This post has been republished via RSS; it originally appeared at: Microsoft Research.
Episode 124 | June 16, 2021
In the world of economics, researchers at Microsoft are examining a range of complex systems—from those that impact the technologies we use to those that inform the laws and policies we create—through the lens of a social science that goes beyond the numbers to better understand people and society.
In this episode, Senior Principal Researcher Hunt Allcott talks with Postdoctoral Researcher Evan Rose about Allcott’s work exploring the everyday decisions people face, like buying fuel-efficient cars or taking out payday loans, and how a clearer understanding of these decisions can shape meaningful public policy. Allcott shares how his and others’ research shows that policy can often have complex outcomes resulting in hidden benefits and drawbacks, as in the case of taxes on sugary beverages. The researchers also discuss why individuals often feel the competing motivations of making bad versus good decisions—a tension that often lies front-and-center in scenarios primed for behavioral public economics research.
HUNT ALLCOTT (TEASER): So you can only measure the effects of payday loans on certain outcomes that you can measure in the data that you have. And so we can measure impacts on bankruptcy and overdraft ’cause you can get data on those things. But what about stress and anxiety? What about your utility being shut off or getting evicted? What about tension with your family ’cause you were or were not borrowing money at some time? What our work is doing is different. We’re trying to look for direct evidence of market failures or direct evidence of behavioral bias and basically use that for evaluation of these policies. And I think some combination of, you know, our style of research and the impact-evaluation style of research will hopefully allow us to triangulate and make better policy.
EVAN ROSE: Welcome to the Microsoft Research Podcast, where you get a front-row seat to cutting-edge conversations. I’m Evan Rose. I’m a Postdoctoral Researcher at the New England lab here in Cambridge, Massachusetts, and I’ll be your host today as we speak with my colleague, Hunt Allcott, a Senior Principal Researcher at Microsoft Research. Now, Hunt is a leading figure, I would say, in the emerging field of behavioral public economics, which is the science of how to set policy, like environmental regulation and taxes, when we acknowledge that, first of all, people are people, not computers, and as such are subject to all sorts of really interesting cognitive and behavioral biases and mistakes. Hunt’s work has touched on a ton of fascinating subjects, from sugary drinks to cigarettes to social media, and I think really shows off the power of economic reasoning paired with careful empirics to tackle questions that impact millions of people’s lives every day. Hunt is also my boss, I think, or at least he’s one of the people who hired me at MSR. But even if he wasn’t, I would still tell you that he’s one of the most stand-up and smartest people I’ve met in the profession, so I’m looking forward to our conversation, Hunt, and excited to have you on the podcast.
ALLCOTT: It’s great to be here. I’m not sure that I’m actually technically your boss, but I’m thrilled to get to work with you.
ROSE: Effective boss. So, Hunt, I want to start just by getting to know you a little bit. So let’s start with your background. You grew up on the West Coast, as I understand, and you majored in engineering at Stanford. How on Earth did you end up as an economist?
ALLCOTT: Yeah, so I grew up on the West Coast, specifically in Eugene, Oregon, the track capital of the world, and I’m a proud Eugenian. You know, even as an engineering major, I was kind of in the middle of engineering economics and public policy. I was, you know, interested—still am interested—in environmental issues, and I did an individually designed major, which some colleges let you do, uh, called energy engineering, which focused on things like designing energy-efficient buildings and wind turbines, stuff like that. When I was at Stanford, I did a master’s degree in management science and engineering, what used to be called engineering economic systems and operations research. That kind of sits at this engineering-economics policy interface. And so even as I was doing these degrees that had different names that weren’t always economics, you know, I was using these same basic tools that economists use, constrained optimization and game theory, statistics, just writing computer code. And so it was natural when I eventually, uh, came to be a full-time economist.
ROSE: Now, engineering is a lot of differential equations and such. Do you remember when you first started to think about the peculiarities of human behavior and how that might affect the questions you cared about?
ALLCOTT: Yeah, I think it was in grad school when I first got interested in behavioral economics per se. Then I continue to be interested in energy and environmental issues, and energy efficiency was and continues to be a key part of that. And it was funny, you know, you read through, what is the rationale that policy makers give for being interested in energy efficiency? And part of it is we want to save the environment. You know, reduced energy use means less climate change. But a lot of it is about saving consumers money. Now, on one level, that’s totally obvious. If we use less energy, we pay less for energy, and so we’re saving money. But on a deeper level, I realized—and many other people have realized—that this is actually a totally fascinating thing because, why is it that we need government regulation to, in a sense, force or encourage consumers to save their own money on energy costs? So usually, of course, we think of government regulation as being justified by market failures. Consumers don’t have good information, or there’s some missing price, there’s a climate-change externality, firms have market power, et cetera. And so what exactly was the market failure? And it turned out that when you go read the regulatory documents, it really was about behavioral economics and saying, “We think that consumers are not fully accounting for these energy costs that they could save for themselves,” and that’s what our market failure is that we’re trying to correct with regulation. And so that’s how I got interested in behavioral economics. Now, in parallel with my interest in energy efficiency, behavioral economics was really growing, and it’s grown a lot in the last 20 years. And so I became interested in this idea that I, and many other people, could merge this behavioral-economics thinking with these energy-efficiency policy questions.
