Financial Concerns and the Marginal Propensity to Consume in COVID Times: Evidence from UK Survey Data — IMF WP/22/47

Summary

IMF Working Paper WP/22/47, dated March 2022, is by Bruno Albuquerque and Georgina Green and is titled Financial Concerns and the Marginal Propensity to Consume in COVID Times: Evidence from UK Survey Data. The abstract says the authors study how household concerns about their future financial situation may affect the marginal propensity to consume (MPC) during the COVID-19 pandemic, using a representative survey of UK households to compute the MPC from a hypothetical transfer of £500. It says households concerned about not being able to make ends meet have a 20% higher MPC than other households. The abstract suggests that policies targeted to vulnerable and financially distressed households may prove more effective in stimulating demand than stimulus payments to all households.

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

Issuer: Source documents
Document type: Financial Concerns And The Marginal Propensity To Consume In Covid Times Evidence From Uk
Case: Financial Concerns And The Marginal Propensity To Consume In Covid Times Evidence From Uk Survey Data

Cite this document

“Financial Concerns and the Marginal Propensity to Consume in COVID Times: Evidence from UK Survey Data — IMF WP/22/47.” Original at doi.org. Archived at Pandemic Darlings, https://pandemicdarlings.com/source-documents/financial-concerns-and-the-marginal-propensity-to-consume-in-covid-times-evidence-from-uk-survey-data/, SHA-256 533088e3f0d64db73f59fa25ef296e26e36ccb3eba306fe8a58ccc48cb361fad.

Full text

IMF Working Paper

Financial Concerns and the Marginal Propensity to Consume in COVID Times: Evidence from UK
Survey Data*
Prepared by Bruno Albuquerque† and Georgina Green‡

Authorized for distribution by Martin Čihák
March 2022

IMF Working Papers describe research in progress by the author(s) and are published to elicit
comments and to encourage debate. The views expressed in IMF Working Papers are those of the
author(s) and do not necessarily represent the views of the IMF, its Executive Board, or IMF management.

ABSTRACT: We study how household concerns about their future financial situation may affect the marginal
propensity to consume (MPC) during the COVID-19 pandemic. We use a representative survey of UK
households to compute the MPC from a hypothetical transfer of £500. We find that household expectations
play a key role in determining differences in MPCs across households: households concerned about not being
able to make ends meet have a 20% higher MPC than other households. Our findings suggest that policies
targeted to vulnerable and financially distressed households may prove more effective in stimulating demand
than providing stimulus payments to all households.

JEL Classification Numbers: D12, E21, E62, G51, H31

COVID-19; Marginal propensity to consume; Survey data;
Keywords: Household behaviour; Expectations; Financial concerns; Fiscal
policy.

Author’s E-Mail Address: BAlbuquerque@imf.org

*The views in this paper represent only our own and should therefore not be reported as representing the views of the Bank of
England, the International Monetary Fund, its Executive Board, or IMF management. We would like to thank Andreas Fuster,
Dimitris Georgarakos, Jeremy Franklin, Lukasz Rachel, Matt Waldron, and Sarah Venables for useful discussions that helped us
shape the main research idea of the paper. We are also thankful for comments and suggestions from Mark Flanagan.
†International Monetary Fund, and Univ Coimbra, CeBER, Faculty of Economics, balbuquerque@imf.org
‡Bank of England, georgina.green@bankofengland.co.uk
©International Monetary Fund. Not for Redistribution
1 Introduction

The COVID-19 pandemic has brought renewed interest to understanding how households’ con-

sumption and saving respond to income changes. Micro datasets, including timely household

survey data and transaction-level datasets from financial budget applications, have allowed

economists to estimate the marginal propensity to consume (MPC) out of income shocks quite

swiftly during the pandemic (Baker and Kueng 2021, Vavra 2021). The available evidence points

to households mostly saving or paying down debt when receiving a one-off payment during the

pandemic (Armantier et al. 2020, 2021, Christelis et al. 2020, Coibion et al. 2020, Crossley et al.

2020, Cox et al. 2020, Baker et al. 2021). This suggests that typical untargeted direct trans-

fers to the average household may not stimulate aggregate demand as much during pandemics

as during other crises. In this context, understanding which household characteristics predict

a higher MPC out of income windfalls is of key importance for the design of fiscal support

measures.

Research during the pandemic has found that the MPC out of positive income shocks is

largest for low-income and liquidity-constrained households, and for households who suffered

greater income falls relative to their pre-pandemic income (Armantier et al. 2020, 2021, Coibion

et al. 2020, Cox et al. 2020, Baker et al. 2021). There is, however, less empirical evidence and

consensus about the link between household expectations and the propensity to consume during

the pandemic. On the one hand, precautionary savings models predict that households who are

more concerned about their future financial situation tend to have lower MPCs, so as to build up

savings to mitigate future negative income shocks (Aiyagari 1994, Jappelli and Pistaferri 2014).

For example, Baker et al. (2021) find that households who expect employment losses and benefit

cuts display a smaller MPC out of US stimulus checks. Christelis et al. (2020) find similar results

for a sample of six euro area countries, indicating that households who are concerned about the

effects of COVID-19 on their finances tend to exhibit smaller MPCs for durable goods out of

a hypothetical unexpected bonus payment. On the other hand, using a large-scale survey from

the Nielsen Homescan panel, Coibion et al. (2020) find little role for individuals’ macroeconomic

expectations – future unemployment rate, inflation rates, and mortgage rates – in explaining

differences in MPCs. And some evidence for the UK suggests that individuals who expect their

financial situation to worsen or a job loss in the next three months actually report a higher MPC

out of an hypothetical positive transfer (Crossley et al. 2020). We believe that the fast-changing

nature of the pandemic, amid an environment of heightened economic and health uncertainty,

1
©International Monetary Fund. Not for Redistribution
warrants further study of the link between household financial concerns and their propensity to

consume. This paper aims to help fill that gap in the literature.

Against this background, we use survey data for UK households to study how household

expectations about their future financial situation may affect their short-term MPC out of

positive income shocks. We focus on the MPCs elicited directly from a representative sample of

the UK population in the COVID-19 special surveys of the Understanding Society (Institute for

Social and Economic Research 2020). The MPC is extracted from a question that asks survey

participants how much they would spend if they were to receive a one-off £500 payment. This

is a similar question to that asked in previous surveys to US households (Fuster et al. 2021).

We build a balanced panel of households that replied to the MPC question in surveys carried

out in July 2020, November 2020, and March 2021. We find that the average elicited MPC

across surveys stands at only 11%.

Our paper contributes to the literature on the MPC out of unexpected and transitory positive

income shocks elicited from household responses to survey questions (Sahm et al. 2010, Jappelli

and Pistaferri 2014, 2020, Bunn et al. 2018, Christelis et al. 2019, 2020, Armantier et al. 2020,

2021, Andreolli and Surico 2021, Fuster et al. 2021). These MPCs extacted from ‘reported

responses’ have been shown to be reasonably similar to the ‘revealed preference’ approach based

on actual data (Parker and Souleles 2019). Moreover, we contribute to the literature studying

the link between household expectations and the willingness to consume during the pandemic

(Christelis et al. 2020, Crossley et al. 2020, Baker et al. 2021).

Our main measure of household expectations, or financial concerns, is the self-reported

probability of not being able to pay usual bills and expenses in the next three months. We define

financially concerned households as those households whose probability of financial distress is

above the median in our sample. In our empirical exercises, we first study the characteristics of

households that are more concerned. We then focus our analysis on understanding the household

characteristics that best explain the heterogeneity in the elicited MPC. Our main interest will

be to investigate how financial concerns relate to decisions to spend out of the transfer of £500.

Let us state upfront that we focus on the first moment of expectations, i.e. how households’

perception of their ability to meet their spending commitments in the short term affects their

MPC. Unlike Coibion et al. (2021), our survey does not allow us to comment on the second

moment of expectations, i.e. the role of uncertainty around their responses.

Our main findings are as follows. First, we find that households who are concerned about not

2
©International Monetary Fund. Not for Redistribution
being able to pay their bills in the short term are more likely to be: already concerned about

their current financial situation; liquidity constrained; belong to low-income groups; renters

or mortgagors; younger, male, and ethnic minorities; furloughed; on benefits or employed in

high-contact industries. We also find a positive correlation between financial concerns and

health-related concerns due to COVID-19, in that a higher perception of being infected in the

next month is associated with higher financial concerns.

Second, using panel regressions with the elicited MPC as the dependent variable, our main

empirical finding suggests that financial concerns over the short term play a key role in ex-

plaining differences in MPCs across households during the pandemic. Our novel result shows

that financially concerned households have an MPC out of a transfer that is more than 20%

higher than households who are not concerned about their ability to pay their usual bills in

three months. This result remains robust to adding several household-specific characteristics,

and to controlling for current financial difficulties, so as to tease out the role of expectations

about future financial difficulties. In addition, we explore qualitative questions from the survey

which asked individuals what they would do with the amount they would not spend. We show

that concerned households are more likely to rebuild their balance sheets by paying off more

debt, but they are less likely to save, and receive less financial help.

Third, we show which other household characteristics are also associated with larger MPCs:

being a mortgagor, which underscores the role of illiquid assets for MPCs out of transitory

income gains (Kaplan et al. 2014, Cloyne and Surico 2017); younger age groups relative to

older groups (Jappelli and Pistaferri 2014, Fagereng et al. 2021); self-employed relative to non-

working households; and households with more children. By contrast, we find no evidence that

liquidity constraints, once we control for financial concerns and other covariates, are important

in explaining differences in MPCs across households. Although we cannot rule out that some

households characteristics are correlated, a possible reason why the relationship between liq-

uidity and MPCs might have broken down during the pandemic may be linked to the massive

build up in household savings across the population. This provides further evidence that it

is the expectation of financial stress rather than liquidity constraints that may be driving our

main result that financially concerned households display a larger MPC.

Fourth, we run alternative specifications to shed more light on the reasons why households

who are more concerned about not meeting their future spending commitments are associated

with a larger MPC. However, we do not find evidence that potential drivers including past

3
©International Monetary Fund. Not for Redistribution
spending cuts, negative income shocks, borrowing constraints, or the labour market situation

explain why financially concerned households have larger MPCs. We could only find some

tentative evidence that part of our result may be driven by different shares of discretionary

spending and a general reliance on benefits, but this is unlikely to play a large role.

Fifth, we explore non-linearities in our financial concerns variable by making use of the

fact that it is a probability bounded between 0% and 100%. Among those households who

report having a non-zero probability of being in financial distress, we find that households who

are moderately concerned and appear uncertain, in the 1%-50% range, are driving our main

results. This goes against the precautionary savings literature and suggests that, as long as the

subjective probability of being in financial distress in the future is not that large, households will

tend to spend a larger fraction of the income windfall than all other households. By contrast,

households who are certain they will not be able to pay they bills (100% probability) display

the smallest MPC; these households save a larger fraction of the transfer to prepare for more

challenging times ahead.

We subject our baseline results to a battery of robustness checks. We find that our main

finding that financially concerned households are associated with a higher MPC is robust to:

(i) alternative measures of financial concerns; (ii) controlling for health-related concerns; (iii)

controlling for several specific shocks at the regional, local, and industry level; (iv) Tobit and

Probit models; (v) and to small changes to the design of the MPC question.

We offer several alternative explanations for our novel result. One possibility is that financial

concerns reflect households’ perceptions of the nature of the COVID shock. Alternatively, our

results might be better rationalised through behavioural models of consumption, by placing a

focus on prospect theory and mental accounting (Duxbury et al. 2005, Milkman and Beshears

2009, Kahneman and Tversky 2013). For instance, households that expect to be in financial

difficulty in the near future may choose to overconsume now because they are less sensitive to

news about future consumption than to news about current consumption. Mental accounting

may also play a role in determining MPCs. People compartmentalise income and spending

into different mental ‘accounts’, such as ‘current income’, ‘current assets’ and ‘future income’

(Shefrin and Thaler 1988, Duxbury et al. 2005, Milkman and Beshears 2009, Baugh et al. 2021).

Households may use budgets within these accounts to facilitate making trade-offs between com-

peting uses for funds and to act as a self-control device. Financially concerned households might

be more likely to ‘budget’ and treat funds within each tagged mental account as distinct and

4
©International Monetary Fund. Not for Redistribution
imperfectly substitutable. That might explain why they are more likely to increase their spend-

ing in response to small windfalls. Relatedly, it is also possible that preference heterogeneity

across households, self-control issues and low financial sophistication, which tend to correlate

with low liquid assets, low savings and a higher probability of becoming financially distressed,

may lead more concerned households towards over-consumption (Laibson 1998, Aguiar et al.

2020, Jørring 2020, Vihriälä 2021). We also cannot rule out that there may have been some

ambiguity in the interpretation of the MPC question, particularly that some households may

have had trouble in distinguishing what constituted ‘spending’ and ‘paying off debt’ (Sahm

et al. 2010).

Our results also tentatively support findings that households who expect not to be able to

make ends meet have larger consumption responses to negative income shocks. Although we

cannot compute an MPC from a scenario of negative income shocks – and thus not compare it

with the elicited MPC from a hypothetical transfer – our findings highlight the role that house-

hold expectations play in determining the responsiveness of consumption to both hypothetical

expansionary and contractionary fiscal policies.

The rest of the paper is organised as follows. In Section 2 we describe the survey data

and show descriptive statistics. In Section 3 we run probit panel regressions to study the

characteristics that predict which households are more financially concerned. Section 4 presents

our main results where we uncover the relationship between the elicited MPCs and financial

concerns. In Section 5 we test whether financially concerned households are more likely to

cut consumption in the face of negative shocks relative to unconcerned households. Section 6

concludes the paper.

2 The Understanding Society COVID-19 survey

2.1 Background on survey and elicited MPC

In this paper we exploit granular data collected in the Understanding Society COVID-19 Study,

henceforth COVID survey (Institute for Social and Economic Research 2020). The COVID-19

survey is a new component of Understanding Society: the UK Household Longitudinal Study,

henceforth Main survey.1 Understanding Society is the UK’s main longitudinal Household
1
Understanding Society is built on the British Household Panel Survey (BHPS) which ran from 1991-2008
and included around 10,000 households. Understanding Society includes around 8,000 of the original BHPS
households.

5
©International Monetary Fund. Not for Redistribution
Survey. From April 2020 participants from the Main survey were asked to complete the more

frequent COVID-19 survey to capture experiences during the pandemic.

The participants in the COVID survey are a subset of those in the Main survey which enables

us to link data across the two surveys on the same respondent. This is useful because the Main

survey includes important background information on participants and their households, such

as their balance sheet positions on the eve of the pandemic. Table A.1 in Appendix A contains

the full list of variables used in this paper.

