Consumer spending during unemployment: evidence from US bank account data

Main article

How does consumer spending evolve during a spell of unemployment? This column reports research showing that unemployed people’s consumption drops sharply at the large and predictable fall in income that happens when their unemployment insurance benefits expire, typically after six months in the United States. The findings, based on analysis of data from nearly 200,000 bank accounts, suggest that a behavioral model, in which many people live ‘hand-to-mouth’, does a better job of explaining their spending while unemployed than a rational model. 

Nearly all of us experience unemployment at some point in our careers. For example, the Bureau of Labor Statistics finds that 90% of baby boomers have been out of work at least once in their lives.

Unemployment is stressful in part because many people do not have enough savings to maintain their standard of living after a drop in income. To help people manage this risk, developed economies provide most workers with unemployment insurance (UI), which replaces a fraction of their wage while they look for a new job.

In nearly all countries, UI is temporary and expires after a number of months. In the United States, benefits last six months under normal labor market conditions. This reflects the fact that most unemployment spells are temporary: when labor market conditions are normal, the median unemployment spell in the United States is about three months.

But during a recession – when it is much harder to find a job – unemployment lasts much longer. Policy-makers usually respond by extending the duration of benefits. In the Great Recession, for example, benefits were extended to as long as 23 months, and the appropriate duration of benefits became a hotly debated policy topic.

The appropriate duration of benefits depends in large part on two forces: the ‘consumption-smoothing’ value of extending benefits (enabling people’s month-to-month spending to be stable even when their incomes are volatile); and the extent to which benefits discourage job search. A vast number of economic studies, as well as the policy debate on benefits, focus on whether and how much job search is discouraged. Yet we know far less about the consumption-smoothing effects of extending UI.

How does consumer spending evolve during a spell of unemployment?

Why is it hard to measure the consumption-smoothing benefits of UI? Studies of spending during unemployment in the United States prior to ours largely rely on surveys of a random sample of the population. These surveys have two limitations.

First, although the experience of unemployment is ubiquitous, at any given time, only a small fraction of US workers are unemployed and it is challenging to conduct a survey only of unemployed households. Thus, analysis of unemployment through the lens of survey data yields small samples. For example, a prominent study by Jonathan Gruber (1997), which uses survey data, analyzes a sample of 1,600 households.

Second, it is hard to survey people repeatedly, and so survey datasets usually capture consumption just once during unemployment.

Our study (Ganong and Noel, 2019) uses bank account records to examine the role of UI in consumer spending decisions. In what follows, we describe the data, present our empirical findings, and then discuss two lessons that arise from those findings.

Evidence from bank account and credit card data

We worked with the JPMorgan Chase Institute (JPMCI) to construct a ‘de-identified’ panel dataset that tracks monthly income and spending for a large sample of UI recipients. The dataset is based on the universe of Chase consumer checking and credit card accounts, aggregated for households to the monthly level.

Spending is measured from debit and credit card transactions, cash withdrawals, and electronic transactions captured through the bank accounts. The JPMCI dataset measures income and spending for about 20 million households each month. A household is classified as receiving UI benefits if it receives income from the state’s UI agency via direct deposit.

Our analysis sample includes 182,000 households who received UI benefits in 20 US states between January 2014 and June 2016. Thus, our dataset is about 100 times larger than previously available US data, and we observe spending every month, rather than just once during unemployment.

Experiences of the typical benefit recipient

Can bank account and credit card data teach us about the experience of the typical UI recipient? Since UI recipients have substantial prior labor force attachment and 90% of UI recipients have a bank account, our sample is likely to be representative of UI recipients. We compare the JPMCI sample to nationally representative surveys on several dimensions. On the dimensions that we are able to analyze, the JPMCI sample is representative of UI recipients across the United States.

Figure 1 shows the path of income (labor income plus UI benefits) for someone who is unemployed for one month, two months, etc. Income drops at the start of unemployment and recovers when the unemployed person returns to work.

But not all unemployed people return to work quickly. Some see a second drop in income when their UI benefits expire. In most US states, benefits expire after six months. Thus, for people who do not find a job within six months, there is a second, even larger drop in income.

Our central empirical finding is that spending closely tracks UI benefits, such that benefit exhaustion causes serious hardship. Figure 1 shows that the path of month-by-month spending is a mirror image of the path of month-by-month income. This figure relies crucially on the rich data described above. It is only possible to construct this figure because JPMCI has monthly data for a large sample of UI recipients.

Sharp drops in spending when unemployment insurance benefits expire

The largest drop in spending occurs after six months, when UI benefits expire. We measure 22 different types of spending; all 22 drop when benefits expire.

