The Impact of the European Emission Trading Scheme on Multiple Measures of Economic Performance

Abstract

The European Emission Trading Scheme (EU ETS) has introduced a price for carbon, thus generating an additional cost for companies that are regulated by the scheme. The objective of this paper is to provide empirical evidence on the effect of the EU ETS on firm-level economic performance. There is a growing body of empirical literature that investigates the effects of the EU ETS on firm economic performance, with mixed results. Differently from the previous literature, we test the effect of the EU ETS on a larger set of indicators of economic performance: employment, average wages, turnover, value added, markup, investment, labour productivity, total factor productivity and ROI. Our results, based on a large panel of European firms, provide a broad picture of the economic impact of the EU ETS in its first and second phases of implementation. Contrarily to the expectations, the EU ETS did not affect economic performance negatively. Results suggest that firms have reacted to the EU ETS by passing-through costs to their customers on the one hand and improving labour productivity on the other hand.

Appendix A: Estimation of Markup and Total Factor Productivity

Market power is the ability of a firm to profitably raise the market price of a good or service over the marginal cost of production (i.e. in charging a markup on marginal costs). As discussed in De Loecker (2011), the empirical search for estimates of markup has mostly focused on estimating demand side components for markup. It has been noted, however, that the “demand approach” to markup is extremely demanding in terms of data requirement (consumers’ preferences, market prices, quantities sold and bought, information on production technology of firms, etc.) and is characterized by methodological difficulties when estimating demand elasticity and cross-elasticities.

De Loecker (2011) and De Loecker and Warzynski (2012) propose a “production approach” to recover firm-level and time-varying markup, starting from firm-level production data and the mild assumption of cost minimization. In presence of two types of production inputs—variable inputs that freely adjust in the short term (e.g. materials and labour) and dynamic inputs that can be considered fixed in the short term (e.g. fixed capital)—it is possible to derive a marginal cost function for the variable inputs for given levels of the dynamic inputs. De Loecker and Warzynski (2012), by simply assuming cost minimization, derive the markup $\mu _{it}$ for firm i in time t. The markup results then in being equivalent to the wedge between the elasticity of output with respect to the variable input (in their case, labour), $\theta _{it}^L$, and the cost share over total revenues of the same input, $\alpha _{it}^L$:

$$\begin{aligned} \mu _{it} =\theta _{it}^L /\alpha _{it}^L \end{aligned}$$

(A1)

The elasticity is calculated from estimates of the production function (starting from production data) under the only assumption of cost minimization, whereas cost share of labour can be retrieved directly from the P&L account of firms.

We recover estimates of the elasticity of output with respect to input by estimating sector-specific (capital letters of the NACE Rev. 2 classification, 2 capital letters for manufacturing sectors) translog production functions. De Loecker and Warzynski (2012) suggest using a translog production function rather than a Cobb–Douglas production function to allow the elasticity to vary across firms, otherwise a constant elasticity for all firms would result in measures of markup that will solely depend on the cost share of employees. We estimate the following translog production function:

$$\begin{aligned} \log \left( {VA_{it} } \right)= & {} \gamma _{jt} +\phi _s +\beta ^{1}\log \left( {L_{it} } \right) +\beta ^{2}\left[ {\log \left( {\hbox {L}_{\hbox {it}} } \right) } \right] ^{2}+{\upbeta } ^{3}\log \left( {K_{it} } \right) +\beta ^{4}\left[ {\log \left( {K_{it} } \right) } \right] ^{2}\nonumber \\&+\,\beta ^{5}\log \left( {L_{it} } \right) \times \log \left( {K_{it} } \right) +\omega _{it} +\epsilon _{it} \end{aligned}$$

(A2)

where $\log \left( {VA_{it} } \right)$ is the log of gross value added in euro for firm i in year $t, \log \left( {L_{it} } \right)$ is the log of total employees, $\log \left( {K_{it} } \right)$ is the log of total fixed assets in euro, $\omega _{it}$ is total factor productivity and $\epsilon _{it}$ is the idiosyncratic error term.

Table 14 Estimated elasticities of value added with respect to labour and capital and estimated markup (by sector, NACE rev 2)

To consistently estimate the production function as well as total factor productivity and the idiosyncratic error term (that is required for the calculation of the markup), we adopt the estimator proposed by Ackerberg et al. (2006). Following Ackerberg et al. (2006) and De Loecker and Warzynski (2012), we first estimate a pooled OLS regression in which log value added is a non-parametric function of labour and capital. This is needed to retrieve the estimate of the idiosyncratic error from the production function. We then estimate with generalized method of moment the translog production function, where we use ’potential value added’ (partialled out of the idiosyncratic error component) as dependent variable and where we instrument the flexible input (labour), in all its interactions, with its lag. From this estimate, as a residual, we retrieve our measure of total factor productivity.¹⁵

Following De Loecker and Warzynski (2012), to estimate markup we first need to estimate firm-specific elasticity of value added with respect to labour is given by:

$$\begin{aligned} \frac{\partial \log \left( {VA_{it} } \right) }{\partial \log \left( {L_{it} } \right) }=\widehat{\theta }_{it}^L =\widehat{\beta ^{1}}+2\times \widehat{\beta ^{2}}\times \log \left( {L_{it} } \right) +\widehat{\beta ^{5}}\times \log \left( {K_{it} } \right) \end{aligned}$$

(A3)

As pointed out by De Loecker and Warzynski (2012, p. 2449), what we observe is not $VA_{it}$ but ${\widetilde{VA_{it}} } =VA_{it} /\exp \left( \hat{\epsilon }_{it}\right)$ where $\hat{\epsilon }_{it}$ is the idiosyncratic error term component of the production function. The share of labour costs over total revenues to be inserted into equation 1 therefore is (we refer to equation 16 of De Loecker and Warzynski 2012):

$$\begin{aligned} \hat{\alpha } _{it}^L =\frac{P_{it}^L L_{it} }{VA_{it} /\hbox {exp}\left( \hat{\epsilon }_{it} \right) } \end{aligned}$$

(A4)

with $P_{it}^L L_{it}$ being unitary labour cost, $VA_{it}$ being total observed gross value added and $\hat{\epsilon } _{it}$ being the idiosyncratic error term of the estimated production function.

Table 14 reports some descriptive statistics (by sector, NACE rev 2) about our estimates. We report average and median elasticities of value added with respect to, respectively, labour and capital, as well as average and median estimated markup.