective of a methodology, this paper is closely related to Berman and Bui (2001) and more recent study by Greenstone (2002) as they empirically examined the employment effect using a difference approach. Berman and Bui analyzed the effect of the local regulation of air pollution in the manufacturing sector in Los Angeles3 by comparing it with Texas and Louisiana. They exploit the variation in stringency of regulation as LA imposed a stricter local regulation in addition to federal and state regulations where Texas and Louisiana did not. They found a positive employment effect, resulting in the creation of a couple of thousand jobs. On the other hand, Greenstone showed that an implementation of the Clean Air Act Amendments (CAAA) led to roughly 590,000 job losses in the manufacturing sector in nonattainment counties from 1972 and 1987, which he claimed was a relatively small negative effect when compared to the size of the entire manufacturing sector in the United States. Even though both studies focus on the employment effect in the manufacturing sector, the estimated results support the opposite effects. In this literature, mostly some types of pollution regulations in the United States are 3 To be precise, their region of interest is South Coast Air Basin, which includes Los Angeles County, Orange County, Riverside County, and the non-desert portion of San Bernardino County. 4 investigated such as CAA and CAAA. The CAA is a “command-and-control” regulation, whereas a carbon tax is a “market-based” regulation. An important prediction of economic theory is that these two types of regulation will produce different behavioral responses with consequences for policy cost (Goulder and Schein, 2013). The only empirical study that examined the effect of a carbon tax on economic activity is by Rivers and Schaufele (2013). However, their focus was on the trade implication of the BC carbon tax in the agricultural sector. Having said that, this study would be the first to empirically examine the employment effect of a market-based regulation. The rest of the paper proceeds as follows. Section 2 describes the BC carbon tax. Section 3 presents a simple theory to explain the mechanism through which an imposition of a carbon tax affects a labor market. This framework predicts that an employment effect of the BC carbon tax varies across industries based on four factors: an elasticity of substitution between inputs, a own price elasticity of demand, a cross price elasticity of demand, and an income elasticity of demand. Then section 4 contains my empirical analysis where I explain data and methodology, and provide results and robustness checks. Lastly, I discuss an implication on wages in section 5, and then conclude the paper in section 6. 2. Overview of the BC Carbon Tax A carbon tax is believed to be the most cost-effective way to reduce greenhouse gas (GHG) emissions as it is a type of Pigovian tax (Elgie and McClay, 2013). Its purpose is to give incentives to households and businesses to consume less of the fuels that release GHG emissions into the atmosphere. The BC carbon tax was introduced with the objective of reducing emissions by a minimum of 33% below the 2007 levels by 2020 (Ministry of Finance, 2013). On February 2007, BC’s new climate policy agenda to prioritize the environmental concerns was unveiled by Premier Gordon Campbell in the throne speech (Harrison, 2012). As 5 the Liberal government, led by Premier Campbell, had undergone deep cuts to the environmental budget and supported offshore oil and gas exploration, the speech was a surprise to many, including members of his own party (Harrison, 2012). In October 2007, the Ministry of Finance publicly acknowledged that a carbon tax was under consideration, and officially announced in their budget plan in February 2008 that a carbon tax would be implemented. The enforcement of the carbon tax was rather quick since it was implemented in just five months following the official announcement. Although the announcement took the public by surprise, polls indicated that 72% felt that the introduction of a carbon tax was a positive step (Duff, 2008). Others, including businesses and industry associations, worried about the adverse effect on their operational costs and requested tax relief or exemption (Ministry of Finance, 2013). Overall, the implementation of the carbon tax has been supported by the residences of BC as the Liberal party has won the majority seats in post-carbon tax elections, 2009 and 2013. Carbon taxes often include some form of exemptions, mostly for energy intensive and trade exposed industries, to protect the domestic economy, which makes the tax relatively narrow-based. However, the BC carbon tax is intended to be broad-based, i.e. exemptions were applied to none of the industries initially.4 In addition, the tax-base is broad in terms of the types of fuels that are subject to the tax. It taxes the carbon content of all fossil fuels including gasoline, diesel, natural gas, coal, propane, and home heating fuel (Ministry of Finance, 2008; Ministry of Small Business and Revenue, 2008). In the budget and fiscal plan for 2014 (Ministry of Finance, 2014), the carbon tax revenue raised in the 2012-2013 fiscal year was $1.1 billion and is estimated to be $1.2 billion for 2013-2014. There are three different ways to deal with these tax revenues from the carbon tax: revenue-raising, revenue-neutral, and deficit-reduction. BC took the revenue-neutral approach. In these cases, all of the carbon tax revenue will be returned to individuals and businesses through reductions of other taxes such as income and corporate taxes (Ministry of 4 This has changed as of January 1st, 2014. To protect agricultural industries, a carbon tax relief is granted to commercial greenhouse growers. For further information, see http:www.gov.bc.ca/agri/. 6 Finance, 2008). The corporate income tax rate and the two lowest personal income tax rates were reduced by 5 percent, which resulted in the BC’s corporate and personal income tax rates the lowest in Canada5 (Elgie and McClay, 2013). In addition, to protect low-income households, the government provides these households a lump-sum credit. In 2012-2013, $1.4 billion was credited back to individuals and businesses. In fact, tax credits exceeded the tax revenue by $260 million. This excess is estimated to decline to only $20 million in 2013-2014. In the initial year of 2008, the tax was set at $10/t CO2 e emissions from burning fossil fuels and set to increase by $5/t CO2 e annually until 2012 (Ministry of Small Business and Revenue, 2008). This means that the carbon tax would increased 2.41 cents per litre for gasoline, rising gradually to 7.24 cents a litre by 2012 (Ministry of Finance, 2008). The gradual increase of the tax rate intends to allow consumers to adjust their fuel usage rather slowly to minimize the financial burden from the tax. As of 2014, the tax rate is expected to stay at $30/t CO2 e. Aside from the economic impact of the BC carbon tax, a recent report (Elgie and McClay, 2013) showed that the BC carbon tax has reduced a per capita use of fossil fuel by 19% more than that of the rest of Canada during the first four years following the implementation. Similarly, per capita GHG emissions have declined by 10% in BC from 2008 to 2011. It has only been several years since the implementation, but we have already seen a substantial reduction of use of fossil fuel and GHG emissions. The BC carbon tax has been fulfilling its purpose so far. 3. How Could a Carbon Tax Raise Employment? In this section, I establish a simple theory that explains the mechanisms through which an employment effect of a carbon tax come about and shows that both positive and negative 5 In fact, BC has tied with Alberta and New Brunswick for the lowest corporate tax rate in Canada, and BC has the lowest personal income tax rate in Canada, but for only those earnings up to $119,000. 