UNIVERSITY OF GRONINGEN FACULTY OF ECONOMICS AND BUSINESS
Accounting for CO
2
emissions in
the measurement of productivity
performance
MSc/MA Thesis
Annechina Harmina Schipper S2763575
Abstract
This thesis extends the traditional multifactor productivity framework to account for CO2 emissions in an effort to offer a more reliable metric of the efficiency with which resource inputs are employed in production activity. This framework rests on the estimation of the implicit price of CO2 emissions otherwise not observable due to the absence of markets for this sort of bad output. The framework is applied to a wide range of developed and developing nations representing 56% of global GDP over the 1995-2007 period. While accounting for CO2 emissions leads to a downgrade of multifactor productivity performance, the results are subject to a large degree of uncertainty which points to the need to enhance the modelling strategy.
Table of Contents
1. Introduction ... 3 2. Literature Review ... 6 3. Framework ... 11 3.1. Measurement Framework... 11 3.1.1 Shadow values ... 12 3.1.2 Multifactor Productivity ... 13 3.2. Econometric Implementation ... 15 3.2.1 Regularity Conditions ... 17 3.3 Source Data ... 18 3.3.1 Variables ... 19 3.3.2 Industries ... 20 3.4 Trend Analysis ... 21 3.5 Descriptive Statistics ... 24 3.5.1 Diagnostics checks ... 25 4. Empirical Results ... 26 4.1 Hypothesis testing ... 27 4.2 Robustness ... 27 4.3 Interpretation ... 284.3.1 Effects of CO2 emissions on the multifactor productivity framework ... 28
4.3.2 Do firms care about CO2 emissions and corporate social responsibility? .. 33
5. Conclusion ... 34
6. References ... 36
Appendix A ... 42
Appendix B ... 54
1. Introduction
The long-term global pattern in economic development in the course of the recent 200 years has been unambiguously upwards, notwithstanding short-term downturns and setbacks. Developed nations experienced remarkable growth during the post-war period, originating from the wave of innovations through specific sectors (Hulten, 2009; Maddison, 1997). Although the long-term growth was not spread evenly, it has been steady over time. One of the most striking features of the global growth in recent years has been the increase in the role played by developing economies.
Through most of the twentieth century only high-income developed economies, representing one-fifth of the world population (World Bank, 2015a), enjoyed the fruits of economic growth. However, developing nations have transformed over the past thirty-five years from technologically backward and poor, to relatively modern and prosperous economies. The emergence of China and India as major forces in the global economy has been one of the most significant economic developments of the past quarter century (Bosworth and Collins, 2007). In addition, some of the East Asians nations experienced more than a fourfold increase of economic growth. In contrast it took the United Kingdom, the U.S., and Germany more than eighty years to achieve such growth. The growth performance of these developing economies has vastly exceeded those of virtually all other economies that had comparable productivity and income levels half a century ago (Nelson and Pack, 1999).
Economists have undertaken the task to increase the understanding of economic growth and sort out the relative importance of factors that drive this long-term growth path of nations. The growth accounting framework, developed in the late 1950s by Robert Solow (1957), became the primary empirical tool in tracking the sources of economic growth1. In the two decades that followed, the framework was refined and set the stage for incorporating multifactor productivity as an official economic statistic (Arnaud, Dupont, Koh and Schreyer, 2011) 2. Most of the developed nations had already implemented numerous multifactor productivity programs by the time that the System of National Accounts (SNA, United Nations) endorsed the construction of a production account suitable for multifactor productivity measurement in 2008. The
1
Growth-accounting breaks down growth of output into the growth of factors of production (e.g. capital, labour), relating the aggregate value of final output to the total value of the used input factors (Solow, 1957; Hulten, 2009). 2
international efforts to develop a consistent set of production accounts have now been extended to emerging economies as part of the World KLEMS initiative.
Although over time productivity accounts have addressed some longstanding gaps as economists attempted to record and quantify the sources of economic growth, it became clear that other important sources of economic growth remained missing (Vouvaki and Xeapapadeas, 2008; Brandt, Schreyer and Zipperer, 2013; Brandt, Schreyer and Zipperer, 2014). The case of the environment constitutes a compelling case in point. With the production of goods and services (“good output”) from a given set of inputs, nations also produce undesirable externalities, such as pollution (“bad output”). Nonetheless the conventional productivity metric rewards an economy for producing more good output but not for less bad output. As pointed out by Ball, Färe, Grosskopf and Zaim (2005) and Repetto, Rothman, Faeth and Austin (1997), ignoring these different performance indicators will most likely provide distortions in cross-country analysis of relative economic performance. Striking differences in the production of bad output exist between nations of similar levels of development but also and perhaps even more essential, between nations of different levels of development.
Of the numerous categories of bad outputs, greenhouse gases (GHG) are those on which an intense debate is focused due largely to their potential devastating effects. Among scientists, policy makers and business leaders consensus is growing that concrete action is needed to face rising GHG emissions, as they encounter challenges of where and how emissions reductions can best be achieved, at what costs, and over what periods of time (McKinsley and Company, 2007).
between nations are gaining ground. Nonetheless doubts still remain about their effectiveness (Earnhart, Khanna, and Lyon, 2014; Kozluk and Zipperer, 2015).
Attention has shifted to Corporate Social Responsibility (CSR), as in recent years there has been an increased interest in the association between measures of environmental performance and traditional measures of firm performance such as profitability and stock price (Färe et al, 2006; Khanna, 2001; Paton and Siegel, 2005). CSR provides businesses with incentives to control their emissions on their own terms due to an increase in demand for environmental transparency from stakeholders and an increase in demand for environmental quality by the consumers (Harrison, Hyman, Martin, and Nataraj, 2015; Khanna, 2001; Nishitani et al, 2014).
My paper builds on and extends various strands of active literature, focussing on the interplay between the environment and economic performance and the competing line of research that has a long-standing tradition in accounting for unpriced environmental effects. Through this paper I want to further bridge the gap between the two perspectives by establishing a link between productivity and measures of environmental performance as the current literature is still incomplete.
This paper examines the empirical support of such an argument by asking the following two specific questions: What is the impact of accounting for emissions on the productivity performance? Do businesses care about emissions despite the absence of any form of regulation of CO2 emissions? To answer these questions, I formulate a cost-function characterization of sectoral production of a wide range of economies. I represent sectoral production costs augmented with unpriced CO2 emissions. Their shadow values, which are inferred from the cost benefits that generate additional emissions of CO2, may be interpreted as the marginal benefits of being able to use the environment freely. These shadow values are then utilized to adjust the standard measure of productivity unpriced environmental CO2 impacts.
2. Literature Review
My paper builds on and extends various strands of active literature. One aspect of the literature is the interplay between the environment and economic performance, highlighting along the way the importance of the related shadow value. Examples of this approach include Halvorsen and Smith (1984, 1986) and Lasserre and Ouellette (1991) who estimated shadow values of the stock of depletable resources. Another line of research represented by Morrison Paul, Ball, Felthoven, Grube and Nehring (2002) estimated the shadow value of unpriced damages such as bad output. Either approach exploited an econometric analysis to estimate otherwise unobservable prices, an approach that I pursue in my own paper.
