Smartphones Are Gold Mines Of Economic Data
An economist taps into payment apps for their every financial transaction
Steven Cherry: Hi, this is Steven Cherry for IEEE Spectrum’s “Techwise Conversations.”
I don’t know about you, but when I leave the house, I need to have three things these days: my phone, my wallet, and my keys. Soon that’ll all be one thing, and it’ll look a lot like a smartphone.
Are you ready for your phone to be your wallet? Millions of people are aligning, if not actually combining, the two, with software that pays their bills and helps them budget their money. There are several apps for this now, but almost 7 million people use one called Pageonce.
That’s a lot of bill paying and budgeting—and a lot of economic data all in one place. So with the company’s permission, economists from the University of Michigan and the University of California at Berkeley dove into the data to try to get more of a real-time handle on the state of the U.S. economy.
My guest today is one of those researchers. Dan Silverman is on the faculty at both the University of Michigan and Arizona State University. He has a master’s degree in public policy from Harvard and a Ph.D. in economics from the University of Pennsylvania. He joins us by phone from Arizona.
Dan, welcome to the podcast.
Dan Silverman: Thank you very much. It’s a pleasure to be here.
Steven Cherry: Dan, you and your fellow researchers were looking to track consumer sentiment. What’s the problem with the way consumer sentiment is currently tracked? And maybe just start by telling us what you mean by consumer sentiment.
Dan Silverman: Well, consumer sentiment is a subject of keen interest among not just academic economists but, [as] you know probably from just reading the newspaper, a much wider public as well. So you say, “Well, why is that of interest?” Well, there are many reasons, I believe, why that draws a lot of attention, but I think the two primary reasons are, first, that it helps predict the future of the economy at least over the relatively short term. So if you wanted to make a guess about the future growth of the economy and you wanted to use fairly limited amounts of data to make that guess, measures of consumer sentiment would be a good tool for that. Much more interesting and more tantalizing, but also much harder to establish, is an idea first, I think, or at least made famous by Keynes, which is some notion of “animal spirits”—in other words, that consumer sentiment itself not just predicts the economy but actually causes the future growth of the economy. And there the idea is that if consumers are feeling uneasy about their current state in the economy and, more importantly, in the future, they save a bit more, they restrict their consumption, they take fewer trips or go out for dinner less or buy less expensive clothes, etcetera, and then businesses see when they look at their bottom line some depressed demand. The growth in their business wasn’t doing quite as well as they thought, and what do they do? They hire a few less people or take their people to work a few less hours or make less investment in the firm itself, and then it seems the consumers were right to be uneasy. In other words, it’s a self-fulfilling belief that the economy is not doing very well.
Steven Cherry: And I guess it’s pretty pervasive. I mean, businesses want to know how much—whether to have a lot of inventory or a little inventory; banks want to know whether people are going to be saving a lot of money or spending a lot of money. It really goes on and on. So there’s something called the “consumer sentiment index,” and I guess it comes from one of your institutions—actually the University of Michigan. How does it work, and what’s wrong with it?
Dan Silverman: That’s right. In fact, there are many measures of consumer sentiment, but the longest-standing and maybe the most famous is something called the Consumer Sentiment Index, and it’s fielded by the University of Michigan and drawn from something called the Survey of Consumers. So, that’s a survey done every month from a group of about 500 randomly selected people called on the phone, and they’re asked these questions about how they think about the time now as a good time to buy a major purchase, what do they think about how they’ll be doing economically in the next six months or so. And so that’s 500 people, and that’s been going on for more than 50 years. So that’s been an extremely productive enterprise, but it obviously has some important limitations. The first is accuracy. So we’re going to talk later, I hope, about what we are doing, and that includes study of whether consumers are breaching the limit on their credit cards: so if I asked you, called you up on the phone somewhat out of the blue and said, “Did you have a credit card breach last week?” Well, first you might not answer the phone because you’re busy. Second, you might answer the phone and say, “It’s none of your business.” Third possibility is, you answer the phone, and you’re trying to answer the question truthfully but you actually don’t know. A fourth possibility is that you do know, but you’re a bit embarrassed about the fact that it happened, and you’d rather not say. So all of these sources of, I would say, “static” lead to inaccuracy in people’s responses to these kinds of questions. The other thing is, it’s quite expensive, As a result, you get an important delay oftentimes in getting results because of the expense. And I want to compare that to what we can get with these new kinds of data, which are being drawn from an application like we’re working with in Pageonce.
Steven Cherry: Yeah. So let’s talk about what data does Pageonce have, and how did you access it?
Dan Silverman: So Pageonce, as you mentioned at the beginning, is an application, and it can be used on [a] mobile phone, on a tablet, and also over the Web. And if you’ve ever tried it, what you do is basically link through your credentials that you use over the Web, say, your checking account or your savings account or your credit card bill or your investment accounts or even your utility bills, etcetera. And what this company Pageonce does is provide for you [in] a relatively simple way a balance sheet every day of what your finances look like. So, it’ll tell you your balance in your savings account and your checking account and your investments; it’ll tell you which bills are due when and how much; and it will also, as of last fall, give you the option to pay some of those bills directly from your phone. So it becomes something like an electronic wallet, as you mentioned at the beginning. The kinds of data that Pageonce comes upon simply by providing this service is perfectly accurate descriptions of how you use your credit card and your bank account to buy, in this case, clothing and food outside the home.