7 employment effects are possible. I do this by taking several steps towards a theory that incorporates most of the characteristics of the BC carbon tax. I. Partial Equilibrium: One Industry Case I first start with the simplest case, a partial equilibrium analysis with only one industry. To simplify the analysis, I assume that a firm only use labor and fossil fuel as inputs and takes the price of fossil fuel as given. Then a perfectly competitive firm would minimize its cost by solving: min c = w1 l + w2 e subject to q = f (l, e) e,l where l denotes labor, e denotes fossil fuel, and w1 and w2 are corresponding factor prices. By solving this, I can drive conditional input demands for l and e as l∗ = l(w1 , w2 , q) and e∗ = e(w1 , w2 , q). Then using these equilibrium input demands, I can define a cost function c = c(w1 , w2 , q). By assuming a constant return to scale production function, I can show that c(w1 , w2 , q) = qc(w1 , w2 ). With this cost function, I can derive a labor demand function by Shepard’s lemma as follows: ld = qc1 (3.1) where ci indicate the partial derivative with respect to ith argument. c1 is the amount of labor required to produce one unit of q, which I redefined it as aLQ . Then taking logs and total differentiating (3.1) yields labor demand decomposition: lbd = + qb |{z} Output effect b aLQ |{z} (3.2) Factor effect where lbd = dld /ld , and so on (i.e. “ b ” denotes percent change). This labor demand decomposition implies that the percentage change in labor demand is determined by two effects, the output and factor effect. Now let me investigate one effect at a time. From a demand side, the demand function is given by q = D(p), holding income fixed. 8 Taking logs and total differentiating yields: dq 1 ∂D(p) = dp q D(p) ∂p (3.3) Then by algebraic manipulation, I can show (3.3) as follows: qb = εDP pb (3.4) where εDP ≤ 0 is a price elasticity of demand. Now to figure out an expression for pb, a firm maximize its profit by solving: max π = pq − qc(w1 , w2 ) q Then the first order condition of firm’s maximization problem yields p = c(w1 , w2 ). Now let me introduce a carbon tax on fossil fuel as follows: w20 = w2 (1 + t) Then p = c(w1 , w20 ). Totally differentiate and dividing by p yields: dp c1 dw1 c2 dw20 = + p p p (3.5) With a simple manipulation, I can show (3.5) as follows: \ pb = θl w b1 + θe (w b2 + (1 + t)) where θi refers to the cost share of input i (i.e. θi = (ci wi )/c). However, in this paper we are only interested in an effect of a carbon tax on employment, which leads to an assumption 9 that w b1 = w b2 = 0. Then I am left with, \ pb = θe (1 + t) (3.6) Finally, by plugging (3.6) back into (3.4), I have an expression for the output effect. \ qb = εDP θe (1 + t) (3.7) (3.7) shows that the change in output from a carbon tax is determined by two factors: the price elasticity of demand and the cost share of fossil fuel. Similarly, let me drive an expression for the factor effect. As aLQ = c1 (w1 , w2 ), total differentiate and divide both side by aLQ yields, daLQ c11 c12 0 = dw1 + dw aLQ aLQ aLQ 2 (3.8) where c1i refers to the partial derivative of c1 with respect to ith argument (i.e. second derivative of the cost function). By simple algebraic manipulation and w b1 = w b2 = 0, I can show (3.8) as follows: c c12 \ θe (1 + t) c1 c2 \ = σθe (1 + t) b aLQ = (3.9) (3.10) where σ ≥ 06 is the elasticity of substitution between labor and fossil fuel. The steps from (3.9) to (3.10) are shown in appendix A. (3.10) shows that a percentage change in labor requirement by a carbon tax is determined by two factors: the elasticity of substitution between labor and fossil fuel and cost share of fossil fuel. Finally, putting (3.7) and (3.10) 6 When we allow for more than two inputs, σ can be negative if labor and fossil fuels are complement. 10 back into (3.2) yields, \ \ lbd = εDP θe (1 + t) + σθe (1 + t) | {z } | {z } Output effect (3.11) Factor effect \ = (εDP + σ) θe (1 + t) (3.12) From (3.11), the sign of the employment effect of a carbon tax is determined by the degree of the negative output and positive factor effect. The size of the negative output effect depends on the price elasticity of demand. The size of the positive factor effect, on other hand, depends on the elasticity of substitution between labor and fossil fuel. (3.12) makes it clear that the sign of the employment effect comes down to a fight between the price elasticity of demand and the elasticity of substitution between factors. Even in the one industry case, both positive and negative employment effects are possible. For example, for a tradable goods industry facing highly elastic demand and limited factor substitutability, the negative effect from the price elasticity of demand outweighs the positive effect from the elasticity of substitution. On the contrary, a non-tradable industry like the service sector face relatively inelastic demand and their factors are highly substitutable, and the positive effect from the elasticity of substitution outweighs the negative effect from the demand elasticity. In addition, if the size of the output and factor effects are the same, then an employment effect of a carbon tax could even be nil because these two effects would exactly offset each other. II. Partial Equilibrium: Many Industry Case One caveat with regard to equation (3.12) is that it does not allow for heterogeneity in employment effect across industries. To investigate heterogeneous industry-specific employment effects, I need multiple industries to see how interactions between industries affect the labor market. To simplify the analysis, I work with two industries. As employment effects of a carbon tax would differ by energy intensity of each industry, I denote these two industries 11 as clean and dirty. With these two industries, I define a demand function more generally as follows: qi = Di (pi , pj ) where pi denotes the price for industry i, say the dirty industry and pj denotes the price for industry j, say the clean industry. This demand function assumes a quasi-linear preferences as it is independent of income. By making the same algebraic manipulation that derive (3.4), I have, qbi = εii pbi + εij pbj where εii ≤ 0 is a own price elasticity of demand for industry i and εij ∈ [−∞, ∞] is a cross \ price elasticity of demand. Then similar to (3.6), pbi = θei (1 + t), which yields, \ qbi = (εii θei + εij θej )(1 + t) (3.13) where θei denotes an cost share of fossil fuel for the industry i. As there is no interaction of two industries on the supply side, the factor effect would be the same as the case with one industry, σi denoting the elasticity of substitution between labor and fossil fuel for industry i. This gives an expression for the heterogeneous industryspecific employment effect of a carbon tax as follows: \ \ + t) lbid = σi θei (1 + t) + (ε θ + εij θej )(1 | {z } | ii ei {z } Factor effect Output effect \ = (εii + σi )θei + εij θej (1 + t) | {z } | {z } Own industry effect (3.14) Cross industry effect By the symmetry, the percentage change in labor demand for industry j can be shown as: \ lbjd = (εjj + σj )θej + εji θei (1 + t) 12 (3.15) Figure 1: Shifts in Labor Demand Based on (3.14) and (3.15), an industry-specific employment effect of a carbon tax with two industries depends on two factors: own industry and cross industry effect. The labor demand decomposition under two industry case is almost same as (3.12) except that now it has an output effect from the other industry, which means the cross price elasticity of demand and cost share of fossil fuel from the other industry also affect own labor demand. In addition to the own industry effect, cross industry effect adds additional complications to the determination of the sigh of the employment effect. As now a percentage change in labor demand depends on three factors, it is impossible to figure out the sign of the changes in labor demand in a response to the BC carbon tax a priori. However, there are two cases where the signs can be determined analytically. The first case is when the own industry effect is negative, and clean and dirty industries produce complementary goods. In this case, the changes in labor demand would be negative because both own and cross industry effects are negative. The second case is when the own industry effect is positive and goods are substitutes. In 13 this case, the employment effect would be positive as both own and cross industry effects are positive. Other than these two cases, an employment effect could be positive, negative, or even nil. Then the aggregate employment effect of the BC carbon tax could also be positive, negative, or zero, which is demonstrated in Figure 1. In addition to the sign of each effect, the size of each effect also determines the employment effect. The size of each effect depends on how much an industry spends on fossil fuels relative to labor. This will be an important variation to identify the employment effect of the