A competing line of research, rooted in a large class of stochastic frontier estimation and data envelopment models, has a long-standing tradition in accounting for unpriced environmental effects (e.g. Färe, 1975). In using a joint-product specification of a cost function to assess how multifactor productivity is affected by bad outputs, this paper is also related to Ball et al. (2005) which examined the same question, investigated by the Malmquist cost productivity index. Their approach does not reply on the estimate of a shadow value while it represents a central element of this paper’s theoretical underpinning, namely accounting for the environment in the multifactor productivity framework by demonstrating how productivity growth can be amended to account for non-traditional output, such as negative externalities.
Muller, Mendelsohn and Nordhaus (2011) took the approach of accounting for the environment one step further by developing a framework to include environmental externalities into a system of national accounts. They compared the marginal damages of air pollution to the value added per industry by measuring the marginal external damages (price of pollution, in terms of pollution permits or pollution taxes) times the quantity of pollution at each source location. They find that the value added per industry is biased and should be adjusted for the external damages.
Their procedure allows to model joint production of good output and bad output without requiring data on shadow values of the externality and to specify an applied measure of improved multifactor productivity which can be used as a benchmark for corporate social behaviour. They find that measures of productivity growth that ignore environmental externalities are biased upward when the production of bad outputs is increasing; conversely, the measure is biased downward when the production of bad output is decreasing.
In line with this research, Vouvaki and Xeapapadeas (2008) show that when a factor of production generates an environmental externality, using CO2 emissions as a proxy, that a part of the multifactor productivity estimates regarded as technological contribution could be attributed to the use of environment in the output production. This is because the environmental externality is not internalized due to the lack of environmental policy.
In contrast, Brandt et al. (2014) find that adjusting the traditional productivity growth measure for bad outputs, considered as CO2, SOx and NOx, leads to small differences, by using a productivity growth measure that implements shadow values for bad outputs. Their model accounts for the environment through natural capital as an input factor and through bad outputs as an output of the production process. In addition, they find that the growth rate of emissions was lower than the GDP growth, implying that traditional multifactor productivity growth measures underestimate productivity growth because they do not account for the fact that countries have employed some inputs to reduce emissions growth rather than increase GDP growth.
While current economic literature has paved the way to account for the environment in the productivity framework, the gap between economic and socio-economic perspective on environmental performance still exists. This leads to the second building block of my literature, the socio-economic perspective, represented by the Corporate Social Responsibility (CSR).
possible if firms were required to reduce environmental externalities (Morrison Paul et al., 2002). Despite the benefits of using the environment as a “free input” and the following environmental damages, that are costless, firms are increasingly adopting a more environmental friendly approach of doing business.
Firms of no longer ask how much environmental management costs but how much it benefits them. CRS provides firms with incentives to control their emissions on their own accord, resulting from an increased demand for transparency from stakeholders but also from consumer demand for environmental quality and products (Chapple, Morrison Paul, and Harris, 2004). Firms’ initiative to take responsibility for its effects on the environment and social well-being that are beyond what may be required by regulators, institutions or environmental protection groups, allows them to try to gain both reputational and market benefits from environmental-friendly behaviour. This attitude may possibly arise from sheer altruism but also, and more likely, in an attempt to gain both reputational and market benefits from environmental-friendly behaviour (Harrison et al. 2015; Khanna, 2001; Nishitani et al, 2014).
It is expected that voluntary incorporating environmental-friendly management is more flexible and effective in reducing bad outputs than direct and indirect regulations. For example, Khanna (2001) analyses environmental organizational measures from a more proactive approach involving voluntary and often ‘business-led’ initiatives to self-regulate their environmental performance. Firms have incentives for self-regulation if they can increase profits by differentiating their products or use their market power to charge premium prices for green products or increase their market value through superior environmental performance. In line with the findings from Khanna (2001) and Bloom, Genakos, Martiono and Sadun (2010), Nishitani et al. (2014) find that firms enhance their economic performance through increases in demand for green output and improvement of the productivity, when participating in emissions management.
and Environmental Sustainability Index, is that it considers the relation between the environment and the economy. It can be applied to measure the economic growth path in relation to less resource consumption and pollution, both being key variables of sustainable development.
Eco-efficiency has given the business sector the possibility to quantify and contribute to sustainable development while enhancing both economic and environmental benefits. While eco-efficiency is a useful tool for the business sector to achieve greater value with lower adverse environmental impacts, it should also be applied beyond the business sector and production patterns (Escap, 2009; Hellweg, Doka, Finnveden, and Hungerbühler, 2005). The government of Japan assessed its own eco-efficiency in terms of the environmental burden per unit of economic activity (CO2 per GDP), which was used as a yardstick in comparing the performance of other OECD countries (Tatsuo, 2010; Bleischwitz, 2002).
Development in the issue to account for the environment into the productivity framework and environmental management has resulted in research studies that aimed at understanding how to accurately calculate the price of CO2 emissions and if and how they reflect the economic costs of the environment (Berman and Bui, 2001; Earnhart et al., 2014; Khanna, 2001; Muller et al, 2011; Muller, 2016). One paper that establishes a first link between productivity and a measure of environmental performance is by Färe, Grosskopf and Pasurka (2006). By using an environmental performance index (EPI) they account for the production of both good and bad outputs, based on output-emission ratios to assess the performance of firms.
Through this paper I want to further bridge the gap between the economic and socio-economic perspective by establishing a link between productivity and measures of environmental performance as the current literature is still incomplete. For example, the measure of Färe et al. (2006) only requires information about good and bad output production, unlike adjusted measures of productivity that require data for inputs, good outputs, and bad outputs. My approach will also be related to Ball et al. (2005), their approach however does not reply on the estimate of a shadow value for unpriced externalities.
3. Framework
3.1. Measurement Framework
In this paper, I develop a basis for integrating the existing literature and extending the current models to account for the effects of bad outputs - CO2 emissions - on the multifactor productivity growth and the economic performance on a global scale. I focus on the private benefits for economic regions and industries for using the environment (measured through CO2 emissions) to produce higher levels of output, or to produce at lower input costs for a given amount of production. Therefore actions reducing CO2 emissions impose private costs on the industries in a market economy. If a producer assigns a negative value to bad outputs, the implied costs of bad output do not enter the firm’s total costs, although, as discussed further below, it may be derived from it.
Measuring the costs and benefits of CO2 emissions and the associated environmental damage (measured as CO2 emissions) involves unambiguously modelling the production structure, and hence distinguishing the cost structure through output (revenue) and input (cost) patterns exhibited in the data. The cost function approach has an advantage over the production function approach in that is provides estimates of productivity effects in terms of both cost-saving and output enhancing measures (Berndt, 1991; Morrison Paul, 1999).
The costs and benefits of using CO2 emissions and the associated environmental damage (measured as CO2 emissions) are recognized in the cost function model. The production of one good output Q and the associated bad output (CO2 emissions), represented by B in the equation, next to the use of three inputs, capital (K), labour (L) and intermediate inputs (M), are considered in the cost function, leading to the following equation (1):
𝐶 = 𝑓(𝑄, 𝑤𝑗, 𝐵, 𝐷, 𝑡)
where C is total costs, wj ( j = K, L, M) is a vector of the prices of inputs j, and t is a representation of the shift in technology which captures the multifactor productivity. For empirical implementation, this cost function is augmented by industry fixed effects to accommodate the differences across industries and time periods, through a fixed-effects vector D.
