Income, Liquidity, and the Consumption Response to the 2020 Economic Stimulus Payments

Scott R. Baker, R. A. Farrokhnia, Steffen Meyer, Michaela Pagel, and Constantine Yannelis

NBER Working Paper No. 27097

May 2020, Revised in September 2020

JEL No. D14,E21,G51

ABSTRACT

The 2020 CARES Act directed large cash payments to households. We analyze house-holds’

spending responses using high-frequency transaction data from a Fintech non-profit, exploring

heterogeneity by income levels, recent income declines, and liquidity as well as linked survey

stimulus. Relative to the effects of previous economic stimulus programs in 2001 and 2008, we

see faster effects, smaller increases in durables spending, larger increases in spending on food,

and substantial increases in payments like rents, mortgages, and credit cards reflecting a short-

term debt overhang. We formally show that these differences can make direct payments less

effective in stimulating aggregate consumption.

Scott R. Baker

Kellogg School of Management

Northwestern University

2211 Campus Drive

Evanston, IL 60208

Steffen Meyer

University of Southern Denmark

Campusvej 55

5230 Odense M

Denmark

Michaela Pagel

Columbia Business School

3022 Broadway

Uris Hall

New York, NY 10027

Chicago, IL 60637

and NBER

Income, Liquidity, and the Consumption Response to the 2020

September 15, 2020

JEL Classification: D14, E21, G51

Keywords: Household Finance, CARES, Consumption, COVID-19, Stimulus, MPC, Transaction Data

The authors wish to thank Sylvain Catherine, Arpit Gupta, Jonathan Parker, and Joe Vavra for helpful discus-

sion and comments as well as seminar participants at the MoFiR Banking Workshop, CEPR New Consumption Data

Conference, Virtual AFFECT seminar, Toulouse School of Economics, Columbia Graduate School of Business, Uni-

versity of Chicago Booth School of Business, and the Federal Reserve Bank of Philadelphia. Constantine Yannelis is

grateful to the Fama Miller Center for generous financial support. R.A. Farrokhnia is grateful to Advanced Projects

and Applied Research in Fintech at Columbia Business School for support. We would like to thank Suwen Ge, Spy-

ros Kypraios, Rebecca Liu, and Sharada Sridhar for excellent research assistance. We are grateful to SaverLife for

University of Southern Denmark (SDU) and Danish Finance Institute (DFI); [email protected]

Columbia Business School, NBER, and CEPR; [email protected]

University of Chicago Booth School of Business, NBER; [email protected]

1

1 Introduction

In three recent instances, the US government made direct cash payments to households in response

to economic downturns. These payments are generally meant to alleviate the effects of a recession

and stimulate the economy through a multiplier effect, i.e., by increasing households’ consumption

describe how household MPCs vary across categories of consumption and how these categorical

responses differ from those seen in previous recessions. Furthermore, this paper links transaction

data to user survey data in order to study how expectations impact household responses to stimulus

payments. Understanding these MPCs is key to targeting policies to households where effects will

be largest, as well as distinguishing between different models of household consumption behavior.

MPCs are important to both policy and economic theory as they determine fiscal multipliers

in a wide class of models. More specifically, heterogeneity in MPCs impacts which households

are most responsive to stimulus payments. In turn, targeting can have large impacts on the effec-

tiveness of stimulus payments on consumption and the aggregate economy. This paper shows that

liquidity is a key determinant of MPC heterogeneity during the 2020 contraction, with highly liq-

to August 2020 for these users. The fact that we observe inflows and outflows from individual

accounts as well as balances in this dataset allows us to explore heterogeneity in levels of income,

changes in income, and liquidity. The sample consists primarily of lower- and middle-income

2

households, and we are able to link the bank account transactions data to survey data about eco-

nomic expectations.

We use this detailed data to look at the CARES Act stimulus payments distributed in April and

May 2020. The first stimulus payments were made in mid April via direct deposit from the IRS,

and we can observe the user-specific stimulus amounts as well as spending daily before and after

stimulus payments are made. We see sharp and immediate responses to the stimulus payments;

within ten days, users spend over 20 cents of every dollar received in stimulus payments. The

largest increases in spending are on food, non-durables, and payments like rent, mortgages, and

These heterogeneity results are important in terms of targeting stimulus policies towards groups

most impacted by them. The theory behind stimulus payments links MPCs directly to the ultimate

fiscal multiplier effect, i.e., the effectiveness of the payments in stimulating aggregate consump-

tion. The results of this study suggest that targeting stimulus payments to households with low

levels of liquidity in a type of recession where large sectors of the economy are shut down will

have the largest effects on MPCs, and hence on fiscal multipliers.

We further explore how beliefs about personal and aggregate outcomes impact the response to

stimulus payments, utilizing a survey of our users which we can then link to the transaction data.

