Policy Research Working Paper 9907 Illicit Schemes Fossil Fuel Subsidy Reforms and the Role of Tax Evasion and Smuggling Jun Rentschler Nobuhiro Hosoe Urban, Disaster Risk Management, Resilience and Land Global Practice January 2022 Policy Research Working Paper 9907 Abstract This study develops a computable general equilibrium differentials with neighbouring countries, subsidy reform model for Nigeria, which accounts for informality, tax reduces incentives for fuel smuggling. Overall, the results evasion, and fuel smuggling. By studying the impact of show that considering illicit activities reduces the welfare fuel subsidy reform on consumption, tax incidence, and losses of fuel subsidy reform by at least 40 percent. In addi- fiscal efficiency, it shows that the presence of illicit activ- tion, fuel subsidy reductions (and by extension energy tax ities substantially strengthens the argument in favour of increases) have a strong progressive distributional impact. subsidy reform: First, fuel subsidy reform can shift the tax The findings hold under different revenue redistribution base to energy goods, which are less prone to tax evasion mechanisms, in particular uniform cash transfers and the losses than for instance labour. Second, by reducing price reduction of pre-existing labour taxes. This paper is a product of the Urban, Disaster Risk Management, Resilience and Land Global Practice. It is part of a larger effort by the World Bank to provide open access to its research and make a contribution to development policy discussions around the world. Policy Research Working Papers are also posted on the Web at http://www.worldbank.org/prwp. The authors may be contacted at jrentschler@worldbank.org. The Policy Research Working Paper Series disseminates the findings of work in progress to encourage the exchange of ideas about development issues. An objective of the series is to get the findings out quickly, even if the presentations are less than fully polished. The papers carry the names of the authors and should be cited accordingly. The findings, interpretations, and conclusions expressed in this paper are entirely those of the authors. They do not necessarily represent the views of the International Bank for Reconstruction and Development/World Bank and its affiliated organizations, or those of the Executive Directors of the World Bank or the governments they represent. Produced by the Research Support Team Illicit schemes: Fossil fuel subsidy reforms and the role of tax evasion and smuggling Jun Rentschler 1, 2, 3 , Nobuhiro Hosoe 3 1 The World Bank, Office of the Chief Economist for Sustainable Development, USA 2 University College London, Institute for Sustainable Resources, UK 3 National Graduate Institute for Policy Studies, Japan Keywords: Fossil fuel subsidies, tax evasion, smuggling JEL classification: C68, O13, Q58, H23 Acknowledgements: The authors would like to thank Raimund Bleischwitz, Richard Da- mania, Paul Ekins, Florian Flachenecker, Michael Grubb, St´ephane Hallegatte, Dirk Heine, Masami Kojima, Samer Matta, Victor Nechifor, Lars Nesheim, Gregor Schwerhoff, and Adrien Vogt-Schilb for helpful comments and discussions on previous versions of this paper. JR is grateful to the National Graduate Institute for Policy Studies in Tokyo for hosting him as a Visiting Researcher for the purpose and duration of this study. The authors gratefully ac- knowledge that this work was supported by a JSPS KAKENHI Grant (No. 16K03613). Listed affiliations include former ones that were active while the study was being conducted. ∗ Corresponding author: jrentschler@worldbank.org 1. Introduction Global subsidies to fossil fuel consumption amounted to $493 bn in 2014 (IEA, 2015) – i.e. exceeding the climate finance commitment of $100 bn agreed under the Paris Agreement by a factor five. As the IMF (2013) outlines, the true economic and societal cost is bound to be far higher than this figure, since fossil fuel subsidies (FFS) cause countless adverse effects, including the lock-in of inefficient technology and behaviour, crowding out of funds for public spending such as education and health care, erosion of competitiveness, exacerbating environmental pressures, and regressive wealth transfer to the rich (Rentschler and Bazilian, 2016; Coady et al., 2017). In addition to these adverse effects, FFS are also frequently argued to be associated with illicit activities, including corruption, fuel smuggling, and tax evasion. However, no systematic study exists exploring the role of such activities in determining the outcomes FFS reforms. Against this background, this study develops a dedicated computable general equilibrium (CGE) model to study FFS reform, calibrated to reflect the char- acteristics of Nigeria. It introduces several innovative model features, includ- ing a large informal sector, evasion of labour and production taxes, and fuel smuggling. More specifically, this study addresses the following interrelated questions: (i) What are the effects of FFS reform on key economic parameters, including income distribution, consumption, and output? (ii) How do effects differ when illicit market activities are taken into account, in particular tax eva- sion and rampant fuel smuggling? (iii) How do effects differ when FFS are not only removed, but replaced by fuel taxes? (iv) How do different methods of revenue re-distribution affect households? Drawing on the environmental and carbon tax literature (in particular the “dou- ble dividend” field), this study also considers the role of labour tax reductions as a way of re-distributing FFS reform revenues. This is in addition to consider- ing cash transfers, which the existing literature on FFS reforms has focused on predominantly. Conceptually, these two approaches have different advantages: Reducing FFS and redistributing revenues using cash transfers has a strong progressive distributional effect. Reducing FFS and redistributing revenues by cutting labour taxes increases fiscal and economic efficiency, as distortive pre- existing taxes are reduced. By introducing the novel model features on fuel smuggling and tax evasion, this study is able to analyse additional perspectives on these arguments: • Fuel smuggling: FFS cause rampant fuel smuggling to neighbouring coun- tries, meaning that a significant fraction of FFS benefits leaks out of the country. Replacing FFS with cash transfers is not only progressive, but terminates smuggling; i.e. the government can disburse the same aggre- gate benefit to the population at a lower cost. • Tax evasion: Labour taxes not only distort incentives to work, but are associated with high evasion rates and incentivise informal economic ac- tivity. Reducing FFS and using revenues to lower labour tax rates not only mitigates labour market distortions, but reduces overall tax evasion; i.e. the government can earn the same level of tax revenues, while charging lower tax rates. 2 Overall, this study finds that replacing FFS with cash transfers results in sub- stantial progressive redistribution. While the top income quintile incurs a re- duction in consumption, all other income quintiles gain, with the poorest income quintile benefitting from the largest relative consumption increase. Moreover, using revenues from FFS reform to cut labour taxes results in an improvement of fiscal efficiency. Crucially, this study finds that taking into account illicit ac- tivities can lower the social welfare costs of FFS reform by up to 40% in Nigeria. Moreover, by going beyond FFS reform and also considering the introduction of fuel taxes, this study finds that the above mentioned benefits can be extended further: A revenue neutral shift of the tax base from income to fuel taxes can significantly reduce illicit activities and their associated welfare costs. However, this move may also incentivise a reversal of the fuel smuggling direction from out- to inbound. The remainder of this paper is organised as follows: Section 2 discusses the relation of this study to prior literature on FFS, environmental tax reform, tax evasion, and smuggling. Section 3 provides a brief conceptual outline of the social welfare implications of tax evasion and fuel smuggling. Section 4 provides information on the Nigerian FFS scheme. Section 5 offers a detailed account of the CGE model set-up. Section 6 outlines data sources, while Section 7 outlines the simulation scenarios considered in this study, and the measures used for evaluating social welfare and fiscal efficiency. Section 8 presents the results, and Section 9 concludes. 2. Relation to prior literature 2.1. Fossil fuel subsidies Often used as a political tool, FFS have been justified by objectives such as alle- viating poverty, redistributing natural resource revenues to citizens, moderating energy price shocks, and promoting industrialization and economic development (Strand, 2013; Kojima, 2016; Yau and Chen, 2021). However, numerous case studies have helped to document that FFS are a very ineffective way of achieving these objectives, and cause a wide range of adverse side effects pertaining to all dimensions of sustainable development: economic, social, and environmental (Rentschler and Bazilian, 2016). These side effects include economic and technological inefficiency, fiscal imbalances, crowding out of public funds for innovation and social services, pollution, climate change, fuel smuggling, and corruption (IMF, 2013; Coady, Parry, Sears and Shang, 2015; Coady, Flamini and Sears, 2015; ADB, 2015). In addition to the fiscal pressure they place on national budgets, FFS are also particularly problematic because of their highly regressive nature. In reviewing FFS schemes in a range of developing countries, Arze del Granado et al. (2012) found that the top income quintile receives on average six times more in absolute subsidy benefits than the poorest income quintile. In this way, FFS continuously reinforce and exacerbate existing patterns of income and wealth inequality. A comprehensive overview of FFS definitions, adverse side effects, reform progress and barriers is provided by Rentschler and Bazilian (2016). In light of mounting evidence in favour of FFS reform, a key question for policy makers has been how to design and implement reforms in a way that ensures 3 public support for reform and protects livelihoods of vulnerable households. In this context, ex-ante simulations of FFS reforms have been particularly impor- tant in understanding the potential socio-economic consequences of reforms. This in turn has been critical for designing effective social protection and com- pensation measures alongside FFS reforms. Broadly speaking, such studies on FFS removal have used either econometric approaches or general equilibrium models. 