This story was originally published on June 17, 2020, and some details may be out of date. For Vox’s latest coverage on the pandemic, visit our coronavirus hub.
The coronavirus pandemic ripped through the American economy at an incredibly rapid pace — so rapidly that it’s been difficult for economists and others to understand what exactly is going on.
Our best data sources about the economy are wildly out of date: Unemployment data comes out just once a month, and GDP data only four times a year. However, a new data source put together by a research group at Harvard, drawing on a variety of corporations’ private data, now allows economists to track what has happened to the economy in real time.
The data they collated shows that the economic crash has been driven disproportionately by the actions of high-income Americans, whose consumer spending has crashed more than that of poorer Americans, devastating low-income workers and small businesses in rich areas.
The data also suggests that economic relief measures have done little for small businesses: Stimulus spending tended to go to Amazon or Walmart, not small local stores, and small businesses eligible for Paycheck Protection Program (PPP) loans are generally not any better off than ones that were not eligible.
And researchers who developed the data found official orders “reopening” states do not increase economic activity, and so appear to endanger public health without any economic benefit.
The picture that emerges in a new working paper based on the economists’ findings is of an economy frozen in place. Simply declaring the economy “reopened” does not seem to do anything to spur high-income people to spend more, and it’s not clear that anything can until the real threat passes.
The tool, the Opportunity Insights Economic Tracker, was launched by the Harvard-based Opportunity Insights group. The research was led by Raj Chetty, Nathaniel Hendren, John N. Friedman, and Michael Stepner, and the tool was assembled by a team of 39 collaborators. It aims to provide a service that has never existed before, but is badly needed during the pandemic: a real-time, day-by-day, ZIP-code-by-ZIP-code snapshot of how the American economy is performing.
The tool works by collating administrative data from a variety of corporations, including Affinity Solutions (a company tracking consumer spending), Burning Glass (a job market analytics company), Earnin (an app that offers advance loans on paychecks), and HomeBase (which sells punch-card and work check-in software), among several others.
The data Opportunity Insights gathers from these companies is anonymized; the Economic Tracker can’t track individual people for privacy reasons. But the data from these companies matches gold-standard government surveys to a remarkable degree. Affinity captures about 10 percent of credit and debit card transactions in the US, and its records are biased toward goods that tend to be purchased with cards rather than cash.
Nonetheless, Chetty and his co-authors found that Affinity’s data aligned almost perfectly with the Monthly Retail Trade Survey conducted by the Census Bureau, which serves as the basis for the Bureau of Economic Analysis’s official GDP numbers.
The blue lines below are consumer spending as measured by Affinity, for all consumer spending (on the left) and just food services (on the right). The green lines are the MRTS. If you have a hard time telling the lines apart, that’s the point:
Similarly, using data from ADP, a large payroll processor, alongside Earnin and HomeBase data provides a decent approximation of the Current Employment Statistics survey used by the Bureau of Labor Statistics. It’s by no means perfect, and the authors are clear that its imprecisions may be more important in normal times (when jumping between 4 and 6 percent unemployment is a huge deal) than now (when the difference between 14 and 16 percent unemployment feels rather less meaningful).
But the data nonetheless allows for striking analyses that you would typically only see conducted many years after a crisis like this.
You can, for instance, compare how small business revenue has fallen between neighborhoods. Here’s what DC looks like, for instance. Businesses in the wealthy central business area have seen huge declines, while poorer ZIP codes like 20020 (which contains the historic but poor Anacostia neighborhood) and 20011 (which contains both poor and gentrifying neighborhoods in the northern part of the city) have seen business revenue actually increase:
You can look up small business revenue in your own area using the interactive below, best viewed on a larger screen:
And you can do the same for employment data here:
The development of the OI Economic Tracker in some ways parallels the creation of “national accounts” data like GDP statistics. That effort began in earnest in the United States in the early 1930s at the National Bureau of Economic Research in Cambridge, Massachusetts, under the supervision of Penn professor and future Nobel laureate Simon Kuznets. Eventually, Kuznets’s methods were adopted by the federal government and they form the basis of our official economic statistics today.
“In the Great Depression, they wanted to measure things with more regularity and Kuznets decided to take this on,” Chetty told me. “That’s the basis of what you see in the Bureau of Economic Analysis data. Going forward, this is the basis of how you could do this without surveys.”
Chetty tells me OI is in talks with Intuit, whose TurboTax and Mint programs are home to vast reams of individual financial data, and Mastercard to provide even more sources on consumer spending and incomes. He hopes eventually that a tool like this, with bigger and more representative data sources, could become a province of the government, just as Kuznets’s idea of regularly measuring the total national income of the United States went from a private academic endeavor to the official business of the federal government.
