Abstract
We take stock of what is known about the consequences of actual value-added tax (VAT) systems, nearly 100 years since its conception. We benchmark it to the tax it most often replaced—the retail sales tax (RST). We do this not because we advocate a return to the RST, but because it allows us to evaluate the performance of two tax instruments in actual practice instead of comparing them to their idealized versions, which are never achieved. Doing so highlights different areas of focus than traditional comparisons to the “good” VAT. Like the RST, the VAT is weak at the final stage and technological advances that ameliorate VAT issues could also be applied to the RST. VAT introduces new evasion opportunities not possible in an RST such as invoice mills and carousel fraud. We propose new metrics and a research agenda to evaluate the “health” of the real VAT.
The Birth Story
The value-added tax (henceforth VAT) was first conceived about a century ago. The idea is usually attributed to the German businessman Wilhelm von Siemens, although some argue that the American academic economist Thomas S. Adams had the same idea at about the same time (Adams 1918). Von Siemens thought of it as a better consumption tax, while Adams thought of it as a better corporate income tax (see James 2015). The VAT remained just an idea until 1948, when France adopted it in a limited form. The first full-fledged VAT systems appeared in Denmark and Brazil around 1967, after which the pace of adoption accelerated in Europe and Latin America, followed by Asia and Africa in the 1980s. Its adoption in Europe was partly due to the belief that it would be easier than a retail sales tax (henceforth RST) to harmonize across countries. There were some early adopters in Africa; Cote d’Ivoire and Senegal both introduced the VAT in the 1960s—and the state of Michigan in the United Sates introduced a form of VAT in 1953. As of 2020, it has become a tremendous success in terms of ubiquity of adoption (160 + countries) and revenue raised (OECD 2018). It accounts for about 30 percent on average of total tax revenue in countries where it has been adopted (ICTD, 2019).
Although its ubiquity suggests a certain robustness, it is only prudent to conduct a comprehensive health assessment of any centenarian, including the VAT. But the 100th anniversary is not the only reason for a health check-up. In the last few years, the technology for assessing its health has dramatically improved, with the burgeoning availability of micro firm data including transactions and tax administrative data, coupled with promising new methodologies that take advantage of these data such as randomized controlled trials and bunching analysis. So, too, have there been advancements in the technology for dealing with VAT health problems, such as the proliferation and monitoring of data from electronic payment systems. Despite its prevalence, there seems to be little consensus on the best practices for VAT administration, and choices seem to reflect short-term goals of policymakers and revenue administrations (Gendron 2017). New research on the VAT that incorporates the shortcomings of real VAT systems through the lens of a modern optimal tax framework can get us closer to designing VATs that achieve socially desirable goals of efficiency and equity. This paper catalogues various legislated and structural issues with VATs, documents existing, rigorous empirical evidence, and points out gaps where more evidence is required.
Conducting a health assessment raises the immediate question of a benchmark against which to assess the VAT’s health. Although a hundred years ago it was offered as an alternative to existing consumption taxes and the corporate income tax, in practice VATs have come to replace other forms of consumption tax and exist simultaneously with a corporate income tax. For these reasons, and because some argue that the VAT addresses—and maybe even solves—some of its key structural problems, we will loosely frame our VAT assessment against the benchmark of a RST, a tax levied in the United States in forty-five states, Washington, DC, and thousands of local jurisdictions, as well as in twenty-six other countries (Annacondia 2018). 1 Cnossen (2019) briefly compares VAT administration to RST, noting theoretical arguments for and against the VAT, and concludes that “On balance, the arguments in favour of VAT outweigh the arguments for RST,” and that “VAT scores highly on neutrality and feasibility grounds.” We point out in this paper that more needs to be known about the extent of VAT collected at each stage of production, the loss of revenue through fake invoices, invoice mills, and refunds, and the efficiency consequences of such distortions in the VAT, to evaluate its success.
There is an existing literature that discusses challenges with the VAT, each focusing on specific issues. Barbone, Bird and Vázquez-Caro (2012) reviews the evidence on compliance and administrative costs in the VAT, Bird and Gendron (2007) discuss VAT designs in developing economies drawing from the experiences of practitioners, Gendron (2017) highlights specific administrative challenges in real VATs, and Ebrill et al. (2001) detail issues in real VAT administrations. Our contribution is to identify systematic evidence that quantifies the magnitude and prevalence of these issues and to apply the lens of academic theory. For example, it is generally accepted that enforcement of the VAT is strongest at the import stage. We ask what kinds of metrics countries should generate to capitalize on this insight for VAT enforcement and design, and what it implies for the efficiency and equity of the VAT. We then describe new empirical research on the VAT that has exploded in the recent decade and identify gaps in the literature.
We begin by outlining how a VAT differs from an RST and considering whether VATs do in fact resolve the RST’s structural problems. Note that we are not advocating a return to RST. Indeed—few countries (with the notable exception of Malaysia) have switched from a VAT to an RST. Rather, we use the contrast to highlight under-studied policy issues in real VAT systems.
Under a “pure” RST, all sales of goods and services to domestic final consumers are subject to a uniform tax rate that must be remitted by retail businesses. A “clean” RST allows final sales of certain goods and services to be exempt from taxation. The idealized VAT differs from an RST only in its remittance rules.
