Public finance: theory, evidence and policy

Raj Chetty interviewed by Romesh Vaitilingam, 06 March 2009

Unfortunately the file could not be found.

Open in a pop-up window Open in a pop-up window


Download MP3 File (10.9MB)




View Transcript


Romesh Vaitilingam interviews Raj Chetty for Vox

January 2009

Transcription of an VoxEU audio interview


Romesh Vaitilingam: Welcome to Vox Talks, a series of audio interviews with leading economists from around the world. My name is Romesh Vaitilingam, and today's interview is with Professor Raj Chetty of Harvard University. Raj and I met at the American Economic Association's annual meetings in San Francisco in 2009, where we spoke generally about his research program on public finance. He began by giving us an introduction to the issues that arise in his research.

Raj Chetty: So, I'm broadly interested in the role that the government should play in the economy, both on the side of tax policy – so, raising money to fund things like public goods and social welfare programs – and on the design of how we should be spending that money – so, how should we structure programs like unemployment insurance, various other social insurance programs, disability insurance, and so forth. And if there's a theme, in terms of methodology, it's connecting the theory with evidence. So, there's traditionally been a lot of interest in mathematical theory in economics. And those theoretical models gave us some insights, for example, the work of Jim Mirrlees, and he has been very influential.
But Mirrlees work on the optimal tax system, essentially, in the end, said that optimal tax rates should be somewhere between 0 and 100 percent, which is actually mathematically a difficult conclusion to prove, but from a practical policy perspective, obviously, you want to know much more than that, right?
And so, in the past 20 or 30 years, as micro data has become more widely available, there's been expansion of empirical research that focuses on the effects of various policies and economic sort of state variables, the macroeconomy, on behavior and economic outcomes. And that empirical research has evolved largely independently of the theoretical research, at least in the US. Actually, in the UK, UCL and LSE have been leading the field to some extent in combining theory with evidence.
And that's, then, from a methodological perspective, a lot of what my research has been about – thinking about areas in which the existing theoretical models that are used in economics, which are often very stylized and ignore many aspects of reality, don't fit the data very well, and thinking about how to connect the data with the theory to improve the theory and then make different statements about policy.
So, let me talk about a couple of specific examples. One set of topics relates to social insurance and optimal design. In particular, take an unemployment insurance program. So, a very simple question which is important for any economy is, what's the optimal level of unemployment benefits? This is something that we're hearing quite a bit about now, especially because of the recession, questions about whether we should extend unemployment benefits in the US, how high should we set the level of benefits and so forth.
So, there's a classic trade-off in this area, as with any social insurance program, which is, when you provide more insurance, you help people in a state of need. So, when they lose their jobs, they have less consumption, less income, and they're suffering, and you want to give them a transfer of that state. We think that that improves social welfare.
But, the cost of doing that is that you reduce incentives to find a job, right? So, if you provide too high an unemployment insurance benefit level, you might increase the unemployment rate and thereby end up sort of hurting people on net, even though you're providing them this transfer, by lowering GDP, in essence. So, a difficult question is, you want to be somewhere between a zero unemployment benefit and a full 100% replacement of your wages, but where do you want to be in the middle?
Now, traditionally, economists have had the view that you don't want to have a very high level of unemployment benefits because there's some pretty compelling evidence that when you raise unemployment benefits, people take longer to find a job, so unemployment durations go up. And that's a fairly powerful effect. A 10% increase in unemployment benefits, in various countries, has been estimated to raise unemployment durations by something like 5 to 8%, which is quite substantial.
And so, from that perspective, a view emerged that, because of this distortionary cost of the program, we don't want to have a very high level of unemployment benefits. And some calculations by MIT economist Jon Gruber suggested that the optimal benefit level may be very close to zero.
So, one aspect of reality that the standard model ignores that's used to make those calculations is the idea that people may not have sufficient liquidity while they're unemployed. So, a lot of calculations in economics are based on a traditional permanent income level, where people are smoothing consumption relative to lifetime income. But, when you're unemployed, that's precisely a time when you're unlikely to be able to smooth relative to permanent income because, if you think about what's necessary to be able to maintain a high level of consumption when you're unemployed, you have to be able to borrow, right? Or you have to have the high level of savings.
Now, it turns out that the median job loser in the US has savings of just $150, net of unsecured debt. So, they essentially have no savings at the time of job loss. Moreover, the primary condition for being able to borrow is having a job. So, it's exactly when you don't have a job or when you're sick that you can't borrow, right?
And so, now, if you consider a model where people are liquidity constrained, when they've lost their job, you get a second effect, a second reason why providing unemployment benefits might increase durations, which is what I call in this research I've been doing "the liquidity effect". And that's the idea that, if you have more cash on hand, you're not under as much pressure to rush and search and find a job, right? So, suppose you're really short on cash, you've had to drastically lower you level of consumption, and you're making a mortgage payment and you're behind your bills. You're going to do anything you can to just get a job as quickly as possible – sacrifice the quality of the job in order to put food on the table in the short run, right?
Now, if I give you unemployment benefits, you're going to have a little bit more time to find the job that you like, maybe search in a less costly way. And all that is not a distortionary incentive effect. It's not that you are thinking that the net return to work is lower than it actually is. It's just that you're able to take longer, do the more socially efficient thing.
Now, I go to the data, and I ask how much of the effect of unemployment benefits on durations is due to the liquidity effect, which is a socially beneficial thing, versus the moral hazard effect, which is the incentive distortion traditionally emphasized in the literature. And using various methods, basically comparing the effects of severance payments, which give you cash on hand but don't distort your incentives to the effects of increases in unemployment benefits, I conclude that roughly 60% of the effect of benefits on durations is a liquidity effect rather than a moral hazard effect. And using this type of evidence, I show that the optimal benefit level, viewed from this perspective, taking the liquidity consideration into account, is much closer to the level you see in the US – actually, 50% or something like that – and that it makes sense, probably, to extend the duration of employment benefits in a recession like this. Although you don't want to go too far, you don't want to go to a system like Sweden, which has sort of indefinite unemployment benefits. But, there's definitely significant value in having short-run liquidity provision when people face shocks.
So, more broadly, this research has similar implications for health insurance, disability insurance, workers' compensation – the same kind of effects are going to arise in all of those programs. And so there are various other research studies now following up on this that look for liquidity effects in these other programs and trying to think about how they should be decided.
So, that's one example of combining theory with evidence to reach a different policy conclusion.