The COVID survey was conducted at an individual level, whereas the Main survey followed

both individuals and households. We conduct the analysis at an individual level and treat

household variables as attributes of individuals. For the sake of simplicity, and for comparability

with other studies, we will refer to households as a loose definition for individuals. To create our

dataset we merge the individual and household Main surveys from Waves 8, 9 and 10 and then

link this to the COVID survey. Most of our pre-COVID covariates are from the most recent

responses to either Wave 9 or 10. We exclude the year 2020 to avoid that our pre-COVID

variables are polluted with information from the pandemic.

We use weights in our data to adjust for unequal selection probabilities, differential non-

response, and potential sampling error Institute for Social and Economic Research (2020).

Specifically, we use the inverse-probability weights provided with the COVID survey, which

correct Main survey weights for non-response. Of the eight available waves of the COVID survey

(from April 2020 to March 2021) we focus on the fourth, sixth and eight waves conducted in

July 2020, November 2020, and March 2021 as these included questions on MPCs.2 Over 13,000

individuals have responded to at least one of the COVID surveys. Our sample is smaller as

we must drop some households from the dataset. First, we drop all households who were not

present in at least one of the two most recent Main survey Waves (Waves 9 or 10), so that

we can extract important pre-COVID information. This effectively means that we only keep

households with non-zero survey weights. Second, we drop all households who did not respond

to all three of the aforementioned COVID surveys. Our final sample is comprised of a balanced

panel of 7,313 individuals, totalling 21,939 observations for the three surveys.

Our variable of interest, the MPC, is extracted from a series of questions asking households

what they would do over the next three months if they were to receive a one-time hypothetical

2
The July survey was carried out between 24-31 July, the November 2020 survey between 24 November and
1 December, and the March 2021 survey between 24-31 March.

6
©International Monetary Fund. Not for Redistribution
transfer of £500. These questions were adapted from a survey conducted by the Federal Reserve

Bank of New York (Fuster et al. 2021). More specifically, survey participants were asked the

following:

Now consider a hypothetical situation where you unexpectedly receive a one-time

payment of £500 today. We would like to know whether this extra income would

cause you to change your spending, borrowing and saving behaviour in any way over

the next 3 months.

1. Over the next 3 months, I would spend more than if I hadn’t received the £500

2. Over the next 3 months, I would spend the same as if I hadn’t received the £500

3. Over the next 3 months, I would spend less than if I hadn’t received the £500

A follow-up question was then asked to those that replied that they would increase or

decrease spending:

You indicated that you would [increase/decrease] your spending/donations over the

next 3 months following the receipt of the £500 payment. How much [more/less]

would you spend than if you hadn’t received the £500?

[Numeric textbox] Pounds

If the quantitative questions indicate that the respondent would not spend all of the £500,

respondents would answer a follow-up question about what they would do with the amount

they would not spend:

You have indicated that you would not spend all of the £500 payment. What would

you do with the amount that you do not spend?

Please select all that apply.

1. Over the next 3 months, I would pay off more debt (or borrow less) than if I

hadn’t received the £500

2. Over the next 3 months, I would save more than if I hadn’t received the £500

3. Over the next 3 months, I would receive less financial help from friends or family

than if I hadn’t received the £500

4. Over the next 3 months, I would give more financial help to friends or family

than if I hadn’t received the the £500

7
©International Monetary Fund. Not for Redistribution
5. Other

In the November 2020 and March 2021 waves respondents were split randomly into two

groups. The first group were asked the July 2020 question and the second group were asked

a very similar question, but where the wording is altered slightly so that the government un-

expectedly gives everyone a one-time payment of £500. The same follow-up questions were

asked.

The elicited MPCs from the above hypothetical questions, also known as the direct survey

approach, has some advantages over other techniques in computing the MPC (Jappelli and

Pistaferri 2014, Bunn et al. 2018, Fuster et al. 2021). For example, under the assumption

that there is no systematic discrepancy between the intention to spend and actual spending, it

overcomes the usual econometric identification problems by isolating an exogenous temporary

shock to income without the need of distributional assumptions.3 And, as it provides the MPC

for each household, it also allows for a more detailed analysis of population sub-groups. The

COVID survey also has an advantage to other surveys which only provide qualitative responses

about their willingness to spend (Shapiro and Slemrod 2003, 2009).

2.2 Descriptive statistics

Figure 1 combines the responses to both the personal and public windfall question to show the

fraction of households who would spend more, spend the same or spend less upon receiving the

£500 transfer. We find that around 78% of households would not change their spending at all

in response to a one-time payment of £500. Around 18% would spend more, whereas roughly

4% would spend less. The responses are relatively stable across the three surveys.

We now extract the implicit MPC for all households following the £500 payment. More

specifically, we compute the individual’s MPC as the reported pound consumption change di-

vided by £500. The MPCs vary between zero and one, so that households who reported they

would spend less or the same are re-coded as having an MPC of zero. We can see that the av-

erage MPC across our balanced sample is 0.11, with very small differences across waves (Table

1). This is in line with Crossley et al. (2020) but low even in comparison to other estimates

from the pandemic. For instance, the average MPC out of stimuls payments made as part of

the CARES ACT in the United States is estimated to range between 25% and 40% (Armantier
3
While stated responses may differ from actual spending responses, Parker and Souleles (2019) compare self-
reported spending responses with revealed-preference estimates and find the former is highly predictive of the
latter. They also find that the two measures imply similar average MPCs.

8
©International Monetary Fund. Not for Redistribution
et al. 2020, Chetty et al. 2020, Coibion et al. 2020, Baker et al. 2021). This falls in the lower

range of previous empirical evidence on the MPCs out of government transfers in 2001 and 2008

(Shapiro and Slemrod 2003, 2009, Parker et al. 2013). Moreover, Armantier et al. (2021) find

that MPCs averaged 29% in April 2020, decreasing slightly over time with additional policy

support measures.4 For euro area households, Christelis et al. (2020) point to an average MPC

of 18% out of a hypothetical unexpected bonus of e3,000.

Figure 1: Households’ response to a hypothetical payment of £500

17.4

Spend more 18.9

18.7

78.0

Spend the same 77.1

77.8

4.6

Spend less 4.0

3.5

0 20 40 60 80
Percent

July 20 Nov 20
March 21

The low MPC reflects to a large extent the fact that roughly 85% of the respondents would

save the full £500 transfer. This implies that for an additional £1 of income, households would,

on average, increase their spending by only £0.11. The table also shows the distribution of

MPCs based on the percentage of households who fall into each MPC group. Two additional

results stand out. First, the second most common reported MPC is one, at around 7-8%.

Second, there is limited variability in the distribution of MPCs across time. At face value, this

is somewhat surprising given that the surveys capture distinct phases of the pandemic, especially

the stringency of the restrictions on spending and social activities.5 The low MPCs today may
4
The average MPC from the three rounds of stimulus checks declines from 29% in April 2020, to 26% in
December 2020, and to 25% in March 2021.
5
In July 2020, the UK had come out of a national lockdown, many social-distancing restrictions had been
eased, and the number of COVID-19 cases and deaths had fallen substantially. In November 2020, the sharp
rise in the number of cases and deaths led to a second national lockdown, with non-essential stores closed, and

9
©International Monetary Fund. Not for Redistribution
therefore not be fully accounted for by the constraints faced by households in allocating resources

to discretionary spending, such as restaurants, travel, and accommodation, in a context of social-

distancing and restriction measures.

Table 1: Summary statistics for MPCs

N Mean SD Zero (0-0.25) [0.25-0.5) [0.5-0.75) [0.75-1) One
Jul-20 7,313 0.11 0.28 84.82 1.57 2.02 3.60 0.74 7.25
Nov-20 7,313 0.11 0.29 84.70 1.73 1.76 2.97 0.99 7.86
Mar-21 7,313 0.12 0.30 84.50 1.58 1.65 3.08 1.03 8.15
Full sample 21,939 0.11 0.29 84.67 1.63 1.81 3.21 0.92 7.76
Notes: Shares calculated as percentages of total MPC sample.

Households who had an MPC smaller than one were asked a follow-up question about what

they would do with the £500 that would not be spent. Table 2 shows that the majority of

households would save more. Saving seems to have become more popular over time, increasing

to 75% in the March 2021 survey, from 69% in July 2020. Roughly 20% of individuals would

pay off more debt, a share that remained relatively stable across surveys. Finally, around 9%

of individuals who would not spend all of the £500 would give more financial help.

Table 2: Use of amount not spent

Pay off more debt Save more Receive less fin. help Give more fin. help
Jul-20 20.55 68.64 1.94 9.15
Nov-20 22.44 67.51 1.57 8.49
Mar-21 20.14 74.49 1.49 8.65
Full sample 21.05 70.18 1.67 8.76
Notes: Shares sum to over 100% as respondents could give more than one response.

In the remainder of this section we look at univariate correlations between the elicited MPC

and population subgroups. We first link the COVID surveys to the Main survey to extract

pre-crisis balance sheet information, such as mortgage debt, and cash-on-hand, which play an

important role in life-cycle models. We follow Jappelli and Pistaferri (2014) and measure cash-

on-hand with the sum of household disposable income and liquid savings net of unsecured debt,

all taken from the Main survey. Table A.2 in Appendix A shows that households in the first two

quintiles of cash-on-hand tend to have a slightly higher MPC, in line with the literature that

finds a negative relation between MPC and liquidity (Parker et al. 2013, Jappelli and Pistaferri

2014, Kaplan et al. 2014, Bunn et al. 2018, Christelis et al. 2020, Coibion et al. 2020, Baker

et al. 2021, Fagereng et al. 2021).

discretionary spending restricted. The last available survey in March 2021 marked a period of a gradual easing in
restrictions, with children returning to school, alongside a substantial fall in the number of infections and deaths.

10
©International Monetary Fund. Not for Redistribution
On housing tenure, we show that mortgagors are associated with a larger MPC than renters

or outright owners, highlighting the role of illiquid assets for MPCs out of transitory income

gains (Kaplan et al. 2014, Cloyne and Surico 2017). For instance, Cloyne and Surico (2017) show

that households with mortgage debt exhibit large and significant consumption responses to tax

changes in the UK, whereas homeowners without a mortgage do not adjust their expenditure.

We also show that mortgagors are more likely to report an MPC of one. When we condition on

the level of mortgage debt, we find that individuals who fall in the top quintile of the loan-to-

income (LTI) ratio distribution are associated with larger MPCs, compared to the first quintile

(Kaplan et al. 2014, Misra and Surico 2014, Baker and Yannelis 2017, Cloyne and Surico 2017,

Bunn et al. 2018, Kovacs et al. 2018).

MPCs also differ across labour market sub-groups. Both employees and the self-employed

tend to have higher MPCs than households who are not working, such as those who are retired.

Moreover, we also find that younger age groups are associated with higher MPCs relative to

older groups (Jappelli and Pistaferri 2014, Fagereng et al. 2021).

Households who had experienced a fall in spending over the previous four weeks compared

to the same period the year before exhibited larger MPCs than households who had not, hinting

at some spending normalisation. Conversely, households whose earnings had fallen by 25% or

more since the start of the pandemic had smaller MPCs.

Finally, we check how the MPC varies according to COVID-19 health-related concerns:

In your view, how likely is it that you will contract COVID-19 in the next month?

1. Very likely

2. Likely

3. Unlikely

4. Very unlikely

When we transform this question into a binary variable – very likely plus likely, against

very unlikely plus unlikely – we find that both groups are very similar, suggesting that health

concerns do not directly affect willingness to spend. It is possible that households concerned

that they would contract COVID in the near future might have had an incentive to spend more

on some goods and services, such as taxis, supermarket supplies, and medical supplies. But,

on the other hand, the precautionary saving motive would work in the opposite direction, with

concerned households saving more as a buffer.

11
©International Monetary Fund. Not for Redistribution
2.3 Financial concerns during the pandemic

In this section we shift our analysis to the MPC across several measures of household expec-

tations. These expectations relate to households’ financial situation in the next three months,

which aligns with the same time horizon as the MPC question. Our first measure of financial

concerns deals with the subjective perception of being able to pay bills:

On a scale of 0-100% how likely do you think it is that you will have difficulty paying

your usual bills and expenses in the next three months?

[Numeric textbox] %

The second measure looks at the individual’ subjective financial situation:

Looking ahead, how do you think you will be financially 3 months from now, will

you be...6

1. Better off

2. Worse off than you are now

3. Or about the same?

The third measure was only asked to those who were working at the time of the survey:

On a scale of 0-100% how likely do you think it is that you will lose your job or shut

your business in the next three months?

[Numeric textbox] %

Table A.2 in Appendix A shows that financially concerned households are associated with

a larger MPC than unconcerned households, regardless of the survey question we use. For

instance, our main baseline measure in this paper – the probability of not being able to pay bills

in the next three months – indicates that those individuals who assign a non-zero probability to

this event occurring have an unconditional MPC of 0.15, compared with 0.10 for the unconcerned

individuals. In Table A.3 we inspect some of the characteristics of these households in an

univariate setting. Concerned households are more likely to be finding their current situation

difficult and to expect their financial situation to be worse in the short-term. They also assign

a much higher probability to losing their jobs in three months. Balance sheet positions of

6
The July 2020 survey focused on a month from now, instead of three months.

12
©International Monetary Fund. Not for Redistribution
concerned households also appear to differ from those who are not concerned. For instance,

they tend to have substantially lower cash-on-hand, a sign of liquidity constraints, and are

more likely to be mortgagors. The average concerned household also tends to be about ten

years younger than unconcerned households.

In the following sections we investigate more formally, in a multi-variate setting, the charac-

teristics of financially concerned households, and how financial concerns may affect the reported

propensity to consume.

3 What determines financial concerns?

We turn to the analysis of the characteristics that correlate with financial concerns by running

several probit panel regressions across the three surveys, July 2020, November 2020, and March

2021. We transform our financial concerns variable – the probability of not being able to pay bills

in the next three months – into a binary variable, assuming the value of one if the household’s

expected probability of financial distress is above the median in the sample, and zero otherwise.

Given that the median value is zero, we are in effect comparing those that assign a non-zero

probability to this event occurring to those that assign a zero probability.7 Figure 2 shows the

probability distribution for financially concerned households i.e. those who assign a non-zero

probability. We can see that the distribution is skewed to the left, with the mean at around 27%

and the median at 20%. Most of the responses fall between 0-30%, with then a slight increase

in the frequency at around 50%.

We use a large set of household characteristics in our probit model. The socio-demographic

variables refer to: binary variables for households belonging to the age groups 18-39 and to 40-64

(65 and older is the omitted group), whether the individual is male, the number of children, the

household size, and whether the individual identifies as being from the White ethnic group. The

financial characteristics of households are captured by: the housing tenure status, i.e. mortgagor

or renter (outright owner is the omitted category), the logarithm of pre-crisis household income

and quintiles of the cash-on-hand ratio, both from the Main survey. We capture current and

future perceived financial subjectivity with the two variables on financial situation described

in the previous section (‘Finances now’, and ‘Finance future’). We transform them into binary

variables that assume the value of one for households who report finding it difficult or very

7
Our headline results are robust to using the mean sample value of 8.74, instead of the median, as the
threshold for the financial concerns variable.

