Brave et al., Blending data to understand the economic impact of COVID-19 (2022-02-23)

Summary

This Business Economics article, based on a session held at the NABE TEC Conference on November 8, 2021, collects talks by Scott A. Brave, Rebecca Hutchinson, Christopher J. Kurz, Michael Stepner and Ron Jarmin on blending private and public data to track the pandemic economy. The efforts it covers are the Chicago Fed's weekly retail sales estimate, the Census Bureau's work on higher-frequency state-level retail sales data, the Federal Reserve Board's computations of business closures and weekly payrolls, and the Opportunity Insights team's estimates of spending, business revenues and employment by income and ZIP code. Brave opens with the Chicago Fed Advance Retail Trade Summary, an index built by blending economic statistics and Big Data.

Summary drafted by a model from the document's text below and checked by script against that text before publication. It is a navigation aid, not a reading of what the document proves. Where AI is used

Cite this document

“Brave et al., Blending data to understand the economic impact of COVID-19 (2022-02-23).” Original at link.springer.com. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/brave-et-al-blending-data-to-understand-the-economic-impact-of-covid-1-46ca1b6969/, SHA-256 21185f870e78a7d42e21c2075907a01ffb26891090bd34b4b8171009622cc0b0.

Full text

Business Economics
https://doi.org/10.1057/s11369-021-00250-3

ORIGINAL ARTICLE

Blending data to understand the economic impact of COVID‑19
Scott A. Brave1 · Rebecca Hutchinson2 · Christopher J. Kurz3 · Michael Stepner4 · Ron Jarmin2

This is a U.S. government work and not under copyright protection in the U.S.; foreign copyright protection may apply 2022

Abstract
The data demands during the pandemic heightened the need to blend information from numerous sources to get a more timely
and granular picture of economic developments. Ongoing efforts include the Chicago Fed’s weekly retail sales estimate, the
Census Bureau’s work on higher-frequency state-level retail sales data, the Federal Reserve Board’s computations of busi-
ness closures and weekly payrolls, and the academic Opportunity Insights team’s estimates of spending, business revenues
and employment by income and ZIP code.

Keywords Blended data · Pandemic · Private data · Public data · Weekly data

1 Ron Jarmin A lot of people have applied some very creative and
clever ways to blend different data sources together to build
This is an immensely important topic. Blending data to something more than what could have happened if you only
understand the world is a topic in and of itself, and I think had access to one data source. We're going to get started with
especially important during something like the COVID-19 Scott Brave, from the Federal Reserve Bank of Chicago.
pandemic. I speak from experience about what happens to
our data collection infrastructure during those times, with
the Census and other things we do at the Census Bureau. 2 Scott Brave

Today I'd like to talk about a project that we've been working
Based on the session Blending Data to Help Understand the
Economic Effect of COVID19, held at the NABE TEC Conference, on at the Chicago Fed for about the last two years now. It
November 8, 2021. All views expressed are those of the authors, really started when the pandemic started–blending economic
and do not necessarily reflect those of the U.S. Census Bureau, the statistics and Big Data into something that we call the Chi-
Federal Reserve Bank of Chicago, or the Board of Governors of cago Fed Advance Retail Trade Summary.
the Federal Reserve.
For us the COVID-19 pandemic really heightened the
* Christopher J. Kurz need for timely measures of consumer spending. If we think
christopher.j.kurz@frb.gov back to February and March 2020, just how quickly things
Scott A. Brave changed during that period and how the consumer was front
scott.a.brave@chi.frb.org and center at that time. It really highlighted some weak-
Rebecca Hutchinson nesses for us on data collection, and really brought about
Rebecca.J.Hutchinson@census.gov this project.
Michael Stepner At the Chicago Fed we developed a new weekly index for
michael.stepner@utoronto.ca this purpose that combines high-frequency Big Data with the
Ron Jarmin U.S. Census Bureau's Monthly Retail Trade Survey. What
ron.s.jarmin@census.gov I'm going to talk to you today is how this index works, and
1
then some of the applications that we've come up with.
Federal Reserve Bank of Chicago, Chicago, USA
What do we do? We take high-frequency data from five
2
US Census Bureau, Washington, DC, USA private companies and one federal agency to construct a
3
Board of Governors of the Federal Reserve System, weekly measure of retail and food services sales, excluding
Washington, DC, USA
4
University of Toronto, Toronto, ON, Canada

Vol.:(0123456789)
S. A. Brave et al.

