we study the mechanisms through which COVID-19 affected the economy by analyzing heterogeneity in its impacts. We first show that high-income individuals reduced spending sharply in mid-March 2020, particularly in areas with high rates of COVID-19 infection and in sectors that require in-person interaction. This reduction in spending greatly reduced the revenues of businesses that cater to high-income households in person, notably small businesses in affluent ZIP codes. These businesses laid off many of their employees, leading to widespread job losses especially among low-wage workers in affluent areas. High-wage workers experienced a “V-shaped” recession that lasted a few weeks in terms of employment loss, whereas low-wage workers experienced much larger job losses that persisted for several months. Building on this diagnostic analysis, we use event study designs to estimate the causal effects of policies aimed at mitigating the adverse impacts of COVID-19. State-ordered reopenings of economies have small impacts on spending and employment. Stimulus payments to low-income households increased consumer spending sharply, but little of this increased spending flowed to businesses most affected by the COVID-19 shock, dampening its impacts on employment. Paycheck Protection Program loans increased employment at small businesses by only 2%, implying a cost of $300,000 per job saved. These results suggest that traditional macroeconomic tools – stimulating aggregate demand or providing liquidity to businesses – have diminished capacity to restore employment when consumer spending is constrained by health concerns
Pointer from Tyler Cowen.
Arnold, I have yet to read the paper. Indeed, it’s very important that the Chetty team has been able to build such a large database. Reading the abstract, I’m surprised that in the study of the first stage of the pandemic (what they call the diagnostic analysis), they don’t address the relative importance of state-ordered closings of their economies in residents’ responses. The results they present for the second state, the one characterized by state-ordered openings, may be determined by the relative importance of the closings. Also, at least in the abstract, for the first stage they mention only some counties (the ones where high-income residents were affected, and I assume the results are conditioned by NY/NJ/Conn), but in the second stage it seems they refer to a different set of counties. As I said, I have yet to read the paper.
Arnold, I have not been able to go into the details of the paper and I don’t expect to do it in the next few days. I hope you and all your readers at least read carefully the introductory section I (it’s a good summary of sections III and IV on COVID-19) and the conclusion (except for the first 3 paragraphs, the others refer to issues not discussed in the text). Section II and most of the conclusion is about the database built by the Chetty team.
In sections III and IV they use the new database to complement the macroeconomic analysis based on national accounts. Their Keynesian approach means that they start with the assumption that (nominal and real) GDP is determined by aggregate demand and in the first 3 months from mid-March the collapse of GDP is “explained” by a sharp decline in consumption (in the first month) and slow, partial recovery. Taking advantage of their database they claim that this decline was largely due to high income (often called “rich”) people that suddenly “developed” health concerns (I’d say that they have a very biased view of the health concerns of the rest of the population). Since the composition of their consumption differs from that of the rest of the population one has to pay special attention to goods and services that are demanded mainly (only) by the rich people. Their detailed data allow them to argue that rich people’s consumption showed a V pattern: it recovered sharply after mid-April, but other people’s consumption didn’t recover enough to return to previous levels. They argue that this slow, partial recovery was due to changes in rich people’s behavior and the response of their suppliers. Note: close to the end of page 3 they say “Together, these results suggest that consumer spending in the pandemic fell because of changes in firms’ ability to supply certain goods (e.g., restaurant meals that carry no health risk) rather than because of a reduction in purchasing power” which in my view openly contradicts their Keynesian assumption and therefore it deserves further scrutiny.
The rich people’s sudden health concerns were shown in their database before the state-ordered closings and have been the driving force of economic performance since mid-March. Thus, they argue that “State-ordered shutdowns and reopenings of economies had modest impacts on economic activity” (their words on page 5). Their argument starts with a review of Colorado and New Mexico and the results are then extended by relying on a statistical method to all states (see section IV). Given the large differences in the timing of shutdowns and reopenings and the short period under consideration, their argument would require a detailed analysis. It’s surprising, however, that in section IV there are no references to NY state and city, and no context-specific analysis of their distress in the 3 months analyzed in the paper.
Yes, Arnold, the “Macro” struggle continues.
“State-ordered reopenings of economies have small impacts on spending and employment.”