13
©International Monetary Fund. Not for Redistribution
difficult to manage financially these days (‘Finances now’), and the value of one for households

who believe they will be worse off financially in three months’ time (‘Finance future’).

Figure 2: Distribution of financial concerns

Percent

0
0 20 40 60 80 100
Prob not able to pay usual bills/expenses in next 3 months

We also add the individual’s employment situation with a binary variable for being an em-

ployee or self-employed (relative to not working), a dummy indicating whether an individual

has been furloughed,8 two dummies indicating whether an individual started receiving Uni-

versal Credit prior to the pandemic (including those who continued to receive it during the

pandemic) or only applied for and started receiving Universal Credit since the pandemic be-

gan,9 and another dummy that captures whether the individual works in a high-contact industry

(accommodation and food service activities, administrative and support service activities, and

arts, entertainment and recreation).10 Finally, we add two variables that proxy health concerns

related to COVID-19: a dummy variable that assigns the value of one to individuals who think
8
Our furloughed dummy is equal to one if someone is currently furloughed or has been furloughed at some
point in the past.
9
Universal Credit (UC) is a means-tested benefit which is available to those who are in work but on low
incomes, as well as to those who are unemployed or whose capability for work is limited by sickness or disability.
In the early stages of the pandemic there was a sharp increase in UC claims. The total number of people on
Universal Credit in Great Britain surged from 3 million in March 2020 to 5.2 million in May, gradually increasing
to 5.8 million as of November 2020.
10
We define high-contact industries as the three industries that reported the largest impact on
sales/employment between 2020Q2 and 2021Q1 relative to what would have otherwise happened. These statistics
come from responses to the Decision Maker Panel (DMP), one of the main representative surveys of businesses
in the UK.

14
©International Monetary Fund. Not for Redistribution
it is likely or very likely that they will get infected in the next month (‘Risk of infection’), and

a dummy variable with the value of one for households who are sure, or think that it is very

likely they have had COVID-19 (‘Had Covid-19’).

We report the Probit results in Table 3.11 In column (1) we find that individuals who are

having difficulties in managing their finances, and who believe they will be worse off in three

months’ time, are more likely to be concerned about their ability to pay their bills in the next

three months. Liquidity constraints are a strong predictor of households’ financial concerns:

being in the bottom quintile of the cash-on-hand ratio increases the conditional probability

of becoming financially concerned about the future by 11 p.p. relative to the top quintile.

Along the same lines, income is negatively associated with the probability of being financially

concerned, pointing to the important role that liquidity and income play in determining financial

expectations. Our findings align well with those in Christelis et al. (2020), who study euro area

households’ concerns about the impact of COVID-19 on their financial situation.

Renters and mortgagors are more likely to be financially concerned relative to outright own-

ers. We rationalise this result with the fact that renters tend to be more financially vulnerable

and liquidity constrained. For mortgagors, the reason may be related to the large share that

mortgage payments typically play in mortgagors’ spending commitments, implying less financial

flexibility when hit by a negative income shock (e.g. unemployment).

Of the socio-demographic variables, individuals identifying as male, and those belonging

to the 18-39 age group tend to be more financially concerned, with those 65 and above being

the least concerned. Ethnicity also plays a role; being White reduces the likelihood of being

financially concerned. This is in line with evidence that the UK’s minority ethnic groups have

been disproportionately affected by COVID (Platt and Warwick 2020). It is also worth noting

that individuals perceived higher levels of financial distress in July 2020. This is somewhat

surprising, as several restrictions had been eased around that time, while the UK was under a

second national lockdown in November 2020. One of the explanations could be related to the

prevailing uncertainty about the continuation of income support schemes into the Autumn 2020.

In addition, uncertainty about how the pandemic would evolve, and the absence of vaccines at

that time could have arguably made people more uncertain about their short-term financial

situation.

11
We apply the delta method in the Probit model to convert the coefficients into marginal effects of a given
variable on the conditional probability of changing the dependent variable.

15
©International Monetary Fund. Not for Redistribution
Table 3: Probit model: determinants of expected financial concerns

(1) (2) (3) (4) (5) (6)

Finances now 0.361*** 0.361*** 0.357*** 0.347*** 0.345*** 0.371***
(0.034) (0.032) (0.032) (0.046) (0.033) (0.040)
Finance future 0.225*** 0.219*** 0.218*** 0.255*** 0.218*** 0.245***
(0.014) (0.014) (0.014) (0.021) (0.014) (0.020)
Cash-on-hand Q1 0.106*** 0.103*** 0.108*** 0.068* 0.100*** 0.112***
(0.024) (0.024) (0.024) (0.040) (0.024) (0.030)
Cash-on-hand Q2 0.079*** 0.078*** 0.080*** 0.037 0.076*** 0.086***
(0.024) (0.024) (0.023) (0.042) (0.024) (0.032)
Cash-on-hand Q3 0.052** 0.047** 0.051** 0.006 0.047** 0.055*
(0.023) (0.023) (0.023) (0.041) (0.023) (0.030)
Cash-on-hand Q4 0.058*** 0.058** 0.059*** 0.042 0.058*** 0.076**
(0.022) (0.022) (0.022) (0.041) (0.022) (0.031)
Mortgagor 0.094*** 0.095*** 0.100*** 0.092*** 0.098*** 0.086***
(0.016) (0.015) (0.016) (0.022) (0.015) (0.020)
Renter 0.150*** 0.150*** 0.152*** 0.138*** 0.141*** 0.122***
(0.020) (0.020) (0.019) (0.030) (0.020) (0.027)
Log HH income -0.030** -0.031*** -0.027** -0.010 -0.026** -0.020
(0.012) (0.011) (0.011) (0.017) (0.011) (0.016)
Age 18-39 0.133*** 0.125*** 0.143*** 0.098** 0.117*** 0.084**
(0.021) (0.020) (0.023) (0.040) (0.020) (0.035)
Age 40-64 0.080*** 0.075*** 0.090*** 0.047 0.067*** 0.031
(0.016) (0.016) (0.019) (0.037) (0.015) (0.031)
Male 0.030** 0.030** 0.030** 0.041** 0.030** 0.037**
(0.012) (0.012) (0.012) (0.018) (0.012) (0.016)
No. children 0.008 0.007 0.008 0.028* 0.006 0.021
(0.011) (0.011) (0.011) (0.015) (0.011) (0.014)
Household size 0.014* 0.015* 0.014* 0.002 0.016* 0.011
(0.008) (0.008) (0.008) (0.011) (0.008) (0.010)
White -0.091*** -0.093*** -0.089*** -0.112*** -0.089*** -0.114***
(0.024) (0.024) (0.024) (0.033) (0.024) (0.030)
Jul20 survey 0.047*** 0.048*** 0.048*** 0.038*** 0.050*** 0.057***
(0.008) (0.007) (0.007) (0.012) (0.007) (0.010)
Nov20 survey -0.003 -0.007 -0.007 -0.001 -0.006 0.013
(0.008) (0.008) (0.008) (0.012) (0.008) (0.011)
Risk of infection 0.097*** 0.102*** 0.133*** 0.098*** 0.128***
(0.022) (0.022) (0.030) (0.022) (0.028)
Had Covid-19 -0.091 -0.088 -0.247* -0.094 -0.219**
(0.097) (0.093) (0.129) (0.096) (0.107)
Employee -0.033* -0.023
(0.017) (0.022)
Self-employed 0.000
(0.022)
Furloughed 0.077***
(0.021)
UC pre-Covid 0.128***
(0.046)
UC post-Covid 0.093**
(0.043)
High-contact 0.063***
(0.024)
Observations 21,932 21,801 21,801 9,679 21,801 11,391
Notes: Marginal effects of probit estimates at the individual level computed with the delta
method. The dependent variable is the financial concerns dummy, taking the value of one if
households’ concerns about not being able to pay their bills in three months is above the median
sample value, and zero otherwise. Standard errors in parentheses clustered at the individual
level. Constant is not reported. Asterisks, *, **, and ***, denote statistical significance at the
10%, 5%, and 1% levels.

16
©International Monetary Fund. Not for Redistribution
Health concerns also play an important role in the perception of households about being able

to meet their financial commitments. In column (2) we find that the perception of contracting

COVID-19 in the next month is associated with an increase of 10 p.p. in the probability of being

more concerned about not being able to make ends meet. But there is some indication that

individuals become less financially concerned once they think that they have had COVID-19,

suggesting a link between health and financial concerns (columns 4 and 5).

In column (3) we find that employees are less likely to be financially concerned than in-

dividuals who are not working. In column (4) we note that furloughed individuals are much

more likely to be financially concerned. The sample, however, drops by around half since the

furloughed dummy can only be computed for employed individuals. In column (5) individuals

who have received Universal Credit, either before or during the pandemic, are also shown to

be more likely to be financially concerned. Finally, the last column focuses only on individuals

who are working. We find that individuals who work in high-contact industries are more likely

to be financially concerned. This result points to the disproportionate effect that the pandemic

has had on industries that rely mostly on face-to-face contact.

4 Financial concerns and the MPC: Panel regressions

This section constitutes the centre stage of our analysis. We use the probability of not being

able to pay bills in the next three months as the main measure of financial concerns. We shed

light on which household characteristics best explain the heterogeneity in the elicited MPC,

focusing on how financial concerns relate to decisions to spend out of the transfer of £500.

4.1 Main results

We run several panel regressions to uncover the heterogeneity in MPCs during the pandemic

across a representative sample of UK individuals. Our dependent variable is the elicited MPC

from the survey questions explored in previous sections. The MPC varies between zero (not

spend anything) and one (spend the full £500 over the next three months).

We model the unobserved heterogeneity with random effects rather than with fixed effects

for the following reasons. First, several important covariates are only available in the Main

(pre-COVID) survey. This means that the fixed-effects model would not be able to estimate

how time-invariant variables, such as cash-on-hand, and household income, correlate with the

17
©International Monetary Fund. Not for Redistribution
MPCs.12 In addition, the same would happen with other variables, such as gender, race, and

the number of children. Second, there is little time variation within households in some of the

key variables, particularly the financial concerns variable. This makes the (within) fixed-effect

estimator inefficient in estimating the effect of variables that have very little within variation.

In addition, demeaning can be problematic for estimating fixed effects when the number of time

periods is small.

We report our main specifications in Table 4. In column (1) we find that our financial con-

cerns variable is positively associated with the elicited MPC, and highly statistically significant.

Individuals who are more concerned than the the median individual in the sample about not

being able to meet their financial commitments are associated with a larger MPC of 3.2 p.p over

the next three months. In our preferred specification, in column (2), we include the same set of

household characteristics as in Table 3. We also control for current financial difficulties – finding

it difficult or very difficult to manage financially – so as to tease out the role of expectations

about future financial difficulties. Our coefficient of interest declines a bit to 0.023, but remains

highly statistically significant. This suggests that financial concerns matter for explaining the

heterogeneity in the MPC across individuals beyond the effect of their current financial sit-

uation, and of other household-specific characteristics. While the economic magnitude seems

small at face value, we note that the average MPC is only 11%. This means that financially

concerned households have an MPC that is more than 20% higher than the sample average.13

In column (3) we do not find any evidence that individuals who are finding their current

situation difficult and are more financially concerned have larger MPCs. This contrasts with

Crossley et al. (2020) who, using data from the first COVID survey in July 2020, find evidence

that the highest MPCs, when nothing else is controlled for, are for those who are currently

finding it difficult and expect things to deteriorate further. In column (4) we test the robustness

of our findings to controlling for the expectation that the government will be supporting private

sector wages in eight months’ time.14 Our previous results remain qualitatively unchanged.

Moreover, we use two alternative measures of financial expectations that we have explored in
12
While the COVID surveys asked about household income and earnings, they are unfortunately not well
populated in every COVID wave.
13
In Table B.1 in Appendix B we run two probit models of this specification with the dependent variable
taking the value of one when the elicited MPC is zero (column 1), and taking the value on one when the elicited
MPC is one (column 2). We find that more concerned households tend to be less likely to report an MPC of
zero, while being more likely to report an MPC of one (fully spending the £500 income transfer.)
14
The question on the probability of wage support was not asked in the July 2020 survey. Individuals were
asked the following: ‘On a scale of 0-100%, what is the chance that the government will be supporting wages of
private sector workers, either directly or through payments to employers, in 8 months’ time (i.e. at the end of
November 2021)?(...) ’.

18
©International Monetary Fund. Not for Redistribution
Section 2.3. The first asks individuals about the probability of losing their job or shutting their

business in the next three months (Job security). To be in line with our main measure of financial

concerns, we construct a dummy equal to one if individuals’ expected probability of losing their

job is above the sample median. The sample size decreases by one-third as this question is

only asked to those currently in paid work or self-employed. The second indicator takes the

value of one for individuals who believe they will be worse off financially in three months’ time

(Finances future).15 Columns (5) and (6) indicate that our main finding – households who are

more financially concerned over the short term tend to have a larger MPC – is not driven by

the choice of the expectation variable.

In the rest of this section we look at the other household-specific covariates. Starting with

the cash-on-hand quintiles, we do not find that individuals with more limited household liquid

assets are associated with different MPCs relative to the top quintile. We actually find some

evidence that the individuals in the top cash-on-hand quintile tend to have larger MPCs than

individuals who fall in the third quintile. Along the same lines, we do not find that pre-COVID

household income is a statistically significant determinant of MPCs during the pandemic. In

models with rational (or near-rational) expectations, individuals who are close to their liquidity

constraint exhibit a significantly higher MPC out of income increases because these individuals

are unable to smooth consumption before they receive the income by borrowing or using savings

(Parker et al. 2013, Jappelli and Pistaferri 2014, Kaplan et al. 2014, Bunn et al. 2018, Christelis

et al. 2020, Coibion et al. 2020, Baker et al. 2021, Fagereng et al. 2021). But, on the other

hand, there is a strand of the literature that has found high MPCs for high-liquidity households

(Sahm et al. 2010, Kueng 2018, Olafsson and Pagel 2018, Baugh et al. 2021).16 The break in

the relationship between cash-on-hand and MPC during the pandemic may be linked to the

massive build up in household savings due to forced consumption cuts. In addition, a number

of policy measures, such as income support schemes, and the extension of universal credit, were

targeted at those more financially vulnerable. These measures may have alleviated the negative

income shock experienced by individuals at the bottom of the liquid asset distribution.

Across all specifications we find that mortgagors have larger MPCs than outright owners,

underscoring the role that illiquid assets have played in determining differences across households

in MPCs out of transitory income gains (Kaplan et al. 2014, Cloyne and Surico 2017).

15
The July survey had a shorter window of ‘a month from now’, instead of three months.
16
In behavioural models agents have a self-control problem (Laibson 1998) or consume according to a rule-of-
thumb or mental accounting (Campbell and Mankiw 1989, Duxbury et al. 2005, Milkman and Beshears 2009).
Ilut and Valchev (2020) argue that bounded rationality may explain this behaviour.