2.2. Fossil fuel subsidy reform simulations Econometric reform simulations typically focus on changes in household con- sumption, and are useful for understanding the welfare and distribution effects of FFS reform (Verme and El-Massnaoui, 2015; Rentschler, 2016). Various em- pirical ex-ante impact assessments of FFS reforms have provided detailed assess- ments of the welfare impacts of FFS reforms, and highlight the need to provide adequate social protection (e.g., in the form of cash compensation) along with FFS reform (Coady et al., 2010; Zhang, 2011; Anand et al., 2013; Araar et al., 2015; Bah and Saari, 2020)1 . Both the IMF and World Bank provide analytical toolkits for the empirical sim- ulation of FFS reforms using household expenditure surveys and input-output models (Verme and El-Massnaoui, 2015; IMF, 2016). The approach taken by these models enables swift and consistent analyses with relatively few data re- quirements (Rentschler, 2016). However, as Plante (2014) points out, these models also strongly simplify com- plex interactions, as they overlook the fiscal policy and general equilibrium perspective on subsidies: These models focus exclusively on the fact that the removal of FFS leads to energy price shocks, which in turn reduce purchasing power and aggregate consumption of households. Thus, FFS removal necessar- ily and exclusively has negative consequences, while benefits such as reduced deadweight losses or increased economic efficiency are not taken into account. Similarly, general equilibrium effects on key macro-economic parameters can- not be captured in models based on household surveys, including changes in government expenditure, output, and tax revenues. To address the above mentioned shortcomings, Plante (2014) developed a gen- eral equilibrium model aimed at capturing the fiscal policy aspect of FFS re- form. It is a standard small open economy model with two sectors, and a single representative household. FFS reductions are balanced in the government’s budget constraint through tax adjustments or lump-sum cash transfer provi- sion. Using this model, it is shown that FFS cause distortions of relative prices, which are identified as the main reason for substantial aggregate welfare losses. Moreover, FFS are shown to aggravate fiscal imbalances, crowd out non-energy consumption, and cause inefficient allocation of labour across sectors, regardless of whether the country is an oil importer or exporter (Plante, 2014). Several general equilibrium models have followed: Durand-Lasserve et al. (2015) offer an analysis for Indonesia based on the OECD’s ENV-Linkages model, a 1 For additional empirical impact assessments of FFS reforms, see Jiang et al. (2015) and Ouyang and Lin (2014) for China, and Solaymani and Kari (2014) for Malaysia. 4 global CGE model. This study focusses in particular on distributional effects of FFS reform, considering redistribution of reform revenues through cash trans- fers or food subsidies. Siddig et al. (2014) use the CGE model by the Global Trade Analysis Project (GTAP) to analyse subsidy reform in Nigeria. They distinguish 12 households, each representing a region rather than an income group. Using this set-up they consider the impact of FFS reform on several standard macro-economic parameters, such as consumption and GDP. For Pak- istan, Feltenstein and Datta (2020) assess electricity subsidy reform and revenue redistribution through targeted transfers, and find increases in the real incomes of poor households. Shehabi (2020) develops a general equilibrium model for Kuwait featuring an oligopolistic industrial structure and studies how subsidy reform affects sectoral diversification. While the above mentioned models differ in the complexity of their modeling set-ups, they coincide in their focus on standard macro-economic parameters, in particular output, consumption, trade and fiscal balances, and income distri- bution. They do not consider illicit activities in any way. 2.3. The double dividend There is a prominent strand in this literature suggesting that there is a “dou- ble dividend” – i.e., that if environmental taxes are increased, but other dis- tortionary taxes are reduced (while maintaining revenue neutrality), then not only can environmental benefits be reaped, but also fiscal efficiency can be in- creased (Goulder, 1995a; Fullerton and Metcalf, 1997; Bovenberg, 1999). How- ever, Parry and Williams (2010) also show that there is a considerable trade-off between efficiency and distributional effects. That is, income tax reductions are economically efficient, but regressive, while cash transfers are progressive, but less efficient. Since FFS are (at least theoretically) equivalent to a negative carbon tax, the double dividend argument deserves thorough consideration in the context of FFS reforms. Especially given the significant economic efficiency costs of labour and consumption taxes (Bovenberg and Goulder, 1996; Parry, 1997; Goulder, 1995b), there may be a rationale for using FFS reform revenues to reduce these taxes. However, instead of reducing pre-existing taxes, past reforms and asso- ciated analyses have focused almost exclusively on cash transfers for revenue redistribution. 2.4. Tax evasion and smuggling More recent studies have contributed an additional perspective, which strength- ens the double dividend argument. Liu (2013) observes that energy (or carbon) taxes are more difficult to evade than labour or income taxes. Using a simple CGE model, the author shows that a revenue neutral shift from labour to car- bon taxes can substantially reduce tax evasion and the welfare cost of climate change mitigation policy2 . Especially when pre-existing tax evasion is high, this argument can significantly strengthen both the environmental and fiscal cases for carbon taxes (Liu, 2013; Bento et al., 2013; Carson et al., 2014). 2 Liu (2013) shows that costs can be lowered by 89% in China and 97% in India. 5 Even though tax evasion tends to be particularly high in developing countries – large informal sectors are symptomatic of this – the “tax evasion effect” has not been studied in the context of FFS reform. Similarly, despite being a frequently cited side-effect of fossil fuel subsidies, smuggling has also received little atten- tion from researchers. As subsidised fuels are smuggled out of a country, the government is directly subsidising fuel consumption in neighbouring countries; in other words public funds intended for domestic beneficiaries are continuously leaking out of the country (ADB, 2015). A study by Mlachila et al. (2016) is likely to be the ony one that systematically analyses the magnitude and impli- cations of fuel subsidies on smuggling activities. They show that fuel smuggling can severely undermine the effectiveness of fuel price adjustment mechanisms and energy tax collection, when neighbouring countries subsidise their domestic fuel consumption. The model presented in this study contributes to the literature on FFS reform and the double dividend, by providing a systematic account of tax evasion, the informal sector, and fuel smuggling in the case of Nigeria. These issues are of great significance in developing countries, especially when considering that the informal economy in Nigeria is estimated at 50% of GDP, and that 85% of fuel consumed in Benin is smuggled from (and thus subsidised by) Nigeria (Mlachila et al., 2016; Hassan and Schneider, 2016). 3. The societal cost of illicit activities This section provides conceptual detail on the argument underlying this paper. More specifically, it briefly discusses the welfare costs associated with tax evasion and smuggling activities. A key assumption by Liu (2013) is that tax evasion (and legal tax avoidance) activities incur real costs. The reason for this is that tax evasion efforts require real resources, and thus compete for production factors with real productive ac- tivities (ADB, 2015). Bribes paid to tax officials, the employment of tax lawyers and advisors, labour hours spent on avoiding tax audits, or the complex shifting of profits between international subsidiaries are examples of costly tax avoid- ance or evasion activities. In addition, taxes may reduce economic efficiency by distorting business decisions and production processes. In practice, the in- creased need for fiscal audits and monitoring may also impose a costly burden on authorities. Overall, tax evasion imposes real costs on society, not only be- cause productive resources are allocated to unproductive evasion activity, but also because of foregone tax revenues. Analogously, fuel smuggling activities also represent a loss to society. Fuel sub- sidies that are intended for the benefit of a country’s own citizens are smuggled abroad; in other words, the government is directly subsidising the fuel con- sumption by foreign consumers. In addition, the smuggling activity itself incurs real economic costs, as significant resources are used on transporting fuel, and avoiding and bribing border and customs controls (Mlachila et al., 2016). A key argument considered in this study is that reducing FFS and using revenues to cut distortionary labour taxes can be welfare-enhancing, since welfare losses associated with smuggling and tax evasion are also reduced. In line with Parry 6 and Williams (2010), this study contrasts such labour tax cuts with other ways of recycling reform revenues: increasing public expenditure, or the provision of uniform lump-sum transfers to citizens. 4. Background: Fuel subsidies in Nigeria As a developing country with substantial fossil resource wealth and a mixed track record of fiscal prudence and transparency, Nigeria is a frequently cited case for studying FFS and natural resource management (Rentschler, 2016). Nigeria is the fifth largest oil exporter in the world (World Bank, 2015; IMF, 2013). Yet, only 55% of Nigerians have access to electricity (34% in rural areas); annual per capita electricity consumption in 2012 was 155 kWh (compared to 4,405 kWh in South Africa). Chronic underinvestment and corruption in the electricity sector mean that the average Nigerian enterprise experiences over 36 power outages a month. Problems also plague the country’s four national oil refineries, which operate at just 20% to 30% capacity. While over 70% of fuel consumption is met by imports, shortages are endemic (World Bank, 2015; IMF, 2013). Through the Petroleum Products Pricing Regulatory Agency, Nigeria maintains artificially low fuel prices, specifically petrol. The gap between fuel import costs and regulated prices are financed through the Petroleum Support Fund, which administers fuel subsidies3 . Nigeria does not provide significant subsidies to fuels other than petrol. At about 3% of GDP in 2011 subsidies are a significant expense for the gov- ernment, and fail to reach Nigerians in more than one sense (NEITI, 2013; IMF, 2013; Soile and Mu, 2015): As with all FFS schemes, the direct financial benefits to households are concentrated on the rich, thus failing to benefit the absolute poor (who constitute 61% of the population4 ). In addition, a complex and opaque system of intermediary dealers and political influence means that, instead of lowering the market price, subsidies are often privately appropriated before the fuel reaches the market. Finally, rampant fuel smuggling means sub- sidy benefits are leaking out of the country. Facing mounting fiscal pressures and recognising the inefficiencies of its subsidy scheme, Nigeria attempted a radical subsidy reform in 2012. While the need for such reform was pressing, the government failed to garner sufficient public support for its reform efforts. Public opposition to the reform had two key reasons in particular: (i) A lack of credibility and transparency with respect to the handling of reform revenues, and (ii) inadequate plans for compensation and social protection, resulting from a poor understanding of the needs and vulnerability of affected consumers. Subsidy removal was met with extensive 3 The Petroleum Support Fund is managed by the Petroleum Products Pricing Regulatory Agency, and receives a set allocation in the federal budget. Contributions to the fund are made by the federal, state, and local governments. Moreover, the fund is supplemented by subsidy “surpluses”, which essentially occur when international market prices exceed the government-set fuel price (GSI, 2012). 4 This figure is based on the national absolute poverty definition, using an absolute poverty line of N54,401 (National Bureau of Statistics, 2010) 7 strikes and violent public protests, and prompted the government to swiftly reintroduce subsidies (Bazilian and Onyeji, 2012; Siddig et al., 2014). 