Romesh: What light do these shed on the actual policies that are out there? I mean, do they show that they are overgenerous or undergenerous, or does it vary across countries?

Raj: Yeah. It varies across countries, because the policies vary a lot across countries. So, the US is much less generous than European countries. Previously, economic theory was sort of, I think, way off in the sense that it said we should be having no social safety net, even much less than what the US has. And so, if the US is in the middle and the Europe is very high, economic theory was saying we should be very low. And I would say my work is suggesting that we should be more somewhere around the US, maybe a little bit above the US, probably not as generous as some of the European countries, which do have very high unemployment rates, and behavior that clearly looks like it's inefficient. So, you see people on the system until benefits expire, and then they go work for a little while, and then they come back on the social safety net. That doesn't seem like the kind of thing that you want to be encouraging.

Romesh: You were going to talk about some other examples of the kind of things that you look at.

Raj: Right. Right. So, another example, a very different style of work, is bringing behavioral economics into the public policy discussion. Now, that's traditionally a very difficult thing to do, because economists always like rationality. And particularly, in thinking about policy and normative questions and optimal policy, rationality is very helpful because, as soon as you abandon the assumption of rationality, it becomes very easy to justify any policy. It becomes completely undisciplined.
So, if I make the claim that, as a government, I know better how much people should save than what the individuals know, then I have no discipline in my model, in the sense that I can say, paternalistically, that I know you should be saving 30% of your income or 20% of your income. There's no way to sort of guide optimal policy.
So, one of the things I've been working on is trying to have a more disciplined approach to optimal policy design when people are making mistakes, so allowing for the fact that people don't optimize perfectly but still using the tools of revealed preference, which is what economists like to do, to say something about optimal policy.
So, the particular context in which I've been doing this is thinking about tax policies, where the standard assumption that we make when analyzing the effects of taxes on the economy is that everybody's fully aware of all aspects of the tax policy and they optimize relative to those incentives. So, in particular, say you have a progressive income tax system, where your marginal tax rate is changing – as your income goes up, your income tax rate is going up, right? That's the typical structure in most developed economies.
Now, in practice, if you survey people, you find that most people have very little sense of how earning an extra dollar affects their tax dollar, affects their tax refund – to take another very simple example, when you buy commodities in the US, say you go to the grocery store and buy something like razor blades. The price on the shelf says $8.99, but in fact you pay something like $9.75 when you go to the checkout register, because you have an 8% sales tax added and that's not saliently shown in the price. Now, the standard economic model would say that's irrelevant because people always pay attention to the full price and know about what's going on.
So, I set out to first test whether that assumption is true. In this very simple context of the grocery store, I did a field experiment, in collaboration with a major grocery chain here in northern California, where, for 1000 products over a one month period, we posted tags below the tags that you see on the shelf, which tell you the tax-inclusive price of the good. So, for example, with the razor blade, it would say, "Total price $8.99, plus California sales tax, equals $9.75."
And so we had half an aisle of the store where we had these tags, and then we had a set of control groups where we didn't put up these tags, other stores, which we had identified as being very similar to this treatment store ex ante, and so forth. And the basic finding is that demand for these products falls by about 8% during the treatment period, relative to other control products, just by putting up these tags. Now, of course, the null hypothesis based on the traditional model is it should have no effect, right? It's just redundant information.
Now, we also pursue a second strategy, which provides complementary evidence sort of on the long-run salience. And what we do there is compare the effect of a one dollar increase in the price of a good with a one dollar increase in the tax. So, traditionally, we would say it should make no difference if I'm increasing the price by a dollar or increasing the tax by a dollar, right? Because you care about the price plus the tax.
So, we test this by looking at alcohol consumption, because alcohol is subject to two state level taxes in the US – one that's included in the price and one that's added at the register or added on a restaurant menu. You wouldn't see it on the restaurant menu. The one that's included in the price is the excise tax, and the one that's added later is the sales tax.
And we have historical variation at the state level in the US, where California raises its excise tax or its sales tax and Arizona does not, so you have lots of sort of quasi-experiments. And you can estimate the effects of an increase in the sales tax or an increase in the excise tax on demand, for beer, say.
And so, you can see this very easily in a graph. You plot changes in the excise tax, the one that's included in the price, on beer consumption. You see a very sharp, negative effect. A 1% increase in the price through the beer excise tax reduces beer consumption by about 1%. So, it's an elasticity of one. For the sales tax, we get something that's like 0.1, one tenth of that, whereas you would predict that the two should be roughly similar. And again, that's consistent with the view that people are reacting to the price and much less to the thing that's added later.