BC carbon tax in the estimation as it relates to industry’s emission intensity. Now to think of aggregate employment, simply adding (3.14) and (3.15) yields: \ b = lbd + lbd = (σi + εii + εji )θe + (σj + εjj + εij )θe (1 + t) L j i i j (3.16) By using the Slutsky equation, I can show that εij = εji . Then I can simplify and rearrange (3.16) as follows: \ b = (σi + εii )θe + (σj + εjj )θe + (θe + θe )εij (1 + t) L j i j i (3.17) (3.17) makes it clear that the sign of the aggregate employment effect of the BC carbon tax is ambiguous. III. Income Redistribution As a part of the BC carbon tax policy, tax revenues were redistributed back to the residence of BC via reductions of income and corporate taxes as well as lump-sum transfers. To investigate the effect of the income redistribution on labor market, I go back to the simplest case, the partial equilibrium with one industry, and now incorporate a lump-sum transfer. To see how an inclusion of the lump-sum transfer could change the employment effect, I 14 need to relax the assumption on the demand function as follows: q = D(p, I) where I denotes income. Once again, taking the same algebraic manipulation that derive (3.4) yields, \ qb = (εDP θe )(1 + t) + εDi I Ib (3.18) where εDi I ≥ 0 is an income elasticity of demand. Now to accommodate the lump-sum transfer from the government, I define income as follows: I0 = I + T where T is the lump-sum transfer. Then by simple algebraic manipulation, I can show that Ib0 = φTb, where φ denotes a share of the lump-sum transfer in after-transfer income. By plugging this back into (3.18), I have, \ qb = (εDP θe )(1 + t) + εDI φTb Then, putting this together with the factor effect yields: \ \ + t) + εDI φTb lbd = σθe (1 + t) + (ε θ )(1 | {z } | {z } | DP e{z } Factor effect Output effect (3.19) Income effect Now from this carbon tax, government receives tax revenues and gives them all back to the residence of BC, which yields the following budget constraint: T = tw2 e(w1 , w20 )q 15 Taking logs and totally differentiating above yields: \ Tb = b t + εew20 (1 + t) + qb (3.20) where εew20 < 0 is a price elasticity of input demand for fossil fuel, i.e., the changes in demand \ for fossil fuel as inputs in a response to the changes in its price. As qb = εDP θe (1 + t) from (3.7), I can rearrange (3.20) as follows: \ Tb = b t + (εew20 + εDP θe )(1 + t) (3.21) (3.21) shows that the size of the tax revenue, equivalently the lump-sum transfer, depends on three factors, the size of the carbon tax, the price elasticity of factor demand, and the price elasticity of demand. The lump-sum transfer could increase as the carbon tax rate goes up. At the same time, it could decrease because the firm uses less fossil fuel now that a price of fossil fuel is higher, which leads to an erosion of a tax-base, or the demand for fossil fuel is less as the firm produces less, which requires less of inputs. One point to be careful about (3.21) is that it is never negative. Although these two elasticities in (3.21) are negative, Tb being negative means either money is taken away from consumers or the amount of the lump-sum transfer is smaller than the previous year. However, the latter case is irrelevant here as there is no dynamics in this simply theory. Therefore, it does not make sense for Tb to be negative. Finally, plugging (3.21) into (3.19) yields: \ \ \ + t) + εDI φ(b t + (εew20 + εDP θe )(1 + t)) lbd = σθe (1 + t) + (ε θ )(1 } | | {z } | DP e{z {z } Factor effect Output effect (3.22) Income effect \ = ((σ + εDP (1 + εDI φ))θe + εDI εew20 φ)(1 + t) + εDI φb t As income elasticity of demand is always positive7 , the income effect would also be 7 Here I assume goods are normal. 16 positive. Then, the chances of the employment effect being positive are higher because now in addition to positive factor effect, there is the positive income effect that could outweigh the negative output effect. From (3.22), I show that revenue-neutrality of a carbon tax could increase the likelihood of positive employment effects of a carbon tax as it pushes the labor demand curve further out to the right. One possible criticism that this subsection may receive is that the income redistribution from the BC carbon tax does not take the form of only lump-sum transfers. Even so the results from this subsection serves as a lower-bound for an employment effect of the revenue-neutral carbon tax policy because lump-sum transfers are roughly 20% of the total redistribution (Ministry of Finance, 2008). IV. Employment Dividend So far, my simply theory has focused on the labor demand. The employment effect of a carbon tax could also arise from the changes in labor supply, especially when the tax revenues are refunded back to the residence of BC. The positive employment effect from a revenue-neutral carbon tax is known as “employment dividend” in the double dividend literature (see Goulder (1995), Bovenberg and Goulder (2002), and Bovenberg (1999)). It basically means that by recycling the revenue from the carbon tax in a form of reductions of distortionary taxes, not only does the carbon tax bring environmental benefits, but also it lessens the sizes of distortion in the other tax system. As there are many theoretical studies in the current double dividend literature, I would not provide theories here. Rather I would simply point out that the employment effect of the carbon tax could also come from the supply side. In particular, if an employment dividend exits, reducing the personal income tax and providing a lump-sum transfer would increase labor supply. Then the overall employment effect would likely to be positive unless the employment effect from the labor demand is negative and large. After considering changes in both labor demand and supply in a response to a revenue17 Figure 2: Shifts in Labor Demand and Supply neutral carbon tax, it is not clear which direction the equilibrium level of labor would go a priori. Both directions are possible, which is illustrated in Figure 2. Even with my simple theory on labor demand and the double dividend literature on labor supply, the sign of the employment effect is ambiguous. 4. Empirical Analysis I. Data Sources To examine the employment effects of a carbon tax, I use industry-level data for employment and GHG emission intensity. Employment is measured as a number of all employees in an industry whereas emission intensity is measured in tonnes per thousand current dollars of production. Both data are from the Canadian Socio Economic Information Management System (CANSIM) at Statistics Canada and categorized based on the North American Industry Classification System (NAICS). NAICS is used by federal statistical agencies in classifying 18 Table 1: Descriptive Statistics: 2001-2012 Mean Overall (N = 5,100 , J = 83) Employment 20323 Emission Intensity 0.4979 Employment BC (N = 816 , J = 71) Overall 21689.3 High 6074 Low 18668 Non-BC (N = 4284 , J = 83) Overall 20063 High 7083.4 Low 13021.4 Std. Dev. Median Min Max 53689 0.50118 4601.5 0.35 24 0.12 675227 3.26 40885.8 6730.7 9005 6776.5 2958 18329.5 333 333 3832 262568 21338 34685 55796.1 10857.8 23235.9 3923.5 3313 3478 24 97 110 675227 57855 133551 Note: N refers to the number of observations. J refers to the number of industries. High refers to industries whose emission intensity is in 90th percentile and Low refers to industries whose emission intensity is in 10th percentile. business establishments based on the types of economic activity in Canada, Mexico, and United States (United State Census Bureau). Its numbering system consists of a six-digit code at the most detailed industry level. The employment data has 19 two-digit industries, 65 three-digit industries, and 170 four-digit industries, whereas GHG emission intensity data has 19, 53, and 54 industries respectively. After merging these two data by matching the digits, in total, 83 industries are in the sample. List of industries and corresponding NAICS codes are reported in Appendix B. Descriptive statistics are reported in Table 1. A characteristic of British Columbia (BC) that is worth noting from Table 1 is that BC is relatively less emission intensive compared to the rest of Canada as the employment ratio between high and low emission intensive industries is smaller in BC than the rest of Canada. More concisely, industry composition in BC is dominated by low emission intensive industries relative to the rest of Canada. CANSIM provides employment data