MASTER THESIS Double Degree Programme

(1)

MASTER THESIS

Double Degree Programme

M.Sc. International Development and Globalization

University of Groningen, Faculty of Economics and Business

M.Sc. International Economics

University of Göttingen, Faculty of Business and Economics

The Distributional Effects of Taxing

Greenhouse Gases in India

– An Input-Output Analysis

Dorothea Drees

Student number: s3975002

Email: d.drees.1@student.rug.nl

Supervisor

Prof. Dr. H.W.A. Dietzenbacher

(University of Groningen)

Co-assessor

Apl. Prof. Dr. Jann Lay

(University of Göttingen)

(2)

Abstract

India's recent economic growth has been accompanied by rising greenhouse gas emissions which contribute to global warming. The implementation of necessary climate policy measures such as emission taxation is however hampered by concerns about adverse distributional ef-fects. Previous research indeed confirms regressive effects of emissions taxation in high-in-come countries but presents mixed findings for low- and middle-inhigh-in-come countries. Given that lack of consistent evidence, I will assess the distributional outcomes of a hypothetical emission tax in India and thereby contribute to the existing literature by considering multiple greenhouse gases, namely carbon dioxide (CO2), methane (CH4) and nitrous oxide (N2O). The combination

of household survey data and national input-output information allows to estimate the burden imposed by emission taxation on different income groups. While I find regressive outcomes for the taxation of CH4, N2O and direct CO2 emissions, the tax imposed on indirect CO2 emissions

shows progressive distributional effects.

(3)

I

List of figures

Figure 1 The combination of household survey and input-output data ... 10

Figure 2 Lorenz curve of household MPCE ... 26

Figure 3 Relative tax burden of the taxation of indirect and direct CO2 emissions ... 30

Figure 4 Relative tax burden of the taxation of indirect CH4 and N2O emissions ... 32

Figure 5 Relative tax burden of the simultaneous taxation of multiple direct and indirect GHG emissions ... 33

List of tables

Table 1 Reference periods for the different MPCE measures ... 11

Table 2 Cutoffs for deciles and average MPCE ... 26

Table 3 Shares of selected consumption categories in total expenditures ... 27

Table 4 Sectoral emission coefficients of the three greenhouse gases ... 29

Table 5 Emission factors of items causing direct emissions ... 30

Table 6 Percentage change in pre-tax expenditure due to different emission taxes ... 31

Appendix A Table A - 1 Consumption items of the CES and their allocation to the WIOD2016 sectors .. 36

Table A - 2 Sector description of the WIOD2016 ... 42

Table A - 3 Correspondence between sectors in WIOD2013 and WIOD2016 ... 43

Table A - 4 Monthly expenditures on energy-related items (in thousand US Dollar) ... 45

Appendix B Table B - 1 Percentage change in pre-tax expenditure due to different emission taxes, based on non-normalized CES data ... 46

Table B - 2 Percentage change in pre-tax expenditure due to direct emission taxes, exclusion of other fuels ... 46

(5)

III

Abbreviations

ASEAN Association of Southeast Asian Nations CES Consumer Expenditure Survey

CH4 Methane

CO2 Carbon dioxide

GHG Greenhouse gas

HICs High-income countries

IPCC Intergovernmental Panel on Climate Change

ISIC International Standard Industrial Classification of All Economic Activities LMICs Low- and middle-income countries

MMRP Modified Mixed Reference Period

MOSPI Ministry of Statistics and Programme Implementation MPCE Monthly Per Capita Expenditures

MRP Mixed Reference Period N2O Nitrous oxide

NDC Nationally Determined Contribution NIOT National input-output table

NSSO National Sample Survey Office

UK United Kingdom

UN United Nations

UNEP United Nations Environment Programme

UNFCCC United Nations Framework Convention on Climate Change URP Uniform Reference Period

US United States

(6)

1

1. Introduction

One of the major challenges currently facing humanity is the one of human-induced climate change, which is generating severe and pervasive impacts for people and ecosystems. Policy action is therefore needed at both global and national scale. A crucial step to take is to reduce the amount of greenhouse gas (GHG) emissions as these are the main driver of global warming (IPCC 2014a). A way to achieve such a reduction, often recommended by economists, is the pricing of GHG emissions, especially of carbon dioxide (CO2). A pricing system for emissions

can for example be established by implementing a tax on each unit of GHG emitted. Countries like the United Kingdom, Sweden or Chile already introduced some form of carbon taxation. However, the desirable emission-reducing effects of a GHG tax do not come as a free lunch. Taxes must be paid by someone. Concerns about adverse distributional implications of emission taxation on low-income population groups belong to the most important political challenges for such policies (Dorband et al. 2019). The rationale underlying these concerns is that poor people might be hit more heavily by the tax burden because they spend a larger share of their income on emission-intensive products and lack possibilities for substitution. Earlier empirical research indeed identifies GHG emission taxes to produce regressive outcomes in most high-income countries (HICs), whereas the effects found for low- and middle-income countries (LMICs) are less consistent (Wang et al. 2016). Moreover, distributional outcomes can change depending on whether only CO2 emissions are charged or whether multiple GHGs are included in the tax

base. From these ambiguous findings it can be deduced that the impacts of a GHG emission tax on a country’s income distribution are not generalizable. They depend on the respective pro-duction and demand structure, which makes it necessary to examine the effect of emission pric-ing on different income groups for each country individually.

This thesis aims to provide exactly such a country-case examination by analyzing the effects of a hypothetical GHG emission tax in India across various income groups. India as a representa-tive of the LMICS is of particular interest not only because the ambiguity of previous research on emission taxation especially concerned this country group. Additionally, LMICs tend to be more vulnerable to climate change risks compared to HICs because of their prevalent location in high-temperature regions, the importance of climate-dependent agriculture and reduced ad-aptation capacities (Mertz et al. 2009). For India, which is among the five nations most severely affected by extreme weather events according to the Global Climate Change Index 2020 (Eck-stein et al. 2019), further global warming would increase the risk of heavy monsoon seasons with extensive flooding as well as heatwaves. Therefore, it is in the country’s own interest to reduce GHG emissions and hence contribute to the prevention of further global warming. But it is not only the elevated vulnerability to climate change that motivates the choice of India as the study subject of this thesis. As an emerging market, India is experiencing growing pres-sure by the international community to implement low-carbon policies because its rapid GDP growth in the past years resulted in a significant increase in CO2 and other GHG emissions

(Rong 2010). The current global agenda on climate change mitigation, the Paris Agreement, emphasizes that the responsibility to combat climate change is shared by all countries. This is why India, the third largest emitter of CO2 worldwide, should undertake political efforts to limit

(7)

2 Of course, the necessary environmental policy measures must not be at the expense of the poor sections of the population. This consideration is essential for a country like India, in which nearly 22% of the population lived below the national poverty line in 2011, according to latest official estimates available (Planning Commission 2013). Before a strategy of emission taxation can be implemented in the country, policy makers must thus ensure that the poorest are not the most burdened by the measure.