Of course, real-life RSTs are not pure and some impurities are legislated while others are not. What we know about failures of the RST in recent years comes from the United States, the largest economy that still implements the tax (See Mikesell and Kioko 2018). A major issue with the RST is that the difficulty of distinguishing between final and intermediate sales in practice means that some business-to-business transactions are taxed (Ring 1999). One study estimates that approximately 40 percent of sales tax revenue comes from non-labor business inputs (Foster 2014). In addition, final sales masquerading as intermediate sales are one avenue of evasion. Another drawback is that services, including financial services, are usually excluded from the RST base, distorting production choices.
There is no systematic nationwide analysis of the sales tax gap in the United States, but individual states have occasionally published estimates of the extent of evasion with their RST. Most recently, the state of Washington estimated that in 2018 non-compliance with sales tax was about 1 percent of total liability and with use tax (the tax due on purchases made out of state, including remote sales) to be about 15 percent of total liability (Washington State Department of Revenue, 2018). Older estimates from other states are higher. Minnesota estimated a gap of 4.9 percent in 2000, Tennessee had a gap of 4.5 percent in 1989 and Idaho estimated a gap as high as 19.5 percent for sales taxes in 1996 (Mikesell 2012).
These drawbacks of the RST in the United States illustrate some of the reasons why most nations have turned to a VAT. Nevertheless, as we describe, the VAT creates new challenges as it resolves issues with the real RSTs.
Fundamentals of a VAT Health Assessment
We turn now to the patient, the VAT. We will consider a destination-based VAT, which is by far the most common type in use and is the analogue to an RST in the sense that a pure uniform RST and a pure uniform rate destination-based VAT are “textbook-equivalent.”
Under a pure VAT, all firms are obliged to remit tax at a uniform rate on their value added, which is equal to their domestic sales minus their purchases from other businesses. This is usually implemented by an invoice-credit system under which firms remit tax on their sales and receive credits from taxes shown on purchase invoices. 2 In a sense, a pure VAT and a pure RST differ only in their remittance system: indeed, a VAT can be thought of as an RST with a pre-retail system of “withholding” taxes that are credited at later stages of the production and distribution process.
Under a destination-based VAT, exports are not subject to tax, while imports are taxed at the same rate as domestic sales. Exports are not simply exempt from the tax; they are zero-rated, which means the exporter owes no tax on their export sales but can claim input tax credits on purchases, and thus are usually due a refund. VAT exemptions, as opposed to zero-rating, are usually commodity based (i.e., sales of particular goods are exempt) or size-based (i.e., firms below a certain revenue-threshold). It is well-documented that some firms who meet the size-based criteria for exemption nevertheless voluntarily register for the VAT (see e.g., Liu et al. 2019). That a firm would voluntarily subject itself to a tax may seem counterintuitive and is due to the incentives created by input tax credits. VAT-unregistered firms cannot claim input tax credits on their purchases, and registered firms who buy from unregistered firms cannot claim input tax credits on those transactions. Firms that rely heavily on inputs from other registered firms, or that sell mainly to registered firms, therefore benefit from being VAT-registered. 3
VAT Health Problems—Legislated
Just as actual RSTs are neither pure nor clean, actual VATs are neither pure nor clean. VAT systems vary substantially across countries in terms of key structural elements, to the point where some forms of “VAT” actually closely resemble a RST (Schenk and Oldman 2001). For example, Brazil has a single-stage manufacturers’ tax where liability is determined by the final-stage price as in an RST, except that the remittance responsibility lies solely with the manufacturer. This tax combines the relatively low compliance burden of an RST by requiring remittance only at a single stage, with the improved enforcement under a VAT by requiring upstream firms to remit.
When a VAT is introduced, it is often done so in stages or only in some sectors, to ease the transition. For example, in China, input tax credits for machinery were first permitted only in three provinces in 2004 and expanded to the rest of the country in 2009 (Cai and Harrison 2019). The state VAT in India did not allow input tax credits on any automobile purchases by firms. These limited introductions are a far cry from the broad base of the ideal VAT.
Even mature VAT systems have legislated deviations from the ideal. We describe some of these deviations common to most countries, below.
Commodity exemptions
Like the RST, the VAT is not a comprehensive consumption tax—education, healthcare, and financial services are often exempted. Standard consumer items that are considered necessities are usually exempted to generate progressivity, and entire sectors could be excluded for reasons of complexity. For example, electricity was only later included in the VAT base in Pakistan, and it is exempt from the GST base in India. As a result, it can be difficult to clearly categorize goods as exempt or taxed, creating an opportunity for evasion by mislabeling taxable items as exempt. Excluding upstream goods like electricity breaks the chain of accountability at these stages, undermining compliance, an issue we discuss in the section “VAT Health Problems—Structural.”
Size-based exemptions
Because of the substantial administrative and compliance costs of the VAT, many tax authorities allow a revenue-based exemption from mandatory VAT registration. This exemption threshold can vary. Some countries, such as Bulgaria, require all firms to register while the United Kingdom has one of the highest exemption thresholds in the EU at £85,000. As mentioned, because exempt firms cannot claim input tax credits or provide tax credits to buyers, some firms voluntarily register. But others are incentivized to either limit or underreport production to avoid VAT liability, creating inefficiency. Although such exemptions are optimal under certain conditions (Dharmapala, Slemrod and Wilson 2011), there is no tractable model yet to identify the optimal level. Keen and Mintz (2004) provide a framework for the optimal VAT threshold, which is broadly determined by the tradeoff between revenue foregone and the administrative and compliance costs saved (appropriately adjusted by the marginal cost of public funds). A more general rule must also consider the behavioral response of firms at the margin and the inefficiencies created by taxing inputs of exempt firms.