for all provinces from 2001 to 2012. GHG emission intensity data is only available at the national level and covers from 1999 to 2008. Ideally, 19 data for all provinces and more post-carbon tax periods would be used to exploit emission intensity variations across time and regions. Therefore, I impose an assumption that national GHG emission intensity level for each industry can serve as a proxy for each industry in all provinces. While the emission intensity level might be different across provinces for each industry, the relative emission intensity level across industries within provinces is likely to be the same. For instance, a relatively dirty industry in Alberta, such as oil and gas extraction, would also be a relatively dirty industry in other provinces as well. Therefore, GHG emission intensity data at national level is sufficient. To examine the employment effect of the carbon tax, only emission intensity variation across industries in the pre-carbon tax period is required. With data for more post-carbon tax periods, one can examine an effect of the carbon tax on a level of GHG emissions. However, that is not the purpose of this study. GHG emission intensity is likely to decline in BC after the implementation of the carbon tax; yet, the relative emission intensity across industries is not likely to change (i.e. a dirty industry will not suddenly become a clean industry solely due to the implementation of the carbon tax). Therefore, not having data for post-carbon tax periods is not an issue. I can simply use an emission intensity data at national level from pre-carbon tax period like 2007. Before quantitatively evaluating the employment effect of the carbon tax, I plot the percentage change in employment for high and low emission intensive industries to visualize the data. Unfortunately, as the carbon tax in BC was implemented in the same year as the global financial crisis, employment might fall regardless of the financial burden from the carbon tax. To isolate the adverse employment effect of the financial crisis, there needs to be employment fluctuations around 2008 in BC that are different from the rest of Canada. To more closely examine this concern, changes in employment for BC and the rest of Canada are assessed for two industries: the high emission cement industry and the low emission insurance industry (Figure 3). Both sectors experienced a decline in employment right before the implementation of the carbon tax. After the implementation of the carbon 20 Figure 3: Comparison of % changes in employment over time between high emission intensive and low emission intensive industry Note: Cement and concrete product manufacturing is one of the high emission intensive industry and insurance carrier is one of the low emission intensive industry. Source: Author’s calculation tax, however, cement and concrete product manufacturing sector in BC experienced a smaller increase in employment than the rest of Canada. Conversely, insurance carrier sector in BC experienced a larger increase in employment than the rest of Canada. These findings suggest that the changes in employment in BC are correlated with the implementation of the carbon tax through the heterogeneity in industry’s emission intensity level. The financial crisis has caused a decline in employment in all provinces as it is more a nation-wide incidence. The fact that changes in employment in BC compared to the rest of Canada are different between these sectors is an evidence for the existence of the employment effect of the carbon tax. II. Methodology In this section, the econometric design to estimate the employment effect of the carbon tax is discussed. The simple theories presented in the last section are used to motivate the estimation equation as follows: ln Lipt = λi + ηp + φit + γt + β1 (EIi × BCp × τt ) + β2 (BCp × τt ) + ipt 21 (4.1) where ln Lipt is the natural log of employment for industry i in province p at time t.8 I isolate the employment effect of the BC carbon tax into two part, an industry-specific and industry-common effects. I do this so that I can identify which industries gain or lose from the implementation of the BC carbon tax. The first interaction term captures the industryspecific employment effect and the second interaction term capture the common employment effect of the carbon tax across industries. Let EIi be a GHG emission intensity level for an industry i in 2007, BCp be an indicator for British Columbia, and τt be carbon tax variable, defined as follows: τt = 0 10 15 if t ≤ 2007 if t = 2008 if t = 2009 20 25 30 if t = 2010 if t = 2011 if t = 2012 λi is an industry fixed effect that captures time-invariant industry heterogeneity, such as industry factor intensity. ηp are province fixed effects that control for time-invariant province heterogeneity such as geographic characteristics9 that are common across industries. φit are a industry-specific time trend that controls for industry specific shocks across provinces at given year. Controlling for industry-specific shock is particularly important as this will control for incidences like financial crisis, exogenous changes in prices of natural resources and commodities, and any shocks that are specific to industries but common across provinces. γt are time fixed effects that capture aggregate shocks that are common across industries and provinces, such as an economic growth at the national level. ipt is an error term that captures idiosyncratic changes employments. 8 I employ the natural log of employment to remove the skewness in the distribution of employment. After the transformation, employment is approximately distributed normal. 9 For example, as BC is closest to a coast, employment might fluctuate because of the level of export. If exports go up in BC, productions need to increase. This increases the demand for labor. Province fixed-effect will control for any fluctuation in employment due to the effect like this. 22 β1 and β2 are the coefficients of interest. Then let αit ≡ (β̂1 EIi + β̂2 )∆τt 10 , and it is interpreted as an approximation of percentage change in employment for an industry i at time t in a respond to the carbon tax. The estimated coefficient β̂1 is identified from across industry × province comparisons over time and β̂2 is identified from across province comparisons over time. Equation (4.1) will be my fundamental equation to identify the employment effect of the carbon tax. To properly identify the employment effect, the underlying identification assumption requires that there be no other factors other than the carbon tax creating a difference in changes in employment for industries in BC that were affected differently by the carbon tax. In other words, there are no unobserved industry × province shocks that are correlated with the carbon tax. This assumption will be violated if the government of BC concurrently implement other policy induced by the carbon tax that affect all industries in BC differently while no other province implement similar policy. III. Results The results are presented in the following subsections. By estimating the regression equation (4.1), the heterogeneous employment effect across industries are discussed in subsection A. Then in subsection B, I constructed a counterfactual based on the estimation in the subsection A to calculate the changes in employment for each industry and the aggregate employment effect. A. The Industry-Specific Employment Effect The results of five specifications based on equation (4.1) are reported in Table 2. The first specification, reported in column (1) in Table 2, is a simple OLS with no controls. This specification is the most vulnerable to an omitted variable bias (OVB) as there is no control variables. To lessen the concern of OVB, I include fixed-effects in the rest of specifications, reported in the column (2) through (5). These estimates are identified with 10 ∆τt ≡ τt − τt−1 23 clustered standard errors (clustered on industries11 ), and interpreted as the average treatment effect of the carbon tax on employment. The second specification, reported in the column (2), adds time fixed effects to control for aggregate time-varying factors, while the third specification, reported in the column (3), includes industry fixed effects that control for time-invariant industry characteristics. In addition to the time and industry-fixed effects, the specification in the column (4) controls for any provincial characteristics. The