By applying input-output techniques, I aim to provide evidence on the short-term expenditure effects of a GHG emission tax across various income groups in India. In other words, I calculate the costs of maintaining current consumption with a hypothetical GHG emission price in place. Doing this for different income quantiles allows to examine whether the tax burden can be expected to fall more heavily on poorer or richer population sections. Consequently, I can ad-dress the following research question with my thesis:

How would the burden of a GHG emission tax, either on CO2 alone or on

multiple GHGs, be distributed across different income groups in India?

To answer this question, I combine the Indian national input-output table and environmental accounts from the World Input-Output Database (WIOD) (Timmer et al. 2015) with Indian household survey data. The latter is taken from the Household Consumer Expenditure Survey (CES) for the period 2011-2012 conducted by the National Sample Survey Office (NSSO). The results of this analysis will shed light on the immediate effects of an emission tax on house-holds’ expenditures, before any adaptation of behavior by individuals or firms can take place. Even though this static scenario is unlikely to capture the definite burden carried by the Indian households in the longer run, it can serve as a first guide for policy makers who consider an emission tax implementation but worry about potential adverse distributional effects.

The remainder of this thesis is organized as follows: The literature review first discusses the current situation regarding global climate change mitigation and highlights the emission tax as an effective instrument. It will present evidence on the distributional implications of emission taxation in other countries and subsequently address the Indian case. In section 3, the data sources used, i.e. the WIOD and the household survey information, are described and their merging is explained. Subsequently, section 4 sets out the methodological basis by explaining how direct and indirect household emissions as well as the associated tax payments can be determined in an input-output framework. The results are presented in section 5, while the last section concludes and offers some policy implications and suggested fields of future research.

2. Literature Review

(8)

3 industrialized countries to reduce GHG emissions were laid down for the first time in history. The six GHGs covered in the Kyoto protocol are carbon dioxide (CO2), methane (CH4), nitrous

oxide (N2O), hydrofluorocarbons, perfluorocarbons and sulfur hexafluoride (UN 1998). 1 While

the Kyoto Protocol did not hold developing and emerging countries like India accountable, the more recent Paris Agreement commits all nations, from low- to high-income, to making appro-priate contributions to combat climate change. The Paris Agreement aims at limiting global warming to less than two degrees Celsius compared with pre-industrial levels, ideally to 1.5 degrees (UN 2015). Despite these necessary efforts to find a global response to the threat of climate change, the current Nationally Determined Contributions (NDC) that are determined under the Paris Agreement are not sufficient to prevent global temperature from rising to levels clearly exceeding the two degree goal according to the UN Environment Programme’s (UNEP 2019) Emissions Gap Report.

Consequently, ambitious climate action at the national level has still an important role to play on top of the NDCs. There is an ongoing debate on how domestic climate policies should be implemented – either by economic, market-based policies (e.g. taxes or subsidies) or regulatory policies (e.g. governmental environment standards). In this discussion, emission taxes have been an instrument frequently advocated by economists and international organizations (Baran-zini et al. 2000). They belong to the market-based policies because emission-intensive goods and services will have higher market prices and/or lower profits, once the tax rate is set by the administrative authority. As a result, market forces will work to reduce the quantity of emis-sions. That way, an emission tax has the advantage of being cost-effective: While in the short run producers might simply pass their increased costs due to the tax forward to the consumers, they have the incentive to foster cost-reducing innovation in the longer run in order to keep their prices at a competitively low level (Hoel 1998; Boyce 2018). In order to mitigate global warming, emission taxes should focus on GHGs, in particular on CO2 whose increasing

con-centration in the atmosphere is the main driver of climate change (IPCC 2013, p. 13).

Other economic policies like an emissions trading system do also result in cost-effective and emission-reducing outcomes if markets are perfectly competitive. Though, the emission tax has several characteristics that positively differentiate it from other market-based climate policy instruments. The government revenue raised by the tax can be used for other purposes, for example for additional climate protection measures or to reduce other taxes (Hoel 1998). More-over, a carbon tax has lower administrative costs and more predictable effects compared to emissions trading (Jiang and Shao 2014).

2.2 Previous studies on the distributional effects of emission taxation

Despite the advantages of environmental taxes outlined above, environmental taxes like the one on GHG emissions face constant political headwind. One of the main concerns is related to the potential adverse distributional outcomes of the tax (Dorband et al. 2019; Wier et al. 2005; Kallbekken and Sælen 2011; Wang et al. 2016). An emission tax is increasing the costs of the current consumption bundle (hereafter pre-tax consumption bundle) for each household:

1_{In the second compliance period of the Kyoto protocol (2013 – 2020), the UNFCCC included nitrogen trifluoride}

(9)

4 Maintaining the same consumption as before the tax was introduced comes at higher costs be-cause a price is put on the emissions be-caused (directly or indirectly) by the consumption. Obviously, not everyone will be charged the same amount as a result of emission taxation. Provided that producers pass on the extra costs due to the taxation to consumers, the amount of tax paid by each household will depend on the its direct emissions, for example from burning fossil fuels, and on its indirect consumption of emissions embodied in the production of goods and services. Consequently, rich households that, in general, consume more will probably pay a higher amount of emission tax.

However, this does not imply that rich households will necessarily pay more taxes relative to their income. In contrast, there are widespread concerns that low-income households might spend a greater share of their income on emission-intensive but basic needs like house heating (Wang et al. 2016). This would lead to a regressive outcome of the emission tax, meaning that low-income households are burdened relatively more compared to high-income households. The relative tax burden can be calculated by dividing the price increase of the pre-tax consump-tion bundle for each household by the respective household’s income. Many studies on distri-butional tax effects thereby use household expenditures as a proxy for income (Renner 2018; Dorband et al. 2019; Phungrassami and Usubharatana 2019). A regressive tax effect means that the percentage increase in the value of the pre-tax consumption bundle due to taxation is larger for low-income household than for high-income households. If the opposite is true, the tax ef-fect is called progressive, implying larger impacts on richer population groups.2

Recognizing that the concerns about potential adverse distributional implications of emission taxation belong to the major political challenges for their implementation, there are numerous studies examining the distributional effects of such taxes. Many of these studies are based on input-output analyses that identify the price changes associated with the implementation of an emission tax. The input-output approach allows to quantify all direct and indirect inputs needed to produce the goods and services that are finally consumed. This information can be linked to data on emissions by industries in order to quantify the amount of emissions (e.g. tons of CO2)

embodied in each unit of a consumed product. By combining this knowledge with information on consumption patterns of households from different income groups (which is often provided in national household expenditure surveys), inference can be drawn on the distributional char-acter of the emission tax. An underlying assumption of this approach is that neither consumers nor producers change their behavior in response to the tax. Accordingly, the input-output anal-ysis captures rather the short-term effects of the tax. In contrast, computable general equilibrium models aim at explicitly modeling the behavioral responses of consumers and/or producers to the implementation of a tax. For this reason, many authors apply general equilibrium ap-proaches instead of a pure input-output framework in order to estimate long-run distributional consequences of emission taxation.