5. Set-up of the illicit activities model The basic structure of this model builds on Hosoe et al. (2010), who offer a small open economy CGE model featuring a representative household, two sectors, and a government. The model developed for the purpose of this study is a static small open economy model, featuring multiple representative households (representing five income quintiles), firms, and a government. The government collects taxes on production, factor usage, and imports, as well as direct taxes on households, and disburses FFS. Moreover, the model adds several non-standard features: A small group of households is modeled to engage in fuel smuggling activities. The profits of smuggling depend on the price differential between the domestic and interna- tional price of fuel, which is determined through the level of subsidies paid. The model also allows for the evasion of production and factor taxes. In order to represent different policy options for using reform revenues, the model consid- ers (1) government expenditure, (2) direct cash transfers, and (3) reductions of pre-existing taxes, such as labour or production taxes. 5.1. Tax evasion: Representation in the model In this model it is assumed that firms choose to evade taxes on different factors and on production, while incurring a real cost of evasion. Formally, factor taxes f τh,j on factor h and sector j are evaded at the rate ef h,j . Similarly, production z taxes τj are evaded at the rate ezj . In line with Liu (2013), tax evasion activities loss loss incur real costs Fh,j and Xj : Af h,j f loss Fh,j = c(ef h,j )Fh,j with c(ef h,j ) = f (ef h,j ) Nh,j +1 (1) Nh,j + 1 loss Azj z Xj = c(ez j )Zj with c(ez j) = (ez )Nj +1 (2) Nj + 1 j z This setting implies that the cost of tax evasion is measured as a share c(e) of either factor inputs Fh,j (in the case of factor tax evasion) or production Zj (for production tax evasion). Parameters Af z f z h,j , Aj , Nh,j , and Nj characterise the cost function and are determined during calibration. The total net benefit to f firm j from evading factor taxes is h Fh,j (τh,j ef f f h,j ph − c(eh,j )). Evasion losses reflect the extra (unproductive) “self-input” due to evasion ef- forts. Specifically, if a firm engages in evasion activities, a share of its resources will be directed towards this evasion activity rather than production (e.g. labour hours). This unproductive evasion activity means that fewer resources are avail- loss able for producing output, thus resulting in the output loss Xj . 8 Moreover, the marginal costs of evading taxes can be expressed as ∂c(ef h,j ) f ∂c(ezj) z = Af f h,j (eh,j ) Nh,j and z = Az z Nj j (ej ) . (3) ∂ef h,j ∂ej Since these marginal costs are increasing in ef z h,j or ej , firms choose evasion rates and factor input quantities as to maximise profits (Section 5.3). 5.2. Smuggling: Representation in the model In practice, fuel subsidies often incentivise smuggling, as subsidised domestic fuel prices are significantly lower than in neighbouring countries, where fuel prices are unregulated. This price gap presents a lucrative opportunity for smugglers, who buy fuel at the subsidised domestic price and sell it abroad at the unregulated market price. However, in doing so smugglers are likely to incur real costs, e.g. in the form of bribes or transport costs (Mlachila et al., 2016). For the sake of consistency, this model not only allows such outbound smuggling (smuggling “exports”); it also considers the possibility of inbound smuggling (“imports”), which may occur by the same logic when domestic fuel subsidies are removed and fuel taxes imposed. In this model, smugglers are assumed to choose the (outbound ) smuggling quan- SM tity Ej (“exports”) as to maximise their profit. They purchase fuel for smug- q gling in the domestic market at the subsidised price (1 − se e j )pj (where sj is the e subsidisation rate), and sell it abroad at the export price pj . In addition, smug- loss glers incur smuggling costs Ej , e.g. transportation costs. Analogously, they SM may also choose to conduct inbound smuggling Mj (“imports”). In this case they will purchase fuel abroad at the import price pm j , and sell it domestically at the composite good price pq j , which is subsidised (or taxed) at the rate (1 − se j ). Thus, the smugglers optimisation problem is given by SM e q q loss max SM ,M SM πj = pe SM j − (1 − sj )pj Ej − (1 − se j )pj Ej Ej j q + (1 − se m j )pj − pj SM Mj − pm loss j Mj (4) loss ASM j SM r subject to Ej = Ej (5) r loss ASM j SM r Mj = Mj (6) r SM SM Ej (or Mj ) denotes the smuggled quantity from sector j , while the total loss loss cost of smuggling is denoted by Ej (or Mj ) expressed as a share of the total smuggled quantity. The shape of the cost function is characterised by parameters ASM j and r, both of which are determined during calibration5 . 5A sensitivity analysis for these parameters is provided in Appendix C. 9 The smuggler’s profit is maximised under following first order conditions: SM ∂πj e q e q SM r −1 SM = pe j − (1 − sj )pj = (1 − sj )pj Aj SM Ej (FOC 1.1) ∂Ej SM ∂πj q r −1 SM = (1 − se m m SM j )pj − pj = pj Aj SM Mj (FOC 1.2) ∂Mj SM For both FOC 1.1 and FOC 1.2, πj is maximised when the marginal benefit of smuggling (LHS) is equal to its marginal cost (RHS). Note that the sign of the subsidisation rate se j plays a key role in determining whether the smuggler chooses inbound or outbound smuggling. Solving the first order conditions yields SM SM the profit maximising smuggling quantities Ej and Mj . 1 e q SM pe j − (1 − sj )pj r −1 Ej = q SM (7) (1 − se j )pj Aj 1 q SM (1 − se m j )pj − pj r −1 Mj = (8) pm SM j Aj 5.3. Domestic production The model by Liu (2013) is extended to include intermediate inputs, by distin- guishing two stages of the production process. In the first stage, the firm uses primary factors (incl. labour, capital, energy) to produce a composite factor. In the second stage, the firm combines the composite factor with intermediate inputs to produce its output. Based on this set-up, the following set of equations describes the firms’ profit optimising behaviour6 : py j Fh,j = βh,j Yj (9) pf f f f h (1 + τh,j (1 − eh,j ) + c(ej )) Xi,j = axi,j Zj (10) Yj = ayj Zj (11) βh,j Yj = bj Fh,j (12) h Xi,j Yj Zj = min , (13) axi,j ayj 1 1 f f Nz τh,j N h,j pz j z τj j ef h,j = and ez j = q z (14) Af h,j 1 − se j p j Aj Equation 9 describes the optimal factor demand Fh,j for factor h in sector j . Equations 10 and 11 describe the optimal conversion of intermediate inputs 6 For details see Appendix A.1. 10 Xi,j (priced at pq y i ) and of the composite factor Yj (priced at pj ) into output Zj (priced at pzj ). Equations 12 and 13 represent the technological constraints in the first and second production stage in the form of Cobb-Douglas and Leontief production functions. Equations 14 represent the optimal evasion rates for factor (ef z h,j ) and production taxes (ej ), which are increasing in their corresponding tax rates τ . Production factor h is priced at pf h , and the Armington composite good Qi is priced at pq i . The composite factor Yj is produced in the first production stage, and is priced at py j . β h,j is the elasticity parameter in the first-stage production function, and bj an efficiency parameter. Moreover, axi,j (or ayj ) is the input requirement coefficient of the i-th intermediate input (or j -th composite factor) for one unit of the output j . 5.4. Government The government in this model takes the role of levying taxes, consuming goods, and providing subsidies and direct cash transfers. Formally, government con- g sumption Xi is a sum of its revenues from different tax sources, net of subsidy e payments (Sf ) and cash transfers (Cttax l ): g µi f Xi = Tld + Th,j + Tjz + Tjm − e Sj − Cttax pq i j j j j l l h l (15) where the share of the i-th good in government expenditure is denoted µi . Direct taxes Tld are levied on the factor endowment F Fh,l of household l at the rate τld . Tld = τld pf h F Fh,l (16) h f f Factor taxes Th,j are levied on firms’ factor inputs at the rate τh,j and are subject to evasion. f Th,j = (1 − ef f f h,j )τh,j ph Fh,j (17) Production taxes Tjz are levied on output Zj by firm j at the effective tax rate (1 − ez z j )τj after evasion. Tjz = (1 − ez z z j )τj pj Zj (18) m Import taxes are levied on imports Mj at the rate τj . m m Tjm = τj p j Mj (19) g Besides public expenditure on goods (Xi ), the government also provides energy 11 e (i.e. petrol) subsidies Sj at the rate se j. e q p Sj = se j pj Xj,l + loss Xj,i + Xj (20) l i p Note that the subsidy sej is provided for household consumption (Xj,l ) and for energy (petrol) as an intermediate input (Xj,i ). Government and investment g v demand (Xi and Xi ) are not subsidised. Investment demand is specified in Appendix A.2, while cash transfers Cttaxl are defined in section 5.5. 5.5. Households This study distinguishes five households, each representing an income quintile, as well as an additional smuggler, representing a relatively small number of households engaged in fuel smuggling activities. Households are modeled to maximise their utility subject to a standard budget constraint. The optimised p consumption choice Xi,l can be expressed as p αi,l Xi,l = pf tax h F Fh,l + Ctl + Ctz l + Ctf p l,h − Ssl − Tl d . pq i (1 − se i) h h (21) The first term in the round parentheses reflects income from factor income (e.g. wages); the second term reflects direct government transfers for redistributing tax revenues; and the third and fourth terms reflect the benefits of produc- f tion (Ctzl ) and factor (Ctl,h ) tax evasion which ultimately accrue to households. These income sources are balanced by savings (Ssp d l ), direct tax payments (Tl ), p and consumption (Xi,l ) of good i which is determined by αi,l , the share param- eter in the utility function. Government cash transfers for redistributing tax revenues are implied by equa- tion 15 and can be expressed as: f Cttax l tax = Rl Tld + Th,j + Tjz + Tjm l h j j j (22) − e Sj − pq g i Xi j i The overall budget for these redistribution transfers is given by tax revenues from four different tax types, from which subsidy payments and government tax consumption must be subtracted. The redistribution rule Rl determines the share of the overall redistribution budget obtained by each household. Moreover, household income from tax evasion activities are defined as Ctz z l = Rl z z z τj ej p j Zj (23) j 12 for the evasion of production taxes, and Ctf f l,h = Rl,h f τh,j ef f h,j ph Fh,j (24) j for factor tax evasion. The benefits of tax evasion are distributed across house- z f holds according to redistribution rules Rl and Rl . The numerical values of these parameters are chosen to reflect the distribution of consumption and fac- tor endowments. In addition to the five households, this model considers a smuggler who consumes the same goods as all other households, but earns income from fuel smuggling activities. Thus, the smuggler’s budget constraint prescribes that his consump- tion expenditure equals smuggling profits: SM αj SM Xj = πj (25) pq j (1 − se j) j The smuggler’s share parameter αj is calibrated according to the share parame- ter of the 2nd income quintile household. This choice reflects anecdotal evidence that fuel smuggling is typically done by low income households (Mlachila et al., 2016), but the relatively small number of smugglers means that this parameter is inconsequential for overall results. Note that for consistency, the above nota- tion allows smuggling in all sectors j , yet the empirical evidence suggests that it is a relevant consideration mainly in the petrol sector. 