automotive spending. It is, most importantly, benchmarked published in the Chicago Fed Insights Blogpost (Brave et
to the Census Bureau's Monthly Retail Trade Survey. al. 2021b).
We use a statistical framework and mixed-frequency Let me just give you a basic sense of how this works,
dynamic-factor model to constrain this latent measure to rather than going into the nuts and bolts of a dynamic-factor
match the latest Census Bureau data, the latest Monthly model. CARTS is a single common factor that's matching
Retail Trade Survey. This allows us to create something that the Monthly Retail Trade Survey on a monthly basis. We're
is both timely and available more frequently. This is some- starting with the monthly Census Bureau data. We're add-
thing we publish publicly twice a month, so twice as often ing to that weekly data on gasoline consumption from the
as the Monthly Retail Trade Survey. The index covers the EIA, daily data that's aggregated up to the weekly level from
period from January 2018, so it's mostly using very recent three sources of credit and debit card transactions, two that
data, with four weekly values per month. are consumer-facing, Consumer Edge and Facteus, and one
We're using a bit of a non-standard calendar to allow us that is business-facing, Womply.
to replicate the Census structure. We adjust for that with We're adding to that retail foot traffic data that's daily
the way we seasonally adjust our data. The most important and aggregated up to the weekly frequency from SafeGraph;
thing that we found that using this high-frequency data really and, finally, daily surveys of consumer senstiment that are
allows us to accurately predict the Advance Monthly Retail run by the polling for Morning Consult, one of the last bit
Trade Survey, or MARTS. of information to go into this model. All of that is com-
We've been back-testing this model and comparing it bined together in our dynamic-factor model into what we
against consensus nowcasts since February 2020. We found call CARTS.
that out-of-sample this model is producing forecasts that are If you go to our website and take a look at what we're
roughly about 50% more accurate than the consensus fore- producing, you're going to see a picture that looks like this
cast. I don't have a whole lot of time to go over the details (Fig. 1). This is retail and food services sales excluding
today, but all those details are available in a Chicago Fed automotive spending, expressed as billions of dollars on a
working paper (Brave et al. 2021a). seasonally adjusted basis.
We call this measure CARTS, a little bit of a play on What you'll notice from this figure is the importance of
words comparing it to MARTS, the Census Bureau meas- the benchmarking. The line with the Xs: that's actually the
ure. It's really short for "Chicago Fed Advance Retail Trade Census Bureau data, that's the Monthly Retail Trade Survey.
Summary." CARTS contains, like I mentioned, a weekly You can see the weekly index line is basically intersecting
index of retail trade summarizing data on credit and debit the Census Bureau data. That's by construction.
card transactions, retail foot traffic, gasoline consumption, We benchmark to the Census Bureau data so that the
and consumer sentiment. monthly frequency of our measure averages out to it. The
This is all available on our public website: ChicagoFed. interesting thing that comes about, though, in this process
org/CARTS. But you can also find it on the St. Louis Fed's is you can see a fair amount of weekly variation that you
FRED Database, and in Haver Analytics Survey Database as wouldn't necessarily see in the monthly data. I think that's
well. And there's also a nice little non-technical introduction most visible in the pandemic period, where we actually pick
to CARTS and everything that it contains. It's most recently up a few weeks of stockpiling effect early on, prior to the

Fig. 1 Retail and Food Services Retail & Food Services Sales Ex. Auto
Sales Ex. Auto. Billions of $, billions of $, seasonally adjusted
seasonally adjusted 550

500

450

400

350

300
2018 2019 2020 2021 2022

Weekly Index of Retail Trade MRTS & MARTS Projection for MARTS

Sources: CARTS: https://chicagofed.org/carts; MRTS & MARTS: Haver Analytics USECON & AS1REPNA databases
Blending data to understand the economic impact of COVID‑19

shutdowns; and we also see an early turning point when the 4 Rebecca Hutchinson
recovery first started.
The other aspect of what we're producing, if you look at Today I will be presenting on the monthly state retail sales,
the far righthand side of the graph, is that little X. Because a Census Bureau experimental data product that was devel-
we are using higher-frequency data, we're actually able oped and released last year in response to the need for more
to project what we think the upcoming MARTS value, or timely sub-national data during the pandemic, especially in
Advance Monthly Retail Trade Survey, is going to be. the retail sector, which was so impacted by the pandemic.
We can also take a deeper dive into the data and look at These state-level retail data are some of the most-
our weekly units of retail trade, to see what's explaining the requested data from our data users, and we were all really
week-to-week variation in the index. It's about a 50/50 split excited to finally release the first version of these data last
between the contributions of the lagged monthly Census data September. Previously, state-level retail sales informa-
and the high-frequency Big Data. tion was only available once every five years as part of the
I mentioned that one of the things that we've looked at Economic Census. The high cost of survey collection and
is the ability of this index to actually nowcast, or forecast respondent burden has made survey-based state-level retail
the Advance Monthly Retail Trade Survey. We've been pro- sales difficult and costly to obtain. These data are modeled,
ducing this index publicly now since June, so we have four so they required no new data collections.
releases of MARTS to compare our projections against other We are publishing year-over-year percentage changes for
projections that are out there (Fig. 2). each state and the District of Columbia for total retail sales,
Our forecasts have done pretty well over the last four excluding non-store retailers, as well as for some of the retail
months. In fact, in a mean absolute error sense, they've been subsectors. We are not publishing those non-store retailers
about 55% more accurate than the consensus forecast. So, which are primarily e-commerce-type businesses. The best
overall, we think this is proving to be a very useful meas- method and data sources needed to attribute e-commerce
ure. You can go to the website and actually see in tabular sales to state geographies is a bigger challenge that we're
form our most recent projections (https://www.chicagofed. still working through.
org/publications/carts/index). We're hoping that it's as useful Let's take a quick look at some of the data. On this slide
for everyone as it certainly has been for us. (Fig. 3) you can see the year-over-year percentage changes
for total retail excluding nonstore retailers for each state in
April and July of 2020 and 2021. States darkly shaded have
3 Ron Jarmin positive year-over-year percentage changes, and those in
lighter shades have negative changes.
I haven't thought of it, but there's some pun in there about In April 2020 when much of the country was impacted
"cart before the horse" or something like that Now we're with those pandemic-related closures, you can see that
going to go with Rebecca Hutchinson from the Census much of the country had negative year-over-year percent-
Bureau. age changes lightly shaded. And in July 2020, in the upper