Wang et al. (2016) provide an extensive literature review of previous studies on distributional effects of carbon taxation of both methodological types: input-output price models and

2_{Throughout this paper, the terms “regressive” and “progressive” are used to describe the tax burden relative to}

(10)

5 computable general equilibrium models.3 Most of the studies discussed concern HICs and find regressive effects of environmental taxes. Wier et al. (2005), Kerkhof et al. (2008), Verde and Tol (2009), Feng et al. (2010) and Mathur and Morris (2014) discover regressive outcomes when putting a price on CO2 emissions respectively in Denmark, the Netherlands, Ireland, the

United Kingdom (UK) or the United States (US). Some of these studies, like the one by Mathur and Morris (2014) however just assume a taxation of the CO2 emissions from fossil fuels and

consequently, do not take all CO2 emissions caused by private consumption into account.

Authors that explicitly model consumer and/or producer behavioral reactions to emission taxa-tion in a general equilibrium model find more ambiguous evidence on the distributaxa-tional out-comes of emission taxes in HICs. For example, Rausch et al. (2011) conclude that a GHG emis-sion tax would have proportional distributional implications in the US when taking into account the supply side, i.e. the adaptation of factor prices. Dissou and Siddiqui (2014) even find a carbon price to be progressive in Canada when allowing for changes in factor prices. Neverthe-less, the majority of studies (Brännlund and Nordström 2004; Bureau 2011; Pashardes et al. 2014) still finds an emission tax to have regressive outcomes in HICs, also in the general equi-librium framework.

Brenner et al. (2007) argue that it is not clear whether these findings of the regressive character of emission taxation can be generalized from HICs to LMICs. Household expenditure and en-ergy usage might show completely different patterns in countries with lower per capita income. For example, while private transport in cars is common even among low-income households in HICs, just the richest population groups in LMICs can afford this kind of transport. Conse-quently, a tax on the GHGs emitted by travelling would probably hit higher income households harder in LMICs. The considerations how consumption patterns in LMICs might lead to distri-butional outcomes different from those in HICs will be further discussed for the Indian case in section 2.3. In the following, I will focus on presenting empirical evidence on the distributional effect of emission taxes in LMICs.

As suspected, the distributional tendencies of GHG emission taxes found for LMICs are less clear than those for HICs. While input-output analyses for Thailand (Phungrassami and Usub-haratana 2019) and Brazil (da Silva Freitas et al. 2016) reveal regressive outcomes of taxing GHGs, Renner (2018) finds that putting a price on CO2 emissions in Mexico would have

pro-gressive effects. In a consistent input-output analysis for 87 mostly LMICs, Dorband et al. (2019) conclude that a carbon price would have progressive effects on average. General equi-librium models find mixed effects of emission taxation in LMICs: Yang and Wang (2002) and Liang et al. (2013) identify carbon taxation to be regressive in Taiwan and China. In contrast, Brenner et al. (2007) show that carbon charges on fossil fuels in China would result in progres-sive outcomes and Yusuf and Resosudarmo (2015) likewise conclude that a carbon tax in In-donesia would have a progressive character. Nurdianto and Resosudarmo (2016) find mixed distributional implications when analyzing a hypothetical uniform CO2 tax in the Association

of Southeast Asian Nations (ASEAN): The authors’ multi-country computable general equilib-rium model reveals that the effect of the tax would be strictly progressive in Vietnam and strictly regressive in Singapore. For the other ASEAN countries included in their study, the effect is

(11)

6 mixed. Given the very heterogeneous evidence outlined above, it is not possible to make a generalizable statement about the distributional implications of taxing emissions in LMICs. Each country has to be analyzed individually, taking the country-specific consumption and pro-duction structure into account. However, the literature indicates that carbon taxation might have on average more neutral or even progressive distributional effects in LMICs compared to HICs. This makes the emerging country India an interesting subject for research.

One last matter to discuss relates to the decision which emissions exactly to tax in order to effectively combat climate change. As mentioned before, the most urgent concern is to reduce emissions of CO2 since this GHG is the main contributor to global warming (IPCC 2013). That

is the reason why most scientific literature has concentrated on mitigation strategies on CO2 in

order to curb global warming. This focus is also reflected in the literature reviewed above. Nevertheless, there are other GHGs whose impact on global warming should not be underesti-mated. Especially CH4 and N2O contribute significantly to the temperature-rising greenhouse

effect, being together responsible for nearly a quarter of all GHG emissions in 2011 (IPCC 2014b, p. 123). Scheraga and Leary (1992) argue that additionally including non-CO2 GHGs

like CH4 and N2O in an emission tax widens the set of control options. That way, it is possible

to identify more cost-effective policies. In a more recent report, Reilly et al. (2003) support this argument by demonstrating in a general equilibrium model that including multiple GHGs in a emissions reduction strategy reduces the mitigation cost and, furthermore, increases the overall amount of emissions reduction. Similarly, Rao and Riahi (2006) find that multi-GHG taxation allows to achieve a long-run stabilization of global temperatures at significantly lower costs compared to a CO2-only tax. As a conclusion, effective emission taxation should address CO2

as well as other GHGs from both an environmental and an economic perspective.

Some of the studies mentioned above take this conclusion, that taxing multiple GHGs can mit-igate climate change at lower abatement costs than a CO2-only policy, into account and analyze

multi-GHG taxation policies. For example, Kerkhof et al. (2008) and Feng et al. (2010) expand their analyses of the distributional effects of an emission tax in the Netherlands and the UK respectively to CH4 and N2O. Both studies, that found a clearly regressive outcome when

ex-clusively taxing CO2, conclude that the distributional consequences are less regressive when

including the two additional GHGs into the tax base. This implies that in these two HICs, poorer income groups tend to consume relatively less CH4-and N2O-intensive products compared to

the richer ones. Renner (2018) shows that the opposite is true for the middle-income country Mexico: While a pure CO2 tax shows progressive outcomes, the distributional trend reverses

and the tax effect gets regressive when including CH4 and N2O in the tax base. Renner explains

this by arguing that the taxation of additional GHGs would increase in particular food prices, on which poorer households spend a large part of their resources. Similarly, da Silva Freitas et al. (2016) find that the burden from taxing CO2, CH4 and N2O simultaneously would hit poorer

(12)

7 2.3 The case of India

India has been the third largest emitter of CO2 in 2017, with the vast majority of these emissions

arising from the combustion of fossil fuels, especially coal for electricity generation (EY 2018). Additionally, the country contributes significantly to non-CO2 GHGs emissions, like CH4

re-sulting from animal husbandry and rice cultivation, or N2O released among others by the use

of fertilizers or the combustion of biomass (MOSPI 2015). Given the role of India as a major global emitter of GHGs, political measures to curb the country’s emissions are essential. An emission tax could be an effective way to reduce India’s impact on global warming.