So that, it's a simple demonstration, but of something that can be quite important in tax policy, which is, if people aren't optimizing relative to the system, the way you think about the optimal design of these systems is quite different.
So, in the second part of this project, we developed a method of analyzing the efficiency costs of taxation, where people are behaving in this way, basically comparing the effects of responses to prices and responses to tax changes to assess welfare consequences.
The idea being that I have two demand curves, a price demand curve and a tax demand curve, and they have different slopes. The price demand curve tells me about the person's true preferences, because they're paying attention to the price, and so I can use the tools of revealed preference to back out how much they really value a good using the price demand curve. And the tax demand curve tells me how they actually respond to the system.
And so, by putting the two together, I can make statements about how much welfare is lost when I raise taxes. Another example is the incidence of a tax. So, traditionally, if you asked an economist, "Should I levy a tax on the producer – on the firm, or on the consumer?" they would say it's irrelevant, because prices are going to change in equilibrium, and either way, whoever you levy the tax on, in the end they're going to end up being just as well off as if you had levied it on the other party.
So, in this model, that wouldn't be true, because if you think about the example of, say, a cell phone and cell phone surcharges. So, you might see a cell phone contract for $39.99. Typically, you'll have eight dollars of fees or something tacked on later. If you had levied that on the firm, they would have to advertise the price as, say, $47.99. Demand falls, and they end up having to cut the price in order to equilibrate the market. If you levy it on the consumers, they're able to just kind of pass it through, and the consumer doesn't completely pay attention and ends up bearing more of the tax.
So various sort of policy questions like this, you would change your views on that.
And then, more recently ... So, just one last thing on this topic. The sales tax case, I think, is very clean. It's interesting because you can cleanly demonstrate the importance of these things, but it's not particularly important as an overall tax policy, although, actually, in Europe, it's fairly important because of the VAT.
So, the income tax is very important in the US. And I've done a study with my colleague, Emmanuel Saez, looking at the Earned Income Tax Credit, which is an important low income redistribution program in the US, the most important anti-poverty program in the US.
So, we spend roughly $50 billion a year on this program. And the idea of the Earned Income Tax Credit is that we pay you a subsidy to work. So, every $10 you earn, you get $4 from the government. Now, of course, that subsidy can only work if people understand that there's a subsidy. So they have to recognize that their wage rate is actually $14 an hour instead of $10 an hour.
So we did a bunch of surveys, and people have done similar things in the past, of basically single mothers – that's the key target population for this program – at welfare agencies. And what you find is that they have very little knowledge of how their Earned Income Tax Credit refund – they get this as part of their tax refund in April – varies with the amount that they earn. So, what you see is they know that they get a big check when they file their taxes, but they don't know the amount of that check it's typically on the order of $4000 or $2000; it's quite a large sum of money for somebody making $10,000 – how that varies with the amount that they earn. But, that's the critical thing that they need to know for the program to increase the amount that they work.
So, we did a large scale, randomized experiment, in collaboration with H&R Block, which is the biggest tax preparation company in the US, where, for 40,000 low income individuals in Chicago, we split them into two groups randomly. And half of them were given simple information about the structure of this program, telling them something like, "Suppose you earn $10 an hour. Then you should really think of your wage as being $14 an hour, because of this program." And we tailored that information to that person's situation and so forth, and when they came in to file their taxes, the tax preparer was trained to spend about five minutes explaining the program to them. Then we tracked their earnings a year later. So, we have data when they come back to H&R Block one year later, and we asked whether the people in the treatment group increased their earnings relative to the people in the control group.
And our preliminary findings – we've just gotten the data recently – suggest that, actually, the impacts are fairly big, even from this simple, two minute intervention. We find that providing information reduces the fraction of people in extreme poverty, say incomes below $10,000, by something like 10% in this group, which is nontrivial, especially for such a light intervention.
And another way to see the importance of information is we calculate how much you would have to expand the program. The traditional policy tool we talk about is, instead of a 40% match, should we make it a 50% match or should we make it a 30% match? We would have to expand the program by about $10 billion to have the same effect as this two-minute treatment. Doing the two-minute treatment, we spent five dollars per person to do this two minute treatment. There are 20 million people who get the EITC in the US. That would just cost $100 million. Relative to the 10 billion, it's still a dramatic savings.
So, it shows you the power of information, and thinking sort of outside the traditional framework of perfect optimization in thinking about how to design policies.