Fig. 2 Nowcasts for the advance Nowcasts for the Advance Monthly Retail Trade Survey (MARTS)
monthly retail trade survey Retail & Food Services Sales Ex. Auto (m/m % Chg.)
(MARTS). Retail and food 2
services sales ex. auto (m/m % Mean Absolute Error Ratio (CARTS/Consensus): 0.45
Chg.) 1.5

0.5

-0.5

-1
June-21 July-21 August-21 September-21

CARTS Consensus MARTS
Sources: CARTS: https://chicagofed.org/carts; MARTS/Consensus: Haver Analytics AS1REPNA database
S. A. Brave et al.

Fig. 3 Total retail sales excluding nonstore retailers by state

right, there was some improvement, with many states having administrative data, in the form of payroll; national- and
positive and significant year-over-year changes. company-level retail sales data, from the Monthly Retail
In that bottom row you can see that both April and July Trade Survey; and point-of-sale retail sales data, from a
2021 every state has positive year-over-year changes when third-party provider, in this case it's MPD, for a small num-
compared to 2020. It is interesting to note that many states ber of retailers. We used national tabulations of this third-
with positive year-over-year changes in July 2020 also have party data for non-response in our Monthly Retail Trade
year-over-year growth in July 2021. Survey, but it is excellent for state-level estimates. However,
During those early months of the pandemic, most of this data requires retailer consent because it is available at
retail saw negative year-over-year changes. However, food that store-level, down to the ZIP Code.
and beverage stores had a more positive April 2020 (Fig. 4), Getting more retailers quickly was not an option. We had
because we were all home cooking and baking bread. Almost to be a little creative to get more of this data. If retailers
every state has positive year-over-year changes in sales in are grouped together, the third-party provider does not need
April 2020. retailer consent to share. We gave the provider with lists of
Flash forward to 2021, in that bottom row, and you can retailers by mixed codes, and they create monthly totals by
see that, not surprisingly, sales in April 2021 were down states for each of the groups. We gave up that store-level
year-over-year for most states with restaurants reopening. granularity, but this was a really great way to get a large
But it is interesting to look at a state like New York that was amount of valuable data quickly.
up 18% in April 2020 and in April 2021 built on that growth With this data on the figure, a blended approach to the
and was up 4.7% compared to 2020. The same thing happens estimates was the natural choice. Jenny Thompson, a sen-
in Texas in July, where it was up 13.8% in July 2020 and it's ior mathematical statistician at the Census Bureau, and her
up 5.9% in July 2021. methodology team developed a composite estimator that is
I just want to touch briefly on how we created these data. a weighted average of synthetic and hybrid estimates. The
Early in the pandemic a small group of us were a tasked weight used is a ratio of the variance of the synthetic esti-
with assessing quickly the data we had that could be used mate to the total variance of both estimates.
for these estimates, and developing a methodology. Data that The synthetic estimates you can see on the left side of
we had access to that would be useful to this work included Fig. 5 make use of survey data and that payroll data. These
Blending data to understand the economic impact of COVID‑19

Fig. 4 Food and beverage stores (NAICS 445) by state

Fig. 5 Modeling the data

are simple to calculate using a ratio of the state payroll to data that we have at either the store-level or at the aggregate
the national payroll multiplied against the national Monthly level are then added in.
Retail Trade Survey estimates. The hybrid estimates make For those retailers with more than one location that we
use of all the data that were on the previous slide. don't have survey data for, or third-party data, we impute for
If a retailer on the Monthly Retail Trade Survey operates their sales using either the store-level point-of-sale data or
many store locations, but only in one state, or only has a a combination of survey data and payroll data. We currently
single location, we use their survey data. The point-of-sale do not have a good method for imputing retailers with only
a single location. As a final step we apply an adjustment
S. A. Brave et al.

factor to bring the data up to the national level to account some produced by the statistical agencies, like some things
for these locations. we've just seen, in addition to some of the work done by
These data are being improved on an ongoing basis, the regional Feds, location tracking information, credit card
and we have a few research efforts underway. One require- spending information, employment trackers.
ment we have had to date with our third-party data are that A lot of this nontraditional data have come to the fore-
we have to know all of the retailers included in the data. front, but it's still important to know that these official statis-
Because so many data sources with valuable retail informa- tics in the background are the gold standard for tracking the
tion are out there, where this isn't possible we are exploring economy. And the way I view thinking about blending data
ways to look at this data and incorporate it in these models. are, "How do we combine this information to get a better
Currently we are working to see if, rather than only using idea of what's happening in the economy?".
a payroll ratio in the synthetic estimate, could we model that And I think the main takeaway point I'd like to focus on
ratio using a variety of sources, including payment processor here is really that these nontraditional indicators are going
data, state sales tax data, and other data of that nature? We to require a frame of reference. I'm going to tackle this ques-
are also seeing if we can extend this methodology to other tion by really thinking about it through business exits and
sectors of the economy, including services. employment data.
The first one was thinking about a possible business exit
surge in 2020. When we were asking this question mid-last
5 Ron Jarmin year we really didn't know what was happening in the econ-
omy. We knew that the BLS data, the BED that would help
Next up is Chris Kurz from the Federal Reserve Board. us answer this question, wouldn't be out until late 2021, and
the Census data weren't going to be out, maybe, until 2023.
That might also depend on Ron.
6 Christopher Kurz I'm also asking the question, "What did blended employ-
ment data tell us?" And I think it's important to think about
I'm going to be talking about blending data to better under- that: The framework we had for blending employment
stand the economic impact of COVID-19. This work primar- data sort of got called into question in the middle of the
ily goes through three papers that I've been jointly working pandemic.
on (Cajner et al. 2020a, b; Cajner, Crane, Decker, Hamins- I'll first review a couple of small business exit trackers
Puertolas, Yildirmaz, and Kurz. 2020; Crane et al. 2021), (Fig. 6) We had some credit card transaction data that was
and some ongoing research. really valuable to look at firms that were no longer receiving
I don't really have to say that Big Data has played a criti- transactions. That was telling us about 35% of these small
cal role in tracking the 2020 pandemic, but I just want to put businesses were gone, or exited.
that out there as the start for this discussion. It's important And the same thing you're seeing in the Homebase data,
to know that many new things came out, new data sources, which is a clock-in, clock-out tracker. But I think this sort