ROSE: Great. So, actually, I think that’s a perfect segue to where I want to start, which is sort of with a high-level overview of what exactly is behavioral economics, how do you think about it, how do you define it, and how do you use it in your work?
ALLCOTT: So public economics is the study of how governments set taxes and spend money and also includes some aspects of government regulation. Behavioral economics is the study of how people make decisions, often with a particular interest in how we, as normal people, might make decisions that are sometimes not in our own best interest—by smoking cigarettes, why we make risky choices, play the lottery, why we don’t save enough money for retirement, et cetera. So behavioral public economics is the intersection of these two ideas. How do we make public policy when people might not act in their own best interests? So, should we have cigarette taxes, and if so, should they be large cigarette taxes or small cigarette taxes? Is the lottery a great way to raise public funds and also help people have fun? Or is it a regressive tax on people who are bad at math, as some people allege? What’s the right way to encourage people to save more money for retirement? Now, here’s why this is so hard. Economists have this great machinery of benefit-cost analysis. So we’re considering a bridge that would cost a billion dollars to build. Should we build it? Well, we can ask, are there enough people who would be willing to pay $5 to drive over that bridge every day such that that outweighs the 1-billion-dollar cost? The problem is that this machinery—we call it the revealed-preference machinery for benefit-cost analysis—assumes that consumers act in their own best interest. If I’m willing to pay $5 to drive over a bridge, that’s $5 of benefit from that bridge. So, therefore, if I’m willing to pay $10 for a pack of cigarettes, then that pack of cigarettes is giving me a $10 benefit. But notice that something feels incomplete about this revealed-preference assumption here with cigarettes. The average smoker dies 10 years earlier. So maybe I’m getting $10 worth of value from that pack of cigarettes, but maybe I’m not. Maybe there’s something about my future health that I’m not taking into account. And so this standard revealed-preference, benefit-cost machinery, by assumption, can’t even engage with the types of public-policy debates that we want to take on in this space. And so we need a new theory, we need new empirical strategies, new ideas that at least admit the possibility that consumers might make mistakes, and then we want to do benefit-cost analysis of public policies, taking those potential mistakes into account.
ROSE: I see. So, if we’re willing to admit that consumers are making mistakes, that sort of implies that there was a correct or a right action that they could’ve taken, and I imagine a critique you’ve heard before and thought a lot about is whether or not that makes this sort of whole endeavor somewhat paternalistic.
ALLCOTT: Yes. That’s a huge critique, and I think the way that I think about the paternalism critique is just, how do we decide whose preferences to respect? And, you know, I think, when people say that something is paternalistic, the concern that we often have is that there’s someone else’s preferences that are being imposed on someone. The parent has particular views, and those views are being imposed on the kid. The regulator, the policy maker has particular views, and those are being imposed on individuals. And the economic approach actually tries to sidestep that concern. We’re trying to figure out what an individual’s “true preferences” are and then make policies that help to facilitate what those true preferences might be. So, in a sense, it’s paternalistic, but we’re trying to still use an individual’s own “true preferences,” to make policy.
ROSE: I see, but, you know, central to the approach we typically take in econ is this idea that you mentioned, of revealed preference. We assume that when people make a choice, they are doing that because it’s in their best interest, and we prefer that way to learn about people’s preferences over something like asking them, “What do you like? How many utils do you get from consuming an apple?” So if people’s revealed actions don’t reveal anything about their preferences, how are we supposed to do that in this space?
ALLCOTT: So what we’re really looking for is choices where—or situations where—you’re choosing inconsistently. Imagine I ask you to decide whether or not you want to get a credit card versus just use the debit card that you already have. And you say, “Yeah, I’d like to use the credit card. That sounds good. I’ll get a credit card.” But now imagine that we sit down and we look at the credit card’s disclosures and fees, and we talk through how, you know, for many people, they end up spending too much, and they carry an unpaid balance and pay a high interest rate. And after learning this information, you say, “Actually, for me, I changed my mind. I don’t want to use a credit card. I’ll just stick with my debit card. That’ll keep me out of the fees and the interest and everything.” Nothing about this choice changed. It was still the same credit card, the same debit card. But I gave you more information, and you made a different choice. So that’s what I mean by inconsistent choice. Now, we can extend this further, right, because it’s actually easy to understand which of those two choices that you made probably more faithfully reflects your true preferences. It’s the choice that you made after you had more information and were allowed the time to thoughtfully think through things. So that’s where this idea that we call characterization failure comes in. So when you don’t have perfect information, you might not correctly characterize the potential outcomes that result from your, uh, different options. And so what the behavioral-economics approach is really about is looking for cases where you make inconsistent choice, say with versus without information, and then try to use only the decisions that you make when you’re making an active choice with full information, and you have the time to fully think through that decision. And this is something that I think the field is still grappling with. It’s something—there’s a, uh, professor named Doug Bernheim at Stanford, who has really pushed the field on this. Um, but I think that’s approximately how many behavioral economists think about this.
ROSE: Yeah, that’s really interesting, ’cause when I think about my day-to-day, I make tons of decisions constantly with really limited information, and sometimes I do that because I don’t know any better, but often I do that because research is costly—it’s gonna take me a long time to go figure out, uh, all the information I might need to make a fully informed decision. In this framework of characterization failure and other inconsistent choice, how do we distinguish between people sort of rationally just making the best choice they can, given the cost of acquiring more information, versus an actual failure of optimization that would be consistent with a behavioral mistake?