5.6. Exports, imports, and the balance of payments For considering the implications of cross-border smuggling, the use of an open economy model is necessary. This section briefly sets out the interaction between the model economy and the rest of the world. For this purpose a small open economy set-up is used, which implies that import and export prices (denom- inated in foreign currency terms) are exogenously given. Formally, domestic import (pm e i ) and export prices (pi ), are linked to their corresponding world Wm We prices (pi and pi ) through the exchange rate ε. pe We i = εpi (26) pm i = εpW i m (27) The balance of payments condition requires that monetary outflows (i.e. due to imports Mi , and inbound smuggling MiSM and Miloss ) equal inflows (i.e. exports Ei , “foreign savings” or the current account deficit Ssf , and gross earnings from SM the foreign sale of smuggled fuel Ei ). pW i m Mi + MiSM + Miloss = pW e f i Ei + Ss + pW e SM i Ei (28) i i i Details on the substitution and transformation of traded goods are provided in Appendix A.3. 13 5.7. Market clearing Two equilibrium conditions are needed to ensure the equivalence of demand and supply in goods and factor markets. For the goods market: p g v SM loss loss Qi = Xi,l + Xi + Xi + Xi,j + Xi + Xi + Ei (29) l j This condition implies that the supply of the i-th Armington composite good Qi must equal its aggregate demand. Demand is composed of demand by house- p g v holds (Xi,l ), the government (Xi ), investment (Xi ), firms (Xi,j ), and the smug- SM gler (Xi ); in addition some of the goods are lost as inputs to tax evasion loss loss (Xi ) and smuggling (Ei ) activities. For the factor market, the sum of endowments of the h-th factor (F Fh,l ) must equal the aggregate factor demand: loss F Fh,l = Fh,j + Fh,j (30) l j Note that firms’ total factor demand is the sum of standard factor demands for production (Fh,j ) and factors used for the purpose of tax evasion activities (e.g. loss labour), denoted Fh,j . 6. Data 6.1. Economic variables The baseline values for macro-economic parameters have been obtained from the GTAP 9 database – in particular, Nigeria’s social accounting matrix (SAM) for the 2011 reference year. These macro-economic parameters are the size of economic sectors (i.e. output), intermediate inputs, capital and labour inputs, taxes, government expenditure, household consumption, imports, exports, and the current account balance. Four sectors are distinguished: (i) the (subsidised) petrol sector, (ii) the (unsubsidised) energy sector, which excludes petrol, (iii) the formal (non-energy) sector, and (iv) the informal (non-energy) sector. The parameters and coefficients αi,l , βh,j , bi , γi , µi , λi , θi , axi,j , ayj , δmi , δdi , ξei , ξdi , sspl , and ssg have been calibrated on the basis of the 2011 baseline data, and the model equations set out in Section 5. Since the GTAP 9 SAM does not provide information on the distribution of income and consumption, aggregate household consumption figures have been split into income quintiles according to expenditure shares contained in the Harmonised Nigeria Living Standards Survey 2010. Across income quintiles, this household expenditure survey provides details on the level of spending on petrol, other energy, and non-energy consumption goods. It is thus essential for the distributional aspects considered in this study. Data on FFS in Nigeria have been obtained from the International Energy Agency’s World Energy Outlook 2015 Fossil Fuel Subsidies database (IEA, 2015). The IEA provides an estimate of $7.1 bn of fossil fuel consumption subsidies in 2011, of which over $6.5 bn are paid to subsidise oil (primarily 14 petrol)7 . This figure, in combination with the estimated size of the petrol sec- tor, translates to a baseline subsidisation rate of 17.8%. Official (subsidized) petrol prices were set at N65/liter in 2011. This price creates incentives for a variety of illicit activities and commercial malpractice, including smuggling and black market sales which exploit local petrol shortages. As a consequence, actual petrol prices observed by households are often significantly higher than N65/liter (Kojima, 2016).8 Nevertheless, observed petrol prices in Nigeria are still lower than in neighbouring countries, thus preserving an incentive for smug- gling. Finally, population data has been obtained from the World Bank’s World Development Indicators database. 6.2. Tax evasion For the purpose of comparing sectors with high and low tax evasion, Liu (2013) uses the self-employment rate of an economy to approximate the size of the high-evasion sector. This approach works particularly well in developed and emerging economies, for which reasonably reliable estimates of self-employment are available from sources such as the International Labour Organization (ILO). However, the ILO’s Labour Statistics database offers no estimate of the self- employment rate in Nigeria. Instead of relying on uncertain alternative estimates for self-employment, this study uses the size of the informal sector as a proxy for the high-evasion sector. The advantage of this approach is that estimates of informal economic sectors exist for a wide range of countries – including developing countries. Specifically, the formal economy estimates from the GTAP 9 database are supplemented by including an informal sector, which in 2011 measured 50% of GDP according to the comprehensive analysis by Hassan and Schneider (2016). This study uses a very conservative estimate of 2% for the tax evasion rate in the formal economy. For comparison, Liu (2013) uses a 5% evasion rate in a selection of 27 developed and emerging economies, based on estimates by Slemrod (2007). The Swedish National Tax Agency (2008) reports a 4.8% evasion rate for income taxes in its jurisdiction. In line with the notion of informality, this study assumes that the informal sector does not pay any taxes. Moreover, this study uses the conservative assumption by Liu (2013), that 10% of evaded taxes are spent on non-productive evasion activities (see Section 3). The net benefits of tax evasion are assumed to ultimately accrue to households; in line with the distribution shares of regular income, the top income quintile is assumed to benefit disproportionately more than lower quintiles. Based on these numbers, the evasion parameters Af f z z h,j , Nh,j , Aj , and Nj are calibrated to characterise the evasion cost functions (Eq. 1 and 2). 7 To make subsidy estimate comparable across countries, the IEA applies some aggregation and simplification assumptions. This results in a lower subsidy estimate than provided by the Nigeria Extractive Industries Transparency Initiative (NEITI, 2013), which estimates total subsidy payments in 2011 to be $12.4 bn. Of this sum, $7.3 bn were paid through the Petroleum Product Pricing Regulatory Agency (PPPRA) and $5.1 bn through the Nigerian National Petroleum Corporation (NNPC). 8 The National Bureau of Statistics only reports observed petrol prices since 2014, hence we apply a 50% mark-up as a simplifying assumption for household petrol consumption. 15 6.3. Smuggling This section outlines the steps taken to estimate the baseline magnitude of fuel smuggling out of Nigeria to neighbouring countries. The first step is to focus on smuggling activity from Nigeria to Benin (and from there to Togo), as detailed estimates are available from the IMF (Mlachila et al., 2016). While Togo does not share a border with Nigeria, its distance to the Nigerian border is under 120 km, and thus extensive smuggling occurs via Benin. The IEA (2016) reports total gasoline consumption for Benin (616 k tonnes) and Togo (175 k tonnes) in 2011. Mlachila et al. (2016) outline that in both Benin and Togo respectively, gasoline is sold on two separate markets: an official market for the sale of legal and regulated gasoline, as well as an informal market for the sale of gasoline smuggled from Nigeria. Mlachila et al. (2016) estimate that the informal market constituted about 85% of total gasoline consumption in Benin in 2011, and 70.7% in Togo. These figures enable the computation of the absolute size of the informal markets (in physical units), which reflects the quantity of fuel smuggled across the Benin-Nigeria border. The revenues earned by smugglers are then estimated by multiplying the to- tal smuggled quantity of gasoline with the respective price differential between Nigeria’s official subsidised market price and Benin’s (or Togo’s) informal mar- ket price. These informal market prices are also reported by Mlachila et al. (2016). The second step is to extrapolate the smuggling estimates for Benin (and Togo) to Nigeria’s remaining neighbouring countries Niger and Cameroon9 . Using the border length between Benin and Nigeria (773 km), and by assuming that the smuggled quantity is proportional to the length of the external border, a rough estimate of total smuggling can be obtained. In other words, the longer the border between Nigeria and a neighbouring country, the more smuggling activity takes place towards this country. For instance, Nigeria shares a 773 km border with Benin and 1,497 km with Niger, thus the quantity of fuel smuggled to Niger should be roughly twice as large. The third step is to refine this extrapolation by making two further adjustments: • Population: Since the states bordering Benin are particularly populous (and states in Northern Nigeria are sparsely populated), the smuggling estimate is further adjusted proportional to the population size in Nige- rian border states. This reflects the presumption that a larger population means that more smugglers are present and that more smuggling takes place. This also accounts for the fact that population densities on two sides of a border tend to be correlated; thus taking into account the num- ber of foreign consumers demanding smuggled fuel. • Availability of gasoline: The availability of gasoline varies significantly across Nigeria, and directly affects the quantity of gasoline available for smuggling. The states bordering Benin are more developed and urbanised, and offer better access to energy goods; e.g. due to proximity to harbours, 9 Chad shares a 89 km border with Nigeria. This implies that smuggling quantities to Chad are negligible for the purpose of this study. 16 where imported gasoline is landed, and better distribution infrastructure. In more remote states gasoline tends to be less widely available and more expensive (reflecting domestic transport costs), thus reducing smugglers’ profit margins. To reflect these factors, smuggling estimates are further adjusted in line with the average per capita expenditure on gasoline in the relevant bordering states. Based on this method a total petrol smuggling estimate of $ 641 m is obtained, about 43% of which is smuggled to Benin and Togo, 33% to Cameroon, 24% to Niger, and less than 1% to Chad. Appendix B provides further details on this estimation. Mlachila et al. (2016) report that petrol smuggled from Nigeria is sold in Benin with an average mark-up ranging between approximately 20% and 40%. They note that informal prices in Benin are lower (i.e. the mark-up smaller) closer to the Nigerian border. This mark-up contains the cost of smuggling (including transport costs), but also profits by smugglers and middlemen. This study makes the assumption that the cost of smuggling corresponds to 10% of the smuggling value10 . 