last specification, reported in the column (5), controls for any unobserved industry-specific shocks over time. As β̂1 measures the heterogeneous employment effect of the carbon tax across industries, the significant negative results verify that the employment effect of the carbon tax is, indeed, heterogeneous across industries. The negative sign of β̂1 means that high emission intensive industries experience a decline in employment due to the carbon tax because their emission intensity is so large that βˆ1 EIi exceeds βˆ2 , which makes αit negative. Depending on the size of βˆ2 , low emission intensive industries experience an increase in employment because of the similar argument. This confirms the results from some of earlier studies that emphasized the heterogeneous employment effect of environmental regulation across industries (Wendling and Bezdek, 1989; Hollenbeck, 1979; Hazilla and Kopp, 1990; Greenstone, 2002). All results show statistically significant negative heterogeneous effects and positive common effects of the carbon tax. As the last specification in the column (5) includes the full set of fixed effects based on the equation (4.1), my preferred specification is the result in the column (5). To quantify the magnitude of the employment effect from the carbon tax in BC, the value of αit is calculated for each industry based on the values from the column (5) of both coefficients. The threshold level of emission intensity that changes the sign of αit between positive and negative is about 0.57 tonnes per thousand dollar of production. If emission intensity is bigger than 0.57, the employment effect of the carbon tax is estimated to be 11 I also clustered on provinces. The results did not change in terms of significances. The corresponding tables for the results with clustered on provinces are available upon a request. 24 Table 2: Estimated Regression Models for the Percentage Changes in Employment lnL EIi × BCp × τt (β1 ) BCp × τt (β2 ) Time FE Industry FE Province FE Industry × time FE N R2 (1) -0.0504a (0.0167) 0.0495a (0.01000) 5,100 0.011 (2) (3) (4) (5) -0.0504a -0.0332a -0.014a -0.0151b (0.0167) (0.00935) (0.00482) (0.00596) 0.0506a 0.0464a 0.00698b 0.00860a (0.0102) (0.00739) (0.00278) (0.00320) Y Y Y Y Y Y 5,100 0.012 5,100 0.314 5,100 0.878 Y Y Y Y 5,100 0.887 c p < 0.1, b p < 0.05, a p < 0.01 Clustered Robust standard errors in parentheses negative. For example, the emission intensity for cement and concrete product manufacturing sector is 1.99 tonnes per thousand dollar of production. Thus, at a rate $10/t CO2 e, they experience a 22% decline in employment on average from the carbon tax. On the other hand, as the emission intensity for insurance carrier sector is only 0.12, they experience, on average, a 7% increase in employment. The percentage changes in employment from a $10/t CO2 e carbon tax for all the industries are reported in Table 7 in the appendix C, and corresponding scatter plot with 95% confidence intervals is presented in Figure 6 in the appendix C. Based on the estimated standard errors for each employment effect, many industries experienced statistically significant employment effects. The magnitude of the employment effect is relatively small if the effect is positive, as the highest positive employment effect is about 7%. On the contrary, if the employment effect is negative, the magnitude varies largely as it ranges from -0.2% to -40%. This is clearly shown in Figure 7 in the appendix C. In the sample, about a 30% of industries exceeds the threshold (21 out of 71 industries in BC that exceeds the threshold). This means that 70% of industries experienced an increase 25 in employment due to the carbon tax. B. The Aggregate Employment Effect This subsection discusses the aggregate employment effect of the BC carbon tax. To do this, I first need to calculate the changes in employment for each industry in a response to the BC carbon tax so that the aggregate employment effect can be calculated by the sum of these changes in employment across industries. I do this by constructing a counterfactual, calculating the level of employment in an absence of the BC carbon tax, and then compare them with the level of employment from the fitted values in the estimation of my preferred specification. The difference would be the employment effect of the BC carbon tax. To visualize the changes in employment for all industries in response to the BC carbon tax, Figure 8 in the appendix C plots the changes in employment for all industries at $10/t CO2 e. The air transportation sector experienced the largest reduction in employment (about 4,500 job losses) whereas retail trade experienced the largest increase in employment (about 11,000 job creations). To link back with the example in the previous subsection, the cement and concrete product manufacturing sector experienced a decline in employment about 650 per year, and the insurance carrier sector experienced an increase in employment per year by about 860. Although the positive employment effect for the insurance carrier sector is only a 7%, the 7% increase leads to an increase in employment by 860. On the contrary, the 22% negative employment effect for the cement and concrete product manufacturing sector leads to a decrease in employment only by 650. This could reflect the fact that the emission intensive industries are less labor intensive. In fact, the data indeed shows the negative correlation between emission intensity and employment, demonstrated in Figure 4. Therefore, the relatively large negative employment effects lead to rather a small reduction of employment. This explanation is consistent with one of the factors that determine the employment effect in the study of Berman and Bui (Berman and Bui, 2001) — the regulation burden falls on capital-intensive industries disproportionally with relatively little employment. 26 Figure 4: The Correlation between Emission Intensity and Employment Note: NAICS codes are displayed in the figure along with the scatter plot. Source: Author’s calculation Finally, by summing these changes in employment across industries for each case, Figure 5 plots percentage changes in aggregate employment, one with carbon tax and the other with no carbon tax. Based on the estimation and counterfactual, aggregate employment in BC increased by about 165,000 over the post-carbon tax periods (i.e. on average 33,000 per year from 2008 to 2012). This means that the overall employment effect of the carbon tax in BC is significantly positive. During the same period, aggregate employment in BC is about 6.5 million in an absence of the carbon tax. Then 165,000 job creation corresponds to a 2.5% of aggregate employment in BC over the post-carbon tax periods. The positive aggregate employment effect means that increases in employment in cleaner industries exceeded declines in employment in dirty industries. There are two ways the aggregate employment effect resulted in positive gains in employment. One is that the industry composition in data is dominated by relatively less emission intensive sectors. The estimation suggested that less emission intensive industries experienced increase in employment. Then as the fraction of the emission intensive industries in data is small, number of industries that experienced the negative employment effect is outnumbered by number of industries that 27 Figure 5: Percentage Change in Aggregate Employment for British Columbia Source: Author’s calculation experienced the positive employment effect. Second is that while the reduction of employment from each negative employment effect is small, the increases in employment from the positive employment effect are large. The aggregate employment effect may not be positive even if there are many more clean industries than dirty industries in BC because the decline in employment from dirty industries could be so large that the negative effects could still dominate the positive effects from clean industries. Therefore, these two conditions together ensure the positive aggregate employment effect from the BC carbon tax. IV. Robustness Checks This paper has attempted to estimate the causal effect of a carbon tax on provincial employment. Even with a perfect econometric design, a nonexperimental research might be vulnerable to unobserved variations that could confound the causal interpretation. To ensure the reliability of the estimations, I probed the robustness of the estimates in many different ways, but found little evidence that undermines the results reported in the last section. 