Fig. 6 Exit in small business
trackers
Blending data to understand the economic impact of COVID‑19

of contextualization that I'm pointing to is really important. that went up to something like thirteen and a half by Febru-
You need to compare the 2020 moves to the previous year, ary of 2021.
2019, to understand the magnitude of these differences. I'm now going to turn my attention to another case for
So one of the most important things we did in the start blended data. This is going to be similar to the CARTS
of this project was, "Let's look at the BLS and Census tra- we saw earlier, where you're taking different informa-
ditional data to establish statistics, better understand what tion sources and kind of fitting, let's say, an unobserved
annual firm exit rates are,” so that we can then compare that state. This is a common filter approach to employment
to what we were seeing in the current data in 2020. information.
Some stylized facts that you should take with you for the We're taking information from ADP and combining that
rest of your career should be that annual firm exit rates are with information with the CES, so basically creating our
about seven and a half percent; quarterly it's about eight and own data series from the underlying ADP microdata, and
a half percent. Temporary closure is actually really common, combining that with the published BLS data. Those two
about 2.0% of establishments per quarter temporarily close. pieces of information were combined to measure what the
If you look at the latest BED data, we're thinking about underlying state was.
600,000 total deaths and about 150,000 to 200,000 excess The noisy information for both series were being com-
deaths in 2020. This is something that we've only been able bined in the context that the point estimate for a given month
to conclude recently. We'll show how other data sources of, let's say, private payroll employment from the BLS usu-
point us in that direction. ally has a confidence interval around 110,000 employees.
We also looked at the ADP microdata to look at shutdown Combining those two information sources gives us a
based on the length of time that establishments or businesses more certain, let's say, point estimate of what's happening
within that data were not issuing pay. And, interesting, in in private payrolls and improves the current estimate, like I
August and September of last year we were seeing that the said, but also improves, our understanding of what the next
count of businesses that were closed was returning to nor- three months are likely to be. The Kalman Filter smooth
mal. Being able to say that in August of last year was actu- state actually gives you a better idea of what employment's
ally quite surprising, because at that time we weren't sure going to be in the next three months than either independent
what was happening in the overall economy. source (Fig. 8).
We also leveraged SafeGraph data to infer closure infor- Importantly, though, we ran into the pandemic, where
mation based on, let's say, a 65% drop in foot traffic. And this idea of being able to combine two pieces of information
from that information we're able to conclude a lot earlier together might have run into the fact that we were actually
than that latest BED data that a little less than 200,000 estab- seeing large employment declines in extremely fast fashion.
lishments died in excess of normal in 2020. Therefore we were completely reliant on the high-fre-
Thinking about the SafeGraph data I just mentioned, we quency aspect of the ADP/FRB microdata (Fig. 9). You can
were able to use that location data to talk about restaurants see the dramatic employment declines that we were only
that are temporarily closed (Fig. 7). You can see how that aware of having the nontraditional data in hand. The ques-
jumped up in March 2020, and the rate of temperature clo- tion is: Would this sort of need for timeliness become the
sure declined thereafter. Cumulatively, we were looking at norm, and reduce the importance of blending both of those
the percent of restaurants that were permanently closed, and data sources in real time?

Fig. 7 Percent of restaurants 70

temporarily closed
60

0
January February March April May June July August September October November December January February

Source: Author’s calcula ons from Safegraph microdata.
S. A. Brave et al.

Fig. 8 Best case for blended 600

data? Kalman filter approach 400

200

-200

-400

-600

-800

-1000
January 2006 January 2008 January 2010 January 2012 January 2014 January 2016 January 2018
CES ADP-FRB Smoothed State Lower Bound of 90% Confidence Interval Upper Bound of 90% Confidence Interval

Source: BLS, ADP micro data, internal calculaons.

Fig. 9 Payroll employment dur-
ing the pandemic

Private Payroll Employment Leisure and Hospitality Employment
Millions of jobs, change from Feb. 15 Millions of jobs, change from Feb. 15
2000 2000

-3000 -3000 ADP=FRB

ADP=FRB
-8000 -8000

-13000 BLS-CES -13000 BLS-CES

-18000 -18000

-23000 -23000

-28000 -28000
Feb. Mar. Apr. Feb. Mar. Apr.