ALLCOTT: Yeah, so part of what you’re speaking to, I think, relates to models of basically a constrained optimization within the brain, so cognition models. You know, there’s some nice work by Xavier Gabaix at Harvard that basically posits exactly the model that you have in mind, where you say, “There are a lot of different things I could take into account as I make a decision—what car to buy, you know, where to go on my next vacation—and I’m gonna focus only on the things that matter. And I’m gonna make those decisions in a way that is optimal for me given my constraints.” There is a distinction between optimally allocating your cognitive resources and optimally searching for information versus making the optimal decision. So you could say, “Listen, I’ve only got, uh, three hours to decide what kind of computer I’m gonna buy next, and so I’m gonna optimally search for information to do that, and then after three hours, I’m gonna stop and just buy that computer.” So that’s an optimal decision within, you know, your time cost, but it doesn’t mean that you bought the right computer for you in the sense that you bought the computer if you had infinite time to search. And so I think that’s the distinction, and that illustrates how you might still make suboptimal choices even if you’re optimally allocating your mental bandwidth.
ROSE: Right. And, of course, people could invest all the time they want and still fail to understand an incredibly complex product like the fees associated with unpaid credit-card balances in your example.
ALLCOTT: Totally, totally.
ALLCOTT: And then a key distinction, of course, is what is the shape of the mistakes that we end up making in aggregate? So in some cases, you know, some people buy computer A incorrectly, other people buy computer B incorrectly, but there’s no systematic bias. You know, some people drink too much soda pop. Some people drink too little soda pop. There’s no systematic bias. But there’s another class of mistakes which has received a lot more attention in behavioral economics, which is the systematic bias. So I’m systematically drinking too much soda.
ROSE: Yeah, okay, so I think I can understand this so far. And this seems particularly focused on policies that try to protect us from misunderstandings, right, where we don’t have the complete information to make the best decision. But earlier we talked about—and later we’ll talk about more—examples where the regulation is about forcing us to do what’s right for ourselves in the long run, you know, changing our decision even if that’s the decision we might make under full information at the right time. So are those kind of policies also trying to address a characterization failure, or is there a different motivation there?
ALLCOTT: Yeah, it’s a great question that you’re hitting on an issue that, I think, is perhaps the most controversial issue among behavioral economists in this space of how to do welfare analysis when consumers have inconsistent choices. And let me—let me give an example. This is kind of a fun example to build intuition. So Jerry Seinfeld has this great bit—I don’t know if you’ve seen it—about Morning Guy and Night Guy. So Night Guy goes out, has a good time, stays out late, and drinks too much, and then Morning Guy pays for it ’cause Morning Guy wakes up and he hasn’t slept enough and, uh, he’s hung over. So Morning Guy hates this, so he tries to figure out how he can get Night Guy to stop going out. So he says, “Well, I’m just gonna sleep in, I’m gonna lose my job, and now Night Guy can’t go out because he doesn’t have any money.” Now, that’s inconsistent choice because Morning Guy and Night Guy have different preferences on whether to stay out late, but is that a characterization failure, or is that just two different selves that disagree on what to do? And as you were sort of suggesting, this matters a lot for many of the public-policy questions, uh, involving inconsistent choice. So think about the social-security system. The social-security system was set up because, in practice, people were reaching retirement age, and they hadn’t saved enough for retirement. And so, you know, we had this horrible scenario of people becoming too old to work and couldn’t sustain themselves. And so we said, “Well, this is unacceptable. We’ll set up a social-security system.” Now, you could take a hard line and say, no, these are just two different people with different preferences. There’s the 30-Year-Old Guy who just wants to go out and spend a lot of money, and he doesn’t care about Retirement Age Guy. And then Retirement Age Guy arrives and doesn’t have any money. It’s the equivalent of Morning Guy being hung over. And so why are we preferencing the consumption of Retirement Age Guy when Working Age Guy is telling you by revealed preference that he doesn’t want to save for retirement. I’ve intentionally chosen an example where it’s easy to be empathetic with the government-intervention approach, and that’s what we’ve been doing in the US and, you know, other developed countries for a very long time. But there are arguments against that type of approach, especially in other settings, um, and Doug Bernheim, again, from Stanford has, I think, been particularly helpful in pointing out what these arguments might be. You know, nobody ever says on their deathbed that they should’ve worked harder and spent more time at the office and planned more for their futures. And, you know, he points out that, even though we might have the intuition that the regulator should step in and, for example, you know, preference the consumption of Retirement Age Guy, we don’t really have sharp evidence on whether this type of failure to save for retirement or other examples at present focus are actually a characterization failure or just different preferences at different times. This speaks to something really interesting where there’s some emerging research and a lot more to be done, which is, we have a really clear sense that people are present-focused, in other words, that they choose different things for themselves in the moment than they would choose for themselves in advance. You know, we always plan to exercise tomorrow, and tomorrow arrives, and we don’t exercise. We’d like to go on a diet tomorrow, but tomorrow arrives, and we don’t start our diet. We have dessert. Uh, we’d like to save for retirement, but tomorrow arrives, and we don’t save for retirement. There’s a lot of underlying potential reasons why this might be the case, and I think there’s more to do there.