28 Table 3: Probing the Robustness of the Employment Effect of Carbon Tax lnL EIi × BCp × τt BCp × τt (1) (2) (3) -0.0151b (0.00596) 0.0086a (0.00320) -0.0127b (0.00559) 0.0078b (0.00305) -0.0146b (0.00582) 0.0085a (0.00316) -0.0146b (0.0059) 0.0085a (0.00316) -0.0746 (0.0756) 0.03 (0.0382) EIi × BCp × τt BCp × τt EIi × QCp × τt QCp × τt N R2 Threshold Carbon Tax Sample (4) 5,100 0.887 0.56 5,960 0.879 0.61 5,960 0.879 0.58 5,960 0.879 BC Exclude Quebec BC All BC & Quebec All BC & Quebec All c p < 0.1, b p < 0.05, a p < 0.01 Clustered Robust standard errors in parentheses Note: All specifications include time FE, industry FE, province FE, industry × time FE. First I re-estimated the regression equation by including Quebec (QC)12 in the sample to see how the results are sensitive to the different samples. The results are reported in Table 3. All the specifications include time fixed effects, industry fixed effects, province fixed effects, and industry × time fixed effects. The result reported in the column (1) is taken from the column (5) in Table 2, which is the full specification model based on the equation (4.1), and serve as a base specification. The rest of the results, reported in the column (2) through (4), should be compared to the base specification. In this analysis, I first estimate the employment effect of the BC carbon tax including 12 Ideally, I would like to also include Alberta as one of the carbon tax provinces for another robustness check. However, Alberta carbon tax is only for firms that emit more than 100,000 tonnes. With industry-level data, I cannot appropriately estimate the employment effect. 29 QC in the sample, but not as the other province that implemented a carbon tax, reported in the column (2). Then QC is included as the province with a carbon tax. QC implemented its carbon tax in 2007 at the rate of $3.50/t CO2 e. The result is reported in the column (3). Regardless of how QC is included in the sample, the estimated results preserved the statistically significances and the sign of the coefficients of interests. The magnitude also stayed around the reasonable range of values, i.e., the threshold level which determines the sign of the employment effect for each industry is about the same as that of the base specification (0.56 tonnes). Based on the coefficients from the column (2) and (3), the threshold levels are 0.61 and 0.58 tonnes per thousand dollar of production, respectively. In the last column, the employment effect of the BC and QC carbon taxes are separately estimated. The result shows that there is no employment effect of the QC carbon tax while the employment effect of the BC carbon tax still exits. This is an evidence that inclusion of QC in the sample does not affect the analysis of the employment effect of the BC carbon tax. One of the most potential sources of bias arises from an employment effect that is not from the carbon tax, such as the adverse effect from the financial crisis and commodity price shocks. For example, as the effect of the financial crisis on employment is at the national level, other provinces had also experienced the decline in employment around 2008. To determine if my econometric design successfully controls for these effects, a placebo test is conducted. In other words, I have estimated the employment effect of a placebo carbon tax in other provinces. If the effect of the financial crisis is fully controlled, then implementing a placebo carbon tax at other provinces in the same year as the BC carbon tax should not have any employment effects because these carbon taxes are fake. The results are reported in Table 4. Each column reports the estimation result for each province. Aside from the results for Saskatchewan, both coefficients for other provinces seem to be statistically insignificant. For Alberta and Ontario, their βˆ1 are significant at 5% and 10%, respectively, but with opposite signs. However, the fact that their βˆ2 are insignificant 30 Table 4: Placebo Tests by Implementing Fake Carbon Tax in Other Provinces lnL EIi × P rovincep × τt P rovincep × τt N R2 (1) AB (2) MB (3) NB (4) NS (5) ON (6) PE (7) SK 0.0122b (0.00591) -0.00166 (0.00401) -0.00398 (0.00733) 0.00185 (0.00411) 0.00712 (0.0128) -0.00549 (0.00509) -0.00586 (0.00996) -0.000106 (0.00540) -0.0144c (0.00838) 0.00288 (0.00474) -0.000117 (0.0203) -0.00334 (0.00853) 0.0158a (0.00555) -0.00698b (0.00319) 5100 0.887 5100 0.887 5100 0.887 5100 0.887 5100 0.887 5100 0.887 5100 0.887 c p < 0.1, b p < 0.05, a p < 0.01 Clustered standard errors in parentheses Note: Provincial abbreviation is as follows: Alberta(AB), Manitoba(MB), New Brunswick(NB), Nova Scotia(NS), Ontario(ON), Prince Edward Island(PE), and Saskatchewan(SK). All specifications include time FE, industry FE, province FE, industry × time FE. in both provinces means that there is no province-level employment effect of their carbon taxes. In other words, even though the employment effect of a placebo carbon tax is different across industries, the employment effect might be negligible as a whole. The only province that shows the different result is Saskatchewan. However, as this is the only province whose coefficients are both statistically significant, this result might not be compelling enough to conclude that my econometric design is not robust. In addition to checking the reliability of my econometric design by the placebo tests, I would argue that no statistically significant employment effects from the placebo carbon tax in other provinces could serve as an evidence for no leakage effect from the BC carbon tax. Another concern that may arise from estimating the employment effect of the BC carbon tax is that the analysis does not take into account any general equilibrium effects across provinces. More concisely, job reallocations from BC to other provinces are assumed to be zero in the analysis. Due to the stricter environmental regulation in BC, some firms in an emission intensive industry like cement and concrete product manufacturing might relocate to other provinces. Then job losses from this industry in BC would mean job gains for the same industry in other provinces. If the leakage effects are large, then the placebo carbon tax in other provinces should capture these effects. However, this is not the case here unless 31 all the possible reallocations happened in Saskatchewan. 5. Discussion The positive aggregate employment effect could arise by a rightward shift of either labor demand or labor supply. As my empirical strategy exploited the variation on industry emission intensities, the positive employment effect is likely to be from the rightward shift of labor demand. If that is the case, the equilibrium wage should be increased. To verify this conjecture, the effect of the BC carbon tax on wage is investigated by a simple regression13 . Table 5 shows that the imposition of the BC carbon tax has a statistically significant negative effect on provincial wage. It reduces average hourly and weekly wages by 3% at the rate of $10/t CO2 e. Although the results might be biased due to the omitted variables that cannot be controlled by time and province fixed effects, the significant negative wage effect negates the conjecture that the positive aggregate employment effect is from the shift of labor demand alone. In fact, the negative wage effects confirm the existence of an employment dividend as wages would only go down by increases in labor supply when labor demand also increased. This implies that the positive aggregate employment effect is the result of the rightward shifts of both labor demand and supply together. The wages declined because the shift of labor supply was larger than that of labor demand. 6. Conclusion This paper provides new evidence that a market-based environmental regulation can generate positive effects in labor markets. As the existing literature fails to provide clear answers to whether an implementation of an unilateral environmental regulation brings positive or 13 After I collected wage data (hourly and weekly) from CANSIM, I regressed the natural log of average hourly wages on carbon tax (i.e. BCp × τt from (4.1)), reported in the column (1) in Table 5 and the natural log of average weekly wages on carbon tax, reported in the column (2). 