On the other hand, the per-capita CO2 emissions in India are still below the levels of HICs and other emerging countries like China (EY 2018). Moreover, large parts of the Indian population live in poverty and would therefore not be able to bear the price increases through an emission tax. Ananthapadmanabhan et al. (2007) address the issue of climate injustice in the within-country context for India. They argue that the poorest households in India are the most vulner-able to climate change whereas it is the richer sections of the population who are contributing to global warming with their lifestyle. Mukhopadhyay (2009) confirms this claim by showing in an input-output structural decomposition analysis that the recent increase in CO2 emissions

in India is mainly driven by the final demand from households, and that among these households it is the richest that emit most. Thus, from a fairness perspective, policies designed to tackle Indian CO2 emissions should impose a larger burden on the rich. Ananthapadmanabhan et al.

(2007) call for the implementation of a carbon tax on fossil fuels, but do not further discuss distributional consequences of such a policy.

In view of this dichotomy between urgently needed climate action at the Indian level and the danger of placing a non-bearable burden on the poor when implementing an emission tax, the question of the distributional consequences of this very tax appears to be of great importance. So far, there is few scientific research on the distributional effects of GHG emission taxation in India.4 Datta (2010) combines Indian household survey data from the CES 2004-2005 with input-output information to assess the distributional effects of taxing CO2 emissions from fossil

fuels in India. He finds a progressive effect, both for the tax imposed on the direct consumption of fossil fuels for cooking, lighting and transportation, as well as for the indirect consumption of fossil fuels as intermediate industry inputs for other final demand products. However, his analysis is restricted to CO2 emissions resulting from non-renewable, fossil fuel5 combustion

and therefore neglects the effects of burning biomass like wood or dung. Moreover, other GHGs like CH4 or N2O are not considered despite their significant contribution to global warming.

In a more recent study, Rathore and Bansal (2013) examine the distributional effects of carbon taxation using household data from the CES 2009-2010. In contrast to Datta’s study and the approach of this thesis, the authors do not apply input-output techniques but consider solely the direct household consumption of fossil fuels as the tax base. They distinguish the fossil fuels used for domestic purposes and those for transportation purposes. While the tax on domestic fossil fuel consumption has clearly regressive outcomes in cities, it is slightly progressive in

4_{Very recently, in June 2020, a paper by Azad and Chakraborty was published, which analyzes the distributional}

effects of CO2 taxation in an input-output framework for India. Since this publication preceded the submission of

my thesis by a few days only, I do not include the discussion of that paper in my work.

(13)

8 rural areas according to their research. With respect to a CO2 tax imposed on transport fuels,

Rathore and Bansal find it to be progressive at both the urban and the rural level. This confirms the aforementioned idea that, in LMICs like India, transport in cars and other private vehicles is rather a luxury good, reserved for the richer sections of the population.

There is quite an extensive strand of literature on residential, i.e. direct, emissions by Indian households via the combustion of fuels that is not explicitly linked to emission taxation (Kavi Kumar and Viswanathan 2007, 2013; van Ruijven et al. 2011; Ahmad et al. 2015; Hussain et al. 2017). Nonetheless, such studies often examine the amount of direct household emissions across different income groups and consequently, shed light on which distributional effects can be expected from the taxation of direct GHG emissions. Most of these studies hypothesize a so-called “energy ladder”, implying an increase in household income is associated with the switch from traditional biomass like wood to more convenient energy sources like modern fossil fuels or electricity (Leach 1992). A common conclusion of the research on direct Indian household emissions is that the poorer parts of the population do still rely heavily on traditional biomass as their main source of domestic energy. On the one hand, this has immediate detrimental ef-fects on the health of household members due to indoor air pollution (Wilkinson et al. 2009; Kavi Kumar and Viswanathan 2013). On the other hand, the domestic combustion of fuels harms the global climate because of the release of GHGs in the atmosphere. A reduction of direct household emissions is therefore desirable from a local health and global environmental perspective.

In conclusion, the reviewed literature agrees that emission taxation is an effective instrument to reduce emissions, but that concerns about potential regressive distributional effects remain. The evidence from HICs shows indeed that poor income groups tend to be relatively more burdened by emission taxation. In contrast, the fewer studies available for LMICs report mixed results on the distributional consequences of emission taxation, some finding progressive and others re-gressive effects. Given this heterogeneity, no valid statement can be made about the distribu-tional effects to be expected from a GHG emission tax in India without specific knowledge about the country’s consumption and production structure. In this thesis, I aim to provide such a country-specific analysis of the distributional effects of a hypothetical GHG emission tax in India. Based on the literature discussed above, some statements can be made about what ex-penditure effects from GHG emission taxation to anticipate for different income groups. This might give first insights on how my research question might be answered.

First of all, the clear evidence from HICs that a CO2 tax produces regressive outcomes cannot

be assumed to hold also for India. In contrast, since rich Indian households are likely to spend a larger share of their overall expenditures on carbon intensive goods and services like luxury manufacturing products, the taxation of indirect CO2 emissions embodied in the final products

might burden high-income households more heavily. This guess is strengthened by the literature strand on energy regarding the energy ladder which concludes that richer households are more likely to use electricity as their domestic energy source. The substantial CO2 emissions from

(14)

9 The case is different for direct emissions. As discussed above, it is especially the poorer popu-lation segments that use fossil fuels and biomass extensively to generate energy domestically and consequently allocate a considerable part of their budget to residential fuels. In contrast to the study on carbon taxation of fossil fuels by Datta (2010), I include the emissions from the combustion of biomass into my analysis. There is an ongoing discussion whether the combus-tion of biomass like firewood or organic residues should be treated as carbon-neutral or not (Searchinger et al. 2009; Schlesinger 2018). Some argue that, even though the combustion of biomass releases CO2 and other GHGs, these emissions are not newly created but just

re-emit-ting the CO2 that was previously absorbed by the living plant. Searchinger et al. (2009, p. 527)

call this common practice a “critical climate accounting error” if the associated land use is not taken into account, i.e. if it is not considered whether new plants are grown to replace the com-busted ones. Accepting this argument, I will include the use of firewood into the direct house-hold emissions assuming a non-renewability factor of 10% for trees.6_{That way, I expand the}

earlier research on India to direct emissions of biofuels. This implies to include a significant emission contribution by poorer households, so that the resulting tax outcome is expected to be less progressive compared to the effects found by Datta (2010) and Rathore and Bansal (2013). The final, and maybe most important, contribution of my thesis to the existing literature is the extension to CH4 and N2O emission taxation. Non-CO2 emissions are projected to rise

signifi-cantly in the futures, especially in LMICs (Rao and Riahi 2006). Hence, these GHGs should not be neglected when designing climate change mitigation policies for a country like India. CH4 and N2O, the two most important GHG for global warming after CO2, are released on a

large scale in the Indian agricultural sector (MOSPI 2013). The husbandry of livestock is the largest source of non-CO2 emissions in India, but also the cultivation of rice releases and

ferti-lization activities emit significant amounts of both CH4 and N2O (Shukla et al. 2003; Anand et

al. 2005; Oo et al. 2018). An emission tax that includes CH4 and N2O is therefore likely to

increase the prices of agricultural products and food significantly. Since poor households tend to spend high shares of their budget on nutrition, rise in food prices due to the taxation of CH4

and N2O can be expected to hit poor households disproportionally hard.