ROSE: Suppose we take that discussion as it is and we want try to find evidence of mistakes in data. How do we begin doing that?
ALLCOTT: Yeah, so let’s pick up on this energy-efficiency discussion that I alluded to earlier. Let’s talk about Corporate Average Fuel Economy standards, or CAFE standards. So these are regulations that basically require automakers to sell more high-fuel-economy vehicles to consumers. The argument for these, as I mentioned before, actually hinges largely on a consumer-protection argument. The actual regulatory impact analyses that the government has done for fuel-economy standards over the last 10, 20 years basically hinge heavily on this argument that, “We’re gonna save consumers money,” um, and they explicitly say, “Consumers are myopic. They’re not paying attention to the future. It’s hard to think about fuel economy and energy, and so, you know, we need to regulate, uh, to help consumers buy the cars that they would buy if they weren’t making what we call characterization failures.” So, notice that these assertions are testable, and then getting to taking this to data, here’s what you could do. You could get people who are shopping for a new car, you could intercept them at the dealership, and you could run a randomized experiment, right, where the treatment group gets factual, non-persuasive—just give them information about the fuel economy of different cars they’re considering. A control group doesn’t. And then you could measure whether the treatment group buys a higher-fuel-economy car than the control group. Now, if the regulatory-impact analyses assertions are correct, that, you know, we don’t buy high-fuel-economy cars because we’re not paying attention to fuel economy or we have bad information, then this experiment of drawing attention and providing information should cause the treatment group to buy higher-fuel-economy cars. Um, that would be inconsistent choice between the treatment and control groups, and then it would be clear which one is a characterization failure. So that’s an example of how we would take this to data.
ROSE: So, Hunt, you’ve done this, um, many times, so let’s dive into an experiment you recently ran with Chris Knittel, which was published last year. How did you structure that experiment to answer these questions, and what did you find?
ALLCOTT: Yeah, so we hired research assistants who intercepted people at car lots—at eight different Ford dealerships around the country. And then we ran a basically symmetric experiment online with people who said they were shopping for cars, and, indeed, most of them ended up buying new cars in the next year. And treatment group gets information, control group gets nothing. And what did we find in terms of differences between treatment and control? Absolutely nothing. So statistically zero effects. When you hit people over the head with fuel-economy information, they do not buy higher-fuel-economy cars, and our sample was large enough that we can rule out any meaningful systematic misunderstanding of energy efficiency leading to any meaningful, uh, impact on fuel economy. And so that means that, to justify, you know, a CAFE standard that would push average fuel economy from 20 to 25 then up to 40 then up to 56 miles per gallon, which is the progression we’ve had over the last 10 years or so of CAFE standards, you would need something else other than, “Consumers don’t pay attention to fuel economy or they’re poorly informed.” And so you basically have to rewrite that part of the regulatory-impact analysis to take out that assertion ’cause it can’t justify these policies. Something else could, but information and attention cannot.
ROSE: So here, you’re providing information about the fuel economy of the car. Were you also trying to teach consumers about the consequences, globally or individually, of driving a high-emissions vehicle?
ALLCOTT: No, so the regulatory justification here, notice that it was not, “Consumers aren’t buying hybrids because they’re not aware of climate change.” The regulatory justification was, “Consumers aren’t buying hybrids because they’re not aware of the impacts on their own pocketbook.” And so we went out to test that specific idea—that consumers are misunderstanding the implications for themselves. You could imagine a different experiment that disclosed, you know, information about the environmental implication of different cars, and I’ve worked on other experiments like that. And there’s a huge literature of really nicely written papers by many other scholars that do that, but that wasn’t the right research question for us here.
ROSE: Well, let’s switch gears from cars and talk about another interesting topic—somewhat more controversial—that you’ve done a lot of fascinating work on, which is payday lending. So tell us a little bit about the policy landscape in debate in payday lending.
ALLCOTT: Yeah, so just for background, payday loans, of course, are single-payment loans that are due on your next payday. So you go into, you know, historically, it is an actual, physical brick-and-mortar store, often in a strip mall, and you say, you know, “Here’s my proof of income. I’d like to borrow $300. It’s gonna be due on my next paycheck,” typically in two weeks. And you’ll pay a fee of, uh, $15 per $100 that you borrow. So if you take out a $300 loan, you’re gonna owe $345 in about two weeks on your next paycheck. So this makes total sense for occasional situations. It’s easy to tell a story where you don’t have a lot of money, your car broke, you need a loan to get your car fixed so that you can keep your job and then, you know, make up the money, uh, soon after. The challenge here, and the thing that really worries people, is that, in practice, people use these loans over and over and over again. So the Consumer Financial Protection Bureau has done some nice analysis with data from payday lenders showing that more than half of payday loans are dispersed as part of sequences of loans to people that are more than 10 loans long. So paycheck after paycheck after paycheck after paycheck, I’m coming in, I’m paying this 15% biweekly interest and just racking up interest. And so I think that’s what’s generated the real concern, among regulators at the state and national levels. And people say, “Listen, these are predatory. They’re too high-cost. They should be banned.” The problem is that, uh, this is another one of these behavioral-economics questions, and if you don’t think that there’s a behavioral bias, your view then would be, “Well, banning payday loans, or restricting access to payday loans in some other way, is actually limiting people’s access to credit at the exact moment that they need it the most.” So your car breaks, you really need a payday loan, your state bans it, you lose your job, you can’t pay for your car to get fixed ever, et cetera, and that ends up hurting you. And so I think that’s what the real debate is. 18 states have banned payday loans. The Consumer Financial Protection Bureau has been focused on payday loans and has gone back and forth. So the Obama-era appointees finalized a rule in 2017. The Trump-era appointees, uh, rescinded part of that rule in 2020. Now the Biden-era appointees say they want to return to “vigorous regulation.” So there’s just a really active debate, and it’s a fascinating question. And it matters a lot because the borrowers tend to be lower-middle-income Americans.