Financial Services
The primary issue that arises in taxing financial services is deciding the extent to which these services represent consumption or investment (International Tax Dialogue 2013). One principle laid out by Grubert and Mackie (2000), and elaborated on by Weisbach (2019), is that financial services need not be taxed if the resulting consumption can be observed and taxed. For example, the fees incurred in investing in a mutual fund should not be taxed because consumption out of the returns from the fund will be taxed eventually. On the other hand, Auerbach and Gordon (2002) argue that, in order to maintain the equivalence between a VAT and a labor income tax, financial services must be taxed. Failure to do so would exempt income in the financial sector from taxation.
Another issue is that it might be difficult to tax imported financial services as these goods do not have to pass a physical border (Auerbach and Gordon, 2002).
VAT in Federal States
VATs present a unique challenge for federal states, especially large ones like Brazil, Canada, India, and Australia. These countries face a tradeoff between harmonizing VAT rules within a single market and maintaining local autonomy. If sub-national governments can decide their own VAT rules, inter-jurisdictional trade could be distorted by differences in rules across internal borders. On the other hand, giving the central government authority over VAT rules reduces local governments’ autonomy over one of the most common local tax bases—local consumption. Revenue is generally distributed according to some form of the destination principle except in the case of some local business VATs that rely on the origin principle, as is the case with Italy’s IRAP. Some local governments have autonomy over the base, rates, and other VAT rules, which gives rise to efficiency and evasion concerns.
Large federal states have arrived at quite different approaches to trading off autonomy and harmonization. Australia’s system is the most centralized—the GST is centrally administered, and revenue is distributed according to the destination principle, that is, the estimated share of the local consumption tax base. Rates and exemptions are determined centrally. India’s new GST is similar—rates vary across goods but are determined by a GST council composed of both state and central representatives. There are parallel state and central GST administrations with taxpayers assigned to one or the other partly randomly and partly by a formula. Canada’s provinces have much more autonomy—three of them have opted out of the federal VAT entirely and operate their own RST. Brazil has a completely separate federal and state VAT that has different bases (the federal one is mainly industrial goods), rates and even different credit mechanisms. The federal VAT operates on the subtraction method, while the states use the more standard invoice-credit mechanism.
As there are generally no border controls within a country, there are unique opportunities for evasion and loss of efficiency that arise when the center and states operate a VAT independently (Bird 2015). For example, input tax credits may not be transferable across states, which results in cascading taxes. Or, inputs could be taxed at a higher rate than final goods, which would result in excess credits and refund claims. State-level differences in rates and exemption create conditions for undesirable tax competition. Navigating different rules, regulations and administrations adds another layer of complexity for taxpayers. On the other hand, taking away sub-national governments’ authority over VAT policy undermines their powers in a federal system.
VAT Health Problems—Structural
Evasion
An often-cited advantage of the VAT over the RST is in enforcement. Because the VAT generates third-party information on business-to-business (B2B) transactions through the input tax credit mechanism, it raises firms’ cost of underreporting their true revenue especially from B2B sales. In contrast, the RST collects all revenue at the final stage (B2C transactions) where consumers have little incentive to demand an invoice and it is easier for firms to misreport sales, particularly cash sales. Some wrinkles in this narrative arise when we consider that the self-enforcement mechanism of the VAT also breaks down at the retail stage, and that firms still evade even at the B2B stage. For example, “ghost” firms are created for the purpose of generating fake invoices to claim input tax credits. Empirical work such as Pomeranz (2015) that demonstrate the success of enforcement strategies also reveal the existence of evasion in the pre-retail stage among registered VAT firms. We next describe some types of evasion in the VAT.
Empirical studies show that programs designed to raise compliance on sales reporting are often accompanied by increases in cost reporting as well. Carillo, Pomeranz and Singhal (2017) find that when Ecuadorian firms were notified that the tax authority knew of revenue discrepancies in the firms’ tax filings, they increased their reported revenue but also increased their reported costs by nearly the same amount. The findings in Waseem (2019a) suggest that one reason for this increase in reported cost might be that firms wish to report zero, but not negative, tax liability. He shows that manufacturers in Pakistan increased reported sales when an important upstream stage, electricity, was included in the VAT, but not when an important downstream stage, wholesale, was included. When the electricity sector was brought into the VAT, manufacturers’ taxable value added decreased as expenditure on electricity could now be deducted from the tax base. This meant that many firms would now have negative tax liability if they continued to report the same revenue as they did previously, resulting in a refund. But manufacturers requesting a refund come under additional scrutiny, which raises their cost of evasion. So, they prefer to report sales as being at least equal to their input tax credits to target a zero reported tax liability.