Source: FRB Staff calculaons, ADP-LLP,
Source: FRB Staff calculaons, ADP-LLP,
Bureau of Labor Stascs
Bureau of Labor Stascs

I would say probably not. In most cases it takes a lot Then we see the high-frequency employment data are
longer to hit peak changes in a downturn. During the Great invaluable, but it also raised ideas about this methodology,
Recession it took something like 16 months to reach our larg- about how we combine things in real time, and moved us
est employment decline, whereas in the pandemic recession, more towards thinking that the nontraditional data were
it only took two months: things moved a lot faster and sort more important within the context of the pandemic.
of moved the pendulum more towards nontraditional data. I would just walk away from this presentation saying
My conclusions are: Blending data are important and nec- that blending data are one of the most important things
essary; it allows for the proper weighting and benchmarking. to provide context and the ability to understand nontradi-
But you have to understand the context, and that's where tional data sources.
blending data are really important. You can either do it from
external sources to aid in understanding what these tradi-
tional metrics look like by comparing them to the actual exit
rates we see in the historical BED or BEDS data. 7 Ron Jarmin
Or you do it from within: Compare current measures to
what we've seen in the past, so you can contextualize what Our last paper is by Michael Stepner from the Univer-
those changes have been. sity of Toronto talking about some of the work from the
One takeaway point from this two-threaded research has Opportunity Insights team.
been that last year’s business exit was elevated, but it wasn't as
dramatic as a lot of us expected. That's a shocking conclusion.
Blending data to understand the economic impact of COVID‑19

8 Michael Stepner small cells; and combining data from multiple companies
to protect the privacy of companies and their customers;
I'm going to be presenting a large blend of datasets that I put and then benchmarking to the gold standard, the national
together in collaboration with Raj Chetty, John Friedman, statistics, which might be less frequent or less granular, but
Nathan Hendren, and a really fantastic team of young econo- guarantee us a representative sample.
mists working at Opportunity Insights. There are so many From this benchmarked, processed data we can tell two
more papers that have been using transaction data, private stories. Let me start from the first one: What happened at
sector data, to analyze the economic impacts of COVID-19 the start of the pandemic? Here (Fig. 10), by blending data
over the course of the last 18 months. on multiple segments of the economy, we can see consumer
What we really focused on within our team was, I think, spending falling, especially among high-income Americans.
two important things. First was taking private sector data We can tell a story of contagion: How that fall in consumer
and converting it into public statistics, so that we have in spending led to business revenues falling, especially among
common with all of the folks here. But traditionally many those small businesses serving high-income Americans; and
of the researchers will sign a confidentiality agreement and then employment falling, especially so among low-income
then analyze private sector data. workers in high-income areas.
Our challenge here was to take this private sector data, I'm going to walk through that story, and then conclude
make it public in a way that's useful both for our own by talking about what's happening today. At the start of the
research and for the research of others, and then also com- pandemic, folks in the top-income quartile were responsible
bine, blend, this data on spending, employment, job post- for the preponderance of the decline in aggregate spending.
ings, business revenue, and other outcomes, and use that In the first weeks, 40% of the decline in aggregate spend-
to form a more complete picture of the chain of macroeco- ing came from the top-income quartile. By the middle of the
nomic events than any study that focuses on one set of out- summer, and into the fall, they were responsible for 50% of
comes alone. the total spending decline. If you look at the bottom of the
In this project we combined data on spending, revenues, income distribution, only about 12% at the beginning, and
employment, job postings, and education. We're blending 5% in the middle of the pandemic of that aggregate spend-
both within categories and across categories. I want to give ing decline was being driven by the changes in behavior of
a high-level overview of two stories that we can learn about low-income households.
what has transpired over the last 18 months. We can see that translating into business revenue. Look,
We are taking raw data and performing many of the same say, at a map of San Francisco (Fig. 11). If you're familiar
procedures that my colleagues have described. We're clean- with San Francisco you'll see the most affluent areas of San
ing the data; smoothing out seasonal fluctuations; excluding Francisco had the largest declines in small business revenue

Credit and Debit Card Spending
$-3.1 Billion $-0.9 Billion
(40% of Agg. (49% of Agg.
10 Spending Spending
+ $0.1 Billion
Decline) Decline)

8 Top Income Quartile

Per Day ($ Billions)
6 + $1.0 Billion
$-0.1 Billion
$-1.0 Billion ( 5% of Agg.
(12% of Agg. Spending Decline)
Spending Decline)
4
Bottom Income Quartile
2
January July January July
2020 2020 2021 2021

Top Income Quartile Top Income Quartile 2019 Counterfactual
Bottom Income Quartile Bottom Income Quartile, 2019 Counterfactual

Fig. 10 Consumer spending by income quartile
S. A. Brave et al.