The paper covers the period of 03/15/20-06/15/20 and is therefore missing roughly 2.5 months worth of data to bring us to the present. Note: the “crazy” and “irresponsible” states like Georgia and Texas didn’t start re-opening until early/mid May, so this paper is at best picking up only 30-45 days worth of data. At best, this provides a very incomplete picture.
Anecdote: traffic jams are about 80-90% of normal here in North Texas as of today. I definitely prefer that to over what I’m hearing about traffic in California.
Is this potentially more intuitive and reliable? I’m not an economist, so I’ve got no clue. However, seems far more creative than other approaches that I’ve seen.
“We harness cell device signal data to examine the effects of the timing and pace of reopening plans in different states.”
https://www.nber.org/papers/w27235
Arnold, the abstract, and your description of the paper, make the paper appear to me an attempt the minimize the impact of the government-ordered shutdowns while, at the same time, minimizing the impact of the government-allowed business reopenings. Indeed, I find the use of the word “ordered” in the second case to be completely inappropriate.
And, even at that, how much time in data do the authors have after many states in the south and center of the country allowed businesses to reopen? Couldn’t possibly be more than a month’s worth of data- here in Tennessee, we didn’t see wide-spread reopenings allowed here until mid-May. I don’t think Chetty etal could have more than data from early June at this point in time.
From the WSJ today…
New York City to Resume Indoor Dining at Restaurants
“Indoor service can resume Sept. 30 with capacity capped at 25%, following six-month ban aimed at curbing pandemic”
This is Texas about 3.5 months ago…in Chetty’s world there shouldn’t be a quantifiable difference b/w TX and NY as of today…does this even make sense? Why even bother to open up if it doesn’t make a difference?
https://www.wsj.com/articles/new-york-city-to-resume-indoor-dining-at-restaurants-11599672105?st=fl929l75rcqcbu6&reflink=article_copyURL_share
And there definitely is a quantifiable difference between NY and TX today.
July unemployment for NY was 15.9%, second worst state in the union. TX was just 8.0%.
Interestingly enough, Utah has 4.5% unemployment with only 426 deaths so far, with some of the lowest per capita death rates in the US. Has anyone heard what Utah did for the pandemic? Or did it mostly benefit from low population density and a relatively youthful population?
https://www.bls.gov/web/laus/laumstrk.htm
F*ck! Thank you!
I’m just a silly undereducated MAGA conservative vs. MIT/Harvard economists, so please correct me in the comments if I’ve gotten something wrong.
But, this paper make no sense whatsoever from a common sense perspective. As far as I can tell, the basic conclusion is: “no need to open up because it won’t make a difference.”
Perhaps part of the explanation for the low Utah death rate is “clean living”.
Nah, too uncool.
I think the argument is that people’s fear of the virus is what cripples the economy less than the lockdowns themselves, which are mostly telling people not to do what they weren’t going to do anyway, and they mostly started not doing those things before the lockdowns and kept not doing them after.
The difference between NY and TX would be explained as differences in reactions to the virus by people rather than even the current threat posed by it. NY had a much more traumatic experience than anything Texas has seen yet, so even if NY is safer than TX right now lingering paranoia still dominates people’s behavior.
I think this is what Scott Sumner has been arguing as well: lockdowns are largely endogenous, and people underreact to the virus at first then eventually overreact.
But the argument has no support, Mark, and like it or not, states like Texas and Florida are the examples that disprove it. I note that you didn’t even bother to mention California.
Florida’s unemployment rate is actually pretty high. Ohio and Michigan’s unemployment rates are only slightly higher than Texas’s despite having much more stringent lockdowns (and that probably understates it since their economies are already anemic compared to Texas). California is indeed doing worse economically than it ‘should’ be regarding how bad the epidemic has been, but overall if you were to regress unemployment rate on 1) deaths per 1m, and 2) intensiveness of lockdown etc., I don’t think it’s clear the coefficient on 2 would be greater than the coefficient on 1.
There appears to be a discrepancy between the blurb excerpt and the paper at the link. The latter says “Paycheck Protection Program loans increased employment at small businesses by only 3%, implying a cost of $290,000 per job saved.” The former has 2% and $300,000.
