Research

Using Big Data to solve an economic mystery

Economics Professor Daniel Silverman and four co-authors used “naturally-occurring data” — data gathered about behavior as it happens rather than from a survey or an experiment — to explore a mysterious discrepancy between economic theory and reality. The work was published in July in Science magazine.

2014-11-05

In 10 years, when we look back on this second decade of the 2000s, we’ll probably see Big Data as one of the most powerful drivers of discovery and innovation. Yet as it stands today, the promise of Big Data — in particular, its usefulness — remains uncertain. That is what makes research by W. P. Carey School economics professor Daniel Silverman and four co-authors so exciting. The research, published in Science magazine in July, represents a new use of “naturally-occurring data,” that is, data gathered in real time from behavior in the real world rather than from a survey (which gathers data after the fact) or an experiment (which can be artificial). The data Silverman and his co-authors use is from Check, a mobile bill pay and personal finance app. The data had been de-identified by Check before it was turned over to the researchers, and all Check app users whose data was utilized had granted permission first. “We never had access to data with personally identifiable information,” Silverman said. As the authors describe the data set, “Check can link almost any financial account to the app, including bank accounts, credit cards, utility bills and more. The application logs into the web portals for these accounts daily and obtains the user’s primary financial data. The data are organized so users can obtain a comprehensive view of their financial situation.” In their first published research using this data, Silverman and his co-authors tackled a mysterious discrepancy between economic theory and what other researchers had perceived about people’s real-world behavior. Do people consume more on payday than between paychecks? According to economic theory, the answer to that question should be “no.” Silverman explains, “Most theories say you want to keep consumption of goods — coffee, or movies, or fast food or whatever — smooth, otherwise you’re missing out on opportunities to improve your welfare.” So, for example, rather than going to the movies several nights in a row at the beginning of the month when the bank account is flush, economic theory says to spread that spending out over the whole course of the month. That seems perfectly sensible — over the course of two weeks or a month, the timing of a person’s consumption should be relatively independent of the timing of that person’s income. The problem, Silverman explains, is that researchers using other data sets have found that a person’s spending isn’t smooth across the time period; spending does spike significantly at the point that income comes in. Some researchers have interpreted that finding to mean that some people are simply really bad at managing their money, so they tend only to have a latte or go to the movies near a payday. Other researchers suggest that the theory is bad — disconnected from what happens in the real world. “Or maybe,” Silverman suggests, “it’s the data.” The power of “naturally-occurring data” Data that is generated in the course of a person’s everyday behavior is particularly powerful because of its level of detail and its accuracy. Silverman explains, “With this data, we can see much better how much people are spending and on what. What’s so unprecedented is the ability to see accurately not only the amount being spent but also the income coming in, and to see them simultaneously. That level of granularity and timeliness is almost impossible to get from a survey.” Tapping into an application like Check that is already capturing the data, researchers can get much more data, much less expensively, which allows them to answer questions they couldn’t answer before. Clarity: Spending vs. consumption Using this better data, Silverman and his co-authors were able to see that spending does indeed spike upon the arrival of income that is well anticipated in both amount and in timing (such as a Social Security check or paycheck). “When you look at a person’s total spending — including rent or mortgage, childcare, auto payment, utilities, tuition, etc. combined with groceries, gas, entertainment, etc. — all together we see a 70 percent uptick in spending the day the income arrives.” But — here’s what Silverman and his co-authors were able to discover from this new data — when you take out “regular” spending like rent or mortgage, childcare, auto payment, utilities, tuition, etc., then there is only a small uptick in spending at the time that the income comes in. Silverman explains, “When we take out the sizable spending that’s predictable both in the timing and the amount and then examine the sensitivity of the remaining expenditures to the arrival of income, there’s a dramatically smaller response. It’s still there, but it’s a relatively small response.” Silverman and his co-authors went one step further, separating types of consumption. Isolating spending on fast food and coffee shops — purchases that are consumed immediately (most people don’t drink half a latte and save the rest for next week) and imminently discretionary (no one “needs” a latte or a hamburger) — Silverman and his co-authors find that type of consumption spending is smooth over time. It doesn’t spike at all in response to anticipated income. “So this phenomenon that other researchers had perceived — which we were able to see in more detail — doesn’t look at odds with economic theory after all,” Silverman says. The nuance is that “regular” spending on things that are paid once but consumed over the course of a week or a month might be timed to coincide with the receipt of income. You schedule your mortgage payment to come out on or shortly after payday. Same with other big, predictable expenditures that can most often be scheduled. That timing of regular spending is not counter to economic theory. Economic theory talks not about “regular” spending but about true consumption — spending on things like coffee and fast food that are consumed immediately. Economic theory says you shouldn’t drink all your lattes the day after payday and then none the rest of the week. And Silverman’s findings demonstrate that people don’t in fact, do that — their behavior is consistent with economic theory. Maybe it’s not you; maybe it’s the credit market In addition to clarifying the mysterious discrepancy between what previous researchers had perceived about the timing of income and spending and what economic theory suggests, Silverman and his co-authors also find that when there are spikes even in “non-regular” spending (groceries, clothes, entertainment), the people who exhibit those spikes are liquidity constrained. They’re the ones truly living paycheck to paycheck. But what is the cause of that liquidity constraint? “Is it a problem with credit markets or a problem with decision-making?” Silverman asks. “We can’t say for certain based on this research, but it is plausible that liquidity constraints are not caused by poor decision-making, but rather a problem of access to liquidity.” That, in fact, is one of the broader implications of this research. “Our findings should turn the emphasis of research and policy more toward imperfections in markets and less on imperfections in decision-making,” Silverman explains. “Our findings suggest that what other researchers might have perceived as really important problems with budgeting and financial management may not be. Where we see deviations from what economic theory expects it is due to insufficient liquidity and perhaps a problem not of the individual's doing but of the credit market.” “Our findings give us more confidence that people are actually pretty good at managing their money. They’re not behaving in ways that run counter to economic theory, after all.” More exciting research to come Silverman is excited about the research he can do with this new data. First, he says, he’s going to examine with great accuracy how well people manage credit cards and how consequential mismanagement can be. He’s also planning to do some forensic economics. “We’ll look at a cross section of people who had real financial problems and then look into the past to see if there are certain common decisions that led to those problems. That could allow an application like Check to help people preempt financial mistakes.” On the topic of Big Data, Silverman says, “It’s an open question how useful these data are, and for whom. What we’ve done with this paper is show that you can take raw Big Data and use it to conduct productive research.” First published in the W. P. Carey magazine, Autumn, 2014.

Economics