Theoretical work has long noted that expectations can play an important role in the efficacy of

stimulus (Barro, 1974, 1989; Seater, 1993; Galí, 2019). In particular, households may respond to

3

users are relatively pessimistic about the length of the pandemic and their own future income and

employment opportunities.

While we do not find evidence that anticipated tax increases impact

MPCs, we do find that expectations about employment and government benefit cuts play an im-

portant role in determining MPCs. Households that anticipate unemployment or benefit cuts save

a significantly larger fraction of their stimulus checks.

We then show in a macroeconomic model with multiple sectors that non-targeted fiscal stim-

ulus payments in environments like the 2020 COVID-19 epidemic may be less effective than the

payments in response to the 2001 and 2008 economic downturns. Reflecting the current situation,

more debt by borrowing from the higher wage sector. Agents end up using the stimulus payments

to repay debt to high wage individuals who have the lowest MPCs out of income. In short, work-

ers will spend their stimulus payment on mortgages and loan repayments as well as non-durable

essentials which implies that the cash flows immediately to agents with lower MPCs. This tends

to make fiscal stimulus less effective overall. This model thus confirms our empirical results.

There is an extensive literature on households’ responses to tax rebates and previous stim-

ulus payments. Using spending data from the Consumer Expenditure Survey, Johnson, Parker

and Souleles (2006) and Parker, Souleles, Johnson and McClelland (2013) look at the tax rebates

granted in 2001 and the economic stimulus payments in 2008. The authors document positive

effects on spending in both non-durable and durable goods. Broda and Parker (2014) use high-

SaverLife conducted our survey from mid-May to the end of July. This survey also elicited self-reported informa-

tion on the receipt and use of the stimulus checks. In terms of the fiscal stimulus use, our survey results line up nicely

with the empirics. 60% of individuals report that they will not use any portion of the check for durables consumption

and 50% of the users are using at least part of the check amount for food spending. A large majority of users also

reported using at least a portion of the stimulus check for payment of current or past due bills. Finally, 15% of users

are reporting to save most of the check amount and 45% report to save none of the check amount.

4

(2007) work with credit card accounts and find that customers initially saved the tax rebates in

2001, but then increased spending later on. In their setting, customers with low liquidity were

most responsive. Misra and Surico (2014) use a quantile framework to look at the 2001 tax rebates

and the 2008 economic stimulus payments on the distribution of changes in consumption.

In Section 4.1, we discuss some of the differences between our estimates and the previous lit-

erature that analyze past stimulus programs. The existing studies exploit the differences in timing

of the arrival of the payment to infer causal effects. Our results are generally comparable. How-

ever, the three main differences are: 1) during the 2020 stimulus, households spend much of their

ment that responsiveness to rebates is driven by liquid wealth. Households with sizable quantities

of illiquid assets but low liquidity are an important driver of the magnitude of the response. To our

knowledge, our study is the first to look at stimulus payments using high-frequency transaction

data, as such data did not exist in 2008.

The use of transaction data allows us to explore very-

short term responses across categories, minimize measurement error, and explore individual daily

heterogeneity in income declines and available cash on hand.

In this paper, we focus on a very different type of contraction relative to those faced dur-

ing previous stimulus programs: one stemming from an infectious disease outbreak that caused

widespread business and government shutdowns. In comparison to the 2001 and 2008 economic

downturns, the downturn due to COVID-19 was inflicted on households at a much faster pace,

5

While previous studies have pointed out that stimulus payments have positive but heteroge-

neous effects on spending, analyzing the 2020 stimulus program will help us learn more about

effects on spending in different economic circumstances. In particular, this crisis was so fast mov-

ing that households had little ability to increase precautionary savings. Additionally, many sectors

of the economy were shut down due to state and local orders, which can impact the effectiveness of

fiscal stimulus, as discussed above. Some policymakers argued that shutdowns make conventional

fiscal stimulus obsolete.

Our results are also important for the ongoing discussion of Representative Agent Neo-Keynesian

a HANK model to study the impact of the quarantine shock on the US economy in the case of a

successful suppression of the pandemic. In their model, the stimulus payment help stabilize con-

sumption and results in an output decline of less than 3.5%. Additionally, Hagedorn, Manovskii

and Mitman (2019) study multipliers in a HANK framework, whose size can depend on market

completeness and the targeting of the stimulus.

This paper also joins a fast-growing literature on the effects of the COVID-19 pandemic on the

economy, and policy responses. Several papers develop macroeconomic frameworks of epidemics,

e.g. Jones, Philippon and Venkateswaran (2020), Barro, Ursua and Weng (2020), Eichenbaum, Re-

belo and Trabandt (2020), and Kaplan, Moll and Violante (2020b). Gormsen and Koijen (2020) use

stock prices and dividend futures to back out growth expectations. Coibion, Gorodnichenko and

6

pact the social distancing response to the pandemic, and Coven and Gupta (2020) study disparities

in COVID-19 infections and responses.