In theory, tax authorities can verify invoices, but it can be resource-intensive to do so depending on the reporting requirements and the ease with which this information can be processed. Reporting requirements to claim input tax credits vary across countries, and many do not automatically collect or match buyer and supplier invoices before providing input tax credits. Taxpayers often only report total sales and total taxable input costs instead of transaction-level information. So, although the idealized VAT is self-enforcing through the asymmetric incentives for buyers and sellers, these incentives may not hold in practice without invoice-matching. Instead, input tax credits open a new avenue for evasion that is not present in an RST.
In the simplest version of missing trader fraud, firms exploit the zero-rating by exporting goods from firm A in one jurisdiction to firm B (the missing trader) in a different jurisdiction, which sells the goods to a customer but does not remit to the tax authority. A more complex version is carousel fraud, in which firm B sells some good to another firm C in the same jurisdiction, which then sells it back to firm A and claims input tax credits on their purchase from firm B, but remits no tax on the re-export to firm A. In this transaction, the tax authority loses revenue equal to the input tax credit to firm C. In variants of this fraud, there might be a few more unsuspecting or complicit buffer firms between firm B and firm C to obscure the trail. The same goods may also be passed around firms A to C multiple times with input tax credits collected in every round, hence the name “carousel” fraud.
Clearly, carousel fraud would not be possible in an RST, although similar avoidance schemes might be possible. For example, cross-border remote sales in the United States posed a problem for state sales tax as the responsibility to remit a use tax or sales tax on these transactions rested with the consumer. Compliance with the use tax was very low and provided a means for consumers to exploit the different remittance rules, and therefore enforcement, governing local and remote sales to avoid tax, resulting in revenue loss for the state.
To take the chain incentives in the VAT one step further, some countries hold all firms along a supply chain jointly accountable for evasion by any firm in the chain. Many member states of the EU currently apply chain liability to certain goods and sectors, for example electronic goods in the United Kingdom (Walpole 2014). If a buyer has “reasonable grounds” to suspect that VAT on their purchase would not be remitted to the HMRC, they could be held liable for the unremitted amount. 4 This strategy is intended to identify and curtail missing trader fraud in sectors where the HMRC suspects collusion between firms.
How does evasion under the VAT compare to our benchmark of the RST? Though there are no direct comparisons of evasion, a test by Keen and Lockwood (2010) to compare the “effectiveness” more generally of a VAT and RST provides some guidance. They compare the effectiveness of VAT in raising the tax to GDP ratio relative to the tax systems they replaced—mainly RST. VAT adoption raised countries’ tax revenue to GDP ratio increased by a modest amount of 1.7 percent in the short run and about 4.5 percent in the long run, suggesting that they were more effective than the taxes they replaced. Ebrill et al. (2001) compare revenue under the VAT and RST immediately preceding and following VAT adoption and find that, on average, revenue increased by around 1 to 2 percent. One exception is Central Europe—revenue decreased by about 2 percent on average in Central European economies following their VAT adoption, which is likely because the tax reform occurred as part of broader economic transition.
Other Structural Issues
Inadequate refunding
A major practical issue in countries with weaker tax administrations and large export sectors is delays in, or non-payment of, tax refunds due to taxpayers. Because exports are zero-rated under a destination-based tax, exporters often have negative VAT liability and in principle are due refunds. 5 However, concerns with fraudulent credit claims or government budgetary shortfalls often result in delays and non-payment of refunds. In an IMF survey of thirty-six countries, Harrison and Krelove (2005) find that refunds can often be a substantial portion of VAT revenue (e.g., 50 percent in Canada and an average of 38 percent in the EU), making the delay or non-payment of refunds a non-trivial budgetary issue for governments.
Although many countries have statutory deadlines for repayment of VAT refunds, these are not always met. This acts as tax on production as firms are deprived of working capital during the delay and face uncertain monthly liabilities. Sharma (2020) finds that exports from industries that rely more heavily on intermediate inputs in production (and therefore are due more refunds) are hurt by a VAT. A 10 percent increase in intermediate inputs as a share of output is associated with an 8 percent decline in exports. These declines in export due to a VAT occur almost exclusively in low and middle-income countries that adopted the VAT and are stronger in countries where a VAT refund request triggers an audit. Reinforcing the finding that VAT only distorts trade where there are refund claim concerns, Benzarti and Tazhitdinova (2021) found that trade flows in the EU do not respond strongly to VAT rate changes—much less than they respond to tariffs, suggesting that VATs do not distort trade in Europe.
Better enforcement at border than internally
Another presumed advantage of VAT is that it increases revenue collection efficiency by increasing the share collected at the border at the import stage. From a sample of countries where data is available in 2004 and 2005, between 33 and 83 percent of gross VAT revenue was collected on imports (Keen 2008). Among the thirty-three countries surveyed, 55 percent of VAT revenue on average was collected on imports, and in at least thirteen countries VAT revenue on imports accounted for over 60 percent of total revenue. These statistics could be interpreted as signaling the importance of ensuring compliance at the import stage. But they also suggest that non-compliance is much greater on domestic sales, as these shares are higher than the import to GDP ratio in these countries, which would be a rough proxy for the share of VAT revenue that should be collected from imports. Morrow, Smart and Swistak (2022) employ this intuition in a structural framework to estimate the extent of domestic evasion of VAT. They first show that there is a much closer relationship between VAT gap estimates and the ratio of imports to aggregate consumption in countries with weak domestic enforcement environments. As the import ratio increases, they find that overall VAT gap falls, which allows them to separate the “compliance gap” in VAT from the policy gap.