The total cost function as presented by equation (1) measures the total costs made by an industry of a specific economic region, in a specific period for all the factor inputs used to produce one unit of output. I consider a restricted cost function in which CO2 emissions are treated as a quasi-fixed input. The contribution of this extended cost function model is that it accounts for the production of good outputs Q to be produced with the use of several paid inputs, j, but also by the means of the environment, represented by B, or conversely, using emissions as an input allows industries to produce a given amount of Q at lower input costs at the extent of the environment. Hence, the model in this paper treats CO2 emissions as either an input of the production process or as an output whose negative value is not fully integrated into revenues.
3.1.1 Shadow values
My approach focuses on private production costs, and emissions are treated as an output whose negative value is not fully integrated into a firm’s revenues or as an input that allows the production of a given amount of Q at lower paid input costs. The variable B is included in the cost function as emissions are produced jointly with the good outputs Q. To achieve lower growth in CO2 emissions, other inputs factors or technological change must grow faster or good output growth must slow down relative to the bad output growth.
If the behaviour of a firm indicates that it assigns a negative value to bad outputs, such implicit cost do not enter the definition of C, although, it may be inferred from it. The associated private shadow values of the bad output 𝑍𝐵, namely the (input) cost saving on other inputs from allowing CO2 emissions, may be measured as the costs effects, −𝜕𝐶
𝜕𝐵= 𝑍𝐵. By estimating a shadow value, a proxy value for CO2 emissions, the value defines what a firm must give up to gain an extra unit of good output. Hence, the shadow value reflects the marginal amount a firm incurs as a result of a reduction in B. I therefore expect that 𝑍𝐵<0, because avoiding CO2 emissions results in lower good output growth or higher costs on other inputs. In this framework, the shadow values incorporate the behavioural motivations fundamental to cost-efficient production choices as well as possibilities for technological substitution through multifactor productivity.
respect to input prices, giving the demand for input j as Xj =𝜕𝑤𝜕𝐶
𝑗. By Sheppard’s
lemma, the effect of a change in B on the demand for input j (which is inferred as a second order effect) measures the dependence of input j on the ability to dispose CO2 emissions. The shadow value of emissions −
𝜕𝐶
𝜕𝐵= 𝑍𝐵 is a relationship parallel to Shepard’s lemma, as this function is a combination of all parameters of equation (1). It follows from Young’s theorem that the impact on 𝑍𝐵 of a change in price of input j is symmetric to the effect of a change in B on the demand for Xj3:
−𝜕𝑍𝐵 𝜕𝑤𝑗 = 𝜕 2𝐶 𝜕𝐵𝜕𝑤𝑗 = 𝜕 2𝐶 𝜕𝑤𝑗𝜕𝐵 =𝜕𝑋𝑗 𝜕𝐵
Shadow values are inferred from the cost benefits, from estimated elasticities of bad outputs with respect to the cost function characterization that result in additional emissions of CO2 which may be interpreted as the marginal benefits of a firm being able to use the environment freely (and not being socially responsible). Or, conversely, as the marginal cost firms would be willing to pay for unrestricted use of the environment (complying with CSR) in terms of the value of other input factors and services that are used to produce less bad output or the value of goods and services that could otherwise have been produced. The shadow values can be interpreted as reflecting the overall effect of CSR concerning the corresponding bad outputs. These shadow values are utilized to adjust the standard measure of productivity with unpriced environmental CO2 impacts.
3.1.2 Multifactor Productivity
Multifactor productivity growth and technical change coincide only under constant returns to scale (Brandt, Schreyer and Zipperer, 2013; Hulten, 2009). In order to accurately measure and decompose multifactor productivity growth information about the production structure derived in the previous section is required. Total differentiating the total costs function 𝐶 = 𝑓(𝑄, 𝑤𝑗, 𝐵, 𝐷, 𝑡) with respect to time, which can be interpreted as technical change,
𝑑𝐶 𝑑𝑡 = ∑ 𝜕𝐶 𝜕𝑤𝑗 3 𝑗=1 𝑑𝑤𝑗 𝑑𝑡 + 𝜕𝐶 𝜕𝑄 𝑑𝑄 𝑑𝑡 + 𝜕𝐶 𝜕𝐵 𝑑𝐵 𝑑𝑡 + 𝜕𝐶 𝜕𝑡 3
Dividing equation (3) by C and applying Shepard’s lemma generates a framework that allows insights into a variety of cost elasticities that capture the existence of economies of scale in the market economy of the different regions and the impacts of CO2 emissions: 𝑇̇ = 𝐶̇ 𝐶− ∑ 𝑤𝑗𝑋𝑗 𝐶 𝑤̇𝑗 𝑤𝑗 3 𝑗=1 −𝑌̇ 𝑇̇ = 𝜕𝐶 𝜕𝑡 1 𝐶 = 𝑡ℎ𝑒 𝑝𝑟𝑜𝑝𝑜𝑟𝑡𝑖𝑜𝑛𝑎𝑙 𝑠ℎ𝑖𝑓𝑡 𝑖𝑛 𝑡ℎ𝑒 𝑐𝑜𝑠𝑡 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛 𝐶̇ 𝐶 = 𝑑𝐶 𝑑𝑡 1 𝐶 = 𝑡ℎ𝑒 𝑡𝑜𝑡𝑎𝑙 𝑐ℎ𝑎𝑛𝑔𝑒 𝑖𝑛 𝑡ℎ𝑒 𝑐𝑜𝑠𝑡 𝑓𝑢𝑛𝑐𝑡𝑖𝑜𝑛
𝑌̇ measures the aggregated output growth Q, corrected for the growth in emissions, hence B is interpreted as an output and 𝑌̇ is the weighted sum of the market economy good output and the emissions output, Q and B.
𝑌̇ = 𝜀𝐶𝑄𝑄̇ + 𝜀𝐶𝐵𝐵 ̇ where 𝜀𝐶𝑄= 𝜕𝐶
𝜕𝑄 𝑄
𝐶 represents the cost elasticity of good output Q and 𝜀𝐶𝐵 = 𝜕𝐶 𝜕𝐵
𝐵 𝐶 the cost elasticity of bad output B.
Totally differentiating (6) with respect to time and dividing by C, yields (7)
𝐶 = ∑ 𝑤𝑗𝑋𝑗 3 𝑗=1 ∑𝑤𝑗𝑋𝑗 𝐶 𝑤̇𝑗 𝑤𝑗 = 𝑗 𝐶̇ 𝐶− ∑ 𝑤𝑗𝑋𝑗 𝐶 𝑋̇𝑗 𝑋𝑗 𝑗 Where ∑ 𝑤𝑗𝑋𝑗 𝐶 𝑋̇𝑗 𝑋𝑗
𝑗 = 𝐼 ̇, represents the weighted growth rate of all inputs. Substituting this into equation (4) and (5) yields:
Multifactor productivity is defined by standard application as 𝑀𝐹𝑃̇
𝑀𝐹𝑃 = 𝑌̇ − 𝐼̇, equating the sum of weighted outputs 𝑌̇ less the sum of weighted inputs 𝐼̇, which accounts for the growth of aggregate output Q not accounted for by the growth of combined inputs, j (Brandt, Schreyer and Zipperer, 2013; Hulten, 2009). This formula will be used to compute multifactor productivity and decompose the measure into two basic factors: a shift in the cost function resulting from technological change (−𝑇̇) and a movement along the cost function due to overall economies of scale 𝑌̇[1 − ( 𝜀𝐶𝑄𝑄̇ + 𝜀𝐶𝐵𝐵̇)].
Data on inputs and output is needed, combined with data on the various cost elasticities involved in the total cost equation, in order to estimate multifactor productivity growth and decompose it into the various components as discussed above. This data are obtained from econometric estimations of the cost structure 𝐶 = 𝑓(𝑄, 𝑤𝑗, 𝐵, 𝐷, 𝑡).