19
©International Monetary Fund. Not for Redistribution
Table 4: Baseline Random Effects regression

(1) (2) (3) (4) (5) (6) (7)

Financial concerns 0.032∗∗∗ 0.023∗∗∗ 0.023∗∗∗ 0.032∗∗∗ 0.024∗∗∗
(0.005) (0.005) (0.005) (0.006) (0.005)
Finances now -0.013 -0.009 0.016 -0.032∗∗ -0.009 -0.013
(0.011) (0.024) (0.014) (0.015) (0.011) (0.011)
Fin. conc.×Fin. now -0.005
(0.026)
Job security 0.017∗∗∗
(0.006)
Finances future 0.015∗∗
(0.007)
Cash-on-hand Q1 0.000 0.000 -0.008 -0.005 0.002 0.000
(0.010) (0.010) (0.012) (0.016) (0.010) (0.010)
Cash-on-hand Q2 -0.005 -0.005 -0.020∗ -0.012 -0.004 -0.007
(0.010) (0.010) (0.012) (0.017) (0.010) (0.010)
Cash-on-hand Q3 -0.019∗ -0.019∗ -0.031∗∗∗ -0.034∗∗ -0.019∗ -0.019∗
(0.010) (0.010) (0.011) (0.016) (0.010) (0.010)
Cash-on-hand Q4 0.001 0.001 -0.005 0.001 0.001 0.001
(0.010) (0.010) (0.011) (0.016) (0.010) (0.010)
Mortgagor 0.031∗∗∗ 0.031∗∗∗ 0.037∗∗∗ 0.034∗∗∗ 0.033∗∗∗ 0.032∗∗∗
(0.008) (0.008) (0.009) (0.010) (0.008) (0.008)
Renter 0.006 0.006 -0.005 0.004 0.010 0.005
(0.008) (0.008) (0.010) (0.012) (0.008) (0.008)
Log HH income -0.002 -0.002 -0.008 0.010 -0.002 -0.002
(0.005) (0.005) (0.006) (0.008) (0.005) (0.005)
Age 18-39 0.061∗∗∗ 0.061∗∗∗ 0.057∗∗∗ 0.052∗∗∗ 0.064∗∗∗ 0.062∗∗∗
(0.009) (0.009) (0.011) (0.016) (0.009) (0.009)
Age 40-64 0.046∗∗∗ 0.046∗∗∗ 0.046∗∗∗ 0.031∗∗ 0.048∗∗∗ 0.046∗∗∗
(0.007) (0.007) (0.008) (0.015) (0.007) (0.007)
Male 0.021∗∗∗ 0.021∗∗∗ 0.019∗∗∗ 0.025∗∗∗ 0.022∗∗∗ 0.021∗∗∗
(0.006) (0.006) (0.006) (0.008) (0.006) (0.006)
No. children 0.021∗∗∗ 0.021∗∗∗ 0.017∗∗∗ 0.025∗∗∗ 0.022∗∗∗ 0.022∗∗∗
(0.005) (0.005) (0.005) (0.006) (0.005) (0.005)
Household size -0.008∗∗∗ -0.008∗∗∗ -0.006∗ -0.007∗ -0.008∗∗∗ -0.009∗∗∗
(0.003) (0.003) (0.003) (0.004) (0.003) (0.003)
White 0.022∗ 0.022∗ 0.023∗ 0.007 0.020∗ 0.023∗∗
(0.012) (0.012) (0.013) (0.015) (0.012) (0.012)
Jul20 survey -0.009∗∗ -0.009∗∗ -0.009∗∗ -0.006 -0.007∗ -0.009∗∗
(0.004) (0.004) (0.004) (0.005) (0.004) (0.004)
Nov20 survey -0.004 -0.004 -0.004 -0.006 -0.005 -0.004 -0.003
(0.004) (0.004) (0.004) (0.004) (0.005) (0.004) (0.004)
Expect wage support -0.005
(0.005)
Risk of infection -0.010
(0.009)
Had Covid-19 0.006
(0.021)

N 21,939 21,939 21,939 14,251 11,732 21,932 21,808
Notes: Estimates from a random-effects model at the individual level, where the dependent variable is the
elicited MPC. Standard errors in parentheses clustered at the individual level. Constant is not reported.
Asterisks, *, **, and ***, denote statistical significance at the 10%, 5%, and 1% levels.

20
©International Monetary Fund. Not for Redistribution
As for age, our results are in line with empirical findings pointing to older age groups be-

ing associated with smaller MPCs, particularly after retirement (Jappelli and Pistaferri 2014,

Fagereng et al. 2021).17 We find those in the youngest age group (18-39) display the largest

MPCs compared to those aged 65 and over. This is also in line with health concerns during the

pandemic, which may have led older consumers to accumulate precautionary savings (Eichen-

baum et al. 2020). We also find that males, those of White ethnicity, and living in households

with more children are associated with higher MPCs.

In the last column, we investigate whether our main findings are driven by a possible omitted

variable bias related to COVID-19 health concerns. Column (7) shows that all of our baseline

findings remain intact after controlling for the likelihood of contracting COVID-19 in the next

month, and also to taking into account the possibility that households may already have had

COVID-19. This is in line with Christelis et al. (2020), who find that the effect of households’

concerns on MPCs operates mainly via financial, rather than COVID-19 health-related concerns.

We check the robustness of our results by controlling for several fixed effects in our pre-

ferred specification from column (2) in Table 4: region-specific (the 12 regions in the UK),

regional time-varying shocks, shocks specific to the industry where the household works, and

time-varying industry shocks. Finally, we run a Tobit Random-Effects model. Table B.2 in

Appendix B shows that our key findings are strongly robust to these alternative specifications.

In particular, financial concerns, mortgagors, younger age groups, males, and the number of

children within households, are positively associated with a larger MPC.

We also replicate the main results of our preferred specification in column (2) of Table 4

by running cross-sectional regressions for each survey in July 2020, November 2020, and March

2021. We also run another specification which adds the health-related variables of column (7).

We are interested in checking whether the link between financial concerns and the willingness

to spend displays any time variation, conditional on the time the survey was carried out. Table

B.3 in Appendix B shows little variation in the coefficients of interest over time, particularly on

financial concerns. Although there is some indication that the coefficient on financial concerns

is smaller in July 2020, it is not statistically different from those in November 2020 and March

2021. These results align with what we have seen in Table A.2 in Appendix B: the average

MPC remained roughly unchanged across the three waves. This offers some evidence that the

17
For instance, Fagereng et al. (2021) show that the age-earnings profile leads young households to consume
more out of windfall gains to smooth consumption over life, while the bequest motive induces old households to
save a larger share for their offspring.

21
©International Monetary Fund. Not for Redistribution
willingness to spend for the same household may remain constant over a short period of time,

even under fast-changing economic conditions.18 This result is also in line with the notion

that households did not change much their expectations about their own short-term financial

situation across surveys.

4.2 Do spending cuts, income shocks, and borrowing constraints explain the
link between financial concerns and the MPCs?

We explore several conjectures trying to understand why UK individuals who are more finan-

cially concerned exhibit a larger MPC: (i) reversion to the mean effects in consumption as large

negative income shocks may have made concerned households consume less during the pan-

demic; (ii) a smaller share of discretionary spending implies a larger MPC as more concerned

households increase consumption on essential goods; (iii) employment and industry effects; and

(iv) borrowing constraints.

First, individuals who expect financial stress might have experienced larger consumption

cuts, as suggested by probit estimates in columns (1) to (3) of Table B.4 in Appendix B. More

concerned households would then spend a larger fraction of a one-off transfer to restore their

consumption. Along the same lines, more concerned households may have also suffered larger

income shocks and thus it seems realistic that they had to cut back more on spending. These

two hypothesis hint at a possible reversion to the mean effect, as concerned households seek to

close the gap between actual spending and desired spending.

We make use of a survey question on actual spending changes compared to pre-COVID

spending patterns. Participants were asked to report how much they had changed their spending

during the current pandemic, compared to the same period in the previous year. The answers

are reported in ranges: increased over 25%, increased up to 25%, stayed the same, decreased up

to 25%, decreased over 25%.19 Unfortunately, the question on actual spending is not available

in the March 2021 wave, which leads the overall sample to fall by one-third. We expand our

18
In a different exercise, we find that our main results are not sensitive to changes in the survey design.
Specifically, half of individuals were randomly assigned in November 2020 and March 2021 to answer the same
MPC question, but with a qualification that the transfer would come from the government. The survey asked
‘(...) hypothetical situation where the government unexpectedly gives everyone a one-time payment of £500 today
(...)’. We find that the MPC is not sensitive to this small change in the way the question is framed, in line with
Sahm et al. (2010), who show that small changes in survey design do not affect survey responses.
19
The question is:‘Thinking about your household spending (e.g. on food and household goods, products and
services, on commuting, bills, etc.) but excluding housing costs (e.g. mortgage payments, rent). Over the last 4
weeks, has your household spending increased, decreased, or stayed the same relative to the same four week period
last year?’

22
©International Monetary Fund. Not for Redistribution
preferred specification in column (2) from Table 4 by adding different dummy variables for

actual spending decreases. Column (1) in Table 5 shows that households whose actual spending

had decreased over the last four weeks have a larger MPC. This is in line with the view that

households would spend a larger fraction of a hypothetical transfer to restore their consumption

levels. But the interaction term cancels out the average effect on spending decreases, telling

us that more concerned households are associated with the same MPC regardless of changes in

actual spending. When we split the spending decreases into large (over 25%) and moderate (up

to 25%) in column (2), we find that households who are not financially concerned tend to display

a larger MPC if they had to cut actual spending by a large amount. But we do not find that

households who are concerned and who experienced large consumption falls behave differently

than concerned households who did not suffer a negative income shock, or who actually managed

to increase consumption.20

We also control for income changes to investigate whether higher MPCs just reflect the need

to smooth consumption due to negative income shocks. In column (3) of Table 5 we include a

dummy reflecting any loss in household earnings between January/February 2020 and the time

of the survey. Note that the sample size decreases as only households where at least one person

is working and who answered both the baseline and current household earnings question are

included. There is some evidence that households that had income losses have a lower MPC,

probably due to precautionary reasons. In column (4) we split the earnings loss variable into

two – a dummy for a loss of less than 25% and a loss of more than 25% – and also interact

these with the financial concerns variable. We find it is the larger loss that is driving the result,

while there is no evidence that more concerned households who experienced an income loss had

a different MPC than those that did not have an income loss.21

Overall, we find that households who had decreased spending in the recent past have a larger

MPC relative to those that did not cut consumption during the pandemic – a spending nor-

malisation effect. But past spending decreases cannot explain why more concerned households

have a higher MPC than other households. Along the same lines, we find that negative income

20
When we use the binary variable for actual consumption decreases as the dependent variable, we find that
more concerned households are more likely to have experienced consumption cuts in the past four weeks relative
to the same period last year (columns 1 to 3 in Table B.4 in Appendix B). But as we have seen in Table 5, this
has no bearing on our main result that more concerned households have a larger MPC regardless of the changes
in actual spending.
21
We find that more concerned households are more likely to have experienced earning losses during the
pandemic, particularly for income falls larger than 25% (last three columns in Table B.4 in Appendix B). But our
results in Table 5 indicate that, conditional on having income losses, more concerned households are not more
likely to have different MPCs relative to other households.

23
©International Monetary Fund. Not for Redistribution
shocks during the pandemic explain precautionary behaviour in the presence of a temporary

windfall gain for the average household. But income shocks during the pandemic also cannot

explain the differential MPCs between financially concerned and all other households.

Table 5: Controlling for past spending changes and income shocks

(1) (2) (3) (4)

Financial concerns 0.037∗∗∗ 0.037∗∗∗ 0.027∗∗∗ 0.027∗∗∗
(0.007) (0.007) (0.009) (0.009)
Spending decrease 0.021∗∗
(0.009)
Fin. conc.×Spend. decrease -0.020
(0.015)
Spending decrease< 25% 0.016
(0.010)
Spending decrease> 25% 0.049∗∗
(0.020)
Fin. conc.×Spend. decrease< 25% -0.011
(0.016)
Fin. conc.×Spend. decrease> 25% -0.057∗
(0.031)
Income loss -0.013∗
(0.008)
Fin. conc.×Income loss 0.008
(0.012)
Income loss< 25% -0.004
(0.009)
Income loss> 25% -0.025∗∗
(0.010)
Fin. conc.×Income loss< 25% 0.010
(0.015)
Fin. conc.×Income loss> 25% 0.009
(0.015)

Controls X X X X
N 13,305 13,305 10,889 10,889
Notes: Estimates from a random effects model at the individual level, where the
dependent variable is the elicited MPC. Standard errors in parentheses clustered at
the individual level. Constant is not reported. Asterisks, *, **, and ***, denote
statistical significance at the 10%, 5%, and 1% levels.

Next we investigate whether our result might be driven by failing to control for the share

of discretionary spending pre-pandemic. Individuals who spent less on discretionary items

pre-pandemic, such as food expenditure away from home, travel, and other leisure activities,

might be expected to spend a larger fraction of the one-off transfer on essential goods. The

consumption of these goods was less restricted during the pandemic than discretionary spending,

which was curbed by restrictions on social interaction. We might expect individuals that spend

more on essential goods, who are also more likely to be more concerned households, to have a

larger MPC than individuals where non-essential items are a far stronger driver of expenditure.

Similarly to Andreolli and Surico (2021), we use the pre-pandemic share of food consumption

away from home in total income as the proxy for discretionary spending. Our results offer some

24
©International Monetary Fund. Not for Redistribution
statistical evidence that more concerned households who spent less on discretionary spending

pre-pandemic tend to have larger MPCs (column (1) in Table 6). In a second exercise we

control for the size of the transfer, which could serve as an alternative proxy for the share of

discretionary spending. Andreolli and Surico (2021) find evidence from Italian survey data that

MPCs vary with the size of the transfer, and that this relationship is indeed related to spending

on non-essential items. We divide the £500 transfer by pre-COVID household income, and

then construct a dummy equal to one if the size of the transfer is above the sample median. In

line with Fagereng et al. (2021), who find that households with larger lottery prizes have lower

MPCs, we find that the size of the transfer is negatively associated with MPCs, but only for

unconcerned households (column 2).

Table 6: Relative size of the transfer

(1) (2)

Financial concerns 0.030∗∗∗ 0.026∗∗∗
(0.007) (0.007)
Share eating out 0.105
(0.090)
Fin. conc.×Share eating out -0.230∗
(0.136)
Size transfer -0.019∗∗
(0.010)
Fin. conc.×Size transfer -0.007
(0.010)

Controls X X
N 20,079 21,939
Notes: Estimates from a random effects model at the in-
dividual level, where the dependent variable is the elicited
MPC. Standard errors in parentheses clustered at the in-
dividual level. Constant is not reported. Asterisks, *, **,
and ***, denote statistical significance at the 10%, 5%,
and 1% levels.