Effective enforcement at the border with weak enforcement on domestic sales has negative efficiency implications for the VAT because it favors domestic production over imports.
Administrative and Compliance Costs
Because the VAT is applied at every stage of production, many more firms bear compliance costs than in an RST. To get a sense of this, in the United States in 2016 there were about 6 million firms with at least one employee, of which approximately 650,000, or 11 percent, were engaged in retail trade. 6 In addition, there were approximately 25 million non-farm sole proprietorships, of which 2.3 million were engaged in retail trade 7 . The share of retail enterprises in OECD countries is similar, ranging from 9 percent in Iceland to about 35 percent in Switzerland 8 , which means a RST would involve far fewer firms than a VAT. Moreover, the information requirements under the VAT are more cumbersome than under an RST. Taxpayers must often use specific invoice formats and must report every transaction and movement of goods instead of simply reporting total sales.
Existing measures of VAT compliance costs on businesses range from 0.25 to 2.5 percent of GDP, based on a narrow definition of compliance cost as the cost of complying with required regulatory procedures and excluding any cost incurred in tax avoidance (Barbone, Bird and Vázquez-Caro 2012). In both the VAT and RST, compliance costs are regressive in the sense that they make up a higher percentage of revenue for smaller firms and increase with the complexity of the tax in terms of the number of rates, exemptions, and other filing requirements. For example, Denmark has the lowest compliance costs of the Scandinavian countries at 0.03 percent of GDP compared to 0.07 percent in Sweden and 0.06 percent in Norway and has the least rate differentiation and lowest frequency of filing. Estimates of compliance costs under an RST are similar, amounting to between 0.03 and 0.09 percent of GDP in the state of Washington in 1998 (Washington Department of Revenue 2018; and author’s calculations).
These costs are estimated through surveys of firms, and by timing simulations of the filing process that a firm would be required to follow. While this approach has the advantage of being able to directly measure the burden imposed by specific compliance procedures, it may not accurately reflect all of the factors that matter. There may even be some beneficial spillovers of the accounting procedures required. More recently, some studies have used firms’ behavior at the VAT registration threshold to arrive at a more accurate estimate of compliance cost. Harju, Matikka and Rauhanen (2019) show that many Finnish firms report revenue just below the VAT registration threshold mainly because of the associated compliance costs. The size of this response estimated through bunching analysis shows that compliance costs amount to about 17 percent of value-added at this threshold. In a VAT system, small differences in compliance costs per-firm can translate to a substantial fraction of GDP because most firms in the economy incur these costs, unlike in an RST where only the retail sector is involved.
Quantitatively Assessing Real-Life VATs
We have argued that real-life value-added taxes differ from textbook, or pure, VATs in many dimensions. This will come as no surprise to anyone who has even the slightest familiarity with real-life VATs. But this statement applies to all taxes, not just to the VAT.
A more relevant question is how the deviations of real-life VATs affect their performance, as measured by their economic impact. What are the incidence and efficiency costs of real-life VATs?
Incidence and Progressivity
We begin by briefly discussing the incidence of our loose benchmark for the VAT, the RST. Ignoring for now the question of pass-through, some argue that, from an annual perspective, a clean RST is regressive because poorer consumers consume a higher fraction of their annual income. Over a longer horizon, savings are eventually consumed and taxed at the time of consumption. Ignoring bequests, all income is eventually consumed and taxed at the uniform rate, which makes a clean RST proportional.
Deviations from a uniform base might deliver more progressivity if the zero-rated or preferentially taxed goods tend to be those more heavily consumed by lower-income households. Most U.S. states exempt food and clothing from the RST, for example, and luxury items might be subject to a higher tax rate. Expenditure shares of food are declining in income, introducing progressivity to the RST. Deviations from the standard VAT rate usually follow the same principle: a reduced or zero rate is applied to necessities that form a large portion of the budget share of low-income households. A higher rate often applies to luxury items—often electronics and jewelry—that are more likely to be consumed by high-income households.
Both RSTs and VATs are susceptible to evasion, but the nature of evasion differs in the two systems in ways that have implications for incidence. The incidence of a VAT or an RST in the presence of evasion depends on two things: (i) the industrial organization of firms, all firms in the case of a VAT and of the retail sector in the case of an RST 9 , and (ii) how final consumers match up with types of retailers. As previously noted, in real VAT systems some intermediate goods producers underreport sales and over-report costs. This means even formal, registered retailers whose supply chain includes other registered businesses may benefit from evasion. The mapping between final consumption purchased from formal or informal retailers and VAT incidence becomes less clear.