32 Table 5: Estimated Regression Models for the Percentage Changes in Wages (1) hourly (2) weekly Carbon Tax -0.00303a (0.000695) -0.00290a (0.000796) Observations R-squared 108 0.981 108 0.977 ln(Wage) Clustered Robust standard errors in parentheses a p<0.01, b p<0.05, c p<0.1 Note: Both specifications include time FE and province FE. negative effect to its economy, this paper finds that the implementation of the carbon tax in BC leads to both a decline and an increase in employment across industries depending on the level of emission intensity. Employment decreases for industries whose emission intensity is greater than 0.57 tonnes per thousand dollar of production and employment increases for industries whose emission intensity falls below the threshold. As there are many more industries whose emission intensity falls below the threshold, overall employment effect of the carbon tax is significantly positive. The preferred result suggests that, on average, the implementation of the carbon tax generated 33,000 net job gains annually in BC from 2008 to 2012, which is about 2.5% increase in aggregate employment in BC during the period. Although these estimates are not outcomes from a randomized natural experiments, and therefore cannot strictly be interpreted as a causal effect, they provide fairly robust evidence that a carbon tax does not provide an adverse employment effect in a regulated region. As the latest report (Elgie and McClay, 2013) claimed that the implementation of the carbon tax did not bring any apparent adverse impact on BC’s economy, this paper adds another perspective into the evaluation of the BC carbon tax because the report only focused on the impact on BC’s GDP. To my knowledge, this paper is the first to estimate the ex post effect of the BC carbon tax on its labor market. As the structure of a carbon tax can take many different forms in terms of the rate, the coverage of tax-based, and the treatment of 33 the tax revenue, it would bring fruitful contribution to the literature if the implementations of the carbon tax in other regions are also investigated empirically. 34 References “Factbox: Carbon taxes around the world,” 2013. http://www.sbs.com.au/news/article/2013/10/29/factbox-carbon-taxes-around-world K. Chay and M. Greenstone, “Air quality, infant mortality, and the Clean Air Act of 1970,” 2003. J. Currie and M. Neidell, “Air pollution and infant health: What can we learn from California’s recent experience?” The Quarterly Journal of Economics, vol. 120, no. 3, pp. 1003–1030, 2005. “B.C. carbon tax cut fuel use, didn’t hurt economy,” CBC News, 2013. http://www.cbc.ca/news/canada/british-columbia/ b-c-carbon-tax-cut-fuel-use-didn-t-hurt-economy-1.1309766 S. Elgie and J. McClay, “BCs Carbon Tax Shift Is Working Well after Four Years (Attention Ottawa),” Canadian Public Policy, vol. 39, no. s2, pp. S1–S10, Aug. 2013. “Quebec to collect nation’s 1st carbon tax,” 2007. http://www.cbc.ca/news/canada/ montreal/quebec-to-collect-nation-s-1st-carbon-tax-1.684888 Alberta Environment and Sustainable Resource Development, “Climate Change and Emissions Management Fund,” 2014. http://esrd.alberta.ca/focus/ alberta-and-climate-change/regulating-greenhouse-gas-emissions/ climate-change-and-emissions-management-fund.aspx “The Macroeconomic Impact of Federal Pollution Control Programs–1981 Assessment,” The Environmental Protection Agency Data Resources, Inc, Tech. Rep., 1981. E. Berman and L. T. Bui, “Environmental regulation and labor demand: evidence from the South Coast Air Basin,” Journal of Public Economics, vol. 79, no. 2, pp. 265–295, Feb. 2001. R. Bezdek, R. M. Wendling, and J. Jones, “The economic and employment effects of investments in pollution abatement and control technologies,” Ambio, vol. 18, no. 5, pp. 274–279, 1989. R. H. Bezdek, “Environment and Economy,” Environment: Science and Policy for Sustainable Development, vol. 35, no. 7, pp. 6–32, Sep. 1993. R. D. Morgenstern, W. A. Pizer, and J.-S. Shih, “Jobs Versus the Environment: An Industry-Level Perspective,” Journal of Environmental Economics and Management, vol. 43, no. 3, pp. 412–436, May 2002. R. Wendling and R. Bezdek, “Acid rain abatement legislationCosts and benefits,” Omega, vol. 17, no. 3, pp. 251–261, Jan. 1989. 35 K. Hollenbeck, “The employment and earnings impacts of the regulation of stationary source air pollution,” Journal of Environmental Economics and Management, vol. 6, pp. 208–221, 1979. M. Hazilla and R. Kopp, “Social cost of environmental quality regulations: A general equilibrium analysis,” Journal of Political Economy, vol. 98, no. 4, pp. 853–873, 1990. M. Greenstone, “The Impacts of Environmental Regulations on Industrial Activity : Evidence from the 1970 and 1977 Clean Air Act Amendments and the Census of Manufactures,” Journal of Political Economy, vol. 110, no. 6, pp. 1175–1219, 2002. L. Goulder and A. Schein, “Carbon Taxes Versus Cap And Trade: A Critical Review,” Climate Change Economics, 2013. N. Rivers and B. Schaufele, “Carbon Taxes, Agricultural Competitiveness and Trade,” 2013. Ministry of Finance, “Carbon Tax Review,” Ministry of Finance, British Colubia, Tech. Rep., 2013. K. Harrison, “A tale of two taxes: The fate of environmental tax reform in Canada,” Review of Policy Research, vol. 29, no. 3, pp. 383–407, 2012. D. G. Duff, “Carbon Taxation in British Columbia,” Vermont Journal of Environmental Law, vol. 10, pp. 87–108, 2008. Ministry of Finance, “BALANCE BUDGET 2008,” 2008. http://www.bcbudget.gov.bc.ca/2008/backgrounders/backgrounder carbon tax.htm Ministry of Small Business and Revenue, “British Columbia Carbon Tax,” Ministry of Small Business and Revenue, Tech. Rep., 2008. http://www.rev.gov.bc.ca/documents library/notices/British Columbia Carbon Tax.pdf Ministry of Finance, “Carbon Tax Report and Plan,” Ministry of Finance, British Columbia, Tech. Rep., 2014. L. H. Goulder, “Environmental taxation and the double dividend: a reader’s guide,” International Tax and Public Finance, vol. 2, pp. 157–183, 1995. A. Bovenberg and L. H. Goulder, “Environmental Taxation and Regulation,” Handbook of Public Economics, vol. 3, pp. 1471–1545, 2002. A. L. Bovenberg, “Green Tax Reforms and the Double Dividend: an Updated Reader’s Guide,” International Tax and Public Finance, vol. 6, pp. 421–443, 1999. United State Census Bureau, “North American Industry Classification System (NAICS).” http://www.census.gov/eos/www/naics/index.html Statistics Canada, “CANSIM - Canadian socioeconomic database from Statistics Canada,” Aug. 2014. http://www5.statcan.gc.ca/cansim/home-accueil?lang=eng 36 Appendices Appendix A Proof for Theory Elasticity of Substitution in terms of Cost Function Show c c12 =σ c1 c2 From cost function, c(w1 , w2 ), conditional input demand functions can be found by Shepard’s lemma, c1 = l and c2 = e. Then I can write a following equation as follows: l = ln l − ln e e = ln c1 (w1 , w2 ) − ln c2 (w1 , w2 ) ln (A.1) As a cost function is homogeneous degree of 1, input demand function is homogeneous degree of 0. Then I can rewrite (A.1) as follows: ln l w2 w2 = ln c1 1, − ln c2 1, e w1 w1 (A.2) Then take derivative with respect to w2 /w1 , w2 2 ) c22 (1, w ) c12 (1, w d ln el w1 1 = − w2 d w1 c1 c2 (A.3) w2 As ci (w1 , w2 ) = ci 1, w ∀ i = 1, 2, differentiate both sides with respect to w2 yields 1 w2 1 ci2 (w1 , w2 ) = w1 ci2 1, w1 ∀ i = 1, 2. Then by using these in (A.3), d ln el w1 c12 (w1 , w2 ) w1 c22 (w1 , w2 ) − w2 = d w1 c1 c2 Multiplying both sides of (A.4) by w2 /w1 yields, d ln el w2 w1 c12 (w1 , w2 ) w1 c22 (w1 , w2 ) = − 2 w1 c1 c2 d ln w w1 (A.4) (A.5) As an input demand function is homogeneous degree of 0, the following holds. w1 c21 + w2 c22 = 0 (A.6) Then by using (A.6) in (A.5), d ln el c21 −w2 c22 w1 c22 = − − 2 c c c2 d ln w 22 1 w1 −(w2 c2 + w1 c1 ) = −c21 c1 c2 37 (A.7) Lastly, by one of the properties of cost function, homogeneous degree of 1, and the symmetry of second derivative by Young’s theorem, σ= Appendix B c12 c c1 c2 (A.8) Data Table 6: Industry list with NAICS code NAICS 41 44 61 71 72 113 115 213 313 315 316 321 323 324 331 332 333 337 339 481 482 483 484 485 487 491 493 511 512 532 561 562 621 622 811 812 813 2111 2121 2122 Industry Wholesale trade Retail trade Educational services (except universities) Arts, entertainment and recreation Accommodation and food services Forestry and logging Support activities for agriculture and forestry Support activities for mining and oil and gas extraction Textile and textile product mills Clothing manufacturing