Could employment and industry effects drive our headline result that financially concerned

households are associated with larger MPCs? It is possible that self-employed individuals or

those that are furloughed have higher MPCs and are also more likely to expect financial stress –

we have shown in Table 3 this to be the case. In column (1) of Table 7 we include dummies for

the self-employed and employed, with those not working as the baseline group. Individuals who

are self-employed have a significantly larger MPC than those who are not working. But we do

not find any statistically significant difference in the MPCs between those financially concerned

and other households, conditional on being self-employed.

In column (2) we zoom in on the sub-sample of individuals who are employed by including

25
©International Monetary Fund. Not for Redistribution
a dummy indicating whether an individual has been furloughed. We find no evidence that

employees who have been furloughed, relative to those who remain working, tend to have a

smaller MPC if they are financially concerned.22 The fact that furloughed employees have

similar MPCs to employees that continued to work throughout the pandemic may reflect the

wide coverage of the government’s Coronavirus job retention scheme. In addition, the prevalence

of employers ‘topping-up’ furloughed employees’ pay to 100% may have also played a role. It

also suggests that our headline result is not being driven by concerned households’ reliance

on exceptional government programmes, the extent and duration of which was uncertain for a

long period of time, but rather is structural in some sense. For instance, the financial concerns

question may pick out elements of the population who are more impatient and or less financially

sophisticated. Alternatively it could pick out households who are more risk averse but where a

small payment can help compensate for this.

In column (3) we include dummies reflecting whether an individual had started receiving

Universal Credit prior to the pandemic or only since the start of the pandemic, as well as

interaction terms with financial concerns. We do not find any statistical evidence that individ-

uals who only received Universal Credit since the start of the pandemic have different MPCs

compared to other individuals. But we find some evidence that individuals who were reliant

on Universal Credit pre-Covid and who are more concerned tend to have larger MPCs than

concerned individuals who were not on benefits prior to March 2020. This suggests that our

main result is not specific to increased reliance on government support during the pandemic,

but may be structural in some sense. It then follows that the effectiveness of stimulus measures

may be increased if targeted to a subset of the population who had entered the pandemic with

greater financial vulnerabilities.

In column (4) we use a dummy to capture high-contact industries to control for house-

holds working in these sectors having a higher probability of job or income loss in the near

future. We do not find any evidence that individuals that work in high-contact industries have

a different MPC to the rest of the population, nor do individuals who are concerned and who

work in these industries. Controlling for the ability to work from home is an alternative way

of capturing those who work in a high-contact industry. In column (5) we include a dummy

equal to one if the individual said she can never work from home. The coefficient is negative

and statistically significant at the 1% level, suggesting a precautionary motive for individuals

22
Our results remain qualitatively similar when we proxy those currently furloughed using a question on why
hours have changed.

26
©International Monetary Fund. Not for Redistribution
who can never work at home. In addition, the MPC declines further if these individuals are

also financially concerned. This suggests that conditional on working from home, financially

concerned households tend to have a larger MPC than those who are not financially concerned.

Table 7: Controlling for the employment situation and UC benefits

(1) (2) (3) (4) (5)

Financial concerns 0.027∗∗∗ 0.030 0.021∗∗∗ 0.021∗∗∗ 0.034∗∗∗
(0.008) (0.008) (0.005) (0.007) (0.009)
Employee 0.003
(0.008)
Self-employed 0.028∗∗
(0.012)
Fin. conc.×Employee -0.003
(0.010)
Fin. conc.×Self-employed -0.023
(0.017)
Furloughed 0.005
(0.011)
Fin. conc.×Furloughed -0.020
(0.015)
UC pre-Covid -0.033
(0.025)
UC post-Covid -0.033
(0.027)
Fin. conc.×UC pre-Covid 0.045∗
(0.027)
Fin. conc. ×UC post-Covid 0.005
(0.031)
High-contact 0.020
(0.016)
Fin. conc.×High-contact 0.003
(0.021)
Never WFM -0.024∗∗∗
(0.008)
Fin. conc.×Never WFM -0.025∗∗
(0.012)

Controls X X X X
N 21,938 9,756 21,939 11,477 11,762
Notes: Estimates from a random effects model at the individual level, where the de-
pendent variable is the elicited MPC. Standard errors in parentheses clustered at the
individual level. Constant is not reported. Asterisks, *, **, and ***, denote statistical
significance at the 10%, 5%, and 1% levels.

Finally, in Table 8, we restrict the sample to mortgagors to investigate the role of borrowing

constraints. The literature has found that highly indebted households tend to exhibit larger

consumption responses to transitory income shocks (Kaplan et al. 2014, Misra and Surico 2014,

Baker and Yannelis 2017, Cloyne and Surico 2017, Bunn et al. 2018, Kovacs et al. 2018). We

are thus interested in studying whether households who expect financial stress and who are

also highly indebted are associated with a larger MPC. If this is true, then higher MPCs

reported by more financially concerned households may actually reflect their underlying level

of indebtedness, rather than a direct channel between financial concerns and MPCs. Table 8

27
©International Monetary Fund. Not for Redistribution
shows that our main result remains strongly robust to restricting the sample to mortgagors.

In column (3) we find some evidence that more highly indebted mortgagors, as measured

by a high LTI, have larger MPCs. However, we do not find any supporting evidence that bor-

rowing constraints – measured with the DSR, LTI, or LTV – explain why financially concerned

households are associated with a larger MPC.23

Table 8: Mortgagors sample

(1) (2) (3) (4) (5) (6)

Financial concerns 0.018∗ 0.025∗∗ 0.024∗ 0.027∗∗ 0.027∗∗ 0.029∗∗
(0.010) (0.012) (0.013) (0.013) (0.011) (0.012)
Payment holiday -0.049∗∗ -0.046∗ -0.064∗∗ -0.062∗∗ -0.046∗ -0.064∗∗
(0.022) (0.024) (0.025) (0.025) (0.024) (0.025)
Fin. conc.×Payment holiday 0.049∗ 0.054∗ 0.058∗ 0.056∗ 0.055∗ 0.061∗
(0.028) (0.031) (0.032) (0.032) (0.031) (0.032)
High DSR 0.006
(0.019)
Fin. conc.×High DSR 0.007
(0.024)
High LTI 0.037∗
(0.019)
Fin. conc.×High LTI 0.003
(0.024)
High LTV 0.013
(0.019)
Fin. conc.×High LTV -0.007
(0.024)
DSR>= 40% 0.018
(0.053)
Fin. conc.×DSR>= 40% 0.023
(0.073)
LTI>= 5 0.044
(0.039)
Fin. conc.×LTI>= 5 -0.065
(0.048)

Controls X X X X X X
N 6,863 5,964 5,644 5,575 5,964 5,644
Notes: Estimates from a random effects model at the individual level, where the dependent
variable is the elicited MPC. Standard errors in parentheses clustered at the individual level.
Constant is not reported. Asterisks, *, **, and ***, denote statistical significance at the 10%, 5%,
and 1% levels.

It is also worth noting that mortgagors who have been granted a payment deferral, also

known as payment holiday in the UK, are associated with smaller MPCs relative to other

mortgagors.24 This suggests that additional income transfers would be mostly saved and not

23
We define high DSR, LTI, and LTV, as dummy variables that take the value of one if the household belongs
to the top quintile of the respective pre-crisis distribution. In the last two columns we include groups for DSRs
and LTIs that are generally thought to be important: DSRs greater than 40%, and LTIs above 5.
24
The FCA guidance introduced on 20 March 2020 encouraged lenders to grant mortgage payment deferrals
to their clients in order to mitigate the effects of the COVID-19 shock on mortgagors. The FCA guidance,
later updated in June 2020, involved a full suspension of mortgage payments, both principal and interest, for a
maximum period of six months. Our payment holiday dummy takes the value of one when an individual with
mortgage debt had been granted a payment deferral at some point in the past or at the time of the survey.

28
©International Monetary Fund. Not for Redistribution
consumed, given that payment holidays have already provided important liquidity relief for

households to smooth consumption, especially those with low savings and low income (Albu-

querque and Varadi 2022). But the interaction term with financial concerns cancels out the

average coefficient on payment holidays. This provides further evidence that it is financial

concerns rather than policy support, or borrowing constraints, that drive our main result of

financially concerned households displaying a larger MPC.

4.3 Non-linearities in financial concerns

We have found that household expectations related to financial concerns over the short term

play a key role in explaining differences in MPCs across households during the pandemic. We

have found that our results hold irrespective of the way we measure financial concerns, whether

it is the likelihood of not being able to pay bills and expenses, the likelihood of losing a job,

or the expectation of generally being in a worse financial situation. Our results consistently

point to more concerned households having a larger MPC out of a one-off income transfer. We

have also shown that our results remain robust to including a large set of household-specific

covariates, including health concerns from COVID-19. The previous section showed that past

spending cuts, negative income shocks, and borrowing constraints cannot explain why financially

concerned households have larger MPCs. We did show some tentative evidence that this may be

driven by different shares of discretionary spending and reliance on benefits, but this is unlikely

to play a large role.

At face value, this is a surprising result. We might expect that households who are concerned

about being in financial distress in the future would tend to have a lower MPC, i.e. spend less

than other households in order to build up savings to mitigate potentially unfavourable income

shocks (Aiyagari 1994, Jappelli and Pistaferri 2014). However, the empirical evidence on the role

of expectations during the pandemic is not conclusive. Using the July 2020 wave of the COVID

Understanding Society survey, Crossley et al. (2020) find that UK individuals who expect their

financial situation to worsen or a job loss in the next three months have a higher MPC. On the

other hand, Baker et al. (2021) find that households who expect employment losses and benefit

cuts display a smaller MPC out of US stimulus checks. Using survey data from a sample of six

European countries, Christelis et al. (2020) also find supporting evidence for a precautionary

saving effect: households that are more concerned about the effects of COVID-19 are associated

with smaller MPCs out of a positive income shock, but only for durable goods. Using US survey

29
©International Monetary Fund. Not for Redistribution
data, Coibion et al. (2020) find little role for individuals’ macroeconomic expectations – future

unemployment rate, inflation rates, and mortgage rates – in explaining differences in MPCs.

Our analysis has focused on the first moment of expectations – each household provides a

single value reflecting concerns about their ability to pay future bills. The literature that also

looks at the second moment of expectations – the uncertainty surrounding these expectations –

tends to find that uncertainty is a more important driver (Itzhak et al. 2020, Dietrich et al. 2020,

Coibion et al. 2021). Using a new survey of European households during the pandemic, Coibion

et al. (2021) do not find any statistical association between household spending and households’

concerns about the expected path of the economy – the first moment of expectations. They find

that it is the second moment of expectations – the uncertainty about the economic outlook as

measured by the disagreement across professional forecasters – that leads households to reduce

their spending on non-durable goods and services in subsequent months. Using a representative

survey of US households in which respondents are asked about the economic consequences of

COVID-19, Dietrich et al. (2020) also find that expected GDP loss has no significant effect

on reported behaviour. But, unlike Coibion et al. (2021), they find that higher uncertainty is

associated with more COVID-related spending, while also being correlated with higher savings.

Unfortunately, households were not asked about their uncertainty around their responses in

the Understanding Society survey. In addition, there is little time variation within households in

the expectations variables to be able to build a proxy for household-specific uncertainty. In this

section we instead investigate non-linearities in financial concerns, which may shed some light

on the role of uncertainty. For instance, households who assign either a 0% or 100% probability

to not being able to pay their bills in the future may be the most certain, whereas those who

assign a probability in between are likely to be more uncertain.

Column (1) of Table 9 includes our preferred specification from column (2) of Table 4. In

column (2) we estimate separate coefficients for individuals who assigned either a 0% or a 100%

probability of experiencing financial difficulties over the short-term. The omitted category is

all individuals who reported values between 1%-99%. The coefficients are all negative and

statistically significant, suggesting that our headline result is being driven by households who

assign a non-zero probability to being unable to pay their bills, but excluding those that are

100% certain. In particular, the latter households have the smallest MPC: 6p.p. smaller than

those that assign a non-zero probability. In column (3) we split further the financial concerns

variable. We find that households who are moderately concerned, in the range of 1%-50%, have

30
©International Monetary Fund. Not for Redistribution
a larger MPC than households who are not concerned (the omitted category). Although the

coefficient is also positive for those that are more concerned, in the range of 51%-99%, it is not

statistically significant at conventional levels.

Overall, we find that households who are moderately concerned and those who appear to

be uncertain are driving our main results, rather than those who appear to be certain about

their financial situation in the near future (0% or 100% probability). While this results stands

somewhat in contrast with the literature linking higher uncertainty to lower MPCs and higher

precautionary savings (Itzhak et al. 2020, Coibion et al. 2021), we note that we do not focus on a

measure of uncertainty typically used elsewhere; we have just interpreted the extreme responses

as proxies of certainty. For instance, those that replied they were not concerned may already

be consuming optimally, so they are more likely to save the transfer. In turn, those that are

certain they will not be able to pay their bills tend to spend a smaller fraction so as to prepare

for tougher times ahead.

Table 9: Non-linearities in financial concerns

(1) (2) (3)

Financial concerns 0.023∗∗∗
(0.005)
Financial concerns = 0% -0.025∗∗∗
(0.005)
Financial concerns = 100% -0.060∗∗∗ -0.037∗
(0.019) (0.019)
Financial concerns [1 − 50%] 0.025∗∗∗
(0.005)
Financial concerns [51 − 99%] 0.019
(0.013)

Controls X X X
N 21,939 21,939 21,939
Notes: Estimates from a random effects model at the individual
level, where the dependent variable is the elicited MPC. Standard
errors in parentheses clustered at the individual level. Constant is
not reported. Asterisks, *, **, and ***, denote statistical signifi-
cance at the 10%, 5%, and 1% levels.

4.4 Alternative explanations

Our results underscore the role of expectations in determining the MPC during the pandemic.

For instance, some households may perceive the nature of the COVID shock as more perma-

nent, implying, for instance, that they expect to have lower earnings for longer, or to remain

unemployed. These households may thus be certain that they will not be able to pay their bills,

so they save a larger fraction of the windfall. In turn, households who perceive the pandemic

31
©International Monetary Fund. Not for Redistribution
shock to be transitory may be more willing to spend, even if they find themselves moderately

concerned about not being able to make ends meet. Against this background, our results are

reminiscent of the model developed by Carroll et al. (2021), who highlight the role of household

expectations in determining the MPC from the US CARES ACT. Carroll et al. (2021) have

three types of households. The first type of households do not lose their jobs but are associated

with a low MPC. They are expected to increase their savings during the pandemic because of the

inability to consume in some sectors (discretionary spending). These resemble our unconcerned

households. The second type of households, the ‘normal unemployed’, are associated with a

high MPC because they expect to be employed again once the economy starts operating nor-

mally. These resemble our moderately concerned households. Finally, the ‘deeply unemployed’

are associated with a low MPC since they perceive their unemployment to last longer. These

households resemble the households in our survey who report being 100% certain they will not

be able to make ends meet.