ROSE: I see. And so for these groups that advocate for banning them or regulating these loans really harshly, what kind of mistakes do they think consumers are making? Are people being manipulated? Are the lenders deceptive? What’s the theory of harm here?
ALLCOTT: Yeah, it’s a great question, and it’s a different theory of harm than in credit cards. So with credit cards, a big part of the concern has been there are hidden fees, people don’t understand the product, there’s failure of disclosure. With payday lending, it’s actually striking how simple and clear the product is. So when you go into a payday-loan center, they have all of the fees and associated interest rates posted on the wall. You can look on the wall, and it’ll say, “If you borrow $100, you’re gonna need to pay $115 in two weeks, and the associated annualized interest rate is 391 percent.” And in surveys, people say that one of the reasons they really like payday loans is that they understand them. They’re simple, and they’re happy with the service they’re receiving. So this is not about a product that is deceptive in its terms. You need a different theory of harm. And the theory of harm is that people misunderstand how they will use the product. So some people call this use-pattern mistakes. So the theory’s not really that payday lenders deceive you but instead that their product allows you to deceive yourself. Now, there’s actually a second theory of harm, which is that payday loans prey on humans’ natural tendency to focus too much on the present at the expense of our future. So we talked about the social-security system earlier. Just like the social-security system is basically a forcing mechanism that makes us tighten our belts during working age and save so that Retirement Guy is not broke, perhaps banning payday loans could make us tighten our belts this week and avoid borrowing so that Next Month Guy is not broke. So notice that this is, again, as we talked about before, a theory of harm that relies on this controversial present-focus theory.
ROSE: I see. So, ideally, what we’d like to do is tell payday lenders, “You can’t lend to Night Guy anymore. You can only lend to Morning Guy.”
ALLCOTT: [LAUGHTER] Uh, basically, yes. That’s part of the motivation.
ROSE: So tell us about how you tried to tackle some of these issues in the most recent study you worked on.
ALLCOTT: Yeah, so this has been just a really fun project, and I’ve learned a lot. It started about four years ago. I actually went and met with the executive team at a major payday lender, and I sat down with them, and I explained, you know, our thinking of behavioral public economics, actually, probably very similarly to what I, you know, shared with you, Evan, a few minutes ago. And I said, you know, “We’re experts on thinking about regulation when consumers allegedly make mistakes, and we want to do a study with your customers to help calibrate a benefit-cost analysis of payday-lending regulation.” And I kind of expected them to say, “Uh, that’s a little bit risky. We don’t want to be involved in this.” But they said, “Great, let’s do it.” And then I said, “But you know you need to sign a legal agreement that allows us to do this survey of your customers and use your data, and you’re not gonna get to futz with the results or edit the paper or anything.” And they said, “Yeah, we understand. That’s what we want.” And, in fact, there are a number of papers that are written with payday-lending company data, and so I think there’s an interest on the part of at least one payday lender to have lots of research be done. So we pitched them this project, and we basically ran a survey and a randomized experiment with, uh, customers at this company, where we were trying to basically empirically test the two theories of harm that I laid out before, right? So the first theory of harm was people don’t anticipate their high likelihood of repeat borrowing. So to answer that, we, uh, did something very simple. We just—when people were taking out loans, we asked them on a survey, “What’s the chance you think you’re gonna get another loan in the next eight weeks?” And then we can compare that to actual data. So, on average, what’s the likelihood that people take out a loan in the next eight weeks? Uh, it’s actually quite striking. People, on average, are pretty close to calibrated, uh, correctly but not quite. In fact, all of the overoptimism is focused in the inexperienced group. So people who are just getting their first loan in a while from the lender or they’ve only had one or two loans recently from the lender, these folks really, uh, underestimate their likelihood of borrowing again in the next eight weeks. They’re overoptimistic. By the time people have a few loans under their belt, they tend to be correctly calibrated. And so I think that’s sort of partial support but also partial rejection of this theory of harm that people don’t understand what they’re getting into. So the second theory of harm was this Morning Guy/Night Guy thing, right, where, the idea is that people want to motivate their future selves to stay out of debt and that we might want to give priority to the preferences of the advanced self as opposed to in-the-moment, Night Guy-spending self. What we did there’s a little bit complicated. Basically, what we did is we offered people an incentive of $100 if they stayed out of debt for the next eight weeks. So if we saw in the data that they didn’t borrow from our lender or from any other payday lender in the state over the next eight weeks, they would get a $100 bonus. And we used experimental techniques to basically elicit how interested they were in that bonus. And notice what is this? It’s like a commitment device in the sense that it is something that will provide motivation to your future self to stay out of debt. It provides that additional $100 incentive to avoid borrowing. And so people who feel that their future self isn’t gonna tighten their belt enough, those are people who are gonna be really interested in actually accepting that $100 “if you’re debt free” incentive. And so, indeed, what we see in the data is that people really do like that $100 “if you’re debt free” incentive, and it’s consistent with some of their qualitative responses. So about 90% of people say that they would like to have more motivation to avoid payday-loan debt in the future. So I think this is actually fairly strong support for this concern that Morning Guy and Night Guy have different preferences.