Fig. 11 Changes in small business revenues from January to April by ZIP code. San Francisco

Change in Small Business Revenue (%)
Apr 15 ‘20 July 23, ‘21
0% 10% 10%
(3.2m)
Change from Jan-Jul Top Income Quartile 6%
Slope = -13.12%/$1000 (1.8m)

Change in Employment (%)
-13% 2%
(s.e. = 0.57) 0% (-4.3m jobs)
Second Quartile (0.6m)

-20%
-10% Third Quartile

Relative to January 2020

Relative to January 2020
Change from Jan-Apr
-40% Slope = -13.47%/$1000 -20% -19% -22%
(-6.2m) (-7.1m)
(s.e. = 0.35) Bottom Income Quartile
-27%
-30% (-8.5m)

-60% -37%
(-11.7m)
-40%
500 1000 1500 2000 Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug

Median Two Bedroom Monthly Rent in 2014-2018 ($)
Fig. 13 Employment changes by income quartile
Fig. 12 Changes in small business revenues vs. rent, by ZIP code

States (Fig. 12). That was true in April, and we can see that
at the start of the pandemic, whether that's Downtown San continued to be true in July. Everywhere had a recovery, but
Francisco, the Silicon Valley, or the area around Berkeley. you still had this differential loss.
This is just an example. But if we go broader, outside of That translated into employment losses (Fig. 13). Those
San Francisco, and you compare on the X axis, you look at employment losses were concentrated in the lowest income
monthly rents in an area, the idea being that areas with high quartile. The bottom-income quartile that lost 40% of their
monthly rent are areas with a large number of people with employment at the trough of the pandemic, while the top-
a fairly substantial income, those tend to be high-rent areas. income quartile also declined, but by only 13%. And once
You see this consistent pattern, with larger declines in again those losses were concentrated in the areas that were
small business revenue in the higher-rent areas of the United most affluent.
Blending data to understand the economic impact of COVID‑19

Fig. 14 Changes in low-wage (bottom quartile) employment rates by ZIP code. San Francisco

Apr 15 ‘20 July 23, ‘21
75
10%
10% (3.2m) +63.2 %
6% Low Req'd Educ.

Change in Job Postings
Top Income Quartile (1.8m) 50

Change in Employment (%)
-13% 2%
0% (-4.3m jobs)
Second Quartile (0.6m)
25
+15.5 %

Relative to Jan 2020 (%)
-10% Third Quartile Moderate
0 Req'd Educ.

Relative to January 2020
-6.13 %
-20% -19% -22% Highest Req’d
(-6.2m) (-7.1m) -25 Educ.
Bottom Income Quartile
-27%
-30% (-8.5m) -50

-37%
(-11.7m)
-40%
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec Jan Feb Mar Apr May Jun Jul Aug

Fig. 16 Job postings by required education level
Fig. 15 Employment changes by income quartile

missing low-wage jobs. We can ask why hasn't employment
Within a map it's those same areas in San Francisco, the recovered at the bottom.
Bay Area, around Berkeley, that had the largest declines in Blended data tells us that labor demand is unlikely to be
low-income employment (Fig. 14). the cause. We can actually see a surge in job postings, espe-
Let's conclude by talking about what's happening right cially for low-skill jobs. If we look to data on job postings
now. Here we tell a story of really a bifurcated recovery collected by Burning Glass Technologies (Fig. 16), we see
where consumer spending has recovered to pre-pandemic jobs with low required education had much higher job post-
levels, and frequently above. Employment has fully recov- ings than in a typical year by the summer of 2021.
ered for the top three-quarters of the wage distribution, but The question is, "Well, what's going on on the labor
employment has totally stagnated at the bottom of the wage supply side?" There's been research using both our data
distribution (Fig. 15). We have more than seven million and others showing that this high-level of unemployment
S. A. Brave et al.

insurance has not really had a big effect on the labor supply. 11 Ron Jarmin
That leaves a couple open questions about the extent of labor
market mismatch: how the needs of jobs and employers have Let me just follow that up. Obviously the COVID-19 pan-
diverged from the supply of labor for that; or this change in demic was a big deal that everyone was paying attention to,
preferences, the story of the Great Resignation that is now and I think folks were willing to share some data.
so prevalent. But let's think about a more mundane time, like the hur-
What I want you to leave with is that this blended data ricane on the Gulf Coast, where it's not something that's got
can tell a very detailed story of propagation through the everyone's undivided attention, just a relatively small portion
macroeconomy, both in our own analyses and in the other of the population. To have these sorts of datasets in place
work that folks have done using our data. So let me conclude in time to look at that, what are some of the challenges that
there, pass it back. you all have encountered in gaining access, and then making
sense of, the data?

9 Ron Jarmin
12 Rebecca Hutchinson
Let me take the chair's prerogative here. Obviously, the title
of the session was using blended data to understand the I think for us one of the biggest challenges with these data-
COVID-19 pandemic. sets is the cost; they're pretty expensive to acquire. Other
But we could generalize a little bit. And let's say how do issues we ran into with getting, especially the most granular
you all see these being useful in maybe sort of more run- data, down to the retailer, is you do need retailer consent. A
of-the-mill business cycle, or other shocks to the economy, lot of retailers don't want to provide that.
maybe a large-scale natural disaster or something like that? I know as we look to expand this concept to other sec-
To speak to some of the difficulties of working with these tors of the economy, with retail you're pretty lucky with the
data, how important it would be to incorporate them into kind of data that's available out there. Finding similar data
the regular measurement infrastructure, and how useful sources for the service sector, which is a pretty diverse part
they could be to help businesses and policymakers react to the economy, is a challenge we are trying to work through
shocks? right now.