We can draw alternative explanations from behavioural approaches to consumption, such

as prospect theory, and mental accounting. In prospect theory the value function is defined

over gains and losses relative to some reference point (Kőszegi and Rabin 2009, Kahneman

and Tversky 2013). Households that expect to be in financial difficulty in the near future may

choose to overconsume now because they are less sensitive to news about future consumption

than to news about current consumption. Mental accounting may also play a role in determining

MPCs. People compartmentalise income and spending into different mental ‘accounts’, such as

‘current income’, ‘current assets’ and ‘future income’ (Shefrin and Thaler 1988, Duxbury et al.

2005, Milkman and Beshears 2009, Baugh et al. 2021). Households may use budgets within

these accounts to facilitate making trade-offs between competing uses for funds and to act as a

self-control device. There is evidence of non-fungibility of funds between accounts and budgets

(Baugh et al. 2021, Shapiro 2013). Financially concerned households might be more likely to

‘budget’ and treat funds within each tagged mental account as imperfectly substitutable. That

might explain why they are more likely to increase their spending in response to small windfalls.

A related possibility is that our results reflect preference heterogeneity across households,

self-control issues or financial sophistication. For instance, for a given level of wealth, more

impatient consumers may consume a larger fraction of income – higher APC – leaving them

more vulnerable to future shocks, and also consume more of any added income – higher MPC

(Aguiar et al. 2020). Similarly, self-control issues may also explain a tendency towards over-

32
©International Monetary Fund. Not for Redistribution
consumption (Laibson 1998, Vihriälä 2021). And financial sophistication may also explain why

financially concerned households display higher MPCs: unsophisticated households tend to save

less and hold fewer liquid assets, leaving them more vulnerable to income shocks (Jørring 2020).

We also cannot rule out that there may have been some ambiguity in the interpretation of

the MPC question, particularly that some households may have had trouble in distinguishing

what constituted ‘spending’ and ‘paying off debt’ (Sahm et al. 2010). For instance, if financially

concerned households would have had to cut back on their spending to pay their usual bills and

expenses in the absence of the transfer, then they might have responded that it led them to

spend more. Similarly, some concerned households may have responded spending more out of

the transfer because they interpreted paying off bills as spending rather than paying off debt.

Some of this ambiguity may have resulted in higher MPCs for more concerned households. We

indeed find that these households were more likely to report paying off more debt with the

amount they would not spend (Table B.5 in Appendix B).25

5 Actual spending and negative income shocks

The analysis in this paper has centred exclusively on inferring the MPC out of an unexpected

windfall of £500. But there is an important strand of the literature that has focused on

the asymmetric response of consumption to positive and negative transitory income shocks.

Research has found, in particular, that households’ consumption is more responsive to negative

income shocks, in line with the theoretical prediction of models with liquidity constraints,

income risk, and precautionary savings. For instance, Christelis et al. (2019) use the responses

of a representative sample of Dutch households to survey questions and find that consumers

react more to negative income changes than to positive changes. Bunn et al. (2018) use a set

of questions in the Bank of England/NMG Consulting Survey from 2011 to 2014 and find that

British households also tend to change their consumption by significantly more in reaction to

temporary and unanticipated falls in income than to increases in income of a similar magnitude.

In turn, Fuster et al. (2021) use data from the Federal Reserve Bank of New York’s Survey of

Consumer Expectations for US households, in which respondents report how they would adjust
25
We run Probit regressions in Table B.5, with the binary dependent variable being one of the possible answers.
We show that concerned households are more likely to rebuild their balance sheets by paying off more debt, but
they are less likely to save and receive less financial help. The rest of column (1) of the table indicates that
households more likely to pay off more debt tend to be more liquidity-constrained households (indicated by the
first quintile of the cash-on-hand ratio relative to the top quintile), who are currently finding it difficult or very
difficult to manage financially (‘Finances now’), and tend to be mortgagors or renters, in line with the evidence
from the US on the uses of the stimulus checks (Coibion et al. 2020).

33
©International Monetary Fund. Not for Redistribution
their spending over the next quarter in response to receiving or losing a specific dollar amount

ranging from $500 to $5,000. They find that responses to unanticipated losses are much larger

and more widespread than responses to gains. In addition, using US aggregate data, Barnichon

et al. (2021) find that fiscal multipliers tend to be larger for contractionary fiscal policy shocks.

We have shown consistently in this paper that financially concerned households are associ-

ated with larger MPCs out of positive income shocks than unconcerned households. An open

question is whether this finding can be extended to a scenario of negative income shocks. The

evidence from the literature above points to households having larger MPCs out of negative in-

come shocks, especially if they are liquidity constrained. But there is relatively scarce research

on the role of expectations, or financial concerns, in the asymmetric response of consumption

to income shocks. One of the exceptions is Bunn et al. (2018), who find that British households

who were concerned about a fall in income over the next year (a proxy for income risk) had

significantly higher MPCs out of a negative income shock relative to unconcerned households.

In our survey, households were unfortunately not asked about a possible scenario of an

income fall. This prevents us from computing the MPC, not allowing us to compare the size of

the consumption response for a given household across positive and negative shocks. We can,

however, test if financially concerned households are more likely to cut consumption in the face

of negative shocks relative to unconcerned households, once we control for their liquidity position

and for a set of other household-specific characteristics. For this purpose, we use a question

that asked households about their spending change over the previous four weeks relative to

the same period in the previous year, and to another question on income changes relative to

Jan/Feb 2020.26 Specifically, we run probit regressions where the dependent variable – binary

variable capturing whether households cut their consumption expenditures over the previous

four weeks – is regressed on a dummy variable capturing whether households experienced a fall

in their earnings relative to Jan/Feb 2020. We interact this variable with the financial concerns

dummy, and include the same set of controls used throughout the paper. One final caveat: the

sample size is much smaller as the spending question was not asked in the March 2021 survey,

and the question on income was only asked to households who were working.

Our main result from the Probit model in Table 10 offers strong statistical evidence to the

view that concerned households who had negative income shocks were 12 p.p. more likely to

have cut spending than unconcerned households who also experienced negative income shocks

26
We have used these questions before in Table 5.

34
©International Monetary Fund. Not for Redistribution
(column 1).27 This a relatively large difference when placed in the context of the 17% of all

households in the sample who reported spending cuts. Note also that financially concerned

households were more likely to report spending cuts than unconcerned households, even if they

did not experience a negative income shock, in line with findings in Table B.4 in Appendix B.

Table 10: Spending cuts and negative income shocks

(1) (2) (3) (4) (5) (6)
< 0% [−25% − 0%] < −25%

Financial concerns 0.060∗∗∗ 0.059∗∗∗ 0.046∗∗∗ 0.046∗∗∗ 0.014** 0.013**
(0.014) (0.014) (0.013) (0.013) (0.006) (0.006)
Income loss 0.044∗∗∗ 0.038∗∗∗ 0.006
(0.058) (0.011) (0.005)
Fin. conc.×Income loss 0.061∗∗ 0.044* 0.016*
(0.026) (0.025) (0.009)
Income loss< 25% 0.033** 0.036*** -0.005
(0.071) (0.013) (0.006)
Income loss> 25% 0.056∗∗∗ 0.041∗∗∗ 0.014**
(0.016) (0.015) (0.006)
Fin. conc.×Income loss< 25% 0.067∗∗ 0.056∗ 0.012
(0.031) (0.029) (0.012)
Fin. conc.×Income loss> 25% 0.052 0.030 0.017
(0.033) (0.033) (0.011)

Controls X X X X X X
N 6,787 6,787 6,787 6,787 6,787 6,787
Notes: Marginal effects of probit estimates at the individual level computed with the delta method.
The dependent variables are binary variables capturing whether the individual has experienced spend-
ing falls. Standard errors in parentheses clustered at the individual level. Constant and controls are
not reported. Asterisks, *, **, and ***, denote statistical significance at the 10%, 5%, and 1% levels.

In column (2) we break down the income loss by ranges: smaller or larger than 25%. We

show that concerned and unconcerned households display a statistically different likelihood of

cutting spending only when income fell up to 25% (concerned households were ten p.p. more

likely to cut spending in this event). This is not surprising as a large negative shock to income

may imply that even households who are not concerned about making ends meet may also be

‘forced’ to adjust their expenditures downwards given the large size of the income shock. The

remaining columns explore whether the results change when we split the consumption dummy

variable by spending cuts up to or larger than 25%. We find that our main result generally holds:

financially concerned households are more likely to cut spending than unconcerned households,

particularly so when we condition on negative shocks to income.

Overall, our results suggest that larger consumption responses to income shocks of house-

holds who expect not to be able to make ends meet may not be exclusive to scenarios of positive

income shocks. Although we cannot compute an MPC from a scenario of negative income shocks
27
For households who had an income loss, the conditional probability of cutting spending for concerned
households relative to unconcerned households is obtained by 100x(0.060+0.061).

35
©International Monetary Fund. Not for Redistribution
– and thus not compare it with the elicited MPC from a hypothetical transfer – our findings

highlight the role that household expectations play in determining the responsiveness of con-

sumption to both hypothetical expansionary and contractionary fiscal policies.

6 Conclusion

Research carried out during the pandemic has shown that households mostly save or pay down

debt when receiving a one-off payment (Armantier et al. 2020, 2021, Christelis et al. 2020,

Coibion et al. 2020, Crossley et al. 2020, Cox et al. 2020, Baker et al. 2021). This suggests that

untargeted direct transfers to households may not stimulate aggregate demand as much during

pandemics as during other crises. This poses several challenges to policy makers in search of

appropriate policies to stimulate demand.

In this paper we have shown the average MPC out of a hypothetical unexpected one-time

payment of £500 to be 11%, implying an increase of £0.11 for each £1 of fiscal stimulus.

This is relatively low compared to previous empirical estimates. We have, however, made the

case that financially concerned households are associated with higher MPCs. In this context,

policy makers can potentially increase the effectiveness of future fiscal stimulus by targeting

these high-MPC households. The financial concerns variable is not directly observable – the

survey is anonymised – and even if it were, it would come with a lag of a couple of months.

Nevertheless, we have shown that more financially concerned households are more likely to

belong to low-income groups, tend to be renters or mortgagors, younger, male, and belong to

ethnic minorities. They are also more likely to have been furloughed, reliant on benefits, or to

work in high-contact industries. Most of these variables can be observed by policy makers, thus

facilitating the design of policies.

Overall, assessing the response of consumption to income changes is of key importance

for the design of fiscal and monetary policies. This is particularly relevant in a fast-changing

environment where household expectations play a crucial role in determining how consumption

reacts to shocks. In this context, our research suggests that policies that provide support to

households who are low income and more vulnerable to becoming financially distressed may

prove more effective than providing untargeted stimulus payments to all households.

Table A.1: Variable definitions

Variable Code Description

Covid survey

Financial concerns finsec On a scale of 0-100% how likely do you think it is that you
will have difficulty paying your usual bills and expenses in
the next three months?

Job security jobsec On a scale of 0-100% how likely do you think it is that you will
lose your job / shut your business in the next three months?

Financial situation: future finfut_cv3 Looking ahead, how do you think you will be financially 3
months from now? (Better off; Worse off; or about the same)

Financial situation: current finnow How well would you say you yourself are managing financially
these days? (Living comfortably; doing alright; just about
getting by; finding it quite difficult; or finding it very difficult)

Weight betaindin_xw Cross-sectional individual web survey weight.

Age age

Sex sex_cv

Household size hhnum Number of children and adults in household.

Ethnicity racel_dv

Risk of contracting Covid riskcv19 In your view, how likely is it that you will contract COVID-
19 in the next month? (Very likely; likely; unlikely; very
unlikely).

Likelihood of having Covid hadcovid In your view, how likely is it that you have had COVID-19?
(Definitely had it; very likely; likely; unlikely; very unlikely;
don’t know/can’t tell).

Change in spending spend Thinking about your household spending (e.g. on food and
household goods, products and services, on commuting, bills,
etc.) but excluding housing costs (e.g. mortgage payments,
rent). Over the last 4 weeks, has your household spending
increased, decreased, or stayed the same relative to the same
four week period last year? (Increased by more than a quar-
ter; increased by up to a quarter; stayed the same; decreased
by up to a quarter; decreased by more than a quarter; don’t
know; prefer not to answer).

Change in household earnings blhhearn_amount, Change in household take-home pay/earnings (after tax, Na-
blh- tional Insurance and pension contributions have been de-
hearn_period, ducted) between January/February 2020 and the month of
hhearn_amount, the Covid-19 survey.
hhearn_period

Employee or self-employed sempderived
Continue on the next page

Variable Code Description

Able to work from home blwah During January and February how often did you work at
home? (Always; Often: Sometimes; Never)

Furlough furlough, An indicator for whether an individual has received a written
ff_furlough, letter or email from their employer confirming they had been
newfurlough furloughed under the Coronavirus Job Retention Scheme at
some point.

Universal Credit pre-Covid blbenefits1, Was receiving Universal Credit in January/February 2020
blbenefitsb651 (including those who continued to receive payments during
the pandemic).

Universal Credit post-Covid ff_ucredit, Applied for and received Universal Credit since the start of
ucredit2b65 the pandemic.

Mortgage payment holiday morhol, An indicator for whether a household has been granted a
morhol2, mortgage payment holiday at some point.
ff_morhol

Main survey

No. children nkids_dv Number of children in household.

Cash-on-hand svamt, svsj, Total household liquid assets net of total unsecured debt plus
svsk, svpn, household net income. The household liquid assets and un-
debty, debtsj, secured debt totals are calculated by aggregating individual
detsk, fihhmn- savings and unsecured debt responses within the household,
net1_dv accounting for shared accounts. Both variables are from Wave
8 of the Main survey as respondents were not asked about the
stock of their savings and unsecured debt in Wave 9 or 10.

Housing tenure tenure_dv Housing tenure. (Owned outright; Owned with mortgage;
Local Authority rented; Housing Association rented; Rented
from employer, Rented private (unfurnished); Rented private
(furnished); Other rented; Missing).

HH income fihhmnnet1_dv Total net monthly household income from the Main survey.
This is the sum of net monthly incomes from all household
members. It includes net labour income, miscellaneous in-
come, private benefit income, investment income, pension in-
come and social benefit income.

DSR xpmg, fihhyr The last total monthly installment on all mortgages or loans
on their property divided by total monthly gross household
income. Top and bottom coded at the 1st/99th percentiles.

LTI mgtot, fihhyr The outstanding sum of all mortgage debt divided by total
annual gross household income. Top and bottom coded at
the 1st/99th percentiles.

LTV mgtot, hsval The outstanding sum of all mortgage debt divided by the ex-
pected current value of the property. Top and bottom coded
at the 1st/99th percentiles.
Continue on the next page

Variable Code Description

Food outside of home xpfdout_g3, fi- About how much have you and other members of your house-
hhmnnet1_dv hold spent in total on meals, snacks or non-alcoholic drinks
purchased outside the home in the last four weeks? Please
include items bought from takeaways, restaurants, sandwich
shops, work or school canteens but do not include alcohol.