Our related paper, Baker, Farrokhnia, Meyer, Pagel and Yannelis (2020b), studies household

consumption during the onset of the pandemic in the United States using a smaller sample drawn

from the same data source. Carvalho, Garcia, Hansen, Ortiz, Rodrigo, Mora and Ruiz (2020),

Andersen, Hansen, Johannesen and Sheridan (2020), Bounie, Camara and Galbraith (2020), Chen,

Qian and Wen (2020) perform similar analyses as the one in this paper using transaction-level data

from the Spain, Denmark, France, and China. Dunn, Hood and Driessen (2020) uses transaction-

We also join a literature on how expectations affect household’s economic behavior. Macroe-

conomic models since the 1980s have noted that government budget deficits may in the short-

term affect household’s expectations about future taxes, and implicitly transfers (Barro, 1989). A

newer and growing body of recent work also shows that expectations about individual and aggre-

gate outcomes impact behavior, studying households (Giglio, Maggiori, Stroebel and Utkus, 2019;

Kuchler and Zafar, 2019; Armona, Fuster and Zafar, 2019; D’Acunto, Hoang and Weber, 2020;

Manski, 2004) and firms (Landier and Thesmar, 2020; Landier, Ma and Thesmar, 2017; Bouchaud,

Krueger, Landier and Thesmar, 2019; Gennaioli, Ma and Shleifer, 2016). During the debate about

the efficacy of the 2008 stimulus, the role that expectations would play in the program’s efficacy

and stimulating consumption was discussed at length, however, there is little empirical work ex-

7

a simple model to explain how fiscal multiplier effects may differ from prior stimulus programs.

Section 8 concludes and suggests directions for future research.

February. By mid-March, the virus was spreading rapidly, with significant clusters in New York,

pandemic in a number of ways: by issuing travel restrictions, shelter-in-place orders, and closures

of many non-essential businesses.

The federal government soon passed legislation aimed at ameliorating economic damage stem-

25, 2020 as a response to the economic damage of the new virus. The Act deployed nearly $2 tril-

lion across a range of programs for households and businesses. This study focuses on the portion

of the Act that directed cash transfers to the vast majority of American households. These one-time

payments consist of $1,200 per adult and an additional $500 per child under the age of 17. For

of households (single parents with children), and $150,000 for married couples. No payments were

made to individuals earning more than $99,000 or married couples earning more than $198,000.

8

4 Effects of Stimulus Payments

Looking at the raw levels of spending for users receiving stimulus payments, Figure 3 shows mean

daily spending before and after the receipt of a stimulus payment without any other controls or

comparison group. In this figure, we only show spending data for users who receive a stimulus

i

t

k=−7

β

k

[t = k]

it

+ ε

. ‘Time to Payment’ is equal to zero for a user on the day of receiving the

than those that do not in the days before they receive the checks. Upon receiving the stimulus

check, users dramatically increase spending relative to users who do not receive the checks.

There is a sharp and immediate increase in spending following the receipt of a stimulus deposit;

keep spending significantly more than those who have not received checks for the entirety of the

post-check period that we observe. The relative difference in spending declines during weekends,

mostly driven by the fact that observed levels of spending tend to be depressed during these days

for reasons described above.

tive spending on food, household, non-durables, and payments each increasing by approximately

There are substantial intra-week patterns in spending, with Mondays typically seeing the highest levels of posted

transactions and spending as transactions that occurred during the weekend sometimes process only on the Monday

that follows.

15

$75-$150 in the week following receipt of a check. Durables spending sees a significant increase,

but it is much smaller in economic terms with only a $20 relative increase in spending during the

first three days.

Table 2 presents similar information, presenting coefficients from the regression c

= α

α

+

[t = k]

+ ε

. That is, we examine the excess spending among users who

received stimulus payments on each day following the receipt of their stimulus checks, scaled by

the size of their payment. A value of 0.03 can be interpreted as the user spending, on day t, 3% of

their stimulus check (eg. $36 out of a $1,200 stimulus check) more than a user who did not receive

first week after stimulus receipt.

Spending on days during this period is economically and statistically significantly higher for

those receiving stimulus checks and there are no days with significant reversals – days with stim-

ulus check recipients having lower spending than those who did not. Overall, for each dollar of

stimulus received, households in our sample spent approximately $0.35 more in the month follow-

ing the stimulus.