ROSE: So, if there’s demand for this kind of commitment device to say, “I want to commit to behave like Morning Guy now when I’m taking out this loan,” is there some sort of market failure or reason why the lender’s not offering those kind of commitment devices or another party’s not doing that in the market now?
ALLCOTT: You know, in the payday lending case and in many other cases, it’s just actually hard to figure out how you would actually make money off of, you know, a way of limiting people’s future borrowing. And even if you could figure out a way to actually do that logistically, it’s often hard to really convince people that, you know, this is a good idea, that they really want to commit themselves, and that that commitment outweighs all the contingencies that might happen in the future. It might be that I think that I’d like to stay out of debt over the next eight weeks, but, you know, there’s always some chance that I’m gonna lose my job, that something else is gonna go wrong, and so designing a commitment device that also builds in the flexibility that we know that people might want is a real logistical challenge.
ROSE: So in that case, if we don’t expect solutions to some of these issues you identified in your experiment to rise organically, what is the space or scope for policy to intervene in the payday lending space based on what you found?
ALLCOTT: Yeah, so we, in our paper, look at a couple different policies. The first is that we just look at banning payday loans, and that’s, as I said, what about 18 states have done by imposing 36 percent annual interest rate caps that, uh, sounds like a high interest rate, but, you know, it’s low enough that, in practice, all the payday lenders can’t make money and they have to exit the market. So, when you want to evaluate that kind of policy, you basically need to trade—it’s not just a question of is there some consumer bias? It’s a question of is there enough consumer bias in the data to outweigh the value that consumers are actually getting from payday loans? And, you know, in our data and in our modeling exercises, really what you would need to justify a ban on payday lending is persistent over-optimism. Like, it would have to be the case that a borrower comes in, is certain that she is gonna pay off in two weeks. Two weeks arrives. She doesn’t pay off, but she’s certain that she’s gonna pay off two weeks after that. That period arrives, can’t pay off, but she’s certain that she’s gonna pay off two weeks after that, et cetera, et cetera, et cetera. So, in that world, consumers are basically being pumped for money by payday lenders. Of course, in our data, that’s not quite what we see. People are over-optimistic when they start to borrow, but by the time we’re a few loans in, people have wised up, and so it’s not enough of a behavioral bias to justify a ban on payday lending in the context of our model, which makes a lot of assumptions. What does look a little bit better in our model is what some people call a rollover restriction. It’s basically something where the state, or maybe at the national level, there would be a database of borrowing from all payday lenders, and then so you report to that database, and then once an individual person has gotten to, say, three payday loans over a short period, there would be an imposed cooling-off period of say 30 days before you could go get another payday loan from any lender. So that looks better in our model, and the reason is actually the Morning Guy/Night Guy thing that we were talking about before. So in our model, we take the position that people’s long-run preferences are the ones that the policy makers should respect, and what the rollover restriction does is that it forces you to get out of debt after three loans. And that’s something that’s consistent with your long-run preferences that you have at the time that you’re starting to borrow, and so it basically enforces this motivation that people say they want to get out of debt faster.
ROSE: Let’s say, though, that we were unwilling to take a stand on Morning Guy versus Night Guy. We didn’t want to try to preference some choices over others. We wanted to let people do what they do when they have full information. Are there other regulations or changes that you think would be beneficial in this market if we’re not willing to take that normative stance, or is this market basically good to go?
ALLCOTT: Um, you know, I think that the biggest problem in the payday-lending market is just that the cost structure is so imposing. I mean, if you think about what a storefront payday lender is, it is a company that’s paying rent and sitting one, but usually two staff members in that center, all day, five days a week. Now, you know, in my experience, sitting in these payday-loan centers, there’s a lot of traffic on Friday. There is not a lot of traffic Monday through Wednesday and only a little bit more on Thursday. And, so, what’s going on is, uh, that there’s a lot of fixed cost that’s being expended to disperse a small number of loans. And so what we really want to have in this market, I think, is more competition or more innovation that brings down the cost of payday lending. And then, if people want to keep borrowing, that’s actually not quite as bad, ’cause they’re not paying such a high interest rate for it. And, so, there is a lot of movement in that direction. The payday-loan business is moving increasingly online, and then there are also a number of new competitors in this space that are reducing credit risks and bringing down costs in other ways.
ROSE: Despite those changes, though, which do seem very positive, I mean, it just remains such a controversial topic. And I think that part of that stems from, um, studies—you might call them impact evaluations—which try to just estimate the causal effect of getting a payday loan relative to a counterfactual or a state of the world where I didn’t have access to that loan altogether, and those seem to find that sometimes getting access to payday lending can be quite deleterious. So what do we find from those studies? And how do you reconcile that with what you were arguing earlier about the benefits of payday lending?