Industry type jbiindb_dv Industrial classification, two digits.

N Mean SD Zero (0-0.25) [0.25-0.5) [0.5-0.75) [0.75-1) One
Wave
Jul-20 7,313 0.11 0.28 84.82 1.57 2.02 3.60 0.74 7.25
Nov-20 7,313 0.11 0.29 84.77 1.63 1.80 3.00 0.95 7.85
Mar-21 7,313 0.12 0.30 84.51 1.60 1.65 3.04 1.01 8.19
Pay bills in 3m
Unconcerned 16,760 0.10 0.27 87.29 1.25 1.27 2.61 0.75 6.83
Concerned 5,179 0.15 0.32 78.47 2.44 3.15 4.67 1.26 10.01
Current situation
Not difficult 21,213 0.11 0.29 84.80 1.60 1.83 3.31 0.85 7.60
Difficult 726 0.14 0.34 82.53 1.63 1.62 1.18 1.86 11.17
Fin.situation in 3m
Same or better 19,944 0.11 0.28 85.22 1.62 1.80 3.03 0.90 7.43
Worse 1,988 0.15 0.33 80.13 1.42 2.08 4.86 0.76 10.75
Job security
Unconcerned 7,439 0.12 0.30 85.00 1.28 1.36 2.82 0.86 8.67
Concerned 4,293 0.16 0.33 78.23 2.15 2.79 5.13 1.25 10.45
Cash-on-hand
Q1 3,571 0.13 0.31 82.11 1.40 2.40 4.18 0.91 9.01
Q2 3,540 0.13 0.31 82.43 2.46 1.87 3.13 0.94 9.17
Q3 4,314 0.10 0.28 85.67 1.80 1.35 3.13 1.17 6.88
Q4 4,571 0.10 0.28 86.16 1.20 1.85 3.14 0.89 6.76
Q5 5,943 0.10 0.27 87.03 1.15 1.67 2.51 0.60 7.04
Housing tenure
Outright owner 12,020 0.08 0.25 88.75 1.09 1.21 3.02 0.59 5.34
Mortgagor 6,863 0.15 0.33 81.05 1.56 1.77 3.38 1.23 11.02
Renter 3,056 0.12 0.23 82.43 2.51 2.93 3.34 1.02 7.77
Mortgage debt
High LTI (Q5) 1,374 0.17 0.36 78.53 2.05 2.01 2.95 0.71 13.75
Low LTI (Q1) 1,377 0.13 0.31 82.46 1.43 1.69 4.04 1.49 8.89
Baseline HH income
Below median 10,971 0.10 0.28 85.30 1.75 1.94 3.36 0.98 6.67
Above median 10,968 0.12 0.30 84.05 1.43 1.70 3.06 0.82 8.94
Labour market
Employee 9,756 0.13 0.31 82.50 1.64 1.98 3.79 1.02 9.08
Self-employed 2,009 0.14 0.32 82.24 1.42 1.50 3.19 0.97 10.69
Not working 10,173 0.08 0.25 87.96 1.59 1.71 2.51 0.74 5.49
Age
18-39 2,989 0.15 0.32 79.52 2.69 2.66 4.05 1.13 9.95
40-64 10,362 0.13 0.31 83.22 1.33 1.82 3.49 1.01 9.13
65+ 8,019 0.06 0.22 90.89 1.15 1.24 2.17 0.51 4.04
Sex
Female 12,497 0.11 0.28 85.40 1.61 1.83 3.33 0.88 6.96
Male 9,442 0.12 0.30 83.95 1.59 1.82 3.09 0.93 8.63
Ethnicity
White 20,283 0.11 0.29 84.75 1.56 1.85 3.17 0.86 7.81
Ethnic minority 1,656 0.11 0.29 83.95 2.20 1.41 3.77 1.49 7.18
Change in spending
Increase or no change 11,205 0.11 0.28 85.28 1.64 1.70 3.25 0.70 7.43
Decreased 0-25% 1,742 0.14 0.31 81.40 1.40 2.31 4.57 1.65 8.67
Decreased 25%+ 358 0.15 0.33 80.77 1.21 2.76 3.04 1.80 10.42
Change in HH earnings
Increase or no change 6,125 0.14 0.32 80.82 2.01 1.88 4.15 1.16 9.98
Decreased 0-25% 2,354 0.13 0.31 82.60 1.39 2.15 3.51 1.02 9.32
Decreased 25%+ 2,410 0.11 0.28 84.84 1.91 2.23 2.79 0.66 7.58
Risk of Covid-19
(Very) likely 1,011 0.11 0.28 83.77 3.69 1.40 2.98 0.41 7.74
(Very) unlikely 20,899 0.11 0.29 84.78 1.48 1.85 3.23 0.93 7.73
Uc pre-Covid
No 21,612 0.11 0.29 84.77 1.59 1.81 3.24 0.90 7.69
Yes 327 0.14 0.32 82.56 1.83 2.14 2.37 0.87 10.23
UC post-Covid
No 21,640 0.11 0.29 84.63 1.60 1.82 3.25 0.89 7.81
Yes 299 0.08 0.25 88.31 1.55 1.84 1.29 1.45 5.56

Notes: Shares calculated as percentages of total MPC sample.

N Mean SD p10 p50 p90
MPC
Unconcerned 16,760 0.10 0.27 0.00 0.00 0.50
Concerned 5,179 0.15 0.32 0.00 0.00 0.90
Current situation difficult
Unconcerned 16,760 0.01 0.09 0.00 0.00 0.00
Concerned 5,179 0.14 0.34 0.00 0.00 1.00
Worse fin. situation in 3m
Unconcerned 16,756 0.05 0.22 0.00 0.00 0.00
Concerned 5,176 0.21 0.41 0.00 0.00 1.00
More likely to lose job
Unconcerned 8,235 0.19 0.40 0.00 0.00 1.00
Concerned 3,497 0.76 0.42 0.00 1.00 1.00
Cash-on-hand
Unconcerned 16,760 46,005 99,520 -9,311 7,715 142,487
Concerned 5,179 14,187 57,741 -16,300 2,083 46,310
Mortgagor
Unconcerned 16,760 0.30 0.46 0.00 0.00 1.00
Concerned 5,179 0.38 0.49 0.00 0.00 1.00
Baseline log HH income
Unconcerned 16,760 8.04 0.61 7.26 8.09 8.78
Concerned 5,179 7.96 0.63 7.16 8.03 8.70
Age
Unconcerned 16,760 56.31 16.37 32.00 58.00 76.00
Concerned 5,179 46.44 15.57 25.00 46.00 68.00
Male
Unconcerned 16,760 0.48 0.50 0.00 0.00 1.00
Concerned 5,179 0.49 0.50 0.00 0.00 1.00
Universal Credit pre-Covid
Unconcerned 16,760 0.01 0.12 0.00 0.00 0.00
Concerned 5,179 0.07 0.26 0.00 0.00 0.00
Universal Credit post-Covid
Unconcerned 16,760 0.01 0.10 0.00 0.00 0.00
Concerned 5,179 0.04 0.19 0.00 0.00 0.00
Notes: Concerned households assign a non-zero probability to having difficulties paying usual bills and
expenses in the next three months.

Table B.1: Probit model: MPC equal to 0
and equal to 1

(1) (2)
MPC=0 MPC=1

Financial concerns -0.065∗∗∗ 0.017∗∗
(0.010) (0.007)

Controls X X
N 21,939 21,939
Notes: Marginal effects of probit estimates at the
individual level computed with the delta method.
The dependent variables are binary variables tak-
ing the value of one when the elicited MPC is zero
(column 1), and when the MPC is equal to one (col-
umn 2). Standard errors in parentheses clustered at
the individual level. Constant and controls are not
reported. Asterisks, *, **, and ***, denote statisti-
cal significance at the 10%, 5%, and 1% levels.

(1) (2) (3) (4) (5) (6)
Baseline Region Region-time Industry Industry-time Tobit

Financial concerns 0.023∗∗∗ 0.023∗∗∗ 0.023∗∗∗ 0.023∗∗∗ 0.023∗∗∗ 0.345∗∗∗
(0.005) (0.005) (0.005) (0.007) (0.007) (0.062)
Finances now -0.013 -0.013 -0.013 -0.039∗∗∗ -0.039∗∗ -0.196
(0.011) (0.011) (0.011) (0.015) (0.015) (0.132)
Cash-on-hand Q1 0.000 0.000 0.000 -0.005 -0.005 0.018
(0.010) (0.010) (0.010) (0.016) (0.016) (0.132)
Cash-on-hand Q2 -0.005 -0.006 -0.006 -0.005 -0.005 -0.060
(0.010) (0.010) (0.010) (0.017) (0.017) (0.135)
Cash-on-hand Q3 -0.019∗ -0.020∗ -0.020∗ -0.030∗ -0.031∗ -0.248∗
(0.010) (0.010) (0.010) (0.016) (0.016) (0.133)
Cash-on-hand Q4 0.001 0.001 0.001 0.002 0.002 0.050
(0.010) (0.010) (0.010) (0.017) (0.017) (0.131)
Mortgagor 0.031∗∗∗ 0.031∗∗∗ 0.031∗∗∗ 0.031∗∗∗ 0.031∗∗∗ 0.382∗∗∗
(0.008) (0.008) (0.008) (0.011) (0.011) (0.100)
Renter 0.006 0.007 0.007 0.003 0.002 0.188∗
(0.008) (0.008) (0.008) (0.012) (0.012) (0.106)
Log HH income -0.002 -0.001 -0.001 0.011 0.011 -0.081
(0.005) (0.005) (0.005) (0.008) (0.008) (0.067)
Age 18-39 0.061∗∗∗ 0.061∗∗∗ 0.061∗∗∗ 0.049∗∗∗ 0.048∗∗∗ 0.870∗∗∗
(0.009) (0.009) (0.009) (0.016) (0.016) (0.119)
Age 40-64 0.046∗∗∗ 0.046∗∗∗ 0.046∗∗∗ 0.032∗∗ 0.030∗∗ 0.664∗∗∗
(0.007) (0.007) (0.007) (0.015) (0.015) (0.099)
Male 0.021∗∗∗ 0.020∗∗∗ 0.020∗∗∗ 0.030∗∗∗ 0.031∗∗∗ 0.216∗∗∗
(0.006) (0.006) (0.006) (0.009) (0.009) (0.072)
No. children 0.021∗∗∗ 0.021∗∗∗ 0.021∗∗∗ 0.024∗∗∗ 0.024∗∗∗ 0.252∗∗∗
(0.005) (0.005) (0.005) (0.006) (0.006) (0.056)
Household size -0.008∗∗∗ -0.008∗∗∗ -0.008∗∗∗ -0.007∗ -0.007∗ -0.093∗∗
(0.003) (0.003) (0.003) (0.004) (0.004) (0.037)
White 0.022∗ 0.020∗ 0.020∗ 0.020 0.020 0.362∗∗
(0.012) (0.012) (0.012) (0.016) (0.016) (0.147)
Jul20 survey -0.009∗∗ -0.009∗∗ -0.010 -0.006 -0.004 -0.085∗
(0.004) (0.004) (0.029) (0.005) (0.078) (0.049)
Nov20 survey -0.004 -0.004 0.011 -0.007 -0.008 -0.039
(0.004) (0.004) (0.026) (0.005) (0.069) (0.049)

Controls X X X X X X
Region FE X
Region×Time FE X
Industry FE X
Industry×Time FE X
N 21,939 21,930 21,930 11,861 11,861 21,939
Notes: Estimates from a random effects model at the individual level, where the dependent variable
is the elicited MPC. Standard errors in parentheses clustered at the individual level. Constant is not
reported. Asterisks, *, **, and ***, denote statistical significance at the 10%, 5%, and 1% levels.

(1) (2) (3) (4) (5) (6)
Jul20 Nov20 Mar21 Jul20 Nov20 Mar21

Financial concerns 0.029∗∗ 0.049∗∗∗ 0.049∗∗∗ 0.033∗∗ 0.052∗∗∗ 0.050∗∗∗
(0.013) (0.014) (0.013) (0.013) (0.014) (0.013)
Finances now 0.000 -0.011 0.022 0.005 -0.010 0.020
(0.043) (0.041) (0.054) (0.044) (0.042) (0.054)
Cash-on-hand Q1 0.011 -0.003 -0.018 0.011 -0.001 -0.022
(0.016) (0.017) (0.016) (0.016) (0.017) (0.016)
Cash-on-hand Q2 0.021 -0.032∗ -0.011 0.015 -0.031∗ -0.014
(0.016) (0.017) (0.016) (0.015) (0.017) (0.016)
Cash-on-hand Q3 0.000 -0.024 -0.034∗∗ -0.000 -0.023 -0.031∗∗
(0.013) (0.015) (0.013) (0.013) (0.015) (0.013)
Cash-on-hand Q4 0.013 -0.010 -0.004 0.013 -0.010 -0.004
(0.014) (0.014) (0.014) (0.014) (0.014) (0.014)
Mortgagor 0.019 0.032∗∗ 0.034∗∗∗ 0.022∗ 0.032∗∗ 0.037∗∗∗
(0.013) (0.014) (0.013) (0.013) (0.014) (0.012)
Renter 0.021 -0.008 -0.006 0.018 -0.009 -0.005
(0.016) (0.015) (0.014) (0.015) (0.015) (0.014)
Log HH income 0.016 -0.012 -0.001 0.012 -0.010 -0.000
(0.011) (0.012) (0.010) (0.010) (0.012) (0.010)
Age 18-39 0.084∗∗∗ 0.046∗∗ 0.058∗∗∗ 0.077∗∗∗ 0.050∗∗ 0.061∗∗∗
(0.021) (0.022) (0.021) (0.020) (0.022) (0.021)
Age 40-64 0.051∗∗∗ 0.035∗∗ 0.056∗∗∗ 0.050∗∗∗ 0.038∗∗∗ 0.058∗∗∗
(0.010) (0.015) (0.011) (0.010) (0.015) (0.011)
Male 0.022∗∗ 0.016 0.022∗∗ 0.019∗ 0.015 0.025∗∗
(0.011) (0.011) (0.010) (0.010) (0.011) (0.010)
No. children 0.032∗∗∗ 0.016 0.020∗∗ 0.031∗∗∗ 0.018∗ 0.022∗∗
(0.010) (0.010) (0.009) (0.010) (0.010) (0.009)
Household size -0.019∗∗∗ -0.007 -0.009 -0.017∗∗∗ -0.008 -0.013∗∗
(0.006) (0.008) (0.006) (0.006) (0.007) (0.006)
White 0.037 0.008 0.035 0.056∗∗∗ 0.008 0.030
(0.026) (0.031) (0.024) (0.020) (0.030) (0.024)
Risk of infection 0.011 -0.047∗∗∗ -0.029
(0.039) (0.017) (0.023)
Had Covid-19 -0.077∗∗∗ 0.072 0.038
(0.024) (0.101) (0.044)

Observations 7,313 7,313 7,313 7,298 7,288 7,222
Notes: OLS estimates at the individual level for each survey, where the dependent variable is
the elicited MPC. Standard errors in parentheses clustered at the individual level. Constant
is not reported. Asterisks, *, **, and ***, denote statistical significance at the 10%, 5%, and
1% levels.