3. Data

As mentioned above, an emission tax affects households depending on their direct emissions due to the combustion of fuels, and indirectly according to the emissions embodied in their consumption. Information on direct emissions can for example be gained from household sur-veys indicating fuel expenditures. With respect to indirect emissions, the combination of input-output analysis with household survey data is a convenient and effective method to portray the GHG emissions embodied in the consumption of different household types (Kok et al. 2006). This thesis is applying this approach, with the basic strategy displayed in Figure 1. Both the analysis of direct and indirect emissions and the associated tax burden is performed separately for ten income groups that are determined based on information from the CES. The classifica-tion of households into the income deciles will be explained in secclassifica-tion 4.1. The remainder of this chapter focusses on explaining the two main data sources and their merging.

(15)

10 3.1 Expenditure data and direct household emissions

Since 1972, the Indian National Sample Survey Office (NSSO) conducts nationally representa-tive Consumer Expenditure Surveys (CES) at quinquennial intervals (NSSO 2013b). For these surveys, the NSSO selects households randomly in a stratified multi-stage sample design and asks them about their recent consumption expenditures. Since the CES interviews are spread evenly over the year, seasonality effects are eliminated (NSSO 2000).In this thesis, household data from the ninth survey of the CES series is used, the CES of the financial year 2011-2012 (NSSO 2013a).7 The interviews for this survey were carried out between July 2011 and June 2012. The survey reports household expenditures on 346 different consumption items. These items comprise 151 items in the category “food and drinks” (including tobacco), 15 items in the category “energy consumption” (fuel, light, cooking and household appliances), 37 items in the category “clothing, bedding and footwear”, 19 in the category “educational and medical expenses”, 52 different durable goods and 72 other items not elsewhere classified. A list of all consumption items can be found in Table A - 1 of the appendix.

In order to get a simple measure of each household’s standard of living, the NSSO summarizes the expenditures on the 346 different for each household in one variable, the “Monthly Per Capita Expenditures” (MPCE). The MPCE is calculated by dividing the household monthly consumer expenditure (i.e. the sum of expenditures over all 346 items) by the number of house-hold members.

The easiest approach to derive the MPCE is to design the survey questions such that households are directly asked about their expenditures during the last 30 days. This implies that the refer-ence period, i.e. the time period for which the household consumption is recorded, is identical across all consumption items. However, such a uniform reference period might not be

7_{The NSSO conducted a CES in 2017-2018 but did not publish it due to data quality issues.}

Figure 1

The combination of household survey and input-output data

(16)

11 appropriate for different types of items. On the one hand, expenditures for more durable goods like clothing, education or furniture are probably not well captured when just looking at the past month. If, for instance, school fees have to be paid only once a year, there is a high chance of missing this expenditure when exclusively asking for the last 30 days’ expenditures. On the other hand, expenditures on daily and less durable goods like perishable food are unlikely to be accurately reported for long reference periods. For example, an interviewed person might re-member easily how much her household spent on fruits during the last week but fail to recall which value of fruit was exactly consumed during the last month.

Being aware that different item groups require different reference periods, the NSSO conducts the CES in two distinct schedule types. One half of the household sample is subjected to sched-ule type 1, the other half to schedsched-ule type 2. Both types cover the same range of 346 consump-tion items, but they differ regarding the reference periods used for collecconsump-tion of consumpconsump-tion data. The two schedule types result into three different measures of MPCE. In schedule type 1, households are asked to report their expenditures on all 346 items during the last 30 days. Cal-culating the MPCE based on this information yields the “Uniform Reference Period” (URP) measure of per capita expenditures. Households subjected to schedule type 1 are additionally asked to also reveal their expenditures on some more durable goods during the last 365 days. Breaking these yearly expenditures to their monthly counterparts and combining them with the 30-days consumption information results in the “Mixed Reference Period” (MRP) MPCE. The schedule type 2 of the CES differentiates the reference periods even further by asking for the expenditures on high-frequency items like perishable food during the last 7 days instead of 30 days. The MPCE constructed based on this type 2 approach is called “Modified Mixed Refer-ence Period” (MMRP). Table 1 provides an overview on the differing referRefer-ence periods. across the schedule types. In this thesis, schedule type 2 will be used since its reference periods are in line with the recommendation of a NSSO expert group that had been formed for the purpose of suggesting the most suitable reference period for each item of consumption (NSSO 2000).

For the schedule Type 2 of the CES 2011-2012 (simply referred to as CES in the following), 101,651 Indian households with 464,730 household members were randomly selected in a strat-ified multi-stage sample design. The process of sample selection covered almost whole India,

Table 1

Reference periods for the different MPCE measures

Item

cate-gory Item group

Reference period used for MPCE measure

Schedule type 1 Schedule Type 2

URP MRP MMRP

I Clothing, bedding, footwear, education,

institutional health, durable goods Last 30 days Last 365 days

Last 365 days

II Perishable food, pan, tobacco, intoxicants Last 30 days Last 30 days Last 7 days III Other food, fuel and light, miscellaneous

goods and services Last 30 days Last 30 days Last 30 days

(17)

12 both geographically8 and inhabitant-wise9 (NSSO 2013b). Each canvassed household is at-tached a sampling weight, which can be interpreted as the number of households from the pop-ulation that is represented by this sample household. For example, if a surveyed household is assigned the weight 100, it represents 100 Indian households. Taking this weighting scheme into account, the universe covered by the survey is 250,362,592 households with 1,106,309,504 persons. The number of persons, derived as the sum of each household size multiplied with its respective household weight, is slightly lower than the overall population in India in 2011. The consumption of goods that lead to direct household emissions can be taken right from the CES. For this purpose, expenditures related to the combustion of fossil and biomass fuels used domestically (for cooking, heating, lighting) or for private transportation are identified for each household.10 These expenditures are combined with emissions factors, i.e. the amount of GHG gas emitted per unit combusted of the respective fuel. The emission factors are compiled from different sources (Kavi Kumar and Viswanathan 2013; Ahmad et al. 2015; UK Government 2019) and further discussed in the sections 4.2.2 and 4.3.2 of the methodology part.

3.2 Input-output data and environmental accounts

In order to link the expenditures per decile of Indian households to the associated GHG emis-sions, information on the Indian output structure is needed. The respective national input-output table (NIOT) is taken from the most recent release of the WIOD (Timmer et al. 2015), which was published in 2016. The WIOD contains annual time series of world and national input-output tables and factor requirements covering the years 2000 to 2014. The 2016 release of the WIOD (hereafter WIOD2016) distinguishes for each country 56 economic sectors mainly at ISIC revision 4 level, which together cover the overall economy (Timmer et al. 2016). A list of the sectors is given in Table A - 2 of the appendix.