The remainder of the columns in Table 2 decompose the effect that we see in overall spend-

ing according to the category of spending. We find significant increases in spending in all of

these categories, with the largest increases coming from non-durables and payments. We find

muted effects of the stimulus payments on durables spending. In previous recessions, spending

16

(CPS) weights. We show results of this approach in Table 3, we compare our estimates from

our unweighted regressions to those using user weights defined on age, sex, state, and income

bins. Column (4) thus runs our primary MPC calculation using these user weights, finding an

MPC of approximately 0.266 rather than 0.369 in column (3). In general, when weighting our

sample to match the national distribution of households more closely, we see qualitatively similar

results approximately 30% lower in magnitudes when weighting along these dimensions. Given

our sample being younger and having lower incomes and assets than the nation as a whole, it is

unsurprising that down-weighting these types of users produces a somewhat lower MPC.

ine the response of households to economic stimulus programs during the previous two recessions

(2001 and 2008). These programs were similar in nature to the stimulus program in 2020 but were

smaller in magnitude ($300-600 rather than $1,200 checks).

In these previous stimulus programs, households also tended to respond strongly to the receipt

of their checks. For instance, in 2008, Parker, Souleles, Johnson and McClelland (2013) estimated

that households spent approximately 12-30% of their stimulus payments on non-durables and ser-

non-durables and services in the six months following receipt.

able to use Nielsen Homescan data to examine weekly spending responses to the 2008 stimulus

households react extremely rapidly, with household spending increasing by approximately one

third of the stimulus check within the first 10 days.

Another notable difference from the stimulus programs is that we find substantially smaller

17

impacts on durables spending and confirm this in our survey of users. Previous research has found

strong responses of durables spending to large tax rebates and stimulus programs, especially on

automobiles (about 90% of the estimated impact on durables spending in the 2008 stimulus pro-

gram was driven by auto spending). In contrast, despite a sizable response in non-durables and

service spending, we see little immediate impact on durables. In part, this discrepancy with past

recessions may be driven by the fact that automobile use and spending is highly depressed, with

many cities and states being under shelter-in-place orders and car use being restricted. Similarly,

as these orders hinder home purchases, professional appliance installment and spending on home

strongly. In addition, our measure of available liquidity from actual account balances arguably

suffers from much less measurement error than the measures used in previous research on stimulus

checks, giving additional confidence in our estimates.

5 Income, Liquidity, and Drops in Income

balances of SaverLife users, we are able to test whether the consumption and spending responses

differed markedly between users who belonged to these different groups.

In Figures 6-8, we show the cumulative estimated MPCs from regressions of spending on an

indicator of a time period being after a stimulus payment is received. Each figure contains the

18

results of multiple regressions, with users broken down into subsamples according to a number

of financial characteristics that we can observe. That is, the graphs represent the sum of daily

coefficients seen in a regression as in Table 2, by group. In these figures, we divide the samples of

users by their level of income, the drop in income we observed over the course of 2020, and their

levels of liquidity prior to the receipt of stimulus payments.

Figure 6 splits users by their average income in January and February 2020 (prior to the major

impacts of the pandemic). We see clear evidence that users with lower levels of income tended to

respond much more strongly to the receipt of a stimulus payment than those with higher levels of

MPCs on the order of 0.1 while those who had under $100 have MPCs of 0.4 or above.

In Figure 8, we examine whether a similar pattern can be seen among users who have had

declines in income following the COVID outbreak. For each user, we measure the change in

income received in March 2020 relative to how much was received, on average, in January and

February 2020. We split users into those who had a decline in monthly income and those who

saw no decline in income (or had an increase). Here we see a significant difference in MPCs,

but of lower magnitude than the previous splits. Households that saw income declines had an

MPC of just over 0.4 while those with no decline in income registered MPCs of about 0.33. This

smaller difference may be driven by the fact that the federal government had also made generous

unemployment insurance available to nearly all workers, mitigating the potential loss of income

19

6 Expectations and Stimulus Responses

In this section, we explore how household beliefs impact the response to stimulus payments.

Household beliefs may impact MPCs in a number of ways. First, if households anticipate income

declines in the future, they may save more to smooth consumption. Second, if households be-

asset returns and invest more out of current resources.

We surveyed over 1,000 users in our sample and asked them about their beliefs regarding

personal unemployment, income, government benefits and taxes, as well as expectations about

the stock market and the duration of the pandemic. We discuss the survey, which was conducted

via mobile device or email, in more detail in section 3.2. We interact these surveyed beliefs with

stimulus receipt and explore how beliefs impact MPCs.

Figure 9 shows MPCs for subgroups, based on user beliefs regarding personal unemployment,

salary cuts, tax increases, government benefit cuts, stock market increases and the duration of the

pandemic. The figure shows cumulative MPCs estimated from coefficients from regressions of

spending on an indicator of a time period being after a stimulus payment, scaled by the amount

it

t

P ost

it

i

D

year. Individuals who believe that they will be more likely to face unemployment or salary cuts see