10 Michael Stepner 13 Ron Jarmin

I think that alternative data sources, and blended data, pro- Somebody had a question in the chat that's very related to
vide two differentiations from a traditional survey database. this. It's about the transparency that you need. We're using
The first is the timeliness, and the second is the granular- data that maybe not everybody gets to see how the sausage
ity. In a normal business cycle, I suspect the timeliness at was made. How do we ensure the data users, especially deci-
a national level is much less important. Things are pretty sionmakers, whether they be central bankers, whether they
stable at a national level and don't fluctuate wildly like dur- be CEOs, or folks running NGOs, or just average folks: How
ing COVID-19. do we ensure that they understand the transparency? And
I think that that's where the granularity can really shine. then how do you ensure that you have access to this over
If we look at our data, we can see these sharp spikes in early time?
2021 in Texas and Louisiana, when we had that cold snap
that led to the shutdown of the Texas economy. If a hurricane
hits Louisiana or the Eastern Seaboard, you can pick that up
14 Christopher Kurz
and really understand, in almost real time, how that's affect-
ing the economy.
That's just a huge problem. It's a problem, from our per-
I think, going forward, with those types of granular
spective as, let's say, someone in the middle of providing
changes, which are difficult to pick up in surveys, we will
this data, or understanding it, and I think it's even more of
know more and more about what is happening in small
a problem for just your average consumer, who might be
areas, and to specific subgroups of people, whether this is
downloading it from a website.
hurting a specific sector or high-income people versus low-
I know the Board, different Reserve banks, and statisti-
income people, we'll get a much quicker appreciation of how
cal agencies have had a lot of difficulties trying to break
regional or sectoral changes are affecting the economy.
Blending data to understand the economic impact of COVID‑19

through that and better understand the underlying microdata. 17 Ron Jarmin
I think it's imperative for the organizations to realize this
and work with these actors to know that, for example, the Central banks and statistical agencies can't pay every com-
Census Bureau really needs to be able to look under the hood pany in the economy for their data. What benefits do you
of something that they're going to be able to try to provide think that these companies could get from allowing more
to the public. Getting everybody on that same page is really access to the data for these more public uses of their data?
important. And I think only through good user agreements,
understanding how to protect confidentiality, and provide
it in public methodology are we going to get to that point. 18 Michael Stepner

I would say we're in early days for the use of private sec-
15 Scott Brave tor data for public sector statistics. At the beginning of
COVID-19 incentives were aligned: The economy was in
I would second that. This was the biggest hurdle for us to freefall; they had valuable information that they typically
jump, and we would have been able to start making this use just for their own internal business decisions; they
public a year ago but there were a lot of data sources that wanted to make that available to policymakers, to jour-
we had to eliminate because of these concerns. Something nalists, to everyone trying to understand the pandemic.
changed right away in the middle of the process. Because if that helped policymakers diagnose what was
For us to get down to the five that we got down to, these going on and assist the economy, that was in their own
were the companies that were willing to work with us to business interests. Everyone's incentives were aligned.
help us understand what was going on, and willing to give I think, going forward, I expect we will see more part-
that kind of information, and allow us to make these kinds nerships between large companies in the banking sector,
of things public, even if we're not publishing the data, which in the tech sector, who see the value of their data for the
we don't. public good. And I think corporate social responsibility
I think that that's probably unrealistic to expect of them will play into that.
always, if they're making money off the data. But the meth- But the idea that they have data which is valuable for
odology and understanding how all the sausage is made was American policymaking, which they can share without
very important to us. It took us a while to find partners that compromising the privacy of their users or their custom-
were willing to do that, but I think once they understood ers or their own business prospects: I think that message
where we were coming from and what we were trying to get will start to lead to a greater expansion of partnerships.
out of this, and really the policy uses of what we're trying to And hopefully lead to a new era in the construction of
do with the bigger picture, we were able to make headway national statistics, where we have more reliable, more
on that. granular data, even more so than what we are showing in
these early days.

16 Christopher Kurz
19 Ron Jarmin
I just want to jump back to that first point and say that I think
we opened the door to a lot more granularity in the past
Well, I share your hope for that optimistic picture of the
year. I would say timeliness let us know that something was
world.
happening, but the granularity helped us understand what
was happening.
I'd actually say that the verdict is really going to be out
for, unfortunately, the next downturn for us to understand
References
which of those is going to be more important because the Brave, Scott A., Ross Cole, Michael Fogarty, Ezra Karger, and
timeliness was so valuable. If you really jump back to the Liam Puknys. 2001. Introducing CARTS: A new index tracking
Great Recession, how many months did it even take the national retail spending. Chicago Fed Insights June 12. https://
NBER to call that recession, and really understand what www.chicagofed.org/publications/blogs/chicago-fed-insights/
2021/introducing-carts.
was happening in the economy? When we have a "normal", Brave, Scott A., Michael Fogarty, Daniel Aaronson, Ezra Karger, and
and I want to put quotes around that, downturn; when we hit Spencer Krane. 2001. Tracking U.S. consumers in real time with
that sort of peak, then we're going to really understand how a new weekly index of retail trade. Working Paper 2021–2025,
some of this investment into non-traditional data are really Federal Reserve Bank of Chicago. https://www.chicagofed.org/
publications/working-papers/2021/2021-05.
going to pay off.
S. A. Brave et al.