3.2. Econometric Implementation
I have chosen for a transcendental logarithmic (translog) specification to estimate the total cost function, which is a flexible functional form without placing a-priori restrictions on the production technology. This functional form is frequently used in previous literature to analyse the cost function across industries. With its flexibility, the trans-log model provides a second-order Taylor series estimate in logarithmic forms to a random cost function around a certain point (Christensen and Greene, 1976; Onghena, Meersman and Van de Voorde, 2014).
(9) ℓ𝑛 ( 𝐶𝑖𝑡 𝑤𝑚𝑖𝑡) = 𝛼 + ∑ 𝛽𝑖𝐷𝑖 𝐼 𝑖=2 + 𝛽𝑄ℓ𝑛𝑄𝑖𝑡+ 𝛽𝐵ℓ𝑛𝐵𝑖𝑡+ 𝛽𝑡𝑡 + 𝛽𝐾ℓ𝑛 (𝑤𝑘𝑖𝑡 𝑤𝑚𝑖𝑡) + 𝛽𝐿ℓ𝑛 ( 𝑤𝑙𝑖𝑡 𝑤𝑚𝑖𝑡) + 𝛽𝑄𝑄(ℓ𝑛𝑄𝑖𝑡)2+ 𝛽 𝐵𝐵(ℓ𝑛𝐵𝑖𝑡)2+ 𝛽𝑡𝑡𝑡2+ 𝛽𝐾𝐾(ℓ𝑛 ( 𝑤𝑘𝑖𝑡 𝑤𝑚𝑖𝑡)) 2 + 𝛽𝐿𝐿(ℓ𝑛 (𝑤𝑙𝑖𝑡 𝑤𝑚𝑖𝑡)) 2 + 𝛽𝑄𝐵(ℓ𝑛𝑄𝑖𝑡)(ℓ𝑛𝐵𝑖𝑡) + 𝛽𝑄𝑡(ℓ𝑛𝑄𝑖𝑡)𝑡 + 𝛽𝑄𝐾(ℓ𝑛𝑄𝑖𝑡)ℓ𝑛 ( 𝑤𝑘𝑖𝑡 𝑤𝑚𝑖𝑡) + 𝛽𝑄𝐿(ℓ𝑛𝑄𝑖𝑡)ℓ𝑛 ( 𝑤𝑙𝑖𝑡 𝑤𝑚𝑖𝑡) + 𝛽𝐵𝑡(ℓ𝑛𝐵𝑖𝑡)𝑡 + 𝛽𝐵𝐾(ℓ𝑛𝐵𝑖𝑡)ℓ𝑛 (𝑤𝑘𝑖𝑡 𝑤𝑚𝑖𝑡) + 𝛽𝐵𝐿(ℓ𝑛𝐵𝑖𝑡)ℓ𝑛 ( 𝑤𝑙𝑖𝑡 𝑤𝑚𝑖𝑡) +𝛽𝐾𝑡ℓ𝑛 ( 𝑤𝑘𝑖𝑡 𝑤𝑚𝑖𝑡) 𝑡 + 𝛽𝐾𝐿ℓ𝑛 ( 𝑤𝑘𝑖𝑡 𝑤𝑚𝑖𝑡) ℓ𝑛 ( 𝑤𝑙𝑖𝑡 𝑤𝑚𝑖𝑡) + 𝛽𝐿𝑡ℓ𝑛 ( 𝑤𝑙𝑖𝑡 𝑤𝑚𝑖𝑡) 𝑡 + 𝜖
where C represents total costs, wj ( j = K, L, M) is the price of input j, Q is output, B represents bad output, and βz are the parameters to be estimated. The variables are collected for individual industries i = 1, … , 19 at time t where t = 1, . . . , 13. Hence, subscripts i and t represent specifications with regards to industry and time, respectively.
According to Berndt (1991), gains in efficiency can be obtained by estimating the optimal, cost-minimizing input demand equations, transformed into cost share equations. All costs and prices are expressed relative to the price of intermediate inputs, M, to impose homogeneity in input prices. To obtain equations that are susceptible to estimation, I employ Shephard’s Lemma, which states that the optimal, cost minimizing demand for input j can be derived by differentiating the total cost function with respect to input prices, in this research presented in relative prices of the intermediate inputs.
Differentiating equation (7) with respect to the price of input j relative to the price of intermediate inputs M, 𝜐𝑖𝑗 =
𝑤𝑖𝑗
𝑤𝑖𝑚
, and using Shephard’s Lemma gives the share of input j (j = K and L) in total cost:
3.2.1 Regularity Conditions
For the cost function to be well behaved and thus concave in input prices, its Hessian Matrix of second-order derivatives with respect to variable input prices should be negative semidefinite as there is a local maximum. In addition, the cost function should be non-decreasing in output.
For the purpose of this research, there are three constraints that need to be fulfilled for in the input shares, concerning “symmetry”, “homogeneity of degree 0”, and the “adding-up condition”. The first restriction, the symmetry condition, implies that 𝛽𝑖𝑗 = 𝛽𝑗𝑖, 𝑖, 𝑗 = 𝐾, 𝐿, 𝑀. The second constraint on the parameter estimates is that the function must be homogeneous of degree zero in prices, given Q (Berndt, 1991). This implies the following restrictions on equation (7):
𝛽𝐾𝐾+ 𝛽𝐾𝐿+ 𝛽𝐾𝑀 = 0
𝛽𝐿𝐾+ 𝛽𝐿𝐿+ 𝛽𝐿𝑀 = 0
𝛽𝑀𝐾+ 𝛽𝑀𝐿+ 𝛽𝑀𝑀 = 0
As the variables for input prices and total costs have already been transformed and presented as relative prices of intermediate inputs, I do not need to impose the homogeneity constraint again in the regression.
Lastly, as factor cost shares must sum to unity, ∑𝑛𝑗=1𝑆𝑗 = 1, the share of the intermediate inputs is measured residually 𝑆𝑖𝑚 = 1 − ∑ 𝑆𝑗 𝑖𝑗 , since there are only n-1 independent equations in the model. According to Berndt (1991) the “adding-up condition” of the share equation system, which has several econometric implications, one which applies particularly to this paper is that one of the share equations has to be dropped to avoid a singularity problem when estimating the model by Seemingly Unrelated Regressions, as used in this research, because the sum of the value of spending on each input should sum to nominal output, which gives unity. This implies that when there are three shares, the last one is redundant. The remaining n-1 share equations are then estimated.
The system of equation (9) and the associated share equations (10) should satisfy the usual regularity conditions. The total cost function will be estimated by means of Seemingly Unrelated Regressions, which is a linear regression model that for several regression equations, each having its own dependent variable and different independent variables, assumes that the error terms are to be correlated
across the equations. As I expect great heterogeneity across the economic regions addressed in this research, the total cost function and respected share equations will be estimated on a country/regional level. The sources data in presented in the following section; the results are presented in section 4.
3.3 Source Data
The cost function is estimated using data for 19 industries of seven economies during the period 1995 to 2007. To estimate the translog cost function in section 3 I considered the most reliable KLEMS dataset developed as part of the WORLD KLEMS initiative. A complete system of industry-level production accounts for Europe, the U.S., Japan, Canada and Australia was constructed following the international standards established by OECD (2001). The underpinnings of these standards have been developed by Gollop and Jorgenson (1980) and Jorgenson, Gollop, and Fraumeni (1987) who formulated the first set of integrated productivity accounts for the U.S. economy. Efforts have been extended subsequently to develop a set of industry production accounts for emerging economies such as India and Russia following the same set of international standards (Goldar, Das, Aggarwal, AzeezErumban, De, and Das, 2014; Voskoboynikov, 2012).