To illustrate the importance of industrial organization, consider the implications of free entry into all sectors of the economy. This implies that profits, or expected profits, will eventually be driven to zero. Thus, on average firm owners (or others whose well-being is tied to firm profits) do not substantially benefit. If there is heterogeneity regarding firms’ willingness or opportunity to evade, those who successfully evade will benefit, while those who do not will lose. With free entry, then, the private benefit of successful tax evasion is passed through to consumers, who face lower prices depending on the extent of evasion that cascades down to them. If, on the other hand, there is not free entry, firm owners, and those whose well-being is tied to firm profits may on average benefit from firm evasion, with the extent and nature depending on the nature of the imperfect competition. Thus, to understand the distributional implications one needs to know the income level of the owners; one suspects that the owners of informal firms are on average lower-income than the owners of formal firms, but we do not have direct evidence that this is the case. Bachas, Fattal Jaef and Jensen (2019) show that effective rates of formal and informal taxation are rising with firm size. If we assume that larger firms are owned by richer individuals, then it is likely that owners of informal firms are lower on the income distribution than owners of formal firms. On the other hand, shareholders of large, publicly-traded firms might fall anywhere in the income distribution. These same considerations apply to understanding the incidence of RST evasion, but only with respect to the retail sector.
The next question is how final consumers match up with types of retailers. If we continue with the idea that avoiding and evading firms tend to be small, then within the retail sector the smaller informal firms offer lower prices because they and their supplier face effectively lower tax rates. If, for example, lower-income households frequent those informal retailers that offer lower prices, then the firm-level evasion and avoidance could have progressive distributional implications. Some of these differences in consumption patterns can probably be attributed to unobservable quality differences in the products of formal and informal retailers. Nygård, Slemrod and Thoresen (2019) investigate a related question—the distributional implications of collusive evasion in the retail stage of a VAT—and find that, in Norway the evasion-adjusted extent of inequality is higher. This result is in part due to the finding that the share of expenditure spent in the informal sector rise with income, as does income from self-employment, which is heavily underreported. An important consideration in this analysis is how evasion affects the relative price of goods in the hidden economy, which determines whether the gains from collusion accrue to the buyer or the seller. Given the dearth of empirical work on the topic, the authors show that the evasion-adjusted estimates of income inequality are robust to a wide set of assumptions about the incidence of evasion, even though the distribution of the gains from evasion are sensitive.
In contrast, Bachas, Gadenne and Jensen (2021) suggest that the existence of a large informal sector makes the VAT more progressive, based on the evidence from consumption surveys in twenty different countries of varying income levels. They find that expenditure shares in the informal sector are negatively related to total household expenditure, which is their proxy for household income. The use of total expenditure instead of household income prevents us from directly comparing their results with that of Nygård, Slemrod and Thoresen (2019). Instead of using direct reports of informal and formal consumption, Bachas, Gadenne and Jensen (2021) classify reported place of consumption (small stalls vs. large retailers, for example) into those likely to be registered and unregistered. Although these classifications are based on the characteristics of the final retailers only, Gadenne, Rathelot and Nandi (2019) as well as De Paula and Scheinkman (2010) show that formal and informal supply chains are segmented so that final retailers that are unregistered are also likely to be part of unregistered or informal supply chains. The negative relationship between informal expenditure share and total expenditure shares is such that, even if traditionally zero-rated goods such as food items were to be taxed, a uniform VAT would still be progressive. A major contributor to these negative “informal Engel curves,” as the authors call them, are differences in the types of products consumed by rich and poor households within broader consumption categories. Richer households are more likely to purchase from large, formal retailers who are also more likely to sell higher-quality versions of the same products.
Using data on household incomes and expenditures for the Dominican Republic, Jenkins, Jenkins and Kuo (2006) estimated the effective rate of taxation, considering the differential rates of tax compliance across households with different expenditure levels. They find that in 1998 the top quintile of households was subject to almost twice the effective rate of VAT per unit of expenditures than did the bottom quintile of households, because the richer households in both of these quintiles tend to shop relatively more in the formal markets.
To the extent that a tax on the formal sector has no impact on prices in the informal sector, the negative correlation between income and expenditure in the informal sector means that both the RST and the VAT are less regressive at the final stage. The difference in incidence under the two systems arises because, under a VAT, goods sold by informal retailers may still be impacted by VAT compliance further up the production chain and so poor consumers benefit less from non-compliance. This concern is lessened, however, by the empirical fact that firms that purchase from other registered firms are more likely to be registered themselves (DePaula and Scheinkman 2010), and so informal retailers may be more likely to be part of an unregistered chain, which means that incidence under the VAT in the presence of evasion is similar to that under the RST.
Efficiency
Both a pure RST and VAT raise the price of all goods (and services) relative to leisure, which is equivalent to a reduction in the real wage. This introduces a distortion in the supply of labor, which will cause inefficiency the larger is the compensated elasticity of labor supply. It is well-known that, in principle, differentiating the rate of tax across goods might reduce the extent of inefficiency, as it would allow a higher tax on goods with less elastic demand, reducing the behavioral response to the tax and therefore the marginal excess burden. In practice, however, this type of rate differentiation has implementation issues such as the possibility of evasion through misclassification or, in the case of a VAT, a potential for refund claims due to differences in the tax rate between inputs and outputs. It is also administratively more burdensome for the taxpayer and the tax authority.
In addition, some structural VAT imperfections discussed in the section “VAT Health Problems—Legislated” have potentially significant efficiency implications. To the extent that collection is more effective on imports than on internal value added, the real-life VAT resembles a consumption tax plus a subsidy to domestic production, specifically the domestic production that is effectively not subject to tax. This is a violation of the Diamond-Mirrlees (1971) production efficiency theorem, which states that as long as final consumption can be taxed, it is never efficient for a tax system to disturb production efficiency. This will be a continuing theme in what follows—imperfections in VAT generally lead to production inefficiency. Even though it is in principle a tax on consumption, imperfections have cross-firm implications because it is collected from firms, often leading to production inefficiency. This is in contrast to the impact of imperfections in an RST.