Cajner, Tomaz, Leland D. Crane, Ryan A. Decker, John Grigsby, data science training programs, which seek to teach important data
Adrian Hamins-Puertolas, Eric Hurst, Christopher Kurz, and science skillsets to the existing workforce.
Ahu Yildirmaz. 2020. The U.S. labor market during the begin-
ning of the pandemic recession. National Bureau of Economic Christopher Kurz is the section chief of the Industrial Output section in
Research Working Paper 27159. https://www.nber.org/papers/ the Division of Research and Statistics at the Federal Reserve Board.
w27159. His responsibilities at the Board of Governors have ranged from fore-
Cajner, Tomaz, Leland D. Crane, Ryan A. Decker, Adrian Hamins- casting industrial production and motor vehicle sales to coordinating
Puertolas, Ahu Yildirmaz, and Christopher Kurz. 2020. Improv- the staff forecast of GDP. Christopher’s research interests are similarly
ing the accuracy of economic measurement with multiple data broad and cover international trade, international finance, nontradi-
sources: The care of payroll employment data. https://www.nber. tional data, and economic history. Research projects of Christopher’s
org/system/files/chapters/c14272/c14272.pdf have employed microdata to quantify the effects of international trade
Crane, Leland D., Ryan A. Decker, Aaron Flaaen, Adrian Hamins- on employment volatility at the firm level and back tested systemic
Puertolas, and Christopher Kurz. 2021. Business exit during the risk measures using historical banking data. Over the past five years,
COVID-19 pandemic: Non- traditional measures in historical Chris has been working on the Federal Reserve Board’s Expanded
context. Finance and Economics Discussion Series 2020–089r1. Measurement Agenda (EMA). The EMA is an attempt to identify novel
Washington: Board of Governors of the Federal Reserve System, data sources and leverage new techniques that can be used to develop
https://doi.org/10.17016/FEDS.2020.089r1 timely and accurate measures of economic activity that are relevant to
the formation of monetary policy.
Publisher's Note Springer Nature remains neutral with regard to
jurisdictional claims in published maps and institutional affiliations. Michael Stepner is an assistant professor at the University of Toronto
and a research principal at Opportunity Insights. His research examines
the relationship between health and economic inequality, with a focus
on how public policy can improve the health and financial security
Scott Brave Scott A. Brave is the lead consumer spending economist of low-income populations. He also serves as the network leader for
for decision intelligence company Morning Consult, where he lever- Health Trends and Inequalities research at the NBER Center for Aging
ages the company’s high-frequency proprietary data in forecasting and Health Research, and as the executive director of the Opportunity
and predictive analytics to help clients make better business decisions. Insights Economic Tracker. He received his PhD from MIT in 2019,
Previously, Brave led the releases of economic activity and financial and his dissertation research was awarded the top dissertation award
conditions indexes as a senior economist in the Economic Research from the National Academy of Social Insurance.
Department of the Federal Reserve Bank of Chicago. His research
expertise has focused on the use of high-frequency data in forecast- Ron Jarmin is the U.S. Census Bureau’s deputy director and chief oper-
ing applications, mixed-frequency econometric modeling, and sports ating officer and has been since January 2019. He performed the nonex-
economics, with publication in journals including the International clusive functions and duties of the director from July 2017 to January
Journal of Forecasting, International Journal of Central Banking, 2019 and from January 2021 to November 2021 and previously served
Growth and Change: A Journal of Urban and Regional Policy, STATA as the associate director for economic programs. He led the team for
Journal, Journal of Sports Economics, and Journal of Sports Analyt- the 2017 Economic Census, overseeing a move to 100 percent Inter-
ics, as well as Chicago Fed research publications, such as Economic net data collection and leveraging enterprise investments to minimize
Perspectives and the Chicago Fed Letter. Brave earned his bachelor’s system, application, and dissemination costs. Data products from the
degree in economics with honors from the University of Chicago and economic census provide the foundation for key measures of economic
an MBA with concentrations in economics, statistics, and finance from performance, including the nation’s gross domestic product. From 2011
the University of Chicago Booth School of Business. to 2016, Dr. Jarmin served as the assistant director for research and
methodology. He oversaw a broad research program in statistics, survey
Rebecca Hutchinson is the Economic Indicator Division’s Big Data methodology, and economics aimed at improving economic and social
Lead at the Department of Commerce’s Census Bureau. She leads measurement within the federal statistical system. Since beginning his
agile teams that explore the use of alternative data sources including career at the Census Bureau in 1992, he has also served as the chief
satellite imagery, payment processor data, scanner data, administrative economist, chief of the Center for Economic Studies, and a research
data, and web-scraped financial filings to enhance the measurement of economist. Dr. Jarmin holds a Ph.D. in economics from the University
the nation’s economy, either through the improvement of existing data of Oregon. An elected fellow of the American Statistical Association,
products or the creation of new products. Her recent work has focused he has published papers in the areas of industrial organization, business
on using alternative data sources to create a Monthly State Retail Sales dynamics, entrepreneurship, technology and firm performance, urban
data product. Additionally, she serves as a champion for the Bureau’s economics, data access, and statistical disclosure avoidance.

File and source

File: brave-hutchinson-kurz-stepner-jarmin-2022-blending-data-business-economics.pdf
Size: 2,406,874 bytes
SHA-256: 21185f870e78a7d42e21c2075907a01ffb26891090bd34b4b8171009622cc0b0
Our copy: brave-hutchinson-kurz-stepner-jarmin-2022-blending-data-business-economics.pdf
Original: link.springer.com