The complete system of industry-level data for Europe, the U.S., Japan, Canada and Australia is most consistent. Of the set of economies currently in scope under the WORLD KLEMS initiative, Europe4, the U.S., Japan, Canada, Australia, India and Russia represent those which offer the datasets that conform closely to the OECD guidelines (OECD, 2001) 5. These economies have developed differently overtime, however this set of developed and developing economies altogether form a well-established representation of the world economy with an average 56% of global nominal GDP in international prices over the 1995-2007 period (see table II in the appendix A).
4
Europe is represented by the 10 EU member states for which growth accounting could be performed, namely of Austria, Belgium, Denmark, Finland, France, Germany, Italy, Netherlands, Spain and the United Kingdom
5
The KLEMS database has been set up to promote and facilitate the analysis of growth and productivity patterns around the world, based on a growth accounting framework at a detailed industry level (WORLD KLEMS, 2015). From the first source, WORLD KLEMS, growth accounting data on output and inputs were obtained for Canada, India and Russia. The second data source, EUKLEMS, part of the WORLD KLEMS initiative, was used to obtain data for Europe, the U.S., Japan and Australia.
The final source is the industry- and economy-level CO2 emissions data from the Environmental Accounts (EA) constructed by The World Input Output Database (WIOD), recently developed by Groningen Growth and Development Centre (GGDC) (Dietzenbacher, Los, Stehrer, Timmer, De Vries, 2013; Timmer, Dietzenbacher, Los, Stehrer and Vries, 2015). CO2 emissions on industry level is not included in the KLEMS data, therefore for the implementation of coherent data on the environment, CO2 emissions data from EA database are used.
3.3.1 Variables
Where data are available, I used the EU- and WORLD KLEMS data to retrieve gross output, capital compensation, labour compensation and intermediate inputs in current prices in millions of national currency. In addition volumes indices for gross output, capital services, labour services and intermediates inputs are used. For a list of the data variables included see table III and IV in the appendix A.
The model in this paper is based on total costs, which are represented by the sum of the factor inputs Capital Compensation, Labour Compensation and Intermediate Inputs of each economy’s industries. Gross output volume indices in constant prices of 1995 are used as a single output variable. Input prices are calculated based on the value weighted input indices divided by the volume indices for each factor input.
Canada and Russia cover all index data using 2005 as the base year while for India 1980 is used. As this research covers the period 1995-2007, the indices data for Canada, India and Russia are adapted to match that of the EUKLEMS data for Europe, the U.S., Japan and Australia, for which already 1995 is used.
available, I derived the gross output volume index data based on gross output in current and constant prices. Next to that, for Russia data on hours worked were available which were used to compute an labour volume index, as the labour services index is not available at the time of constructing my paper. Intermediate inputs volumes indices for India and Russia are derived based on intermediate inputs in current prices and gross output in constant prices. Despite these differences, these countries are included as they represent a large share of the world GDP (World Bank, 2015b). Hopefully my model is sophisticated enough to comply for the differences. For an overview of the data adjustments see table IV in the appendix A.
3.3.2 Industries
The total cost function is estimated using data for 19 industries of seven economies during the period 1995 to 2007. As I expect great heterogeneity across the economic regions addressed in this research, the total cost function and respected share equations will be estimated on a country/regional level. The fixed effects vector D includes 18 industry-specific intercepts with cross effects for each input price and output or emission quantity. For the purpose of this research, the estimations focus on the market economy of one economic region at a time, existing out of 19 industries which are presented in table 16.
Despite the high integration of the KLEMS dataset in terms of industries, due to differences with the industries used in WIOD, some adjustments in the data need to be pointed out (see table V in the appendix A for the complete construction of industries representing the market economy).
For consistency across regions, some industries have been combined (table V). For example, for Canada, Australia and the data on CO2 emissions ‘Chemical, Rubber, Plastics And Fuel’ represents the combined data of three industries, namely 'Coke, refined petroleum products and nuclear fuel ', 'Chemicals and chemical products' and 'Rubber and plastics' (sectors 23, 24 and 25). ‘Wholesale And Retail Trade’ (industry G) represents the summed data of 'Sale, maintenance and repair of motor vehicles and motorcycles', 'Wholesale trade and commission trade except of mote vehicles' and 'Retail trade, except of motor vehicles and motor cycles' (sectors 50, 51 and 52). The weighted average of these merged industries is used to compute
6
accurate index data for these industries. For an overview of the industry adjustments, I refer to table V in the appendix A.
Table 1: Industries of the Market Economy
Industry WORLDKLEMS CODE
1 Agriculture, Hunting, Forestry And Fishing AtB
2 Mining And Quarrying C
3 Food , Beverages And Tobacco 15t16
4 Textiles, Textile , Leather And Footwear 17t19
5 Wood And Of Wood And Cork 20
6 Pulp, Paper, Paper , Printing And Publishing 21t22 7 Chemical, Rubber, Plastics And Fuel 23t25
8 Other Non-Metallic Mineral 26
9 Basic Metals And Fabricated Metal 27t28
10 Machinery, Nec 29
11 Electrical And Optical Equipment 30t33
12 Transport Equipment 34t35
13 Manufacturing Nec; Recycling 36t37
14 Electricity, Gas And Water Supply E
15 Construction F
16 Wholesale And Retail Trade G
17 Hotels And Restaurants H
18 Transport And Storage And Communication I 19 Finance, Insurance, Real Estate And Business Services JtK 3.4 Trend Analysis
I present a selection of descriptive statistics for the variables used in this research for the seven economies and the assigned 19 industries. The descriptive statistics of the key variables for Europe, the U.S., Japan, Canada, Australia, India and Russia are presented in table 2, which provides average growth rates of total output, the input factors and CO2 emissions for the total market economy. Tables VIa till VIg in the appendix A exhibit the complete material for the individual 19 industries of the market economy of each economic region, structured in terms of average growth of real output and real inputs, multifactor productivity growth and the related information in terms of relative shares during the 1995-2007 period.
Table 2: Descriptive statistics of key variables in percentages (averages over 1995-2007) Growth of gross output Share of capital Share of labour Share of intermediate inputs Growth of
capital Growth of labour
EU 3.0 17.8 27.3 54.9 3.1 0.9 U.S. 3.4 22.1 30.7 47.2 4.0 1.0 Japan 0.9 23.6 27.4 49.1 2.2 -0.2 Canada 3.6 24.1 25.1 50.8 3.9 1.9 Australia 3.4 20.5 24.4 55.1 5.1 2.0 India 4.7 24.7 19.5 55.8 8.4 3.4 Russia 5.8 24.6 25.0 50.4 2.8 0.8 Aggregated average 3.0 21.2 27.8 51.0 3.8 1.0 Growth of intermediate inputs Multifactor productivity Growth of CO2 emissions Partial eco-efficiency Share of group GDP (PPP, current $) EU 3.1 0.5 0.0 3.0 32.9 U.S. 3.8 0.5 -0.4 3.7 37.0 Japan 0.6 0.2 -1.2 2.1 15.2 Canada 4.1 0.1 0.6 3.0 3.0 Australia 3.4 0.1 2.0 1.5 1.7 India 4.6 -0.5 3.6 1.2 8.4 Russia 6.3 1.8 0.9 4.8 1.8 Aggregated average 3.2 0.4 0.0 3.0 100.0
Note: All figures are in percentages. Share of capital, labour and intermediate inputs are given in the share of total gross output for the respected region. Partial eco-efficiency is measured as the differences between gross output growth and CO2 emissions growth. The aggregated averages are weighted by GDP, PPP (World Bank, 2015b).