7. Simulation scenarios and outcome variables 7.1. Simulation scenarios The model is computed for the following scenarios. Scenario 1 – Baseline: This scenario reproduces the baseline economy ob- served in the data. It serves as a reference point for evaluating the results in the subsequent simulation scenarios. It also enables a baseline evaluation of the regressivity of FFS. Scenario 2 – Uncompensated subsidy reform: This scenario simulates an uncompensated petrol subsidy reduction and petrol tax increase (from se P etrol = 0.22 to se 11 P etrol = −0.22) . The government uses reform revenues to increase government spending. Households receive no compensation. Scenario 3 – Subsidy reform with cash transfers: This scenario simulates a petrol subsidy reduction and petrol tax increase, in which reform revenues are redistributed to households uniformly in the form of direct cash transfers. Each household – no matter the income level – receives the same amount. Scenario 4 – Subsidy reform with labour tax reduction: This scenario simulates a petrol subsidy reduction and petrol tax increase, in which reform 10 The overall results of this study are not found to be influenced significantly by increasing the cost of smuggling to 20% or 30%. 11 Note that, as mentioned previously, the baseline subsidisation rate observed in the data is se P etrol = 0.18. This means that the simulation range mainly focusses on the FFS reduction (0.18 to -0.22), but also provides estimates for the effects of an increase in FFS (from 0.18 to 0.22). 17 revenues are used to reduce labour taxes (i.e. a double dividend style fiscal reform). Labour taxes across all sectors are reduced. Counter-factual scenario (Revenue neutral subsidy reform ignoring tax evasion and smuggling): This scenario repeats the simulation of a dou- ble dividend style fiscal reform (Scenario 4), but disregards tax evasion and smuggling activities. This enables an assessment of the size of the evasion and smuggling effects on estimated reform benefits. 7.2. Assessing welfare effects 7.2.1. Fiscal efficiency and social welfare As Liu (2013) shows, a “double dividend” style tax reform – i.e. using envi- ronmental tax revenue to reduce pre-existing taxes – can reduce, but not fully eliminate, the social welfare cost of environmental taxes (Bovenberg and Goul- der, 1996; Goulder, 1995b). The same welfare costs must be expected when FFS reform revenue is used to reduce pre-existing taxes (simulation scenario 4). To confirm this, this study estimates the social welfare cost of a double dividend style FFS reform (i.e. scenario 4) by evaluating changes at the tax base12 . For this purpose, this study adopts the approach taken by Williams (2002), Bento and Jacobsen (2007), and Liu (2013), who use the following expression to measure the welfare effect of a change in the environmental tax rate – in this 12 Note that the term “welfare” here is used – in line with the literature – to refer to the fiscal efficiency benefits of subsidy reform, and thus the associated increase in societal well-being. It does not refer to household level consumption, which is covered by Section 7.2.2. 18 case the subsidisation rate se 13 j : Welfare impact = f ∂Fh,j  f  τh,j 1 − ef h,j ph ∂se      j h j   z ∂Zj  z (1 − ez   + τj j )pj Tax base effects j ∂sej     p loss SM   ∂ l Xj,l + i Xj,i + Xj + Xj e q  − s p   j j ∂se   j j  f  ∂c(eh,j ) f − ph Fh,j   ∂se   j  j h Tax evasion effects  ∂c(ezj) z −   p Zj ∂sj j e    j  loss ∂Ej pq    − j ∂se    j j Smuggling effects loss (31)  ∂Mj − pm   j ∂se   j  j The first line represents the marginal change in factor tax revenues following a change in the subsidisation rate. The second line represents the marginal change in production tax revenues. The third line represents the marginal change in subsidy payments (or petrol tax receipts, in the case of a negative subsidisation rate). The fourth line represents the marginal change in real factor losses asso- ciated with factor tax evasion. Similarly, the fifth line represents the marginal change in real output losses due to production tax evasion. Lastly, lines six and seven represent the marginal change in smuggling losses associated with out- loss loss bound (Ej ) and inbound smuggling (Mj ). Note that Liu (2013) does not consider subsidies, production tax evasion, and smuggling; likewise, this study does not consider the environmental benefits due to emission reduction. 7.2.2. Distribution and household welfare In addition, this study considers the reform’s effects on household welfare, i.e. utility. However, utility, being an ordinal measure, is not a practical measure for the purpose of quantitative policy evaluation. This is especially the case when welfare effects on heterogeneous households are to be quantified and compared. Nevertheless, changes in utility levels can be monetised and thus consistently evaluated and compared by computing Hicksian equivalent variations (Hosoe et al., 2010; Durand-Lasserve et al., 2015)14 . Equivalent variation measures by how much households’ income would need to change (at original price levels) to 13 See Liu (2013) for a full analytical derivation. 14 For further details refer to Mas-Colell et al. (1995). 19 induce the same welfare change as caused by the policy reform. As the original price levels are used to monetise both baseline and counter-factual utility, the equivalent variation measure allows consistent evaluation of fiscal reforms which directly affect prices. 8. Results This section presents the key results from the simulations, while distinguishing the different simulation scenarios wherever relevant or useful. 8.1. Effect on the distribution of petrol consumption In Nigeria FFS are predominantly provided for petrol consumption, thus this section presents evidence on the inequality of petrol consumption across income groups – and how this pattern changes as FFS are removed. The results show that removing fuel subsidies (from a baseline subsidisation rate of 18%) will cause a 21% reduction in national petrol consumption. Increasing a petrol tax to 22% will cause an additional 36% reduction in consumption. Figure 1 shows that – in absolute terms – the reduction in petrol consumption mainly occurs in the top income quintile. Figure 1: Annual per capita petrol expenditure by income quintiles (IQ) for different subsidisation rates (in Naira)15 . 8.2. Effect on subsidy (or tax) incidence From a distributional perspective the key criticism of FFS is their highly regres- sive nature (Arze del Granado et al., 2012). Figure 2 shows that in the baseline scenario (se P etrol = 18%) most of the subsidy benefits are indeed received by the top income quintile. Thus, in absolute terms, removing FFS and moving to fuel taxation predominantly affects the top income quintile. Likewise in absolute terms, imposing a fuel tax will also affect the top income quintile most heavily. This illustrates why FFS reform is considered to be a progressive tax reform – and why rich people and powerful political interest groups are often vocal opponents to reform. 15 N10,000 correspond to roughly US $62 at the 2011 exchange rate. 20 Figure 2: Annual receipt (or payment) of fossil fuel subsidies (or taxes) in N per capita for different income quintiles (IQ). 8.3. Effect on consumption When considering the effects of subsidy reduction and energy tax increase on the consumption expenditure of different income groups, it is essential to distinguish different revenue redistribution mechanisms (i.e. simulation scenarios). All results in this section are presented as consumption gains (or losses) relative to income, as this also enables an insight into the vulnerability and exposure of different income groups. Figure 3 presents relative consumption losses for an uncompensated subsidy reform and tax increase (scenario 2). Reform revenues are used by the gov- ernment to increase public spending. The estimates show that reform induced consumption losses are relatively consistent at around 3-4% of income across the whole income distribution. The reason for this is that in the case of Nige- ria, energy shares in total consumption expenditure are relatively even across income groups (ranging from about 4% to 7%; see Rentschler, 2016) – thus uncompensated FFS removal affects different income quintile to similar extents (relative to income). Figure 4 presents relative consumption losses for a subsidy reform and tax in- crease, with reform revenues redistributed uniformly to all households using cash transfers (scenario 3). Note that this scenario does not simulate targeted cash transfers (i.e. to specific income or population groups), but universal trans- fers. While the highest income quintile (IQ 5) is estimated to incur consumption losses despite the cash compensation, the first, second and third income quintiles experience significant consumption increases. A full FFS removal (se P etrol = 0) is estimated to increase consumption of the bottom income quintile by 3.4%, while the introduction of a fuel tax (se P etrol = −0.22) increases this to 7%. The reason for this progressive effect is that the highly regressive distribution of benefits via FFS is replaced by a uniform distribution, such that post-reform benefits received by low-income households significantly exceed their receipts through FFS (vice versa for high-income households). Overall, this illustrates that replacing (highly regressive) fuel subsidies with uniform cash compensa- tion is a progressive fiscal reform. This observation applies analogously to the 21 Figure 3: Scenario 2: Change in consumption relative to baseline for each income quintile. imposition of petrol taxes, if the revenues are redistributed using uniform cash transfers. Figure 4: Scenario 3: Change in consumption relative to baseline for each income quintile. Figure 5 presents relative consumption losses for a subsidy reform, in which reform revenues are used to reduce pre-existing labour taxes in all sectors (sce- nario 4). Falling in a range between 0.7% (IQ1) and -0.5% (IQ5), the estimated consumption changes are small compared to the other scenarios. The reason is that no significant redistribution of resources takes place across income groups, as in the case with cash transfers. Instead, households benefit from labour tax rate reductions proportional to their pre-reform consumption spending. How- ever, not visible in Figure 5, a significant shift takes place within households’ consumption bundles: As the tax base shifts, the aggregate consumption of petrol falls by 35.5%, while consumption of formal sector goods increases by 1.3%. The net change resulting from shifting consumption bundles is depicted in Figure 5. 22 Figure 5: Scenario 4: Change in consumption relative to baseline for each income quintile. 8.4. Effect on household welfare This section presents the estimated welfare effects of subsidy removal and fuel tax increases for each of the redistribution scenarios, measured as Hicksian equivalent variations. Figure 6(a) shows that households across the entire income distribution incur welfare losses as subsidies are reduced (and fuel taxes increased) without com- pensation (scenario 2). A marginal welfare gain can be observed for all income quintiles for a subsidisation rate of 22%, as it is higher than the baseline sub- sidisation rate of 18%. Welfare losses are presented in absolute terms, and are thus largest for the top income quintile. In addition, Figure 6(b) presents the total welfare loss incurred by the whole population. Figure 7 illustrated the redistribution of wealth associated with the uniform, universal cash compensation scheme (scenario 3). Compared to the baseline scenario, the bottom 60% (i.e. bottom three income quintiles) experience sig- nificant welfare gains, at the expense of the richest 20%. The fourth income quintile is barely affected in this scenario, as cash compensation offsets welfare losses due to energy price increases. Welfare effects in scenario 4 (Figure 8) are less pronounced than in the first two scenarios, as previous results have also suggested (see Figure 5). The reason is that revenue redistribution using tax rate reductions benefits households pro- portionally to their pre-reform consumption expenditure - thus no significant redistribution across income groups takes place, and the reduction of disposable income due to FFS removal is mostly offset. 