In the meantime, the OI Economic tracker can give us important insights into which policies are and are not working during the Covid-19 recovery. The paper by Chetty, Friedman, Hendren, and Stepner suggests that small business loans and stimulus checks haven’t been enough to keep small businesses afloat, especially in affluent areas. Indeed, it suggests that any kind of recovery is likely impossible until the pandemic is over. What’s needed is income support in the form of unemployment insurance or some other program for people put out of work who are struggling to afford food and rent. Only once the pandemic is well and truly behind us is a regular recovery possible.
Any one of the paper’s five key findings could justify an individual research paper; this relatively short paper covers all five, and more.
1) The high-income recession
Spending has collapsed dramatically since the Covid-19 crisis began. But it didn’t fall evenly. The data in the OI Economic Tracker indicates that by May 31, 66 percent of the fall in credit card expenditures since January was concentrated in the top 25 percent of households by income. The bottom quartile, by contrast, was basically back to precrisis spending patterns by the end of May:
High-income people spend more overall than low-income people, as you’d expect, including more on in-person services. Those factors combined with a higher percentage-wise drop in overall spending meant that the large majority of the drop in spending is attributable to the wealthiest segment of the population.
These spending changes seem to be correlated varying levels of Covid-19 infection between ZIP codes, as well. “Spending fell sharply on March 15, when the National Emergency was declared and the threat of COVID became widely discussed in the United States,” the authors find.
Where people changed their spending was heavily influenced by restrictions on in-person interaction.
“Nearly three-fourths of the reduction in spending is accounted for by reduced spending on goods or services that require in-person physical interaction (and thereby carry a risk of COVID infection), such as hotels, transportation, and food services,” the authors find, despite the fact that these categories only made up one-third of consumer spending before the crash. “pending on luxury goods such as installation of home pools and landscaping services — which do not require physical contact — increased slightly after the COVID shock.”
Similarly, ZIP codes with higher rates of Covid-19 infection saw sharper reductions in spending. The mechanism here is simple: People in areas with higher Covid-19 caseloads spent less time outside (as verified by Google cellphone data), and this translated into less spending on in-person services. What’s more, the authors find that regardless of the level of Covid-19 infection in an area, high-income people spent less time outside than lower-income people.
There are a number of reasons why this might be; high-income people are likelier to have jobs where they can work from home and have larger living spaces that they can enjoy. But this all combines to help explain why spending fell so much among higher-income households, in particular.
2) Those who serve the rich are suffering most
Housing in the United States is heavily segregated by income: There are rich neighborhoods and poor neighborhoods of cities, rich suburbs and poor suburbs, more and less affluent rural communities. So it stands to reason that people providing in-person services in richer areas, like baristas or waitstaff or hotel cleaners, might have suffered more than their counterparts in poorer areas, given the sharper reduction in spending by the rich.
So the authors look at data from Womply, a business software company that offers credit card transaction tracking for small businesses. Sure enough, the biggest losses are recorded in the richest neighborhoods of major cities:
Small businesses lost 73% of their revenue in the Upper East Side in New York, compared with 14% in the East Bronx; 67% in Lincoln Park vs. 38% in Bronzeville on the South Side of Chicago; and 88% in Nob Hill vs. 37% in Bayview in San Francisco. Revenue losses are also large in the central business districts in each city (lower Manhattan, the Loop in Chicago, the Financial District in San Francisco), likely a direct consequence of the fact that many workers who used to work in these areas are now working remotely. But even within predominantly residential areas, businesses located in more affluent neighborhoods suffered much larger revenue losses.
Overall, 55 percent of small businesses in ZIP codes with the highest apartment rents closed, compared to 40 percent of small businesses in the lowest-rent ZIP codes. The shock to these businesses was tremendous: Not only did they face higher fixed costs to begin with due to high rents, but the greater collapse in spending by the rich pummeled their revenue more.
This flows through to service workers employed in these high-rent, high-income areas (but who likely live in poorer, lower-rent neighborhoods that they can afford). Hours worked fell by more than 80 percent in the richest ZIP codes of New York, San Francisco, and Chicago, and only 30 percent in the poorest ZIP codes of those cities.
A similar pattern emerges for outright job losses: In the data from Earnin, the paycheck advance company, 36 percent of job losses are in ZIP codes in the top quartile by rent, and 11 percent in ZIP codes that fall in the bottom quartile by rent. Same goes for job postings for lower-skilled workers in high versus low-rent areas. Tellingly, no such pattern in job postings emerges for workers with college degrees, suggesting that the pain here is concentrated among lower-income people who work in rich neighborhoods.
This then flows through to consumer spending. The charts below look exclusively at low-income ZIP codes, and compare ZIP codes where more workers work in high-rent, affluent areas to ones where fewer workers work in rich areas. The more people in the ZIP code worked in rich areas, the worse off they were, both in terms of hours worked and consumer spending:
For more on this phenomenon, see Emily Badger and Alicia Parlapiano’s excellent piece in the New York Times based on the OI Economic Tracker data. They talked to service workers in NYC and DC who worked in higher-income neighborhoods and have seen hours and tips dry up.