During the 1995-2007 period, the economies experienced different growth rates when measuring the CO2 emissions. Only for the U.S and Japan CO2 emissions declined between 1995 and 2007, with 0.4% and 1.2%, while for the remaining five economies CO2 emissions were stagnant or even increased. Again a great deal of heterogeneity arises when measuring industry sizes in terms of CO2 emissions. During the respected period for all regions manufacturing, utility and transportation industries were the largest producers of CO2 emissions. For Europe, the U.S. and Japan on average 65% of the industries show a decline in CO2 emissions, while during the same period on average 45% of the Canadian and Russian industries experience a decline. 70% of the Indian and Australian industries experienced an increase in CO2 emissions between 1995 and 2007.
The substantial difference between gross output growth and CO2 emissions growth leads to eco-efficiency. All economies are eco-efficient7, some more than others, confirming that with increased gross output, CO2 increased less than proportional. What is interesting to point out is that though Russian CO2 emissions increased more than the European CO2 emissions, Russia is more eco-efficient. There is however a diversity in eco-efficiency at the industry level, which is overall retractable to industries that the major producers of CO2 emissions; manufacturing, utility and transportation industries. Nonetheless, for the emerging economies, India and Russia, eco-efficiency positively benefits from the large gross output growth rates, while for the developed nations eco-efficiency is mostly determined by the growth (decline) of CO2 emissions.
Furthermore, factor cost shares differ among the 19 industries. The capital share ranges from 17.8% for Europe to a 24.7% in India, while labour shares ranges from 19.5% for India and 30.7% for the US. For all economies intermediate inputs are the largest contributing factor of production, suggesting offshoring is occurring within these economies. Overall labour is a larger contributor than capital with a few exceptions for particular industries and for India, where for 85% of the industries capital is the largest contributor.
For Europe, the U.S., Japan, Canada and Australia the multifactor productivity varied between 0.1% and 0.5% for the total market economy. This is in line with economic theory, suggesting that more developed economies have lower multifactor
7
productivity growth. Russia on the other hand has a higher multifactor productivity of 1.6% while India shows a negative multifactor productivity of -0.5% between 1995 and 2007. Overall multifactor productivity is negative for all Indian industries except agriculture, mining and machinery. This however is surprising as other papers have found increasing and positive MFP for India (e.g. Enrumban, Das, and Aggarwal (2012) observed increase in MFPG 1995-2005).
3.5 Descriptive Statistics
The system of equations used to estimate the parameters required by the measurement framework consist of the cost function (9) and the share equations (10) for input j (j=K and L). The share equation for intermediate inputs is computed as a residual through the “adding up condition”. I have pooled time-series cross section data for 19 industries of the seven economic regions’ market economies for the period 1995-2007 to estimate the model, amended to estimate the parameters for each economic region. Estimating the model as a pooled system, per region, adds both flexibility to the model and imposes cross-equation restrictions to allow a fully integrated cost structure model, facilitating more efficient estimates. I will perform a Seemingly Unrelated Regression (SURE) Technique for the estimation of the coefficients, as the three equations share common parameters. I will apply only symmetry constraints on the three remaining equations since I already implied the homogeneity assumption by transforming the total cost and input factor variables.
Table 3 presents the means and standard deviation of the key variables of the regression model for the economic regions. The complete industry level breakdown of the means the key variables in the equation are not reported here, but are available upon request.
Table 3: The mean and standard deviation of key variables (averages over 1995-2007)
variable EU US Japan Canada Australia India Russia
log (TC) μ 0.1227 0.1386 -0.0731 0.2352 0.1473 0.2693 0.1434 σ 0.1893 0.1812 0.2011 0.2034 0.1847 0.2119 0.2915 log (Q) μ 4.7366 4.7424 4.5706 4.8401 4.7575 4.8745 4.8337 σ 0.1469 0.1797 0.2008 0.1827 0.1856 0.2119 0.2877 log (B) μ 4.6070 4.6082 4.5636 4.6349 4.6030 4.7465 4.5463 σ 0.0926 0.1509 0.2437 0.1979 0.3260 0.4638 0.4229 t μ 7 7 7 7 7 7 7 σ 3.7493 3.7493 3.7493 3.7493 3.7493 3.7493 3.7493 log (wK) μ 0.1698 0.1712 -0.0254 0.2141 0.2400 0.1951 -0.0399 σ 0.3370 0.2764 0.2962 0.3462 0.3604 0.3394 0.4772 log (wL) μ 0.0440 0.1262 -0.1233 0.1892 0.0821 0.2086 0.1793 σ 0.2017 0.1953 0.1943 0.1994 0.1890 0.2963 0.3305 Share of capital μ 0.1379 0.1753 0.1826 0.2049 0.1625 0.2246 0.1940 σ 0.0993 0.1072 0.1110 0.1382 0.1083 0.1375 0.1098 Share of labour μ 0.2626 0.2772 0.2591 0.2395 0.2355 0.1741 0.2436 σ 0.0777 0.0899 0.0909 0.0876 0.0652 0.1085 0.0858
TC = total costs in relative prices of intermediate inputs (wm), Q = gross output, B = bad output, wK = price of capital in relative prices of intermediate inputs (wm), wL = price of labour in relative prices of intermediate inputs (wm). Share of capital and share of labour are the average cost shares of the inputs with respect to total costs, over the period 1995-2007. μ is the mean, the standard deviation is presented by σ.
3.5.1 Diagnostics checks
As my model imposes a translog model with several quadratic and cross product variables, high collinearity is expected. I use a correlation matrix to analyse collinearity (Hill, Griffiths and Lim, 2012). As expected, the quadratic and cross-product variables show high correlation coefficients (>80.0%). Because the translog cost function is prone to multicollinearity, it is standard practice to estimate it together with its dual expenditure-share functions (Ray, 1982), for which the SURE model complies.
is visually tested by creating a kernel density plot of the residuals (Hill et al., 2012). A Shapiro-Wilk W test confirms what the kernel density plots show: the assumption of normality is not fulfilled for either of the variables. The issue of normality of residuals is likely not critical as the sample size is sufficiently large (>200), as the Central Limit Theorem ensures that the distribution of disturbance term will approximate normality.
The last assumption tested is for heteroscedasticity, through the White’s General Test for Heteroskedasticity (Hill et al., 2012). The results indicate that the null hypothesis of homoscedasticity is rejected for all three equations for all regions; all were significant with a p-value of at least 0.10. In addition formal tests show that autocorrelation were present in the cost and share equations, which may cause biased standard errors. The Wooldridge test for autocorrelation show p-values lower than 0.05, except for the total cost equation of India which is not significant at any level. Therefore, the estimation procedure should control for these statistical phenomena. This indicates that autocorrelation is present in all equation, with exception of India. I corrected for first-order autoregressive disturbances making use of lagged values of the dependent variables in all three equations. Specifically, incorporating the lagged dependent variable into the cost and share equations gave the form 𝐶𝑡= 𝛼 + 𝛽𝑋𝑡+ 𝜌𝐶𝑡−1+ 𝜇𝑡 which implies, after rearranging, the following form 𝐶𝑡 = 𝛼(1 + 𝜌) + 𝛽(𝑋𝑡+ 𝜌𝑋𝑡−1) + (𝜇𝑡+ 𝜌𝜇𝑡) where 𝑋𝑡 refers to the vector of right-hand variables in (9), 𝛼 and 𝛽 refer to the corresponding parameters, and 𝜌 is the coefficient of autocorrelation. A similar formula applies to the share equations (10). An overview of the diagnostic checks can be found in table III of appendix B. Correlation matrices and data used for normality checks are available upon request.