23 (a) Equivalent variation by income quintiles (in N) (b) Total equivalent variation (in mil. N) for different levels of subsidisation Figure 6: Scenario 2: Change in welfare, measured by Hicksian equivalent variation. (a) shows equivalent variation in per capita terms for each income quintile; (b) presents national aggregate equivalent variation in mil. Naira. 24 (a) Equivalent variation by income quintiles (in N) (b) Total equivalent variation (in mil. N) for different levels of subsidisation Figure 7: Scenario 3: Change in welfare, measured by Hicksian equivalent variation. (a) shows equivalent variation in per capita terms for each income quintile; (b) presents national aggregate equivalent variation in mil. Naira. 25 (a) Equivalent variation by income quintiles (in N) (b) Total equivalent variation (in mil. N) for different levels of subsidisation Figure 8: Scenario 4: Change in welfare, measured by Hicksian equivalent variation. (a) shows equivalent variation in per capita terms for each income quintile; (b) presents national aggregate equivalent variation in mil. Naira. 26 8.5. Effect on output Figure 9(a) presents the estimated change in output for all sectors considered in scenario 4. Full subsidy removal is estimated to result in a 10% reduction of the petrol sector, while increasing petrol taxes to 22% would reduce this sector even further to 20%. Estimated output changes are very similar in scenario 3, thus not reported separately. Figure 9(b) shows further that the largest absolute growth would occur in the formal sector of the economy. This illustrates that FFS result in grave misal- location of resources in favour of the petrol and energy sectors, crowding out consumption from all other sectors. (a) Relative output change (%) for different subsidisation rates (b) Absolute output change (mil. Naira) for different subsidisation rates Figure 9: Scenario 4: Relative and absolute change in output in different sectors. (a) presents the change relative to baseline output, while (b) presents absolute change in mil. Naira. 27 8.6. Effect on labour tax evasion As factor and production taxes remain unchanged in scenarios 2 and 3, the modelling results suggest that tax evasion is not reduced significantly in most sectors (Figure 10). A notable exception is the petrol sector, where a significant reduction in sector size (i.e. output) means that its tax burden decreases, and thus necessarily also the amount of evaded taxes. As this observation is valid for scenarios 2 and 3, only results for the latter are presented here (Figure 10). In scenario 4 reform revenues are used exclusively to reduce labour taxes in all sectors. Accordingly, Figure 10 shows a significant reduction in labour tax evasion throughout the economy. Figure 10: Scenario 3: Change in labour tax evasion for different subsidisation rates Figure 11: Scenario 4: Change in labour tax evasion for different subsidisation rates 8.7. Effect on fuel smuggling Note that smuggling is positive for the baseline subsidisation rate, i.e. fuel is being smuggled out of the country. As the subsidy is reduced, and eventually turned into a fuel tax, the price differential between domestic and foreign fuel is reversed. Without measures to prevent smuggling, in-bound smuggling takes 28 place, thus undermining the energy tax. It should be noted that smuggling is not necessarily zero when the subsidisation (or tax) rate se P etrol is zero; the smuggling quantity depends not only on se P etrol , but also on the ratio between q prices pe j and pj (see equations 7 and 8). Figure 12 presents the estimated percentage change in fuel smuggling, while Figure 13 presents the total value of fuel subsidy leakage (or fuel tax under- mining) due to smuggling. Out-bound smuggling implies that fuel subsidies provided by the home government are smuggled (i.e. leaked) out of the country. In-bound smuggling implies that domestic energy taxes are being evaded, as cheaper un-taxed fuel is smuggled in, thus reducing the government’s fuel tax revenue. SM Figure 12: Percentage change in petrol smuggled out of (∆EP etrol ) or into SM (∆MP etrol ) Nigeria in scenario 2. Figure 13: Total net subsidy value smuggled out (for se P etrol > 0), or fuel tax undermined through inbound smuggling (for se P etrol < 0). In mil Naira. 8.8. Fiscal efficiency and social welfare: the role of tax evasion and smuggling Figure 14 presents the social welfare cost of reform, which is used to evaluate double dividend style reforms (see Section 7.2.1) – i.e. scenario 4. For reference 29 the figure presents welfare costs for a counter-factual simulation which omits tax evasion and smuggling (line (a) in Figure 14), and the model which takes these illicit activities into account (b). Figure 14: Scenario 4: Welfare cost of FFS reform (a) without, (b) and with tax evasion and smuggling taken into account (in mil N) for different subsidisation rates. The reduction of welfare costs due to tax evasion effects (grey) is larger than the reduction due to smuggling effects (blue). The results show that taking into account illicit activities lowers the estimated welfare costs of full FFS removal (i.e. se P etrol = 0) by 34% relative to the counter-factual simulation (omitting tax evasion and smuggling). When fuel taxes are further increased to 22% (i.e. se P etrol = −0.22), taking into account illicit activities lowers the welfare cost by 36% relative to the counter-factual. Figure 15 summarises these results: In the simulated range for the subsidisation rate, the effect of tax evasion and smuggling reduces welfare costs by between 34% to 42%. The larger portion of this difference is due to the tax evasion effect (accounting for 69% to 86% of the welfare cost reduction). Overall, these results highlight that accounting for illicit activities, such as tax evasion and fuel smuggling, can make a crucial difference when determining the costs and benefits of FFS reform. Omitting these aspects may cause studies to significantly under-estimate the benefits (or over-estimate the costs) of FFS reform. 30 Figure 15: Scenario 4: Percentage reduction of welfare cost of FFS reform when illicit activities are taken into account (relative to counter-factual scenario). The tax evasion effect (grey) accounts for a larger share of the reduction than the smuggling effect (blue). 9. Conclusion It is widely accepted that FFS incentivise rampant fuel smuggling to neighbour- ing countries, meaning that a significant fraction of FFS benefits leaks out of the country. In addition, labour taxes not only distort incentives to work, but are associated with high evasion rates and incentivise informal economic activity. This study makes the case that such illicit activities can play a key role in determining the welfare costs and benefits of fiscal reform, in particular FFS reform. It develops a CGE model for Nigeria to study the impact of FFS reform – and energy taxes – on key economic parameters, including consumption, income distribution, tax incidence, and fiscal efficiency. Throughout this analysis, the study examines the role of tax evasion and fuel smuggling, and shows that these factors can substantially strengthen the argument in favour of subsidy removal. First, the study makes following key observations on the distributional and welfare implications of FFS and energy taxation: • FFS are highly regressive, with the bottom income quintile receiving 1% of total FFS payments and the top income quintile 75%. • Removing FFS without compensation measures results in significant dis- posable income shocks to households across all income levels. • Removing FFS and redistributing revenues using uniform cash transfers has a strong progressive (i.e. pro-poor) distributional effect. This progres- sive distribution becomes even more pronounced when FFS are replaced by fuel taxes. • Removing FFS and using revenues to cut pre-existing labour taxes reduces fiscal distortions and the associated welfare losses. • Removing FFS causes significant structural shifts in consumption bundles, with overall petrol consumption decreasing by about 20%. The simulated 31 fuel tax can extend this reduction to over 35%. In turn, households in- crease their formal market consumption accordingly. In addition, by considering the role of illicit activities, this study shows that conventional analyses may be overlooking a significant part of the picture: • Regardless of the method of revenue redistribution, reducing subsidies di- minishes the incentives for fuel smuggling, and hence the welfare losses associated with it. The reduction of these welfare losses must be consid- ered when evaluating FFS reforms. • In the case when revenues of FFS reform are redistributed using cash trans- fers, avoided smuggling means that the cash transfer scheme is disbursing the same aggregate benefit to the population as in the FFS scheme, but at a lower cost. • Reducing FFS and using revenues to lower pre-existing labour tax rates not only mitigates labour market distortions, but reduces tax evasion; i.e. the government can earn the same level of tax revenues, while charging lower tax rates. • A conservative estimate for Nigeria is that taking into account illicit ac- tivities can lower the welfare cost of FFS reform by up to 40%. The tax evasion effect accounts for (on average) 75% of this difference, with smuggling effects accounting for the remainder. • The above mentioned benefits of FFS removal (i.e. in terms of income dis- tribution, consumption, fiscal efficiency) can be increased when subsidies are not only removed, but replaced by fuel taxes. Such fuel taxes may reverse the direction of smuggling activities, though this is not enough to undermine the overall benefits. Even though tax evasion tends to be particularly high in developing countries – large informal sectors are symptomatic of this – the “tax evasion effect” has not been studied before in the context of FFS reform. Similarly, despite being a frequently cited side-effect of FFS, smuggling has also received virtually no attention in the literature so far. This study demonstrates that such illicit activ- ities can make a significant difference to the argument in favour of FFS reform; and should be considered when designing and implementing such reforms. Moreover, evidence from Luxembourg demonstrates significant cross-border fuel demand due to large price differentials with neighbouring countries – and thus illustrates that the relevance of such activities is by no means limited to resource- rich developing countries (IMF, 2015). A better understanding of such activities can be vital for designing functioning energy and carbon pricing schemes. 