The Indian NIOT allows to calculate the intermediate inputs required to satisfy the final demand of each household decile respectively. In order to subsequently calculate the GHG emission embodied in each decile’s final consumption, emission data in the same industrial breakdown is needed. The environmental satellite accounts of the WIOD, provided by the EU Science Hub (Corsatea et al. 2019), allow this combination with CO2 data. I use the table with the Indian CO2

emissions for 2011, split into the 56 WIOD sectors. This table furthermore provides the total amount of CO2 that was directly emitted by Indian households in 2011.

In the environmental accounts of the WIOD2016, CO2 is the only GHG reported. In contrast,

the previous release of the WIOD (hereafter WIOD2013) additionally provides sectoral emis-sions of two other GHGs, namely CH4 and N2O, for the period 1995 to 2009 (Genty et al. 2012;

Timmer et al. 2015). As mentioned in the literature review, taxing multiple GHGs can improve the cost-efficiency of the tax and might alter the distributional outcomes. For the sake of in-cluding the two non-CO2 GHGs in the analysis, the sectoral emissions of CH4 and N2O are

8_{Exceptions are isolated villages in Nagaland or on the Andaman and Nicobar Islands.}

9_{Exceptions are i.a. persons without normal residence, foreign nationals or persons residing in military barracks.} 10_{As explained in section 2.3, the expenditure on electricity is considered to be related to indirect emission and}

(18)

13 extrapolated to 2011 based on trends in CO2 emission coefficients. This approach is described

in more detail in section 4.3.1.

Even though the WIOD2016 structures the countries’ economies into 56 sectors in general, in the case of India just 45 sectors show domestic economic activity. In eleven of the WIOD sec-tors, there is neither Indian production recorded nor final demand by household registered in any of the years 2000 to 2014. Accordingly, there are no CO2 emissions reported for these

eleven sectors in the environmental satellite accounts. The concerned sectors are flagged in Table A - 2 of the appendix and excluded from the analysis. Consequently, the number of eco-nomic sectors distinguished reduces to 45 for the remainder of this thesis.

3.3 Matching of survey data and input-output sectors

The information on consumption expenditures from the CES has to be assigned to the 45 WIOD sectors. Combining two conceptually different datasets inevitably brings an element of error into the analysis. The first problem associated with the merging is that the two sources are subject to different reporting periods. While the WIOD provides input-output data as an annual time series, the CES does not refer to a specific calendar year but to the period between July 2011 and June 2012. Therefore, a choice has to be made whether the CES expenditure data should be linked to the NIOT for the year 2011 or 2012, or whether some kind of average input-output table from the two years should be constructed and employed. I will use to the NIOT from 2011 because of the nature of the official Indian country data on which the NIOT is based: The “Input Output Transaction Tables” published by the Indian government serve as the bench-mark for the construction of the NIOT (Erumban et al. 2012). Similar to the CES, these official input-output tables do not cover calendar years but the Indian financial years, running from 1 April to 31 March (see for example MOSPI (2009)). Consequently, the NIOT from 2011 is rather reflecting the Indian economy between April 2011 and March 2012 such that linking it to the CES 2011-2012 ensures an adequate overlap in months covered.

Another well-known problem of matching survey data, based on household interviews, and input-output data, usually based on official national accounts, is that the value of total consump-tion derived from the weighted survey data is in most cases significantly lower than the one reported in the input-output statistics (Steen-Olsen et al. 2016; Renner 2018). This discrepancy can also be observed when matching Indian CES and WIOD data: On the one hand, the total (weighted) expenditures by households are 512,827 million US Dollar according to the CES 2011-2012. On the other hand, the WIOD reports total final consumption expenditures by In-dian households for domestic goods of approximately twice that size, namely of 1,083,194 mil-lion US Dollar which is consistent with official national accounts data (MOSPI 2013).

(19)

14 decision to rely on expenditure shares as reported in the CES, it should be noted that the value of total consumption (or emissions) does not influence the outcome of interest of this thesis, namely the distribution of the tax burden relative to household expenditures. Multiplying both tax payments and total expenditures with a scaling factor of approximately 2 (accounting for the difference in total final demand between CES and WIOD) does not influence the relative tax burden across different income groups.

However, there is one relation in the analysis that will be affected by the total used, namely the one between the magnitude of indirect and direct emission tax payments. As has been shown in Figure 1, the direct household emissions are derived from household consumption of fuels whereas the indirect emissions are estimated using input-output technique. When using the WIOD for normalization of the total final demand by households before computing the indirect emissions embodied in that final demand, the direct emissions can be similarly normalized to the totals of direct household emissions reported in the WIOD’s environmental accounts. This is a strong argument in favor of using the WIOD total as the benchmark because the estimates of direct household emissions based on the CES are very rough approximations as will be illus-trated in section 4.3.2.

Apart from the discrepancy in reporting periods and totals reported, the CES and the WIOD differ from each other regarding the level of detail in which they distinguish economic sectors. The CES reports expenditures on 346 consumption items, whereas the WIOD covers 45 sectors in the Indian case. For most of the consumption items, the allocation to a WIOD sectors is unambiguously possible based on the item description in the survey questionnaires (NSSO 2013b, Appendix E). In general, the allocation follows the principles of the United Nations’ (UN) International Standard Industrial Classification of All Economic Activities (ISIC) Revi-sion 4 (UN 2008), on which the WIOD sectors are built. Nonetheless, I make some assumptions in order to allocate all 346 consumption items from the CES to one of the WIOD sectors:

(i) As explained before, I do not assign any CES items to the eleven WIOD sectors that do not show domestic economic activity in India. If one of these sectors would have been the most suitable choice for a CES item, I use the second most appropriate WIOD sector instead. The main reason for this strategy is that allocating CES items to an inactive WIOD sector would be the same as assuming that these activities do not emit any GHGs. The emissions indirectly caused by consuming CES items that are allo-cated to inactive WIOD sectors would consequently be omitted in the analysis. (ii) I do not allocate CES items to the three retail sectors in the WIOD (G45 to G47), even

though these sectors show economic activity in India. Since retailing is the final step of the distribution of goods and services, each CES item could in principle be allocated to these three sectors. Obviously, mapping all expenditures into just three sectors would disregard the variation in sectors needed for the input-output analysis. Nonethe-less, even though the direct final demand by households for the three retail sectors is assumed to be zero, the retail services are still allowed to enter the household con-sumption indirectly as an input for other items.

(20)

15 the share in total household expenditures of the CES items allocated to a particular sector on the one hand, and the share reported for that same sector in the WIOD on the other hand, I verify whether I did not assign too many or too less CES items to the concerned sector. Doing so, I nevertheless take into account that there are sectors for which differences between WIOD and CES are to be expected. For example, as Deaton and Kozel (2005) explain, surveys do typically show a larger share of food in budget. The detailed allocation of consumption items is given in Table A - 1 of the appendix.

4. Methodology

4.1 Construction of the final demand by income decile

The aim of this thesis is to examine the distributional consequences of a hypothetical emission tax for Indian households. In order to assess whether such a tax would tend to be progressive or regressive, I will evaluate the average tax burden relative to the pre-tax total household ex-penditures, for different income groups. The first step is to determine how much each income group spends on goods and services in each WIOD sector respectively.

