Image by Sonny Side Up! CC 2.0 via Flickr

The famous 1976 paper by Anthony Atkinson and Joseph Stiglitz achieves an elusive goal. It provides a policy relevant result in a general framework that has played an important role in reshaping global tax policy. Four decades later, this paper still acts as an important guide in tax policy debates in both academic and policy communities.

What is the result?

The result looks at the best way to jointly implement direct taxes (income taxes) and indirect taxes (commodity taxes) when both efficiency and equity are important. The result states that if (i) consumption goods are not complementary with leisure and (ii) consumer tastes are independent of their ability to earn income, then the optimal indirect tax should be equal across all goods. In other words, we should have a GST/VAT that applies the same tax rate to all goods. Another way of stating this result is that any equity based goal that could be achieved with a progressive indirect tax is better achieved by changing the income tax (and transfer) system.

How restrictive are the conditions of this theorem? The first condition rules out Corlett-Hague style results, where a government that is able to tax consumption goods, but unable to tax leisure, may worry that they will create a distortion to consume too much leisure. In this circumstance, the government may wish to tax goods that are complements with leisure and subsidise goods that are complementary with working. While of theoretical interest, such issues are rarely the focus of policy discussions, and so will not be discussed further here.

The second condition can be interpreted as consumer patterns not providing information about a person’s affluence that is not observed through that person’s income level. In general, annual income is a good measure of affluence, but this may not always be true. For instance, someone may have a low observed income in a given year because they are a student, or recently retired, or have business losses. However, buying an expensive car might provide an indication of their affluence that is not observed through their income level.

Whether or not this is an important issue for policy is open to debate, and it may be possible to argue that taxes on expensive property (or first class flights, or yacht mooring, etc.) provides this type of insight into personal affluence, but it is a much more restrictive property than just saying that a tax is progressive or regressive.

Two further notes:

  • Taxes designed to address externalities (such as taxes on pollution) are compatible with the Atkinson-Stiglitz framework, and are applied in addition to a uniform tax. While this means that the final taxes are not uniform, the result is not meaningfully changed by externalities.
  • The result only applies to taxation, and not to government expenditure on public goods. Progressive government spending programs (such as libraries and public transport) are an important component of redistributive policy, and the Atkinson-Stiglitz result does not apply to such cases.

The intuition of the result

The key idea behind the result is that an income tax will create an economic distortion that reduces an individual’s incentive to earn income, but a consumption tax will create two distortions. One of these distortions is equivalent to the distortion of the income tax, as this indirect tax reduces the purchasing power of earned income and in effect acts as a tax on earned income. In addition, it will change preferences for consumption between taxed and untaxed goods, creating another distortion along this margin.

To see things in concrete terms, imagine a tax on a product consumed entirely by high income individuals (let’s say Rolls Royce cars). Even if these consumers don’t change their consumption of these goods by a large amount in response to a Roll’s Royce tax, the tax will still reduce the income of this group. In effect, it is acting as an income tax on wealthy people who purchase Rolls Royce cars. Wealthy individuals may now choose to work less and enjoy more leisure as the purchasing power of their income is diminished.

Therefore, the Rolls Royce tax creates the same distortions on earning income as an income tax. Moreover, the tax creates an additional distortion by changing the relative prices of Rolls Royce cars and other consumption goods. Therefore, the Atkinson-Stiglitz result suggests that an income tax will be at least as good as an indirect tax (but they may be very similar if the second distortion is small).


The Atkinson-Stiglitz result represented a major shift in policy advice regarding indirect taxation. Until this point the dominant paradigm was Ramsey taxation, in which inelastically demanded goods are taxed at a higher rate than elastically demanded goods. The shift to a system of equally taxed goods therefore represented a significant policy shift.

In the Australian context, this policy shift was seen in the ANTS package, in which the Wholesale Sales Tax was replaced by the GST. This mirrors trends in the global setting with an increasing number of countries implementing a value added tax. According to the OECD report on ‘Consumption Tax Trends’, around 160 countries have now implemented some form of VAT.

The Atkinson-Stiglitz result is also relevant for a number of ongoing policy debates in Australia. The result suggests that the GST should be applied with as few exemptions as possible, and that exemptions (such as food) that are based on equity reasons should be removed and replaced with changes that make the income tax and transfer system more progressive.