32 References ADB (2015), Fuel-Marking Programs: Helping Governments Raise Revenue, Combat Smuggling, and Improve the Environment, Asian Development Bank, Manila. Anand, R., Coady, D., Mohommad, A., Thakoor, V. V. and Walsh, J. P. (2013), ‘The Fiscal and Welfare Impacts of Reforming Fuel Subsidies in India’, IMF Working Papers 13(128), 1. Araar, A., Choueiri, N. and Verme, P. (2015), ‘The Quest for Subsidy Reforms in Libya’, World Bank Policy Research Working Papers 7225. Armington, S. P. (1969), ‘A Theory of Demand for Products Distinguished by Place of Production’, IMF Staff Papers 16(1), 159–178. Arze del Granado, F. J., Coady, D. and Gillingham, R. (2012), ‘The Unequal Benefits of Fuel Subsidies: A Review of Evidence for Developing Countries’, World Development 40(11), 2234–2248. Bah, M. M. and Saari, M. Y. (2020), ‘Quantifying the impacts of energy price reform on living expenses in saudi arabia’, Energy Policy 139. Bazilian, M. and Onyeji, I. (2012), ‘Fossil fuel subsidy removal and inadequate public power supply: Implications for businesses’, Energy Policy 45, 1–5. Bento, A., Jacobsen, M. and Liu, A. A. (2013), ‘Environmental Policy in the Presence of an Informal Sector’, Journal of the Association of Environmental and Resource Economists . Bento, A. M. and Jacobsen, M. (2007), ‘Ricardian rents, environmental policy and the ‘double-dividend’ hypothesis’, Journal of Environmental Economics and Management 53(1), 17–31. Bovenberg, A. L. (1999), ‘Green Tax Reforms and the Double Dividend: an Updated Reader’s Guide’, International Tax and Public Finance 6(3), 421– 443. Bovenberg, L. and Goulder, L. H. (1996), ‘Optimal Environmental Taxation in the Presence of Other Taxes: General-Equilibrium Analyses’, The American Economic Review 86(4), 985–1000. Carson, R. T., Jacobsen, M. R. and Liu, A. A. (2014), ‘Comparing the Cost of a Carbon Tax in China and the United States’, American Economic Association Coference 2015 Working Paper . URL: https://www.aeaweb.org/conference/2015/retrieve.php?pdfid=634&usg= AFQjCNGhW1bl9fFYsDXU3ujw-Wfyv7BW3A&sig2=DeUCH Coady, D., Flamini, V. and Sears, L. (2015), ‘The Unequal Benefits of Fuel Sub- sidies Revisited: Evidence for Developing Countries’, IMF Working Papers WP/15/250. Coady, D., Gillingham, R., Ossowski, R., Piotrowski, J., Tareq, S. and Tyson, J. (2010), ‘Petroleum Product Subsidies: Costly, Inequitable, and Rising’, IMF Staff Position Note SPN/10/05. 33 Coady, D., Parry, I., Sears, L. and Shang, B. (2015), ‘How Large Are Global Energy Subsidies?’, IMF Working Papers WP/15/105. Coady, D., Parry, I., Sears, L. and Shang, B. (2017), ‘How Large Are Global Fossil Fuel Subsidies?’, World Development 91, 11–27. Durand-Lasserve, O., Campagnolo, L., Chateau, J. and Dellink, R. (2015), ‘Modelling of distributional impacts of energy subsidy reforms: an illustration with Indonesia’, OECD Environment Working Papers 86. Feltenstein, A. and Datta, B. K. (2020), ‘Broad-based subsidies or targeted transfers? distributional equity vs macroeconomic costs’, Journal of Eco- nomic Policy Reform pp. 1–18. Fullerton, D. and Metcalf, G. E. (1997), ‘Environmental Taxes and the Double- Dividend Hypothesis: Did You Really Expect Something for Nothing?’, Na- tional Bureau of Economic Research Working Paper Series No. 6199. Goulder, L. H. (1995a), ‘Effects of Carbon Taxes in an Economy with Prior Tax Distortions - An Intertemporal General Equilibrium Analysis.pdf’, Journal of Environmental Economics and Management 29, 271–297. Goulder, L. H. (1995b), ‘Environmental taxation and the double dividend: A reader’s guide’, International Tax and Public Finance 2(2), 157–183. GSI (2012), A citizens’ guide to energy subsidies in Nigeria, International In- stitute for Sustainable Development: Global Subsidies Initiative, Geneva. Hassan, M. and Schneider, F. (2016), ‘Size and Development of the Shadow Economies of 157 Countries Worldwide: Updated and New Measures from 1999 to 2013’, IZA Discussion Paper Series IZA DP No. Hosoe, N., Gasawa, K. and Hashimoto, H. (2010), Textbook of Computable Gen- eral Equilibrium Modelling, Palgrave Macmillan, New York. IEA (2015), World Energy Outlook 2015, International Energy Agency, Paris. IEA (2016), ‘International Energy Agency: Energy Statistics’. URL: https://www.iea.org/statistics/ IMF (2013), Energy Subsidy Reform: Lessons and Implications, International Monetary Fund, Washington DC. IMF (2015), Luxembourg - Selected issues, International Monetary Fund, Wash- ington DC. IMF (2016), ‘IMF and Reforming Energy Subsidies: Distributional Analysis of Fuel Subsidy Reform’. URL: http://www.imf.org/external/np/fad/subsidies/ Jiang, Z., Ouyang, X. and Huang, G. (2015), ‘The distributional impacts of removing energy subsidies in China’, China Economic Review 33(0), 111– 122. 34 Kojima, M. (2016), ‘Fossil fuel subsidy and pricing policies: recent develop- ing country experience’, World Bank Policy Research Working Papers WPS 7531. Liu, A. A. (2013), ‘Tax evasion and optimal environmental taxes’, Journal of Environmental Economics and Management 66(3), 656–670. Mas-Colell, A., Whinston, M. D. and Green, J. R. (1995), Microeconomic theory, Vol. 1, Oxford university press New York. Mlachila, M., Ruggiero, E. and Corvino, D. (2016), ‘Unintended consequences: Spillovers from nigeria’s fuel pricing policies to its neighbors’, IMF Working Papers WP/15/17. National Bureau of Statistics (2010), Nigeria Poverty Profile 2010, National Bureau of Statistics, FCT Abuja. NEITI (2013), Report on Revenues, Deductions, and Analysis of Disbursement and Utilization of Funds of the Federal Beneficiary Agencies 2007 – 2011 – Petroleum Product Pricing Regulatory Agency, NEITI. Ouyang, X. and Lin, B. (2014), ‘Impacts of increasing renewable energy subsi- dies and phasing out fossil fuel subsidies in China’, Renewable and Sustainable Energy Reviews 37(0), 933–942. Parry, I. W. H. (1997), ‘Environmental taxes and quotas in the presence of distorting taxes in factor markets’, Resource and Energy Economics 19, 203– 202. Parry, I. W. H. and Williams, R. C. (2010), ‘What are the Costs of Meeting Dis- tributional Objectives for Climate Policy ?’, The B.E. Journal of Economic Analysis & Policy 10(2). Plante, M. (2014), ‘The long-run macroeconomic impacts of fuel subsidies’, Journal of Development Economics 107, 129–143. Rentschler, J. (2016), ‘Incidence and impact: The regional variation of poverty effects due to fossil fuel subsidy reform’, Energy Policy 96, 491–503. Rentschler, J. E. and Bazilian, M. (2016), ‘Reforming Fossil Fuel Subsidies: Drivers, Barriers and the State of Progress’, Climate Policy . Shehabi, M. (2020), ‘Diversification effects of energy subsidy reform in oil ex- porters: Illustrations from kuwait’, Energy Policy 138. Siddig, K., Aguiar, A., Grethe, H., Minor, P. and Walmsley, T. (2014), ‘Im- pacts of removing fuel import subsidies in Nigeria on poverty’, Energy Policy 69, 165–178. Slemrod, J. (2007), ‘Cheating Ourselves: The Economics of Tax Evasion’, Jour- nal of Economic Perspectives 21(1), 25–48. Soile, I. and Mu, X. (2015), ‘Who benefit most from fuel subsidies? Evidence from Nigeria’, Energy Policy 87, 314–324. 35 Solaymani, S. and Kari, F. (2014), ‘Impacts of energy subsidy reform on the Malaysian economy and transportation sector’, Energy Policy 70, 115–125. Strand, J. (2013), ‘Political Economy Aspects of Fuel Subsidies: A Conceptual Framework’, World Bank Policy Research Working Papers (6392). Swedish National Tax Agency (2008), ‘Tax Gap Map for Sweden’, Report 2008:1B. . Verme, P. and El-Massnaoui, K. (2015), ‘An Evaluation of the 2014 Subsidy Reforms in Morocco and a Simulation of Further Reforms’, World Bank Policy Research Working Papers 7224. Williams, R. C. (2002), ‘Environmental Tax Interactions when Pollution Affects Health or Productivity’, Journal of Environmental Economics and Manage- ment 44(2), 261–270. World Bank (2015), ‘World Development Indicators’. URL: http://data.worldbank.org/indicator Yau, R. and Chen, G. H. (2021), ‘Assessing energy subsidy policies in a struc- tural macroeconomic model’, Energy Economics 103. Zhang, F. (2011), ‘Distributional Impact Analysis of the Energy Price Reform in Turkey’, World Bank Policy Research Working Papers 5831. 36 Appendix A. Model specifications Appendix A.1. Domestic production: Two stage set-up First stage The firm’s first production stage is described by a Cobb-Douglas production function, which is homogeneous of degree one and exhibits constant returns to scale. It maximises profits by choosing the quantity of its output (i.e. the composite factor Yj ), of its inputs (i.e. production factors Fh,j ), and the factor tax evasion rate ef h,j . y max πj = py j Yj − f 1 + τh,j 1 − ef h,j pf f loss h Fh,j − ph Fh,j (A.1) Y,F,ef h βh,j s.t. Y j = bj Fh,j (A.2) h Standard Lagrangian optimisation yields the following first order conditions: y ∂πj f = Af f h,j (eh,j ) Nh,j f = τh,j (FOC 2.1) ∂ef h,j y ∂πj Yj = pf f f f y h 1 + τh,j (1 − eh,j ) + c(eh,j ) = pj βh,j (FOC 2.2) ∂Fh,j Fh,j FOC 2.1 states that the firm’s marginal cost of tax evasion (LHS) is equal to the marginal benefit (RHS). Similarly, FOC 2.2 states that the marginal cost of an additional unit of input (LHS) equals its marginal benefit (RHS). Moreover, using FOC 2.1 the optimal evasion rate efh,j can be derived: 1 f f τh,j N h,j ef h,j = (A.3) Af h,j f Note that the optimal evasion rate is increasing in the factor tax rate τh,j . By plugging ef h,j into the original cost equation (1), the optimal cost of evasion can be determined. This also allows solving for and calibrating parameter Af h,j , as all other parameters are known. FOC 2.2 allows solving for the optimal factor demand Fh,j : py j Fh,j = βh,j Yj (A.4) pf f f f h (1 + τh,j (1 − eh,j ) + c(eh,j )) Second stage The firm’s second production stage is described by a Leontief production func- tion, which is homogeneous of degree one and exhibits constant returns to scale. 37 The firm maximises profits by choosing the quantity of output (Zj ), of interme- diate inputs (Xi,j ), of the composite factor (Yj ), and the production tax evasion rate ez j. z max πj = pz z z z j Zj + τj ej pj Zj Z,Y,X,ez (A.5) − py j Yj − pq e q e loss i (1 − si )Xi,j − pj (1 − si )Xj i Xi,j Yj s.t. Zj = min , (A.6) axi,j ayj The first term of the profit function denotes sales revenues, the second term denotes the benefit from tax evasion, the third and fourth terms denote the cost of composite and intermediate inputs, while the last term is the cost of evasion. Xi,j and Yj can be replaced using equation A.6, and thus derive an unconstrained maximisation problem: z y max z πj = pz z z z j Zj + τj ej pj Zj − pj ayi Zj Z,e (A.7) q e q − 1 − se z i pi axi,j Zj − 1 − sj pj c(ej )Zj i This yields the following first order conditions: z ∂πj e q z z Nj z = pz z j τj = 1 − sj pj Aj (ej ) (FOC 3.1) ∂ez j z ∂πj z z y q = (1 + τj ej − c(ez z j ))pj = pj ayj + (1 − se i )pi axi,j (FOC 3.2) ∂Zh,j i As in the case with factor tax evasion, the firm chooses production tax evasion such that the marginal benefit of tax evasion (LHS of FOC 3.1) is equal to the marginal cost (RHS of FOC 3.1). This means that the firm’s optimal level of production tax evasion ezj can be expressed as 1 Nz pz j z τj j ez j = q z . (A.8) 1 − se j pj Aj Note that the Leontief production function implies rectangular isoquants, which are prone to computational problems due to their kinks. Thus, as suggested by Hosoe et al. (2010), equation 13 is replaced with a unit cost function for com- putational purposes. This unit cost function can be obtained by transforming 38 z the zero profit condition πj = 0 using functions 10 and 11: y pz j = ayj pj + pq e q e z i (1 − si )axi,j + pj (1 − sj )c(ej ) (A.9) i Appendix A.2. Investments and savings Given the static setting of the model, dynamic aspects such as investment and savings cannot be reflected in their strict sense. However, recognising that these activities can constitute significant shares of final demand, a virtual investment account is incorporated (Hosoe et al., 2010). This account is modeled to use sav- ings from households and abroad to invest these in investment goods. Formally, v investment demand Xi is given by λi v Xi = Ssp l + εSs f (A.10) pq i l Ssp p l = ssl pf tax h F Fh,l + Ctl + Ctz l + Ctf l,h (A.11) h h Savings are denoted Ssf for the foreign sector (at exchange rate ε), and Ssp for households. Moreover, the parameter λi denotes the expenditure share of the i-th good in overall investment; the average propensity to save is denoted ssg for the government and ssp l for households. Household