Steven Cherry: There are some confidentiality issues here, right? Do Pageonce users know their data is being used?
Dan Silverman: Yeah. So this is part of the user agreement, that data can be used in a way that is unidentifiable. In other words, no individual name or address is ever linked to the data, and we’ll never publish anything that could possibly identify or be used to identify individuals and their behavior.
Steven Cherry: One more methodological question: The population of people who have smartphones and want to manage their finances on them is not exactly representative of the population as a whole, so how do you draw conclusions from one to the other?
Dan Silverman: So, I think that’s a really important observation, and I think there’s two things to say about that. The first is that, of course, especially the smartphone population is not representative of the entire population of the United States—and I should mention that this is a company that serves almost exclusively American consumers. But it’s broader than, at least, I would have guessed at first. So we have a broad swath of the income spectrum here, and you have a pretty broad swath of the age spectrum, but you’re right to say this will be disproportionately better educated and disproportionately higher income. And in fact you’ll miss the very low end of the income spectrum, folks who don’t have access to smartphones, but also you’ll miss the very high end, who even if they have a smartphone will be very reluctant to link into an app like this. So your question is, then, “Okay, so what do we learn?” And I think in part you learn simply from the enormous size of the sample. So even if you thought, “I’m just getting a taste of what smartphone users are doing,” this is hundreds of thousands of them—they just can’t be ignored. But beyond that, we have methods for inferring—and we have survey methods to do this—inferring from the selected sample what the larger economy and consumers are doing.
Steven Cherry: Very good. So, let’s get to what you found. You looked for spikes and dips in two things: basically, these breaches—and that sounds bad, but it’s basically the number of times people go over their credit limit on a credit card—and then you also looked at the number of times they overdraw their bank accounts. So, you compared those things to several different things. One of them was the labor market, right? What did you find there?
Dan Silverman: Yeah, so we compared. Let’s just focus [for] simplicity on the credit breaches. As you said, it’s an indicator that a person went over his or her credit limit on a credit card, and what I think is interesting is, it needn’t be bad news, at least for the macro economy. We watched it go up and down over the past 15 months or so, and we saw, for example, a big spike—you know, the highest point is in the spring of 2011, so May, June of 2011, very high levels of these breaches. And from there it declined through most of, I would say, 2011. And then into the winter of 2012, the beginning of 2012, we saw some sort of distressing signals that this indicator of financial distress was starting to rise. And you say, “Well, okay, that’s, I think, intrinsically interesting, at least to economists, but how does it compare to other measures?” And what we saw was two things. The first is, as you say, with employment we saw that it did track, in other words, that it was highly correlated to—at least contemporaneously, at the same time—with new claims for unemployment insurance. So, if you look at new claims for unemployment insurance in the spring of 2011, you would also see a spike. So, we saw in 2011 a spike in this indicator of financial distress; you would also see, if you looked at the data, a spike in new claims for unemployment insurance in the spring of 2011; and you would also see improvement in those numbers in the early part of 2012, just as we saw. And then what we saw were these distressing signals that, in fact, in the spring of 2012, this past spring, some signs of the weakening of the job market again in the spring of 2012. And so that was exciting to us, that we had this indicator of financial distress drawn from, as I would, say from naturally occurring data, these Pageonce data, and it seemed to be well correlated with these very well-established indicators of the macro economy, in particular, new claims for unemployment insurance. The second thing we did is to correlate the same kinds of indices of financial distress, credit limit breaches, with this measure of consumer sentiment where we began our discussion. And the Consumer Sentiment Index isn’t collected as frequently; it’s only done every month. And here we see something, I think, especially exciting, which is that these indices that we’ve crafted from these naturally occurring data in Pageonce are not only correlated with consumer sentiment, they seem to, in other words, to predict it or to lead it. In other words, what we’re seeing is maybe five weeks ahead a spike in credit limit breaches, [and] five weeks later you see a dip in consumer sentiment of approximately, you know, a similar magnitude.
Steven Cherry: Very good. Well, Dan, it’s a great line of research, and thanks for taking the time to tell us about it.
Dan Silverman: It’s my great pleasure. Thank you.
Steven Cherry: We’ve been speaking with University of Michigan economist Dan Silverman about a new way of crowdsourcing data for broad economic indexes. For IEEE Spectrum’s “Techwise Conversations,” I’m Steven Cherry.
Announcer: “Techwise Conversations” is sponsored by National Instruments
This interview was recorded 22 August 2012.
Segment producer: Barbara Finkelstein; audio engineer: Francesco Ferorelli
NOTE: Transcripts are created for the convenience of our readers and listeners and may not perfectly match their associated interviews and narratives. The authoritative record of IEEE Spectrum’s audio programming is the audio version.