Leather and allied product manufacturing Wood product manufacturing Printing and related support activities Petroleum and coal product manufacturing Primary metal manufacturing Fabricated metal product manufacturing Machinery manufacturing Furniture and related product manufacturing Miscellaneous manufacturing Air transportation Rail transportation Water transportation Truck transportation Transit and ground passenger transportation Scenic and sightseeing transportation and support activities for transport Postal service and couriers and messengers Warehousing and storage Publishing industries, information services and data processing service Motion picture and sound recording industries Rental and leasing services and lessors of non-financial intangible associations Administrative and support services Waste management and remediation services Health care services (except hospitals) and social assistance Hospitals Repair and maintenance Personal and laundry services and private households Grant-making, civic, and professional and similar organizations Oil and gas extraction Coal mining Metal ore mining Continued on next page 38 NAICS 2123 2211 2212 2361 2362 3111 3112 3113 3114 3115 3116 3117 3221 3222 3251 3252 3253 3254 3255 3261 3262 3271 3273 3341 3342 3351 3352 3361 3362 3363 3364 3365 3366 5151 5152 5221 5222 5241 5311 5413 5418 5419 6113 Industry Non-metallic mineral mining and quarrying Electric power generation, transmission and distribution Natural gas distribution, water and other systems Residential building construction Non-residential building construction Animal food manufacturing Miscellaneous food manufacturing Sugar and confectionery product manufacturing Fruit and vegetable preserving and specialty food manufacturing Dairy product manufacturing Meat product manufacturing Seafood product preparation and packaging Pulp, paper and paperboard mills Converted paper products manufacturing Basic chemical manufacturing Resin, synthetic rubber, and artificial and synthetic fibres and filaments manufacturing Pesticides, fertilizer and other agricultural chemical manufacturing Pharmaceutical and medicine manufacturing Miscellaneous chemical product manufacturing Plastics product manufacturing Rubber product manufacturing Miscellaneous non-metallic mineral product manufacturing Cement and concrete product manufacturing Computer and peripheral equipment manufacturing Electronic product manufacturing Electrical equipment and component manufacturing Household appliance manufacturing Motor vehicle manufacturing Motor vehicle body and trailer manufacturing Motor vehicle parts manufacturing Aerospace product and parts manufacturing Railroad rolling stock manufacturing Ship and boat building Radio and television broadcasting Pay television, specialty television and program distribution and telecommunications Depository credit intermediation Non-depository credit intermediation Insurance carriers Lessors of real estate Architectural, engineering and related services Advertising, public relations, and related services Other professional, scientific and technical services Universities 39 Appendix C Additional Results Table 7: Industry specific employment effects at $10/t CO2 e with 95% Confidence Interval NAICS 41 44 61 71 72 113 115 213 313 315 321 323 324 331 332 333 337 339 481 482 484 485 487 491 493 511 512 532 561 562 621 622 811 812 813 2111 2211 2361 2362 3111 3112 3113 3114 3115 3116 Industry α Std Error Wholesale trade Retail trade Educational services (except universities) Arts, entertainment and recreation Accommodation and food services Forestry and logging Support activities for agriculture and forestry Support activities for mining and oil and gas extraction Textile and textile product mills Clothing manufacturing Wood product manufacturing Printing and related support activities Petroleum and coal product manufacturing Primary metal manufacturing Fabricated metal product manufacturing Machinery manufacturing Furniture and related product manufacturing Miscellaneous manufacturing Air transportation Rail transportation Water transportation Transit and ground passenger transportation Scenic and sightseeing transportation Postal service and couriers and messengers Warehousing and storage Publishing industries and information services Motion picture and sound recording industries Rental and leasing services Administrative and support services Waste management and remediation services Health care services (except hospitals) and social assistance Hospitals Repair and maintenance Personal and laundry services and private households Grant-making, civic, and professional organizations Oil and gas extraction Electric power generation, transmission and distribution Residential building construction Non-residential building construction Animal food manufacturing Miscellaneous food manufacturing Sugar and confectionery product manufacturing Fruit and vegetable preserving and food manufacturing Dairy product manufacturing Meat product manufacturing 0.0436b 0.0512b 0.0512b 0.0527b 0.0270c -0.0064 -0.0276 0.0133 0.0012 0.0421b 0.0088 0.0315b -0.1215b -0.0578c 0.0164 0.0345b 0.0360b 0.0300b -0.1336b -0.0518c -0.0821b -0.0260 0.0315b 0.0285b 0.0542a 0.0603a 0.0330b 0.0482b 0.0618a 0.0164 0.0572a 0.0588a 0.0421b 0.0572a 0.0648a -0.0790b -0.3910b 0.0254c 0.0330b -0.0684b -0.0321 0.0209 -0.0230 -0.1366b -0.1230b 0.0179 0.0201 0.0201 0.0206 0.0140 0.0144 0.0198 0.0126 0.0132 0.0175 0.0126 0.0149 0.0537 0.0300 0.0127 0.0155 0.0159 0.0146 0.0583 0.0278 0.0388 0.0193 0.0149 0.0143 0.0210 0.0230 0.0152 0.0192 0.0235 0.0127 0.0220 0.0225 0.0175 0.0220 0.0245 0.0377 0.1587 0.0137 0.0152 0.0338 0.0212 0.0131 0.0184 0.0595 0.0543 c p < 0.1, b p < 0.05, a p < 0.01 95% C.I. 0.0086 0.0118 0.0118 0.0124 -0.0005 -0.0345 -0.0663 -0.0113 -0.0246 0.0079 -0.0159 0.0024 -0.2267 -0.1166 -0.0086 0.0041 0.0049 0.0015 -0.2479 -0.1063 -0.1582 -0.0639 0.0024 0.0005 0.0130 0.0152 0.0032 0.0106 0.0157 -0.0086 0.0141 0.0147 0.0079 0.0141 0.0168 -0.1529 -0.7021 -0.0015 0.0032 -0.1347 -0.0736 -0.0048 -0.0592 -0.2532 -0.2294 0.0786 0.0906 0.0906 0.0930 0.0544 0.0218 0.0112 0.0380 0.0270 0.0763 0.0335 0.0606 -0.0162 0.0009 0.0413 0.0650 0.0672 0.0585 -0.0192 0.0027 -0.0060 0.0118 0.0606 0.0564 0.0955 0.1054 0.0628 0.0858 0.1079 0.0413 0.1004 0.1029 0.0763 0.1004 0.1129 -0.0052 -0.0799 0.0524 0.0628 -0.0022 0.0094 0.0466 0.0131 -0.0200 -0.0166 Continued on next page 40 NAICS 3117 3221 3222 3251 3254 3255 3271 3273 3341 3342 3351 3362 3364 3366 5152 5221 5222 5241 5311 5413 5418 5419 6113 Industry α Std Error Seafood product preparation and packaging Pulp, paper and paperboard mills Converted paper products manufacturing Basic chemical manufacturing Pharmaceutical and medicine manufacturing Miscellaneous chemical product manufacturing Miscellaneous non-metallic mineral product manufacturing Cement and concrete product manufacturing Computer and peripheral equipment manufacturing Electronic product manufacturing Electrical equipment and component manufacturing Motor vehicle body and trailer manufacturing Aerospace product and parts manufacturing Ship and boat building Pay television distribution and telecommunications Depository credit intermediation Non-depository credit intermediation Insurance carriers Lessors of real estate Architectural, engineering and related services Advertising, public relations, and related services Other professional, scientific and technical services Universities -0.0018 -0.0624b 0.0042 -0.1684b 0.0466b -0.0124 -0.1139b -0.2153b 0.0557a 0.0527b 0.0224c 0.0224c 0.0557a 0.0270b 0.0679a 0.0663a 0.0542a 0.0679a 0.0285b 0.0648a 0.0618a 0.0633a 0.0482b 0.0136 0.0316 0.0129 0.0718 0.0187 0.0156 0.0508 0.0900 0.0215 0.0206 0.0133 0.0133 0.0215 0.0140 0.0255 0.0250 0.0210 0.0255 0.0143 0.0245 0.0235 0.0240 0.0192 c p < 0.1, b p < 0.05, a p < 0.01 41 95% C.I. -0.0284 -0.1243 -0.0210 -0.3091 0.0099 -0.0431 -0.2135 -0.3918 0.0136 0.0124 -0.0037 -0.0037 0.0136 -0.0005 0.0178 0.0173 0.0130 0.0178 0.0005 0.0168 0.0157 0.0163 0.0106 0.0248 -0.0005 0.0295 -0.0277 0.0834 0.0183 -0.0143 -0.0389 0.0979 0.0930 0.0485 0.0485 0.0979 0.0544 0.1179 0.1154 0.0955 0.1179 0.0564 0.1129 0.1079 0.1104 0.0858 Figure 6: Industry Specific Employment Effects with 95% C.I. 42 Source: Author’s calculation Figure 7: Industry Specific Employment Effects vs. Emission Intensity Source: Author’s calculation 43 Figure 8: Changes in Employment at $10/t CO2 e 44 Source: Author’s calculation