This theme is key to understanding the efficiency implications of VAT evasion, which—putting aside the collusion of final consumers to effect evasion at the retail level—is evasion by firms. Consider the implication of one possible pattern of evasion, that it is concentrated among small firms. (This might result from a possibly optimal allocation of the tax authority’s enforcement resources toward larger firms.) Because the effective tax rate on smaller firms is relatively low (perhaps even zero), there is a tax-related advantage to these firms, resulting in production inefficiency. Measuring that cost must take into account the optimal audit coverage policy of the tax authority, as well as the fact that there are likely to be fixed cost elements of the administrative and compliance costs, as discussed in Dharmapala, Slemrod and Wilson (2011).
The tradeoffs are similar in setting a size-based threshold for the VAT exemption. Keen and Mintz (2004) note that there is an efficiency loss resulting from a VAT registration threshold because some firms will reduce production to remain below this threshold. The threshold is set to trade off this loss in production with the reduction in administrative burden and the marginal social benefit of public expenditure. Under some assumptions, it would be optimal to set a positive and finite VAT registration threshold, allowing for some loss in production efficiency. Some firms that are smaller than the exemption threshold may nevertheless be better off as registered firms. Another dimension of choice for the tax administration is whether to allow voluntary registration and, if so, adjust the potential savings in administrative cost by the share of firms that would be unaffected by the exemption.
The Promise of New Technology-Based Treatment
At the same time that new VAT ailments have cropped up, new technology holds out some promise of treating these, and other related, maladies. At least seven countries now operate an invoice lottery to encourage consumers to obtain receipts for transactions. The earliest known and continuously operating lottery is the Uniform Invoice Lottery in Taiwan that was begun in 1951. Every receipt issued is associated with a unique lottery number that is entered into a bi-monthly prize drawing. Other countries that offer such incentives include Portugal, Slovakia, Malta, Poland, Georgia, and Lithuania. In Portugal, consumers can identify unreported transactions on an online platform and receive a tax rebate as a reward. These programs are designed to reduce evasion at the weakest point in the production chain where the incentives under the VAT and RST are identical.
These programs have apparently been successful in raising compliance at the retail stage. For example, Naritomi (2019) shows that tax rebates and lottery prizes to incentivize consumers to demand and report invoices from retailers substantially raised reported sales and VAT revenue in São Paulo. She finds that VAT revenue increased by about 9 percent net of expenditure in tax rebates and lottery prizes, although the net effect on welfare is less clear, as the administrative costs of the program and the cost to consumers for participating are unknown; 50 percent of customers who requested a receipt and would have been eligible for a reward did not claim it, suggesting that the costs are non-trivial.
Another key development is the rise of electronic payment systems and their use in tax enforcement. Many tax authorities require credit card companies to report transactions, which provides third-party information on firms’ sales. The use of electronic payments can lower evasion in both the VAT and RST systems. In fact, one of the first evaluations of the impact of credit card reporting on compliance comes from the United States, which uses the RST (although the study concerns income tax compliance). Slemrod et al. (2017) show that the businesses raised reported income to at least the amount reported in electronic transactions. If the key advantage of the VAT over the RST is to generate third-party information regarding the true tax base, the rise of electronic transactions and the involvement of consumers might have eroded some of that advantage. In developing countries where cash-based transactions are much more widespread, the VAT may still have an edge over the RST in generating a paper trail, but this is perhaps less relevant in less cash-based economies.
A related intervention is the use of e-billing systems, which are machines that automatically apply the correct tax amount on retail transactions and transmit data to the tax authority in real time. Eissa and Zeitlin (2014) found that a 2013 Rwandan mandate for all VAT-registered businesses to use an electronic-billing machine raised revenue by about 5.4 percent.
Automatic invoice-matching is a promising new system to cut down on fraudulent invoices. Fan et al. (2018) show that VAT revenues increased by about 14 percent when the Chinese government introduced electronic recording and linking of invoices in 2001. Korea has had comprehensive invoice-matching since they adopted the VAT in 1976 (Krever 2014), and many other tax authorities have adopted or are adopting this technology. The downside, of course, is the additional compliance burden on taxpayers to provide detailed invoice information for every transaction in specific formats and the additional administrative burden to manage this volume of information and provide input tax credits without delay. The recent experience of the Indian government introducing automatic invoice-matching in its Goods and Service Tax (GST) reform is illuminating. The filing dates for returns were staggered so that the IT system can handle the information.
Automated matching means that purchasers cannot claim input tax credits without the government receiving information on that sale from both the buyer and the seller. Without this system, buyers can over-report their purchases and sellers can under-report their sales but not be discovered without a careful audit. Invoice-matching improves compliance in a VAT relative to an RST, but the question is whether the gains in compliance are sufficient to offset the compliance and administrative burden.
Recommended Diagnostic Tests—An Agenda for Future Research
In parallel with promising recent treatment technologies have come powerful new diagnostic tests. A wave of recent research, some of which we have drawn on above, has leveraged the enhanced availability of administrative data and complementary innovative empirical methodologies to reveal much about how real-life VATs work and the effectiveness of certain policy initiatives in addressing some of its imperfections.