4. Empirical Results
In the following section the results of the model are presented. First the estimated parameters and the coinciding standard errors as well as the parameter estimates for industry dummy intercepts are presented in table IV of appendix B.
the other six regions, indicating an outstanding fit of the model and suggesting that the model is well estimated.
The coefficients of the model are mostly statistically significant. What is interesting to point out is that the coefficient for bad outputs is only significant for the U.S., Japan and Australia. Overall the coefficients for bad output have a positive sign with exception of Japan, meaning that if CO2 emissions growth in Japan increases, total costs growth decreases. Technological change coefficient is significant for all regions, with exception of Europe and is only negative for the emerging economies in the research, India and Russia, confirming that the effects of technological change are deteriorating for the developed nations (Europe, the U.S., Japan, Canada and Australia).
4.1 Hypothesis testing
Through a log-likelihood chi-square test, the joint hypothesis 𝛼0𝐷= 𝛼𝑄𝐷= 𝛼𝐵𝐷 = 𝛼𝑡𝐷 =
𝛼𝐾𝐷= 𝛼𝐿𝐷 = 0, stating that the dummy industry coefficients are zero (table Va in
appendix B) is tested. The test indicates at a 0.01 significance level that the null hypothesis of no-industry differences is rejected for all regions, pointing out there are inter-industry differences present in the cost-structure of industries within the same region.
A similar test has been conducted to examine the joint hypothesis that the coefficients of bad output are zero in the total cost function (9), which is rejected at the 0.01 significance level. The estimates can be found in table Vb of appendix B.
4.2 Robustness
the lagged dependent variables 𝜌 show less significance. The R2 is nonetheless higher, as more variables are included and the standard error of the model has decreased.
However, as in the original estimation, a lot of uncertainty surrounds the coefficients for India; the R2 is unity and the coefficients seem to have a scaling problem. Both the standard estimation model and the model containing industry slope dummies for good and bad output include lagged dependent parameters. However, diagnostic checks indicated that there is no autocorrelation present in the translog cost equation of India. Due to the scope of this research, this possibility is not further explored.
4.3 Interpretation
The various cost elasticities computed from the estimated parameters, good output Q, bad output B, technological change t, prices of capital K and prices of labour L, are presented in table VI of appendix B for the complete data sample. The reported estimates are both presented at the region specific industry level and as aggregated market economy averages weighted by gross output for the 19 industries and time periods for each parameter, presented in the last row of the table.
The t-statistics are based on a computation of the measures evaluated at the mean value of the data, using the delta method, for which the variance of elasticity was computed according the variance equation as presented in appendix C. The elasticity estimates based on the robust model were also computed (not reported here, but available upon request).
4.3.1 Effects of CO2 emissions on the multifactor productivity framework
100.00 105.00 110.00 115.00 120.00 125.00 130.00 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007
Alternative MFP measure Standard MFP framework
0.00 0.50 1.00 1.50 2.00 2.50 3.00 3.50 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 Alternative MFP framework Standard MFP framework
are presented in figure 1a. Two indices were computed: the first index is based on the standard framework, excluding bad output; a second index represented the adjusted multifactor productivity framework, including bad output.
Figure 1a presents the compounded growth rate for the joined seven regions. The standard framework of productivity growth that ignores bad outputs measures nearly an equivalent level of multifactor productivity as the adjusted framework; the compounded growth rate of the two indices follow a similar pattern. Figure 1b depicts the yearly growth rate of the two aggregated multifactor productivity frameworks. The growth rate that excludes CO2 emissions overestimates productivity growth in the majority of the years (exceptions are 1997, 1999 and 2002), but only by a small difference; less than 0.05%.
Figure 1a: Aggregated multifactor productivity compounded growth rate for Europe, the U.S. Japan, Canada, Australia, India and Russia (1995 – 2007 period)
75.00 80.00 85.00 90.00 95.00 100.00 105.00 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 Robust - Europe Alternative MFP framework Standard MFP framework 100.00 105.00 110.00 115.00 120.00 125.00 130.00 135.00 19 95 19 96 19 97 19 98 19 99 20 00 20 01 20 02 20 03 20 04 20 05 20 06 20 07 Europe Alternative MFP framework Standard MFP framework
The compounded growth rate estimates in figure 2 are based on region specific multifactor productivity indices, with weighted averages based on gross output, of all industries in the specific region. For each region, two graphs are presented: the first graph presents the estimates using the elasticities from the model as describes in 3.5, the second graph is based on the estimates from the robust model. The standard productivity framework, excluding bad output, overestimates productivity for Europe, Japan and Canada, similar levels of multifactor productivity for India and Russia, while for the U.S. and Australia it underestimates productivity.
When accounting for bad outputs, the productivity differences between Europe and the U.S. becomes even more apparent, as it leads to a down grading of productivity. When looking at productivity difference based on the robust model, it suggests that the approach employed was carried out in a consistent manner, as for five out of seven regions, the graph showcase a somewhat similar trend. The productivity gap for Europe however becomes even larger. Considering that Europe already has a lower level of productivity (Havik, Mc Morrow, and Turrini, 2008; O'Mahony, Rincón-Aznar, and Robinson, 2010), accounting for CO2 emissions in the productivity framework has a far larger effect on the European productivity level than that for other regions. However a lot of uncertainty surrounds the estimations of the elasticities, as some measures are not significant, particularly for the elasticities measured with the estimates of the robust model. Nonetheless, from these findings a clear picture arises that accounting for bad outputs in the productivity framework can over- or underestimate the levels of productivity.
4.3.2 Do firms care about CO2 emissions and corporate social responsibility?
For evaluating the marginal benefit for firms using the environment as a free input, measured by CO2 emissions, I use the shadow value (Zb) computed by partial differentiation of the cost function (9) with respect to lnB. Table VII of appendix B presents these shadow values or cost elasticities for good output Q, bad output B, technological change t, prices of capital K and prices of labour L. Cost elasticities for capital and labour are positive and significant at the either 0.05 or 0.01 significance level for most industries across regions, indicating that with increasing the input factors of capital and labour increases costs, which is in line with expectation. Again India stands out in that only the cost elasticity of good output is significant but the t-statistics are very large hinting at an error. The estimated elasticities based on the robust model are not presented here but available upon request. The estimates of the robust model show even less significance; for most regions the elasticities of capital and labour are not significant, suggesting they don’t affect the cost function.
I find that the shadow value of emissions, or private cost, is not significant in any region or for any industry, suggesting that firms in all industries use the environment as a free input when designing their production mix and thus firms do not incur costs from CO2 emissions that go beyond their private input costs.
5. Conclusion
All economic regions investigated in this paper have achieved environmental technological progress in the sense that the growth rate of CO2 emissions was lower than gross output growth over the period 1995 until 2007. When measuring technological progress through the traditional multifactor productivity framework, these emissions, or bad outputs, are ignored.