3) Stimulus checks kept people afloat, but not small businesses
One interesting aspect of this downturn is that while unemployment has skyrocketed, individual incomes have too. In April, personal income (defined as the money Americans receive from wages, government benefits, investments, and so on) grew by 10.5 percent, by far the highest monthly growth rate in the metric’s 60-year history, even as unemployment shot up from 4.4 percent to 14.7 percent that same month.
This is largely due to the CARES Act, Congress’s relief measure that included both a $1,200 per person stimulus check to most Americans and a super-sized unemployment benefit package that boosted UI benefits by $600 a week.
Chetty, Friedman, Hendren, and Stepner are able to see what effect the stimulus checks specifically had because the Earnin data indicates that the large majority of people (over 70 percent) got their $1,200 from the IRS on April 15, exactly; a small minority arrived on April 14 as well. That allows the authors to test how the stimulus affected households by comparing spending on April 13 to April 15. This is a variant on what’s called a “regression discontinuity” approach in social sciences, and it’s one of the higher-quality tools we have for testing what effects a policy actually caused, as opposed to what happened around the same time.
Sure enough, spending jumped modestly for high-income households (by 9 percentage points) and enormously for low-income households (by 26 percentage points) over the two days that the stimulus package was implemented.
The blue line above is the poorest 25 percent of Americans by income, the green line the richest. Both saw spending fall sharply upon the onset of the crisis, but almost instantaneously the stimulus caused spending to bounce back. It just bounced back much, much more for the poorest Americans, nearly back to precrisis levels.
But looking at how people spent their stimulus checks reveals some limitations of the policy’s ability to boost the economy. Spending on in-person services only rose 7 percentage points, whereas spending on durable goods (furniture, cars, TVs, computers, etc.) grew a startling 21 percentage points, accounting for nearly half the recovery in spending.
What’s more, the authors find little effect of the stimulus on small business revenues, and absolutely no effect on small business employment. Small businesses tend disproportionately to sell goods and services in-person, and a disproportionate share of their revenue comes from wealthy people. So the stimulus, which was both less meaningful for wealthy people and came at a time when people were understandably afraid to leave their homes, did little to help them.
4) Paycheck Protection didn’t do much
It’s not necessarily an indictment of the stimulus package that it didn’t help small businesses. It wasn’t meant to — it was meant to help poor Americans, especially, get by and survive amid this calamity. And at that job, it excelled.
The program meant to help small businesses was the Paycheck Protection Program (PPP), which offers forgivable loans to small businesses that keep the majority of their staff on payroll. But Chetty, Friedman, Hendren, and Stepner find little evidence that this program was effective.
PPP is administered to Small Business Administration (SBA)-eligible businesses, which with a few minor exceptions means businesses with 500 or fewer employees. So the authors are able to compare businesses with total employee counts just below and just above these thresholds, who likely closely resembled each other with the exception of one group being eligible for PPP loans, and the other not.
Here is how hours worked at small businesses evolved, at different levels of small business size:
In the immortal words of Pam Beesly: They’re the same picture.
Employers with around 1,500 employees reduced hours ever so slightly more by the end of May than businesses with fewer employees, but the difference is minuscule, and goes away entirely if you limit the analysis to food and hospitality businesses that PPP was particularly designed to help. “We conclude that the PPP had no meaningful effect on unemployment at small businesses, at least as measured through the middle of May,” the authors write.
5) Reopening orders don’t actually reopen the economy
So if stimulus checks and PPP loans aren’t enough to keep employment at small businesses afloat, maybe declaring the crisis over could? That’s the intuition behind politicians nationwide, led by President Donald Trump who are pushing for a “reopening,” and have succeeded in rolling back stay-at-home orders in a number of states.
But this only works if the order to reopen actually spurs consumers, and in particular the wealthy consumers who’ve cut back spending the most, to go out and buy the in-person goods and services they’ve been shunning. The OI Economic Tracker data let the authors test precisely if that’s happening.
First the authors compared Minnesota (which did an early partial reopening on April 27) to Wisconsin (which did so on May 13 in response to a court order). As you can see, despite very different reopening timing, the trajectories of the two states was nearly identical when it comes to consumer spending, and the orders themselves didn’t seem to do much.
Then the authors expanded this analysis to a total of 20 states that issued reopening orders prior to May 4; for each reopening date, they paired these states with control states that didn’t reopen, and that prior to reopening had a similar trajectory in employment or consumer spending. The authors find that consumer spending was growing before formal reopening and kept growing after; it’s possible the reopening caused some of the boost. But in any case, employment did not rise at all in the reopened states.
“The implication of this finding is that restoring confidence in public health may be a prerequisite to a full recovery,” the authors conclude. That was true in 1919 during the Spanish flu, and it seems to be true again today.