I divide the sample of households from the CES into income deciles. Following the approach of the NSSO (2013b, Appendix C), the MPCE per household thereby serves as the proxy for income. This means that the income deciles are constructed based on per capita household ex-penditures.11 Not only does the CES not report income data, but the use of current expenditure for distributional analyses has certain advantages over the use of income data. According to Poterba (1989), total expenditure provides a better measure of long-term household well-being than annual income if the lifecycle-permanent income hypothesis holds, i.e. if households smooth consumption over time by setting it proportional to permanent income. This opinion is agreed on by many authors who are analyzing the distributional effects of emission taxation (Wier et al. 2005; Mathur and Morris 2014; Renner 2018). Moreover, the tax base of the hypo-thetical emission tax is expenditures, which makes it relevant to measure the distributional out-come relative to expenditure (Wier et al. 2005).

In general, the use of deciles allows to split up a population into ten equally sized groups ac-cording to a particular variable, which is the MPCE in this case. In other words, the households are sorted by their MPCE into ten equally sized groups, or deciles, from the 10% of the house-holds with the lowest MPCE to the 10% with the highest MPCE. In case of weighted survey data like the one of the CES, the sampling weights have to be taken into account. As each interviewed household represents a different number of households in the overall population, it is not the quantity of surveyed households that is constant across income deciles, but the quan-tity of households represented. This means that the sum of sampling weights needs be identical for all deciles.12

11_{Since the MPCE serves as the baseline for construction the deciles, it might appear to be more precise to talk of}

“expenditure deciles” instead of “income deciles”. However, since “income decile” is the more common term in the literature, I prefer to use it. This is also consistent with the lifecycle-permanent income hypothesis, which is assuming that current household expenditures is proportional to lifetime income (Poterba 1989).

12_{For this thesis, the deciles are determined using the built-in Stata command for weighted percentile calculation.}

(21)

16 Having assigned each household to its respective income decile, I sum the expenditures of all households belonging to the same decile (𝑑 = 1, … , 10), separately for each of the 45 WIOD sectors (𝑖 = 1, … , 45). As motivated in the previous section, the total of household expenditures derived from the CES is then scaled up in order to match the total of household final consump-tion reported in the WIOD:

(1) ℎ𝑖𝑑 = ℎ_𝑖𝑑𝐶𝐸𝑆 ∑45_𝑖=1∑10_𝑑=1ℎ_𝑖𝑑𝐶𝐸𝑆 ∙ ∑ ℎ𝑖 𝑊𝐼𝑂𝐷 45 𝑖=1

ℎ_𝑖𝑑𝐶𝐸𝑆 indicates the overall (weighted) expenditures according to the CES, of households in in-come decile 𝑑 for all the consumption items allocated to input-output sector 𝑖. ℎ_𝑖𝑊𝐼𝑂𝐷 means the final demand for products in sector 𝑖 as reported in the WIOD. Consequently, the sum of ℎ𝑖𝑑 matches the WIOD total but sectoral and decile shares of ℎ𝑖𝑑 are adopted from the CES.13

4.2 Taxation of CO2 emissions

This section will assess CO2 emissions caused by Indian households. In general, there are two

ways in which households are responsible for emissions (Kerkhof et al. 2008):

(i) Indirect emissions are released during the production of goods and services, i.e. the household causes emissions indirectly by purchasing a product.

(ii) Direct emissions occur during the consumption of a product, i.e. the household causes emissions directly by using the product.

An excellent example of the distinction between indirect and direct household emissions is a household owning a car. Indirect emissions are released during the production of the vehicle, for example due to the metal extraction for the car body or the manufacturing of the rubber tires. In contrast, the household emits CO2 directly when driving the car since fuel is combusted.

Consequently, if one aims at taxing all CO2 emissions caused by households, both indirect and

direct emission should be considered. The sum of both is the so-called household carbon foot-print. Section 4.2.1 will address indirect household CO2 emissions while section 4.2.2 will

ad-ditionally include direct household CO2 emissions in the analysis.

4.2.1 Taxation of indirect household CO2 emissions

In the baseline scenario of this thesis, the GHG emission tax will be exclusively levied on CO2

that is emitted during the production process of the goods and services which the households consume. I will refer to the tax payments related to these indirect emissions as the indirect tax

[continued from previous page] households is sorted by ascending MPCE. Subsequently, the running sum of the weights is taken and serves as the benchmark for assigning the deciles. Ideally, the sum of weights in each decile would be exactly the same, namely 25,036,259 (which is the total number of households represented, 𝑁, divided by 10). But generally, this exact matching is not possible, so the number of households represented per decile differs slightly. The general rule for assigning a decile 𝑑 (𝑑 = 1, … , 10) is that the sum of weights up to the respec-tive decile’s “richest” household must be greater or equal to 𝑑 ×𝑁

10. For example, the richest household in decile 3

must have a value of the weight’s running sum of at least 3 × 250,362,592/10 = 75,108,778.

13_{Formally, these relations can be expressed as: ∑} _∑ _ℎ

(22)

17 payments. The input-output technique is applied to consider the total (infinite) chain of inter-mediate deliveries from industry to industry needed to produce the household consumption goods and services. The methodology of the input-output analysis outlined in the following is based on Miller and Blair (2009).

As indicated in section 3.2, the input-output information for the Indian economy is taken from the Indian NIOT provided in the WIOD. The essential information given in the NIOT is the monetary values of product flows from each industrial sector to each of the sectors, itself and others. According to the NIOT, any production in each sector 𝑖 = 1, … , 45 takes place either

(i) to deliver intermediate inputs 𝑧_𝑖𝑗 (from sector 𝑖 to sector 𝑗) to the same or to other industries,

(ii) or to satisfy final demands 𝑓_𝑖𝑘 in six categories 𝑘 = 1, … , 6.14

Since this thesis examines the burden on households imposed by emission taxation, I solely focus on 𝑘 = 1, the final consumption expenditure by households. I thus replace the final con-sumption expenditure by households, 𝑓_𝑖1, by the overall sectoral household expenditure ob-tained in equation (1):

(2) ∑10_𝑑=1ℎ_𝑖𝑑 = 𝑓_𝑖1

and summarize the remaining five final demand categories in 𝑟_𝑖: (3) 𝑟𝑖 = ∑6𝑘=2𝑓𝑖𝑘

The production of sector 𝑥_𝑖 can be represented as follows: (4) 𝑥𝑖 = ∑45𝑗=1𝑧𝑖𝑗+ ∑10𝑑=1ℎ𝑖𝑑+ 𝑟𝑖

implying that production either takes place for delivering intermediate inputs for further pro-duction, or for enabling household consumption (in the ten income deciles), or for satisfying the final demand by others than households.