It also suggests that where a tax is based on an externality, such as the fuel excise and alcohol and tobacco taxes, the focus of the policy should be on setting a tax that is equal to the size of the externality. If this causes equity concerns, these concerns can and should be addressed through changes to the income tax.

Of course, with all of these taxes, care must be taken to interpret whether this result can be directly applied. With that in mind, the following section discusses a range of circumstances when the result is not valid.

Limitations of Atkinson-Stiglitz

The first case (discussed briefly above) in which the Atkinson-Stiglitz result will not hold is when consumption patterns give some information about the well-being of an individual that is not directly observed from income. For instance, if someone with low income buys a $2,000 bottle of wine, it suggests that they are well off in a meaningful sense and have only reported a low income in that particular year. While there are certainly goods that display this property, it is unclear whether the goods could be identified and incorporated into the tax system, and whether it would achieve a meaningful improvement compared to simply taxing income.

Another practical concern with this result is that it requires changes to multiple parts of the tax and transfer system to be implemented. For instance, in implementing the ANTS tax reform in Australia, a variety of variable wholesales and sales taxes were replaced with a GST, while simultaneous changes were made to the income tax, welfare system and pension payments in order to limit the distributional consequences. However, in most cases, such general reform to the tax system is not possible, and it may only be possible to change one tax instrument. If a tax rate is flattened (which improves efficiency at the expense of equity) without a corresponding adjustment to the income tax system, the Atkinson-Stiglitz result does not directly apply, and it is not clear whether the policy is welfare improving or not.

A further limitation of this result is that it is an abstraction, and ignores the administrative issues regarding the collection of a tax. In some contexts, collecting a commodity tax from a small number of suppliers (such as fuel refineries), may be simpler than trying to collect taxes at all sales. Different taxes also result in different amounts of tax evasion (and tax evasion from different groups), which must be taken into account in policy discussions, and is not considered in this abstract framework.

A final potential limitation of the result is that the framework of the Atkinson-Stiglitz result is built upon the idea that we fully incorporate indirect taxes and act on these as if they were an income tax (meaning that Australian workers would think about the GST they would pay on spent income when deciding whether or not to work). However, there is evidence from the experimental economics literature that suggests that this might not be the case. If this result is confirmed to be broadly true, and the effects are large, then the general premise of the result is questioned.

A comment on ongoing policy debates

To conclude, I will attempt to summarise this discussion into two pieces of policy advice regarding indirect taxes.

The first piece of intuition applies to policymakers considering introducing or increasing an existing indirect tax. In this situation, being progressive or regressive is unlikely to be a key priority. This is not a universal rule. For instance, it may be possible that the tax provides some information not observed in a person’s income. Or, it may be a second best option in a world in which it is not possible to achieve redistribution through the income tax, but this is a much higher standard than simply showing that the tax is progressive.

The second piece of intuition applies to policymakers concerned that existing indirect tax rates are not equal, such as the case of the GST exemption for fresh food. The Atkinson-Stiglitz result shows that unequal indirect taxes create distortions because of substitution between these goods with different tax rates. If this substitution is large, then equating the tax rates should be a priority for reform. However, if there is likely to be little substitution, then although Atkinson-Stiglitz would suggest equalising the tax rates in a first best world, the potential benefits of such a change would be small, and this would therefore represent a low priority for reform. For instance, in the case of the GST exemption for food, to make the case that removing this exemption is important, it is important to show that there is significant substitution between fresh food and other consumption items.

Further Reading

The original Atkinson-Stiglitz paper

An extension of the result by Louis Kaplow that shows that the result does not require the existence of an optimal income tax.

A paper written as part of the Mirrlees review of the UK tax system written by Ian Crawford, Michael Keen, and Stephen Smith.

Uniform VAT: Is It Optimal? Notes by Rufus Pollock

Optimal Taxation in Theory and Practice, by Gregory Mankiw, Matthew Weinzierl and Danny Yagan 2009.

This article has 2 comments

  1. Hi Peter,
    Some golden oldies there with Atkinson and Stiglitz.Have you ever read any work by Ed McCaffery on A Consumed Income Tax or Post Paid consumption tax . See


  2. Indirect tax can be defined as a type of tax where the incidence and impact of taxation do not fall on the same entity. It is collected by the government from an intermediary such as a retailer or a manufacturer. … Some examples of indirect taxes include sales tax, entertainment tax, excise duty, etc.

Leave a comment

Your email address will not be published. Required fields are marked *