Romesh: So, you're coming up with some pretty strong policy implications, focusing on issues around tax and social insurance here in the United States. But, I gather you're looking wider now ...

Raj: Yeah.

Romesh: ... You're looking at issues of finance in developing countries. I'd be interested in what the different issues are there.

Raj: Yes. So, more recently, I'm just starting to get interested in doing work in developing countries, and thinking about the optimal design of various kinds of programs there. And I think the key differences are, in the US, you're usually thinking in a system where the institutional structure is already very well developed. So, if you're thinking about a change in tax policy, here it's something like changing the parameters of the Earned Income Tax Credit or the level of the unemployment benefit.
In African countries or in India, the countries that are just developing, now it's more setting up such a system. It's not so much what's the optimal tax rate. It's what types of taxes should we even be trying to collect? How do we deal with serious problems of evasion? So, I think those are particularly central issues in countries that are now starting to grow fairly rapidly.
So, think of the Indian case, where it really seems like the public sector is a serious hindrance. So, you often hear stories of the private sector doing quite well. There's a lot of foreign investment. Companies want to come into India.
But then, they need to set up their own electricity plant or develop the roads near the company or set up their own sort of residences for people, because all the things that you take for granted in the US, that you know you're going to be able to get electricity, all these public goods aren't being provided by the government. And so that would suggest that, for India to really take the next step and be able to push forward in the future, you really need to develop infrastructure and public goods at a much higher level.
Now, the constraint in doing that is being able to generate tax revenue, right? And so that's kind of a collective action problem. So, no single individual wants to pay taxes, because he'd just like to free ride on everybody else. But, you need to get everybody to coordinate. And everybody's going to be better off if they pay taxes, have more public goods, attract more companies to India, and so forth.
So, I've been trying to think about ways in which you might be able to increase compliance with the tax system, and think about what effects that might have on economic growth. This work is still sort of at a speculative stage, but I'm interested in doing small scale policy experiments in India, trying to experiment with different tax collection systems.
One of the ideas I have is to try to exploit concerns for status as a way of trying to get people to pay their taxes. So, to give you an example, the universities in the US are very good at doing this. Often, in the US, if you make a big donation to a university – for instance, the University of Chicago recently got a $300 million donation, and within a few days the school was renamed the Booth Chicago School, right?
So, you could think of the government trying to exploit similar kinds of concerns, for signaling your status or your name, where you say something like, take a small village. There are a few wealthy individuals. People generally seem to underreport their income in order to evade taxes.
But, you say something like, suppose if you're the highest taxpayer, you get to designate your money to one of 10 public goods say a school or the local hospital or something like that – and if you're the highest contributor to that good, we're going to have a ceremony where we name the thing after you. At least for a period of a few years, we recognize you as the benefactor of that school. You kind of give people a carrot for participating in the system. Rather than, I think, right now, we focus very much on sticks: so we're going to punish you in some way. But enforcement is so difficult, I think.
And this way, I think it would be sort of possibly self enforcing, because, say there are four rich guys in a village and three of them have something named after them, and everybody knows this fourth guy has a big house and throws lavish parties and so forth. Where's his public good? What is he doing for the community? So, I'm interested in seeing whether we might be able to experiment with something like this and see if that increases funding for public goods, possibly as other beneficial outcomes in the world around them.

Romesh: Oh, some very interesting experiments in prospect. Come back and talk about them on a future occasion.

Raj: Yes.

Romesh: Raj Chetty, thank you very much.

Raj: OK. Thank you.

Topics:  Taxation

Tags:  tax policies, social welfare programmes, tax compliance

William A. Ackman Professor of Public Economics, Harvard University


CEPR Policy Research