The input-output technique assumes that each sector 𝑗 uses the intermediate inputs 𝑧_𝑖𝑗 in fixed proportions. This relationship is captured by the intermediate input coefficients 𝑎𝑖𝑗, stating that

one (monetary) unit of output by industry 𝑗 needs 𝑎_𝑖𝑗 (monetary) units of input from industry 𝑖. (5) 𝑎_𝑖𝑗 =𝑧𝑖𝑗

𝑥𝑗

Thus, 𝑧𝑖𝑗 = 𝑎𝑖𝑗𝑥𝑗 can be substituted in equation (4):

(6) 𝑥_𝑖 = ∑45_𝑗=1𝑎_𝑖𝑗𝑥_𝑗+ ∑10_𝑑=1ℎ_𝑖𝑑+ 𝑟_𝑖

The solution of equation (6) for 𝑥_𝑖 can be represented more easily in matrix form. Therefore, I make several assumptions concerning matrix notation. Throughout this thesis, I use lower-case, italicized bold letters for vectors and upper-case, italicized bold letters for matrices. The column vector 𝒊_𝑁 will always refer to a summation column vector of ones with the length 𝑁. The matrix

14_{The six final demand categories in the NIOT comprise final consumption expenditure by households, final}

(23)

18 𝑰_𝑁 will always correspond to an identity matrix of dimension 𝑁 × 𝑁, containing ones in the diagonal and zeros elsewhere. Whenever a prime follows a vector or a matrix, the transpose of that vector or matrix is taken. A hat on top of a vector indicates that I use the diagonal matrix of that vector, e.g. 𝒙̂ represents the diagonal matrix of 𝒙.

With these conventions, the previous equations are transferred into matrix form. Equation (4), the decomposition of output for all sectors, can be summarized in matrix notation as follows:

(7) 𝒙 = 𝒁𝒊₄₅+ 𝑯𝒊₁₀+ 𝒓

𝑯 is the 45 × 10 matrix of sector-wise final demand by Indian households, split into the ten income deciles and normalized to the total of household expenditures in the WIOD. Each ele-ment ℎ_𝑖𝑑 of 𝑯 indicates the final demand of decile 𝑑 for goods and services of sector 𝑖.

The matrix equivalent of the intermediate input coefficients in scalar form, as calculated in equation (5), is the 45 × 45 matrix 𝑨:

(8) 𝑨 = 𝒁𝒙̂−𝟏

Combining (7) and (8), equation (6) can be written in matrix form as: (9) 𝒙 = (𝑰₄₅− 𝑨)−1(𝑯𝒊₁₀+ 𝒓) = 𝑳(𝐇𝒊₁₀+ 𝐫)

The matrix 𝑳 = (𝑰₄₅− 𝑨)−1_{is the Leontief inverse and central to input-output analysis. Each}

element 𝑙_𝑖𝑗 of 𝑳 indicates the amount of output required in sector 𝑖 for a final demand of one unit for sector 𝑗.

While equation (9) determines the overall production of the Indian economy, 𝒙, I am just inter-ested in the amount production needed to satisfy the final demand, 𝑯, of the Indian households. For this case, the production can be represented as follows:

(10) 𝒀 = 𝑳𝑯

The 45 × 10 matrix 𝒀 represents the Indian production which is necessary to satisfy the final demand by households. Each element 𝑦_𝑖𝑑 in 𝒀 indicates the amount of production needed in sector 𝑖 to satisfy, directly or indirectly, the final demand of the households in income decile 𝑑 for goods and services of all sectors.

Equation (10) allows to quantify the amount of production caused by consumption of Indian households but does not yet contain any information about GHG emissions in the production process. Consequently, in the next step, I determine how much CO2 is emitted to produce 𝒀,

i.e. how much CO2 emissions are caused by the households’ consumption. To do so, I link the

NIOT with its environmental accounts, more specifically with the sectoral CO2 emissions data.

By diving the total emissions of sector 𝑖, 𝑢_𝑖 given in kilotons (kt) of CO2, by the respective

sector’s total output 𝑥_𝑖, I compute sectoral CO2 emission coefficients. That is, 𝑐_𝑖 = 𝑢_𝑖/𝑥_𝑖, which

can be written in matrix notation as: (11) 𝒄_𝐶𝑂₂ = 𝒙̂−𝟏𝒖_𝐶𝑂₂

The 45 CO2 coefficients comprised in the column vector 𝒄𝐶𝑂2 depict the environmental impact

in terms of CO2 emissions per monetary unit (here kilotons of CO2 per million US Dollar) for

(24)

19 Knowing about the sectoral CO2 coefficients, the production’s environmental impact can be

quantified along the entire chain of intermediate deliveries from sector to sector up to the point where the household purchases the product. In other words, the CO2 emissions by sector

em-bodied in the Indian households’ consumption can finally be calculated. These indirect emis-sions, 𝑒_𝑖𝑑𝑖𝑛𝑑, can be computed separately for each income decile as follows:

(12) 𝑬𝑖𝑛𝑑 = 𝒄̂𝑳𝑯 = 𝒄̂𝒀

The superscript 𝑖𝑛𝑑 underlines that these emissions are the indirect ones, released in the pro-duction process of goods and services which are finally consumed by the households. In the 45 × 10 matrix 𝑬𝒊𝒏𝒅, each element 𝑒_𝑖𝑑𝑖𝑛𝑑 indicates how many kilotons of CO2 are emitted in

sector 𝑖 in order to satisfy, directly or indirectly as an intermediate input for another product, the final demand of all households in income decile 𝑑.

I consider a tax ρ which is charged per kiloton of CO2 emitted in the production sectors. I make

three main assumptions in order to be able to assess the distributional consequences for house-holds due to the price changes caused by the CO2 tax. First, I assume that the tax is completely

passed forward to consumers in form of higher prices of goods and services. This means that ultimately, the buyers of final products (such as households) bear the full burden of the tax. Second, I assume that that no behavioral response takes place in reaction to the after-tax prices. In other words, I consider a short-term scenario in which neither consumers substitute away from CO2 intensive goods nor producers shift away from CO2 intensive inputs. The third

as-sumption is that just the domestic production sectors are taxed. This implies that just the CO2

emitted in India is subject to the tax. Whether this is the appropriate approach is in the end a political question and beyond the scope of this thesis. My motivations for limiting the tax to domestic production are threefold: I want to focus on a policy that reduces emissions in India, not elsewhere in the world. Moreover, taxing the CO2 embodied in imports would bear the risk

of double taxation if the respective exporting countries also had a carbon pricing system in place. Additionally, import taxes are always accompanied by the danger of sparking a trade war if other countries feel disadvantaged by them and therefore decide to levy import taxes as well. In order to finally assess whether the imposed tax ρ has a regressive or a progressive character, the tax burden carried by each decile relative to the total expenditures of that decile is calcu-lated.

(13) 𝐭𝒊𝑛𝑑 = ρ𝒊₄₅′ 𝑬𝑖𝑛𝑑(𝒊_̂₄₅′ _𝑯)−1

As in equation (12), the superscript 𝑖𝑛𝑑 emphasizes that the formula refers to the indirect tax payments. The 1 × 10 row vector 𝐭𝒊𝑛𝑑 indicates what proportion of the value of its consumption bundle each decile must spend on the carbon tax. If these proportions increase (decrease) with the ascending income deciles, the CO2 tax imposed is progressive (regressive).