Table B.4: Probit model: spending and income losses

(1) (2) (3) (4) (5) (6)
Spending Income
< 0% [−25% − 0%] < −25% < 0% [−25% − 0%] < −25%

Financial concerns 0.051∗∗∗ 0.062∗∗∗ 0.052∗∗∗ 0.011∗∗ 0.036* 0.051∗∗∗
(0.015) (0.013) (0.012) (0.005) (0.019) (0.015)

Controls X X X X X X
N 13,305 13,305 13,305 10,889 10,889 10,889
Notes: Marginal effects of probit estimates at the individual level computed with the delta method.
The dependent variables are binary variables capturing whether the individual has experienced
spending falls (columns 1 to 3) or income/earning losses (columns 4 to 6). Standard errors in
parentheses clustered at the individual level. Constant and controls are not reported. Asterisks,
*, **, and ***, denote statistical significance at the 10%, 5%, and 1% levels.

(1) (2) (3) (4)
Pay debt Save Receive less help Give more help

Financial concerns 0.092*** -0.084*** 0.015*** 0.000
(0.011) (0.014) (0.005) (0.008)
Finances now 0.207*** -0.283*** 0.019*** -0.025
(0.021) (0.027) (0.005) (0.019)
Cash-on-hand Q1 0.201*** -0.167*** 0.012 -0.016
(0.019) (0.020) (0.008) (0.010)
Cash-on-hand Q2 0.145*** -0.110*** 0.007 -0.007
(0.019) (0.018) (0.007) (0.010)
Cash-on-hand Q3 0.103*** -0.076*** 0.013* 0.001
(0.021) (0.019) (0.008) (0.010)
Cash-on-hand Q4 0.024 0.014 0.000 -0.007
(0.019) (0.016) (0.008) (0.008)
Mortgagor 0.129*** -0.065*** 0.000 -0.043***
(0.015) (0.015) (0.004) (0.009)
Renter 0.139*** -0.110*** 0.013*** -0.016*
(0.017) (0.019) (0.005) (0.010)
Log HH income 0.015 -0.012 -0.012*** 0.011*
(0.010) (0.010) (0.004) (0.006)
Age 18-39 -0.009 0.143*** 0.010* -0.081***
(0.020) (0.022) (0.005) (0.013)
Age 40-64 0.027* 0.053*** 0.012** -0.042***
(0.015) (0.015) (0.005) (0.007)
Male 0.009 0.022* -0.001 -0.038***
(0.011) (0.012) (0.004) (0.007)
No. children 0.023** -0.034*** 0.002 -0.001
(0.009) (0.010) (0.003) (0.006)
Household size 0.002 0.004 -0.001 -0.010**
(0.007) (0.008) (0.002) (0.004)
White 0.036* -0.014 -0.007 -0.040***
(0.021) (0.024) (0.005) (0.014)
Jul20 survey -0.004 -0.054*** 0.002 0.007
(0.009) (0.010) (0.004) (0.006)
Nov20 survey 0.014* -0.064*** -0.000 0.003
(0.008) (0.009) (0.004) (0.005)

N 19,521 19,521 19,521 19,521
Notes: Marginal effects of probit estimates at the individual level computed with
the delta method. The dependent variables are given at the top of each column.
Standard errors in parentheses clustered at the individual level. Constant is not
reported. Asterisks, *, **, and ***, denote statistical significance at the 10%, 5%,
and 1% levels.

45
©International Monetary Fund. Not for Redistribution
References
Aguiar, M. A., Bils, M. and Boar, C. (2020), Who Are the Hand-to-Mouth?, NBER Working
Papers 26643, National Bureau of Economic Research, Inc.

Aiyagari, S. R. (1994), ‘Uninsured Idiosyncratic Risk and Aggregate Saving’, The Quarterly
Journal of Economics 109(3), 659–684.

Albuquerque, B. and Varadi, A. (2022), Consumption effects of mortgage payment holidays:
evidence during the Covid-19 pandemic, IMF Working Paper WP/22/44.

Andreolli, M. and Surico, P. (2021), Less is More: Consumer Spending and the Size of Economic
Stimulus Payments, CEPR Discussion Papers 15918, C.E.P.R. Discussion Papers.

Armantier, O., Goldman, L., Koşar, G., Lu, J., Pomerantz, R. and Van der Klaauw, W. (2020),
How Have Households Used Their Stimulus Payments and How Would They Spend the Next?,
Liberty Street Economics October 13, Federal Reserve Bank of New York.

Armantier, O., Goldman, L., Koşar, G. and Van der Klaauw, W. (2021), An Update on How
Households Are Using Stimulus Checks, Liberty Street Economics April 7, Federal Reserve
Bank of New York.

Baker, S. R., Farrokhnia, R. A., Meyer, S., Pagel, M. and Yannelis, C. (2021), ‘Income, Liquidity,
and the Consumption Response to the 2020 Economic Stimulus Payments’, Review of Finance
(forthcoming).

Baker, S. R. and Kueng, L. (2021), Household financial transaction data, NBER Working Papers
29027.

Baker, S. R. and Yannelis, C. N. (2017), ‘Income Changes and Consumption: Evidence from
the 2013 Federal Government Shutdown’, Review of Economic Dynamics 23, 99–124.

Barnichon, R., Debortoli, D. and Matthes, C. (2021), ‘Understanding the Size of the Government
Spending Multiplier: It’s in the Sign’, The Review of Economic Studies (forthcoming).

Baugh, B., Ben-David, I., Park, H. and Parker, J. A. (2021), ‘Asymmetric consumption smooth-
ing’, American Economic Review 111(1), 192–230.

Bunn, P., Le Roux, J., Reinold, K. and Surico, P. (2018), ‘The consumption response to positive
and negative income shocks’, Journal of Monetary Economics 96(C), 1–15.

Campbell, J. Y. and Mankiw, N. G. (1989), Consumption, Income and Interest Rates: Rein-
terpreting the Time Series Evidence, in ‘NBER Macroeconomics Annual 1989, Volume 4’,
NBER Chapters, National Bureau of Economic Research, Inc, pp. 185–246.

Carroll, C. D., Crawley, E., Slacalek, J. and White, M. N. (2021), ‘Modeling the Consumption
Response to the CARES Act’, International Journal of Central Banking 17(1), 107–141.

Chetty, R., Friedman, J. N., Hendren, N., Stepner, M. and The Opportunity Insights Team, .
(2020), The Economic Impacts of COVID-19: Evidence from a New Public Database Built
Using Private Sector Data, NBER Working Papers 27431, National Bureau of Economic
Research, Inc.

46
©International Monetary Fund. Not for Redistribution
Christelis, D., Georgarakos, D., Jappelli, T. and Kenny, G. (2020), The Covid-19 crisis and
consumption: survey evidence from six EU countries, Working Paper Series 2507, European
Central Bank.

Christelis, D., Georgarakos, D., Jappelli, T., Pistaferri, L. and van Rooij, M. (2019), ‘Asymmet-
ric Consumption Effects of Transitory Income Shocks’, Economic Journal 129(622), 2322–
2341.

Cloyne, J. S. and Surico, P. (2017), ‘Household Debt and the Dynamic Effects of Income Tax
Changes’, Review of Economic Studies 84(1), 45–81.

Coibion, O., Georgarakos, D., Gorodnichenko, Y., Kenny, G. and Weber, M. (2021), The Ef-
fect of Macroeconomic Uncertainty on Household Spending, NBER Working Papers 28625,
National Bureau of Economic Research, Inc.

Coibion, O., Gorodnichenko, Y. and Weber, M. (2020), How Did U.S. Consumers Use Their
Stimulus Payments?, NBER Working Papers 27693, National Bureau of Economic Research,
Inc.

Cox, N., Ganong, P., Noel, P. J., Vavra, J. S., Wong, A., Farrell, D. and Greig, F. E. (2020),
‘Initial Impacts of the Pandemic on Consumer Behavior: Evidence from Linked Income,
Spending, and Savings Data’, Brookings Papers on Economic Activity (Summer), 35–69.

Crossley, T., Fisher, P., Levell, P. and Low, H. (2020), MPCs through COVID: spending, saving
and private transfers, ISER Working Paper Series 2020-14, Institute for Social and Economic
Research.

Dietrich, A., Kuester, K., Muller, G. J. and Schoenle, R. (2020), News and Uncertainty about
COVID-19: Survey Evidence and Short-Run Economic Impact, Working Papers 202012,
Federal Reserve Bank of Cleveland.

Duxbury, D., Keasey, K., Zhang, H. and Chow, S. L. (2005), ‘Mental accounting and decision
making: Evidence under reverse conditions where money is spent for time saved’, Journal of
Economic Psychology 26(4), 567–580.

Eichenbaum, M., Godinho de Matos, M., Lima, F., Rebelo, S. and Trabandt, M. (2020), How
do People Respond to Small Probability Events with Large, Negative Consequences?, NBER
Working Papers 27988, National Bureau of Economic Research, Inc.

Fagereng, A., Holm, M. B. and Natvik, G. J. (2021), ‘MPC heterogeneity and household balance
sheets’, American Economic Journal: Macroeconomics 4(13), 1–54.

Fuster, A., Kaplan, G. and Zafar, B. (2021), ‘What would you do with $500? Spending responses
to gains, losses, news, and loans’, The Review of Economic Studies 88(4), 1760–1795.

Ilut, C. L. and Valchev, R. (2020), Economic Agents as Imperfect Problem Solvers, NBER
Working Papers 27820, National Bureau of Economic Research, Inc.

Institute for Social and Economic Research (2020), ‘Understanding Society COVID-19 Study,
2020’. UK Data Service. SN: 8644, 10.5255/UKDA-SN-8644-1.

47
©International Monetary Fund. Not for Redistribution
Itzhak, B.-D., Fermand, E., Kuhnen, C. M. and Li, G. (2020), Expectations Uncertainty and
Household Economic Behavior, Available at https://ssrn.com/abstract=3293399, SSRN.

Jappelli, T. and Pistaferri, L. (2014), ‘Fiscal Policy and MPC Heterogeneity’, American Eco-
nomic Journal: Macroeconomics 6(4), 107–136.

Jappelli, T. and Pistaferri, L. (2020), ‘Reported MPC and Unobserved Heterogeneity’, American
Economic Journal: Economic Policy 12(4), 275–297.

Jørring, A. T. (2020), Financial sophistication and consumer spending. Mimeo.

Kahneman, D. and Tversky, A. (2013), Prospect Theory: An Analysis of Decision Under Risk,
in L. C. MacLean and W. T. Ziemba, eds, ‘Handbook of the Fundamentals of Financial
Decision Making Part I’, World Scientific Book Chapters, World Scientific Publishing Co.
Pte. Ltd., chapter 6, pp. 99–127.

Kaplan, G., Violante, G. L. and Weidner, J. (2014), ‘The Wealthy Hand-to-Mouth’, Brookings
Papers on Economic Activity 45(1 (Spring), 77–153.

Kovacs, A., Rostom, M. and Bunn, P. (2018), Consumption Response to Aggregate Shocks and
the Role of Leverage, Discussion Papers 1820, Centre for Macroeconomics (CFM).

Kőszegi, B. and Rabin, M. (2009), ‘Reference-dependent consumption plans’, American Eco-
nomic Review 99(3), 909–36.

Kueng, L. (2018), ‘Excess Sensitivity of High-Income Consumers’, The Quarterly Journal of
Economics 133(4), 1693–1751.

Laibson, D. (1998), ‘Life-cycle consumption and hyperbolic discount functions’, European Eco-
nomic Review 42(3-5), 861–871.

Milkman, K. L. and Beshears, J. (2009), ‘Mental accounting and small windfalls: Evidence from
an online grocer’, Journal of Economic Behavior & Organization 71(2), 384–394.

Misra, K. and Surico, P. (2014), ‘Consumption, income changes, and heterogeneity: Evidence
from two fiscal stimulus programs’, American Economic Journal: Macroeconomics 6(4), 84–
106.

Olafsson, A. and Pagel, M. (2018), ‘The Liquid Hand-to-Mouth: Evidence from Personal Fi-
nance Management Software’, Review of Financial Studies 31(11), 4398–4446.

Parker, J. A. and Souleles, N. S. (2019), ‘Reported effects versus revealed-preference estimates:
Evidence from the propensity to spend tax rebates’, American Economic Review: Insights
1, 273–90.

Parker, J. A., Souleles, N. S., Johnson, D. S. and McClelland, R. (2013), ‘Consumer spending
and the economic stimulus payments of 2008’, American Economic Review 103, 2530–2553.

Platt, L. and Warwick, R. (2020), ‘COVID-19 and Ethnic Inequalities in England and Wales’,
Fiscal Studies 41(2), 259–289.

48
©International Monetary Fund. Not for Redistribution
Sahm, C., Shapiro, M. and Slemrod, J. (2010), ‘Household Response to the 2008 Tax Rebate:
Survey Evidence and Aggregate Implications’, Tax Policy and the Economy 24(1), 69–110.

Shapiro, J. (2013), ‘Fungibility and consumer choice: Evidence from commodity price shocks’,
The Quarterly Journal of Economics 128(4), 1449–1498.

Shapiro, M. D. and Slemrod, J. (2003), ‘Consumer response to tax rebates’, The American
Economic Review 93, 381–396.

Shapiro, M. D. and Slemrod, J. (2009), ‘Did the 2008 Tax Rebates Stimulate Spending?’,
American Economic Review 99(2), 374–379.

Shefrin, H. M. and Thaler, R. H. (1988), ‘The Behavioral Life-Cycle Hypothesis’, Economic
Inquiry 26(4), 609–643.

Vavra, J. (2021), ‘Tracking the Pandemic in Real Time: Administrative Micro Data in Business
Cycles Enters the Spotlight’, Journal of Economic Perspectives 35(3), 47–66.

Vihriälä, E. (2021), Commitment in Debt Repayment: Evidence from a Natural Experiment,
Available at https://ssrn.com/abstract=3792708, SSRN.

49
©International Monetary Fund. Not for Redistribution
Financial Concerns and the Marginal Propensity to Consume in COVID Times: Evidence from UK Survey Data
Working Paper No. WP/22/27

File and source

File: financial-concerns-and-the-marginal-propensity-to-consume-in-covid-times-evidence-from-uk-survey-data.pdf
Size: 2,262,955 bytes
SHA-256: 533088e3f0d64db73f59fa25ef296e26e36ccb3eba306fe8a58ccc48cb361fad
Our copy: financial-concerns-and-the-marginal-propensity-to-consume-in-covid-times-evidence-from-uk-survey-data.pdf
Original: doi.org