ALLCOTT: There are a number of what you call impact-evaluation studies, including two that are quite recent and quite credible that use what we call regression discontinuity. So, many lenders use credit-score cutoffs to determine who gets loans. So if your score is above, say, you know, 700 on some scale, you get a loan, and if your score is 699 on that scale, you don’t get a loan. And then you can look at those two groups—the 700 group and the 699 group—as they proceed forward into the future. And what these recent studies have found is that getting a payday loan in the, you know, medium to long term will increase your use of overdrafts, so basically other high-cost debt in your bank account. They increase the likelihood that you eventually go bankrupt. These are really credible and really valuable studies. The problem with this style of research is that it’s like looking for your keys under a lamppost. So you can only measure the effects of payday loans on certain outcomes that you can measure in the data that you have. And so we can measure impacts on bankruptcy and overdraft ’cause you can get data on those things. But what about stress and anxiety? What about your utility being shut off or getting evicted? What about tension with your family ’cause you were or were not borrowing money at some time? What our work is doing is different. We’re trying to look for direct evidence of market failures or direct evidence of behavioral bias and basically use that for evaluation of these policies. And I think some combination of, you know, our style of research and the impact-evaluation style of research will hopefully allow us to triangulate and make better policy.
ROSE: This relates a little bit to our conversation earlier about the types of information people have when they’re making decisions about which vehicles to buy. You know, the analog here is, you’re providing information about emissions standards versus the consequences of driving a high-emissions vehicle for the climate and all sorts of stuff. And people may understand the products very well, but was your sense, from talking to people who were heavy users of payday lending, that they also understand some of these potential negative consequences on financial health?
ALLCOTT: It’s a great question. So in our study, we don’t ask that. And I’m actually not aware of a study that’s tried to look at whether people understand the potential, you know, long-run impacts as opposed to understand the products. So I think that’s something for future research.
ROSE: So, let’s switch gears yet again and move to another topic which I think is incredibly fascinating—lies at the intersection of behavioral economics, public policy, and inequality and is also quite controversial—which is taxes on sugary beverages. So, why don’t you give us a quick rundown of the policy landscape in that space.
ALLCOTT: Yeah, so, there’s been a growing realization that sugary drinks are particularly harmful to your health. And just to be clear, the health community thinks that—or my read of the public-health literature is that—the public-health community believes that drinking sugar in water form is more harmful than drinking the same amount of sugar in solid form. So, uh, stepping back into something that I know better professionally, uh, in the policy and econ literature, there are about 40 countries and 7 US cities that have enacted taxes on sugary drinks, uh, to address this public-health concern.
ROSE: So, a prominent criticism of these taxes that have been cropping up all over the country is that they’re regressive because the people who tend to consume these drinks tend to be lower income, on average. Is that something that is actually true in the data?
ALLCOTT: So, it’s true that low-income people drink more soda. It’s correct in a mechanical, financial sense. Low-income people then pay more in soda taxes, and that’s then certainly a higher share of their incomes. And so it’s regressive in the financial sense that you might consider regressivity. But it’s an incomplete story for a couple of reasons. So, first is, in some estimates, including the work that I’ve done with a couple of co-authors—Ben Lockwood, uh, and Dmitry Taubinsky—the elasticity of demand that we estimate is greater than 1. So, in other words, people are just very elastic, very responsive to the price increase that’s generated by a soda tax. And, in fact, they’re so responsive that when the price goes up, people actually reduce their total expenditures on soda, and then that’s happening more for low-income people and it’s freeing up disproportionately more money for low-income people to spend on things other than soda. A second reason this is an incomplete story of regressivity is, there’s a question of what you do with the revenues. So, you can take the tax revenues and you can hand them to rich people, or you can take the tax revenues and you can, you know, devote them to pre-K education in the school district, you know, something that has higher incidence, higher positive incidence, on low-income people. The third thing, which is where the real behavioral economics comes in, is what some people might call the progressivity of bias correction. If you think soda taxes are regressive, what about type 2 diabetes? Right? So the idea is that the health benefits of the soda tax also disproportionately accrue to lower-income people, and that offsets the financial regressivity. Another way to look at this is, in our data with Ben Lockwood and Dmitry Taubinsky, we do a survey of about 20,000 American, uh, households. And we deliver this nutrition-knowledge questionnaire. And it turns out that in our data, the lower-income people tend to have lower nutrition knowledge. And so that’s sort of suggestive evidence that there’s maybe less knowledge in the low-income community, uh, about nutrition that is then offset by the tax. So the soda tax just is progressive in the sense that it’s disproportionately offsetting bias among low-income people.
ROSE: So, backing up for a second, though. The whole motivation for these taxes is that people don’t have correct information, so they don’t internalize the health costs and consequences of drinking sugary drinks, and that’s the reason why we need them?
ALLCOTT: Well, it’s interesting. I mean, the stated motivation by policymakers is that there is a public-health concern. And then you’ve got to say, “Well, how does that translate into economics? What’s the market failure that people are—have in mind?” One part of the market failure is that when I consume more soda, that makes me more likely to get sick—obesity, diabetes, heart disease, et cetera. And since we have insurance pools, either private insurance or Medicare/Medicaid, a lot of that cost of my sugary-drink consumption is imposed on other people. So it’s a type of externality. So, that’s one justification for taxing sugary drinks. The other justification is this kind of internality or consumer-bias behavioral public economics justification. The idea there then has to be that people are not able to make decisions for themselves that trade off how much they like drinking soda with the health consequences that might accrue.