But this appears to assume that the loans will all be forgiven. Scanning the 108 page paper briefly The SBA, however, will forgive the loans only if all employee retention criteria are met, and the funds are used for eligible expenses. SBA did not begin accepting loan forgiveness applications until August 10. Moreover, the PPP Flexibility Act of 2020 lets borrowers extend the period that they have to disburse the borrowed amount up to 24 weeks or Dec. 31, 2020. The program didn’t expire until August 8. A bill introduced to offer blanket loan forgiveness has not moved apparently due to fraud concerns.
The SBA itself reports that through June 30th (Chetty may be using employment data up to July 15 but loan data only up to June 20, the paper isn’t very clear.) had made net loans of $521 billion supporting 51 million jobs reported putting net loans to jobs at about $10,200 per job.
https://www.sba.gov/sites/default/files/2020-07/PPP%20Results%20-%20Sunday%20FINAL.pdf
Chetty gets the $300,000 per job shocker figure by coming up with a projection of what employment would have been like absent the program. Interestingly, no attention is given to an alternative estimate on page 41: “Even at the upper bound of the 95% confidence interval for employment impact, we estimate a cost per job saved of $163,000. If we assume the treatment effect of the PPP program on food services was the same in percentage terms, then we estimate a total of 2.1 million jobs saved by the PPP.”
At June 30th the SBA reported that it still had $132 billion left. Apparently all loan data through August 20 can now be downloaded but I haven’t found any actual analysis of this data online. There is an August 8 report at Treasury but it doesn’t include a jobs supported figure.
At any rate, Chetty’s pronouncements seem premature, and we must await a final word. When complete data on the program are available, it will be interesting to see if anyone revisits his paper. The rush to promote the paper’s headline conclusions is perhaps as informative as the paper itself.
There’s another severe problem: the assumption that small businesses and large businesses would have been equally impacted by COVID in the absence of PPP.
“Employment rates at firms with less than 500
employees (who were eligible for PPP assistance) increased only slightly – by about 3 percentage
points – relative to larger firms that were ineligible for PPP when the PPP program began.”
If relative employment at small firms would have fallen 20% in the absence of PPP, rather than increased by 3%, that would point to a much high cost efficiency of the program.
It’s a counterfactual that we cannot test, but it seems highly unlikely to me that large corporations and small businesses were equally well equipped to navigate the pandemic economy.
Excellent point, thank you.
Sorry to comment again.
“traditional macroeconomic tools – stimulating aggregate demand or providing liquidity to businesses – have diminished capacity to restore employment when consumer spending is constrained by health concerns”
Yes, if people are not buying, employers need fewer employees. Sounds reasonable. Yet doesn’t answer the “Krugman ‘the stimulus was too small’ hypothesis” from a previous recession. Instead of $1,200, would $3,600 have worked? Please, no, but still.
And what about other countries whose macro- interventions appear to have succeeded. Switzerland, for example, paid a portion of furloughed employees’s salaries and now has a 3.4 percent unemployment rate and has ridden out the crisis with less economic loss than most other European countries.
But of course on a scale of 1 to 10 on a government lockdown hysteria scale, 1 being laissez faire and 10 being New York, Switzerland was about a 2 or 3. Looking at zip codes by income, as Chetty does, one might expect scale of lockdown hysteria to have some explanatory power given that there are disproportionately so many high income counties in high hysteria states like New York, New Jersey, California, Maryland and Virginia.
How did low-income workers fare in low versus high hysteria average income states? For example, low hysteria as of Sioux Falls, SD (median household income $61,915 in 2017, population 189,000, 2,323.91 people per sq mi) had a 6% unemployment as of July 2020. And high hysteria Tom’s River, NJ (median family income $76,167 in 2017, population 94,000, 2,253.5 people per sq mi)
had 16.4 percent unemployment as of July 2020.
WalletHub did a virus hysteria scale that you can scan to get a rough idea of the unemployment/income/hysteria connection:
https://wallethub.com/edu/most-aggressive-states-against-coronavirus/72307/
And of course, more hysterical responses are not correlated with any improvement in disease outcomes. Sioux Falls has had 71 deaths, Tom’s River 150.
Again, it appears way to early to be jumping to any conclusions, particularly this paper.