income from cash transfers (Ctl ) are detailed in the following subsection. Appendix A.3. Substitution and transformation of traded goods Substitution between imports and domestic goods An “Armington composite good” is introduced to reflect that imports and do- mestic goods are imperfect substitutes (Armington, 1969). This allows endoge- nous market shares for imported goods, as opposed to a “cheapest takes all” setting. Specifically, profit maximising firms choose a combination of imported and domestic goods to produce the Armington composite – which is then con- sumed by households, firms, and the government. Using the constant elasticity of substitution (CES) production function 1 ηi ηi Qi = γi (δmi Miηi + δdi Di ) (A.12) a standard profit maximisation procedure yields demand functions for imports and the domestic good: 1 ηi γi δmi pq i 1−ηi Mi = Qi (A.13) (1 + τim )pmi 1 ηi γi δdi pq i 1−ηi Di = Qi (A.14) pd i Qi denotes the i-th Armington composite good, which is composed of imports (Mi ) and domestic goods (Di ). The coefficients δmi and δdi denote the respec- 39 tive input shares of the composite good (fulfilling 0 ≤ δmi ≤ 1, 0 ≤ δdi ≤ 1, and δmi + δdi = 1); while γi is the scaling coefficient in the composite production function. Lastly, ηi is a parameter defined by the elasticity of substitution σi (ηi = (σi − 1)/σi , with ηi ≤ 1). Transformation between exports and domestic goods Analogous to the demand side, imperfect transformation on the supply side (i.e. between exports and domestic goods) is reflected using a constant elasticity of transformation (CET) production function. This setting allows that the gross domestic output of a good comprises both exports and domestic supply, the ratio of which is determined by their relative prices. This is modeled by introducing a “virtual” profit maximising firm, which trans- forms the gross domestic output (Zi ) into exports (Ei ) and domestically supplied goods (Di ) according to following CET production function: 1 ϕi ϕi ϕi Zi = θi ξei Ei + ξdi Di (A.15) By solving a standard profit maximisation problem following supply rules for exports and domestic goods are obtained: 1 ϕi 1−ϕi θi ξei (1 + τiz (1 − ez z j ))pi Ei = Zi (A.16) pei 1 ϕi 1−ϕi θi ξdi (1 + τiz (1 − ez z j ))pi Di = Zi (A.17) pd i The coefficients ξei and ξdi are the share coefficient of the transformation pro- cess (fulfilling 0 ≤ ξei ≤ 1, 0 ≤ ξdi ≤ 1, and ξei + ξdi = 1). Moreover, θi is the scaling coefficient characterising the transformation. Lastly, ϕi is a parame- ter defined by the elasticity of transformation ψi (ϕi = (ψi +1)/ψi , with ϕi ≤ 1). Appendix B. Data: Estimating smuggling quantities To summarise the three estimation steps outlined in section 6.3, the total quan- tity of smuggled fuel is estimated as SM EP etrol = SQc c BLc AP Cc P opc = SQBenin (B.1) c BLBenin AP CBenin P opBenin where: • SQBenin = Petrol quantity smuggled to Benin (a share of which is then smuggled further to Togo) 40 • SQc = Estimated petrol quantity smuggled to neighbouring country c • BLc = Length of external border shared by Nigerian states and country c • AP F Cc = Average petrol consumption per capita in Nigerian states shar- ing a border with country c • P opc = Population in Nigerian states sharing a border with country c Nigeria Benin Togo Niger Cameroon Chad Border with Nigeria (km) – 773 – 1,497 1,690 89 Gasoline price (N/litre) 97.4 162.8 182.9 167.4 186.8 204.6 Pop. in bordering – 27.9 – 23.6 16.5 6.1 Nigerian states (m) Av. petrol consumption – 286 – 100 175 insig. in bordering Nigerian states (N/month) Table B.1: Parameters used for refining extrapolated smuggling estimates Appendix C. Sensitivity and robustness This section presents sensitivity tests for key parameters. Appendix C.1. The elasticities of substitution and transformation To test the sensitivity of sectoral output estimates to variation in elasticity values, low and high value cases for the elasticities of substitution and trans- formation are considered. The elasticity of substitution in the CES production function is given by σi (Equation A.12, where ηi = (σi − 1)/σi , with ηi ≤ 1). The elasticity of transformation in the CET production function is given by ψi (Equation A.15, where ϕi = (ψi + 1)/ψi , with ϕi ≤ 1). The low case is defined as a 25% reduction of the elasticity value compared to the base run calibration; the high case is defined as a 25% increase. Table C.2 shows that the variation in elasticities has minimal impact on the estimates. In addition, Figure C.16 demonstrates that also the social welfare cost of reform is robust to variation in the elasticities. 41 σi ; ϕi Base run Low case High case se P etrol = 0.22 3,736,698 3,728,096 -0.23% 3,744,630 0.21% Petrol se P etrol = 0 3,252,824 3,288,715 1.10% 3,217,423 -1.09% se P etrol = −0.22 2,898,255 2,970,800 2.50% 2,827,282 -2.45% se P etrol = 0.22 5,870,983 5,870,212 -0.01% 5,873,555 0.04% Energy se P etrol = 0 5,720,992 5,735,723 0.26% 5,709,553 -0.20% se P etrol = −0.22 5,607,641 5,635,414 0.50% 5,584,303 -0.42% se P etrol = 0.22 142,209,300 142,341,800 0.09% 142,257,400 0.03% Formal se P etrol = 0 143,031,500 143,116,800 0.06% 143,157,200 0.09% se P etrol = −0.22 143,661,100 143,701,300 0.03% 143,849,000 0.13% se P etrol = 0.22 114,992,600 115,065,200 0.06% 115,144,100 0.13% Informal se P etrol = 0 114,866,800 114,996,800 0.11% 115,004,900 0.12% se P etrol = −0.22 114,745,200 114,924,800 0.16% 114,868,500 0.11% Table C.2: Sensitivity of sectoral output to variation in the elasticities of substitution (σi ) and transformation (ψi ). Absolute values represent the total value of output (in mil Naira) for each sector in each case. Percentage values represent the deviation of the low and high case estimates from the base run results. Figure C.16: Sensitivity to variation in σi and ψi : Percentage deviation of the social welfare cost of reform from the base run estimation. Note that illicit activities are taken into account in both cases. Appendix C.2. Parameter r in the smuggling function The smuggler maximises profits by choosing the inbound and outbound smug- SM SM gling quantities (Ej and Mj ), subject to the cost of smuggling (e.g. trans- portation costs, bribes) which in turn depends on the smuggled quantity, and parameters r and ASM j . In the base run analysis the parameter value is set at r = 2, which assumes linear smuggling behaviour. Table C.3 presents the deviation of the estimated smuggling quantities in re- sponse to a variation in parameter r (25% lower and higher than the base run value). Note that the parameter ASM j is calibrated on the basis of r, thus no separate sensitivity analysis is required. 42 r Base run Low case High case se P etrol =0.22 160,376 212,556 +32.54% 145,940 -9.00% Value of smuggling se P etrol =0.1 55,919 25,869 -53.74% 72,264 +29.23% outbound (+) se P etrol =0 -12,048 -1,203 -90.02% -27,116 +125.06% and inbound (-) se P etrol =-0.1 -76,158 -48,018 -36.95% -94,089 +23.54% se P etrol =-0.22 -154,739 -197,849 +27.86% -135,590 -12.38% Table C.3: Sensitivity analysis for smuggling parameter r. Absolute values represent the total value of smuggled fuel (in mil Naira) in each case. Percentage values represent the deviation of the low and high case estimates from the base run results. The percentage deviations in Table C.3 appear large, in particular for subsidis- ation rates close to zero. However, it should be noted that the absolute values are small in all cases. This is illustrated by Figure C.17, which demonstrates that the variation in r mainly affects the curvature of the smuggling estimates. In both the low and high cases of r the smuggling quantity can be below or above the base run estimate, depending on the value of se P etrol . Figure C.17: Sensitivity to variation in r: Estimated change in the outbound (+) and inbound (-) smuggling estimates (in mil Naira) at different subsidisation rates se P etrol . To test the sensitivity of the estimated social welfare cost of FFS reform, Figure C.18 presents the difference between the welfare costs in the illicit activities model and the counter-factual (Section 8.8). Both the low and high cases high- light that the variation in parameter r has no impact on the overall conclusion that illicit activities (in this case smuggling) can play a key role in determining the welfare cost of FFS reform. In both cases the welfare cost is about 40% lower when illicit activities are considered. 43 (a) Low case (b) High case Figure C.18: The percentage difference between social welfare costs of FFS reform in the model considering illicit activities, and a counter-factual (see Section 8.8). Due to the lesser contribution of smuggling to this difference, the variation in r has little effect on the overall conclusions. f z Appendix C.3. Tax evasion parameters Nh,j and Nj As part of its optimisation problem, the firm chooses the optimal level of factor and production tax evasion (equations A.3 and A.8). Besides the effective tax f z rates (τh,j and τj ), the choice of the optimal evasion rate is determined by f z the parameters Nh,j and Nj , which characterise the cost of evasion activities f z (equations 1 and 2). The values of Nh,j and Nj in the base run calibration range between 8.01 and 9.02; thus the elasticities of tax evasion with respect to the tax rate (expressed as 1/Nj ) are between 0.11 and 0.13. These values are in line with the elasticities used by Liu (2013). Table C.4 reports the sensitivity of tax evasion estimates for low and high cases f z of parameters Nh,j and Nj (25% lower and higher than the base run values). Parameters Af z f z h,j and Aj are calibrated on the basis of Nh,j and Nj , thus no separate sensitivity analysis is conducted. 44 f Nh,j Base run Low case High case Total factor se P etrol =0.22 2,416,237 2,343,597 -3.01% 2,463,199 +1.94% tax evasion se P etrol =0 1,577,961 1,494,444 -5.29% 1,631,069 +3.37% se P etrol =-0.22 966,775 895,413 -7.38% 1,012,810 +4.76% z Nj Base run Low case High case Total se P etrol =0.22 2,078,998 2,078,724 -0.01% 2,079,238 +0.01% production se P etrol =0 1,978,902 1,977,739 -0.06% 1,979,653 +0.04% tax evasion se P etrol =-0.22 1,906,655 1,905,962 -0.04% 1,907,141 +0.03% Table C.4: Deviation from the base run estimates for low and high value cases for f z the tax evasion parameters Nh,j and Nj . f z Figure C.19 shows that the variation in parameters Nh,j and Nj notably influ- ences the level of tax evasion taking place, thus the associated evasion losses. However, note that base run parameter values for illicit activities have been chosen conservatively; i.e. the base run is likely to underestimate the role of tax evasion. Figure C.19: Percentage deviation of the social welfare cost of FFS reform to a f z variation in parameters Nh,j and Nj from the base run estimate (see Section 8.8); presented for different subsidisation rates along the horizontal axis. With respect to the social welfare cost of FFS reform, the difference between the illicit activities model and the counter-factual remains large regardless of the f z value of Nh,j and Nj . In the high value case, the social welfare cost of reform is nearly 50% lower in the illicit activities model compared to the counter-factual. Even in the low value case, the welfare cost is at least 30% lower compared to the counter-factual. Thus, the overall conclusion remains unchanged that illicit activities play a significant role in determining the welfare costs of reform. 45 (a) Low case (b) High case Figure C.20: The percentage difference between social welfare costs of FFS reform in the illicit activities model and a counter-factual (see Section 8.8). (a) presents the f z estimate for Nh,j and Nj taking values 25% lower than in the base run (25% higher in (b)). 46