ROSE: So, the first argument seems like a classic case for a tax. There’s some sort of externality to individual behavior, so we should tax that good to bring the private cost closer to the social cost of the behavior. Second one is you mentioned is obviously much more behavioral. But, you know, if this is really about information—people don’t really know what they’re getting into and don’t understand the health consequences of consuming these goods—why wouldn’t we do, like you suggested in the emissions-standards case and just try to get people more information about the sugar content of what they’re consuming and how that might affect their health?
ALLCOTT: You know, of course we already have nutrition-facts panels in the US. Chile and other countries have started to require front-of-package, uh, nutrition labeling. And then there’s also a debate about implementing warning labels, you know, stop-light warning labels or even, uh, graphic warning labels on sugary drinks. So just like in the cigarette case, you might put pictures of, you know, people who have problems with their gums or pictures of people with cancer, you might put pictures of obesity or tooth decay on sugary drinks. And that’s super controversial for lots of reasons, but that’s absolutely part of the policy discussion.
ROSE: And when would that be a better idea than just taxing grams of sugar in every beverage?
ALLCOTT: I think one is, you know, in practice, information provision in this type of labeling doesn’t move demand very much. And if your view is that the externalities are large and that there may be some other internalities that are not fully corrected, there’s some other consumer mistake that’s not fully corrected by information provision, you might think that you want to reduce demand more than, you know, just what an information label would do. So, like, if you think about the cigarette case, when you put warning labels on cigarettes, that reduces demand by a little bit. We could probably add graphic warning labels, uh, on cigarettes in the US, and that would have some additional impact. But the policymakers have decided that we want to have, you know, very large cigarette taxes in some states. There’s a view that information provision isn’t getting us all the way there. So, I think that’s part of the consideration. Another consideration is the targeting of the tax instrument versus information provision, right? So if your view of the world is that the real problem is that people don’t have information, a tax is not a good way to deal with that—as you were suggesting, right? Because it’s gonna reduce consumption. It’s gonna distort consumption by the already-well-informed-and-rational people. And so you might think that information provision might be well-targeted. You could also imagine that it’s the already-well-informed people who would be the most responsive, and it might be that, you know, information provision is actually not well-targeted in the sense of attacking market failures.
ALLCOTT: There’s a third issue, which I think is particularly fascinating, which, you know, I first came to understand in reading a paper by Emmanuel Farhi and Xavier Gabaix, both Harvard economists. In the case of sugary drinks, as we have discussed, there’s more consumption among lower-income people. And the problem with the tax is that you have lower-income people then paying tax money to the government, and we’d rather that not happen. We’d rather low-income people keep their money. What does information provision do? It might generate the reduction in consumption that we want without requiring low-income people to pay money to the government.
ROSE: So, why don’t you sum up for us in this topic, and what do you take away from your research on sugary drinks that you think policymakers should know when they’re deciding whether or not to ban things, how to set taxes? What are the lessons here?
ALLCOTT: Yeah, so Ben Lockwood and Dmitry Taubinsky and I, uh, in our paper, try to estimate what the optimal level of a soda tax would be—taking into account concerns about regressivity, what are the levels of consumer bias and externalities, what you do with the revenues, et cetera, and try to do this in one holistic benefit cost analysis. And our sense is that the optimal level of the soda tax in our models is not far off of the soda tax that’s currently being used in many US cities. So about 1 to 2 cents per ounce. But I think there are some additional insights that you might want to bring to the table. One is that the soda taxes that are implemented in the US so far are only in seven cities. And what happens when you live on the border between, you know, Berkeley and Albany, and Berkeley imposes a soda tax and Albany doesn’t? Well, when you’re shopping in Albany for food, you buy your soda in Albany. And that’s what economists call leakage. Basically, any leakage that happens is reducing the efficiency of the tax that we’ve imposed. Ideally, what you would do is, if you thought soda taxes were a good thing—and our models suggest that they are—you would implement them not at the city level, but you’d implement them at the state level or even the national level to reduce the proportion of leakage that you see. There’s another insight, which I think is actually really low-hanging fruit. The existing soda taxes are all some amount per ounce of sugary drink. So 1 cent per ounce. The issue is that some drinks are really sugary and other drinks are actually not that sugary. And it’s the sugar that generates the harm, not the water. And so if you were starting from first principles, the way that you would design these taxes is, you would have a tax that scales in the grams-of-sugar content in the drink, not the volume of liquid in the drink. And that’s something that a couple countries are doing, but, actually, uh, all cities in the US are not doing. In some of our estimates, there are substantial welfare gains to be had from shifting in that direction.
ROSE: Interesting. Low-hanging-fruit juice, I suppose.
ALLCOTT: [LAUGHTER] Exactly.
ROSE: Hunt, thank you so much for this really fascinating conversation and for being a part of the Microsoft Research Podcast. I’ve learned a ton from your work over the years and I’m really looking forward to seeing what impactful stuff you’re gonna come up with next. If you want to learn more about Hunt, you can use Bing to search his name, and you’ll find his website right at the top of the search results. If you want to learn more about Microsoft Research, you can check out microsoft.com/research. And we’ll see you next time on the Microsoft Research Podcast.