Though the general assumption exists that environmental protection, measured through decreasing CO2 emissions, leads to a decline in the productivity performance, and active development in the measurement of productivity, there have been few attempts to include bad outputs within a joint-production framework. This paper does so. This paper uses a detailed model of the production structure of the market economy for seven of the largest economic regions, namely Europe, the U.S., Japan, Canada, Australia, India and Russia and imposes a detailed cost function characterization to measure productivity differences when augmenting the framework for bad outputs. In an attempt to extend the traditional productivity framework with a measurement for bad outputs, shadow values were inferred that represented the private costs imposes by firms.
The findings on excluding bad outputs in the traditional productivity framework concludes the traditional productivity framework leads in the general case to an overestimation of productivity, but only by a small difference. However, when looking at the scenario on a national level, a different picture arises. Especially the productivity gap between Europe and the U.S. becomes even larger, as accounting for bad outputs downgrades the productivity level of Europe, but increases the multifactor productivity in the U.S.
I find that the shadow value of emissions, or private cost, is not significant in any region or for any industry, suggesting that firms in all industries use the environment as a free input when designing their production mix and thus firms do not incur costs from CO2 emissions that go beyond their private input costs.
Some limitations much be highlighted that surrounded the process of finding statistical support for the complex model used in this paper. Firstly, despite the high consistency of the EU and WORLD KLEMS dataset, there have been limitations surrounding the data of India and Russia as highlighted in section 3.3. The current data situation does seem to influences the results are presented in this thesis. Notwithstanding the international efforts to develop a set of industry production accounts for emerging economies India and Russia following the same set of international standards used for the dataset of Europe, the U.S., Japan, Canada and Australia, some inconsistencies remain.
Secondly, while accounting for CO2 emissions leads to a downgrade of multifactor productivity performance, the results are subject to a large degree of uncertainty which points to the need to enhance the modelling strategy used in this paper. The heterogeneity across regions and industries measured in this study might have been out of the scope of this paper. Future research could be done in the area of finding a suitable modelling strategy to measure a cross-country cross-industry data panel, as used in this paper.
Linked to this issue of the model and the data limitation, is the estimation method used for India. The diagnostic checks for India indicated that other estimation methods, such as a fixed effects model, could be suitable for the estimation as well. Considering the scope of this research, the estimation was limited to using Seemingly Unrelated Regressions model for most specifications.
6. References
Arnaud, B., Dupont, J., Koh, S.H. and Schreyer, P. (2011). Measuring multifactor productivity by industry: Methodology and first results from the OECD productivity database. Paris, France: Organisation for Economic Co-operation and Development. Retrieved February 6, 2016.
Ball, E., Färe, R., Grosskopf, S. and Zaim, O. (2005). Accounting for externalities in the measurement of productivity growth: the Malmquist cost productivity measure. Structural change and economic dynamics. 16(3): 374-394.
Berndt, E.R. (1991). The practice of econometrics: Classic and contemporary. Reading, MA: Addison-Wesley Pub.
Berman, E. and Bui, L.T. (2001). Environmental regulation and productivity: evidence from oil refineries. Review of Economics and Statistics. 83(3): 498-510.
Bleischwitz, R. (2002). Governance of eco efficiency in Japan: An institutional approach (No. w124). Wuppertal papers.
Bloom, N., Genakos, C., Martin, R. and Sadun, R. (2010). Modern Management: Good for the Environment or Just Hot Air?. The Economic Journal, 120(544): 551-572.
Bosworth, B. and Collins, S.M. (2007). Accounting for growth: comparing China and India (No. w12943). National Bureau of Economic Research.
Brandt, N., Schreyer, P. and Zipperer, V. (2013). Productivity Measurement with Natural Capital. OECD Economics Department Working Papers. Retrieved February 6, 2016.
Brandt, N., Schreyer, P. and Zipperer, V. (2014). Productivity Measurement with Natural Capital and Bad Outputs. OECD Economics Department Working Papers. Retrieved April 27, 2016.
Christensen, L.R. and Greene, W.H. (1976). Economies of scale in US electric power generation. The Journal of Political Economy, 655-676.
Dietzenbacher, E., Los, B., Stehrer, R., Timmer, M. and De Vries, G. (2013). The construction of world input–output tables in the WIOD project. Economic Systems Research, 25(1): 71-98.
Earnhart, D.H., Khanna, M. and Lyon, T.P. (2014). Corporate environmental strategies in emerging economies. Review of Environmental Economics and Policy, 8(2): 164-185.
Erumban, A.A., Das, D.K. and Aggarwal, S. (2012). Industry Origins of Aggregate Productivity Growth in India. 32nd general conference of the international association of research in income and wealth. Boston, USA, August (5-11).
Escap, U. (2009). Eco-efficiency Indicators; Measuring Resource use Efficiency and the Impact of Economic Activities on the Environment. Greening of Economic Growth Series, UN Publication.
EU KLEMS. (2011). EU KLEMS Growth and Productivity Accounts. Retrieved February/March, 2016, from http://www.euklems.net/
Färe, R. (1975). Efficiency and the production function. Journal of Economics. 35(3): 317-324.
Färe, R., Grosskopf, S. and Pasurka Jr, C.A. (2006). Social responsibility: US power plants 1985–1998. Journal of Productivity analysis. 26(3): 259-267.
Goldar, B.N., Das, D.K., Aggarwal, S., AzeezErumban, A., De, K. and Das, P.C. (2014). Research Team Research Advisor. Estimates of Productivity Growth for Indian Economy http://rbidocs.rbi.org.in/rdocs/content/DOCs/IKLEMS110614.xls, Retrieved February 3, 2016
Halvorsen, R. and Smith, T.R. (1984). On measuring natural resource scarcity. The Journal of Political Economy. 954-964.
Halvorsen, R. and Smith, T.R. (1986). Substitution possibilities for unpriced natural resources: restricted cost functions for the Canadian metal mining industry. The Review of Economics and Statistics, 398-405.
Harrison, A., Hyman, B., Martin, L. and Nataraj, S. (2015). When do Firms Go Green? Comparing Price Incentives with Command and Control Regulations in India (No. w21763). National Bureau of Economic Research.
Havik, K., Mc Morrow, K., & Turrini, A. (2008). The EU-US total factor productivity gap: An industry perspective (No. 339). Directorate General Economic and Monetary Affairs (DG ECFIN), European Commission.
Hellweg, S., Doka, G., Finnveden, G. and Hungerbühler, K. (2005). Assessing the Eco‐efficiency of End‐of‐Pipe Technologies with the Environmental Cost Efficiency Indicator. Journal of Industrial Ecology. 9(4): 189-203.
Hill, R. C., Griffiths, W. E. and Lim, G. C. (2012). Principles of econometrics (international student version). Hoboken. (4th Edition). Hoboken, NJ: Wiley.
Hulten, C. R. (2001). Total factor productivity: a short biography. In New developments in productivity analysis. University of Chicago Press, 1-54.
Hulten, C. R. (2009). Growth accounting. (No w15341). National Bureau of Economic Research.
Huppes, G. and Ishikawa, M. (2005). Eco‐efficiency and Its Terminology, Journal of Industrial Ecology, 9(4), 43-46.
Jorgenson, D.W., Gollop, F.M. and Fraumeni, B.M. (1987). Productivity and US Economic Growth, Cambridge MA: Harvard University Press.