We believe that the next generation of research should address two tasks. The first is to integrate what is being learned about real-life VATs (e.g., with evasion) into a general equilibrium framework with heterogeneous firms that will allow us to make statements about the efficiency and incidence of alternative VAT systems. How quantitatively important are the deviations from a pure VAT? To be specific, how much do they add to the efficiency costs of a pure VAT, and do they change the textbook assessment of its incidence? This kind of analysis will allow us to evaluate proposed changes to VAT systems. Models with heterogeneous firms have become standard in international trade since Melitz (2003) and more recently in macroeconomics, as in Baqaee and Farhi (2018). Dharmapala, Slemrod and Wilson (2011) develop a model of heterogeneous firms with fixed administrative costs in the Melitz tradition, but do not use it to investigate the VAT. Their model shows that it is optimal under certain conditions to exempt small firms from the tax—a standard feature of most VAT systems.
Liu et al. (2019) characterize the intermediate goods producers as monopolistically competitive with the ability to price discriminate between consumers and businesses (or registered and unregistered buyers). De Paula and Scheinkman (2010), on the other hand, characterize the formal and informal sector as two separate competitive markets where the price of the intermediate good in the two sectors is determined simultaneously with the registration status, demand, and supply in each sector. In both cases, the difference in equilibrium price in the formal and informal sector affects the choice of registration status but come from quite different assumptions about markets. The price-setting mechanism is important because the impact of the VAT on prices in the untaxed sector is one of the determinants of the progressivity of the tax.
Liu et al. provide the theoretical insight that VAT registration depends on the intensity of intermediate input use 10 , sales to final consumers 11 , and the level of competition in output markets. This is also captured in the framework of Gadenne, Rathelot and Nandi (2019), which has only two production sectors. An upstream sector uses only labor to produce intermediate goods for the downstream sector that produces final goods. Consumers have a taste for both the final and intermediate goods. The two-stage production instead of the three stages as in Liu et al. (2019) has some disadvantages. For example, because downstream firms’ production relies only on taxable intermediate inputs (no labor), taxes have no impact on a registered firm’s input mix. 12 If some intermediate inputs were tax-exempt, registration would raise the relative price of exempt inputs. Overall, their model shows that taxes distort supply chains in the presence of evasion.
We now have empirical evidence of more imperfections in real VAT systems that could be incorporated into an optimal VAT framework. For example, the existence of invoice mills should intuitively weaken the link between the actual production function, demand function, and registration decision of firms. This channel of evasion could theoretically justify zero-rating domestic production that is upstream in the supply chain of exporters, as Pakistan has chosen to do. Such a policy would not be justified by existing VAT models where gaps in the VAT chain created by exemptions decrease the likelihood of registration and compliance. Other features that should be incorporated into a general equilibrium framework are differences between reported and real revenue and costs (i.e., misreporting), differences between input and output VAT rates, and differences in enforcement by sector.
The second task is to integrate what is learned about the impact of policy initiatives on, for example, evasion into a normative framework. Keen and Slemrod (2017) offer a simple framework in the context of an income tax, but notably it presumes a representative agent, so it cannot address the distributional implications of policies, although the framework could be extended to heterogeneous agents (and to a VAT) in a straightforward way and, ideally but non-trivially, to a setting with imperfect competition and firm profits.
In addition to theory, there is a lot more progress to be made empirically. As described in the section “Quantitatively Assessing Real-life VATs,” incidence, efficiency, and redistributive impacts of real VATs should be empirically estimated taking into account observed patterns of evasion, exemptions, and industrial organization. Administrative VAT return data are a new source of information on linkages across firms that can be used to inform policy. As we expand information requirements and interventions to reduce VAT evasion, we need to carefully assess their compliance costs and how they vary by firm size as well as by sector. This information will help us evaluate the tradeoff between imposing additional burden on firms and the social benefit from having firms, and in some cases consumers, actively participate in ensuring compliance.
Conclusions
About 100 years since its conception, the VAT has taken the tax world by storm, the United States being the conspicuous holdout. Like any tax, real-life VATs do not attain their ideal form. It is not hypochondriasis—excessive worrying that one is seriously ill—to move past admiration of the advantages of a pure VAT to address the fitness of the VATs in use today. Rather, it is prudent to do so. Luckily, recent years have seen substantial progress in documenting how VATs operate, diagnosing problems, and testing approaches to ameliorating these problems—through harnessing new technology and through using administrative tax data to understand the incentives and opportunities for avoidance and evasion. Further progress is needed toward constructing a comprehensive theoretical framework for the VAT that can incorporate such issues as chain incentives, exemptions, evasion and generate conclusions about the distributional and efficiency effects of the real-life VATs that exist now.
Footnotes
Acknowledgments
This paper is a revised version of remarks prepared for the Conference on the Value Added of Value-Added Taxation, held at the World Bank, May 9–19, 2019. The authors thank Anne Brockmeyer and Mazhar Waseem for their valuable comments.
Declaration of Conflicting Interests
The authors declared no potential conflicts of interest with respect to the research, authorship, and/or publication of this article.
Funding
The authors received no financial support for the research, authorship, and/or publication of this article.