The drops are particularly severe for goods that are necessities. For example, grocery spending drops by 16%, medical out-of-pocket spending drops by 14%, and drug store purchases drop by 15%. This indicates that people may not be able to purchase the food that they need and may skip medical visits after UI benefits are exhausted.

But is this drop caused by the exhaustion of UI benefits? Perhaps something other than benefit levels (for example, discrimination against the long-term unemployed) also changes after six months of unemployment.

To disentangle correlation from causation, we use state-level variation in UI benefit policy. We compare spending in states where benefits last six months to spending in Florida, which is unique in having very low UI benefit levels and also a shorter maximum duration of benefits. We analyze a time period where benefits in Florida last for four months. Figure 2 shows that spending in Florida drops after four months, exactly when benefits expire. Thus, we conclude that benefit exhaustion causes substantial hardship.

The sharp drop in spending at benefit exhaustion teaches us important lessons about how people make consumption decisions (lesson #1) and about how to design UI benefits (lesson #2).

Behavioral explanations of people’s consumption decisions

The first lesson is that many people who are unemployed set their spending in a way that is inconsistent with a rational economic model, in which consumers plan spending to account for predictable increases or drops in income. Rather, the drop in spending is consistent with a behavioral economics model in which people are ‘present-biased’, as discussed in Laibson (1997), or are myopic for some other reason: they discount the future and spend based on their income today.

Benefit exhaustion offers a unique way to test between a rational model and a behavioral model because it is a predictable decrease in income. The drop is predictable because benefits expire after six months. Although there is always uncertainty associated with when someone will find a job, someone who has already been unemployed for five months knows that there is a very high probability (about 75% for our sample) that benefits will run out before they find a job.

What should someone who is unemployed do if she knows that benefits will most likely run out next month? This depends on how much people plan for bad events in the future. The rational model says that she should cut her spending immediately. If she spends less today, she will have more saved in her bank account for the likely hardship coming next month.

In contrast, in many behavioral models, someone will not cut her spending until the (predictable) income drop arrives. This spending is sometimes called ‘hand-to-mouth’ consumption behavior. Thus, spending in the months before UI benefits expire teaches us about the extent to which people prepare for major economic risks.

Our results are consistent with the behavioral model (and inconsistent with the rational model). Spending is nearly constant in the months leading up to benefit exhaustion and then drop sharply after benefits do run out. Figure 3 shows that the rational model predicts that spending will decline gradually in the months leading up to benefit exhaustion. This is not what we see in the data.

A behavioral model where some people are effectively hand-to-mouth does a better job of explaining spending. Such individuals do not behave rationally when faced with an impending loss of UI benefits, in the sense that they do not save money to cushion themselves when the insurance is about to run out.

Improving the design of unemployment insurance benefits

The second lesson is that UI benefits can be better designed. Economists have long studied how best to set the level and duration of unemployment benefits. One assumption underlying this analysis is a preference for consumption smoothing. That is, economists assume someone would prefer to consume $4,000 a month for two months than to consume $5,000 in one month and $3,000 in the second month. The sharp decline in spending at UI benefit exhaustion suggests that policy-makers can do better.

We consider two hypothetical policy changes: policy A increases the duration of UI benefits for a month; while policy B decreases the level of UI benefits by 1.8% of monthly income. The size of the benefit change in policy B is chosen to have the same cost as policy A. This means that together policy A and policy B are ‘budget-neutral’ – that is, they can be simultaneously implemented with no additional tax revenue.

We find that a budget-neutral policy that extends UI benefits (policy A) while cutting the level of benefits (policy B) may improve welfare. This combination policy may improve welfare because of the preference for consumption smoothing.

There are at least two caveats to this policy recommendation. First, we do not take a stance on what is the appropriate level of total UI expenditure. Instead, we provide a guide for how to redesign benefits given a fixed expenditure level.

Second, the potential welfare improvement from a budget-neutral extension of UI benefits depends on the extent to which job search is discouraged by additional UI benefits. Although there is a welfare improvement at the best estimates of discouragement available in the existing literature, this could change as additional research becomes available.

Conclusions

Overall, the key lessons from our analysis are that UI benefits play a crucial role in stabilizing consumption during unemployment, that the loss of benefits imposes substantial hardship, that at least some people have trouble preparing for economic risks like the expiry of their UI benefits, and that there is scope for policy reform.

These results highlight the urgent need for additional research and policy experimentation to figure out how best to help jobseekers search for high-quality jobs and smooth their consumption while they search.

This article summarizes ‘Consumer Spending During Unemployment: Positive and Normative Implications’ by Peter Ganong and Pascal Noel, published in the American Economic Review in 2019.

Peter Ganong is at the Harris School of Public Policy, University of Chicago. Pascal Noel is at the Booth School of Business, University of Chicago.