Households’ Response to the Wealth Effects of Inflation*

Philip Schnorpfeil„ Michael Weber Andreas Hackethal§

June 12, 2024

Abstract
We study the redistributive effects of surprise inflation combining administrative bank
data with an information provision experiment during an episode of historic inflation.
On average, households are well-informed about prevailing inflation and are concerned
about its impact on their wealth; yet, while many households know about inflation
eroding nominal assets, most are unaware of nominal-debt erosion. Once they receive
information on the debt-erosion channel, households view nominal debt more positively
and increase estimates of their own real net wealth. These changes causally affect actual
consumption and hypothetical debt decisions. Our findings suggest that real wealth
mediates the sensitivity of consumption to inflation once households are aware of the
wealth effects of inflation.

JEL codes: D12, D14, D83, D84, E21, E31, E52

Keywords: Inflation Beliefs, Information Treatment, Consumption, Monetary Policy

* We thank our partners at a bank for their collaboration and the data necessary to conduct this research.
The bank data were obtained under a non-disclosure agreement. The bank did not review the results of the
paper prior to public dissemination. We are also grateful for helpful comments from Adrien Auclert, Paola
Boel, João Cocco, Andreas Fuster, Katrin Gödker, Mete Kilic, Nora Lamersdorf, Christine Laudenbach,
Stephan Luck, Chris Roth, Jan Toczynski, Emil Verner, Felix von Meyerinck, and Johannes Wohlfart,
and from conference and seminar participants at Amsterdam, Bath, BIS, Bundesbank, Cambridge, CUHK,
Edinburgh, Geneva, Georgetown, Glasgow, Goethe/SAFE, the Heidelberg-Karlsruhe-Mannheim seminar,
Imperial, Lancaster, Maastricht, Manchester, Mannheim, Nijmegen, Nova, PHBS, Rotterdam, UT Austin,
WashU Olin, the 2023 Bundesbank International Conference on Household Finance, 2023 CEBRA, 2023
Dolomites Summer Finance Conference, 2024 FIRS, 11th Helsinki Finance Summit, 13th ifo Conference on
Macroeconomics and Survey Data, 2024 ifo Workshop on Macroeconomics and International Finance, 7th
SAFE Household Finance Workshop, 2024 SGF, 2023 Swiss Winter Conference on Financial Intermediation,
and 2023 WFA. The experiment received IRB approval from the German Association for Experimental
Economic Research (Number: 5hjDTgn8) and is preregistered in the AEA RCT Registry (ID: 9735). We
gratefully acknowledge research support from the efl – the Data Science Institute and Leibniz Institute for
Financial Research SAFE. The contribution by Philip Schnorpfeil has been prepared under the Lamfalussy
Fellowship Programme sponsored by the ECB. Any views expressed are only those of the authors and do
not necessarily represent the views of the ECB or the Eurosystem.
„ Goethe University Frankfurt. Email: schnorpfeil@finance.uni-frankfurt.de
University of Chicago, CEPR, and NBER. Email: michael.weber@chicagobooth.edu
§ Goethe University Frankfurt and SAFE. Email: hackethal@safe-frankfurt.de
1 Introduction

After having been dormant for decades, inflation in 2022 reached levels many households had
not witnessed during their lifetime. Unexpected inflation erodes the real value of savings and
debt with fixed nominal interest rates, redistributing wealth from savers to borrowers (Fisher,
1933).1 These redistributive effects can be sizable because households hold large nominal
positions (Doepke and Schneider, 2006); in fact wealth effects dwarfed income effects of
surprise inflation for households with significant net nominal positions during the 2021–22
inflation surge (Pallotti et al., 2024). Theoretically, this redistribution of wealth can also
have sizable stimulative effects on aggregate demand, because those that benefit, consumers
with large negative net nominal positions, have on average higher marginal propensities to
consume (MPC) than consumers who lose (Auclert, 2019).
Yet, little is known about whether households are aware of the distributional consequences
of surprise inflation and how they adjust their economic decisions to the induced wealth
effects. Households might be unaware of the wealth effects of inflation because of money
illusion (e.g., Cohen, Polk, and Vuolteenaho, 2005; Modigliani and Cohn, 1979). Moreover,
even households that are aware of the wealth effects might primarily adjust their consumption
to realized payments rather than unrealized capital gains (e.g., Di Maggio, Kermani, and
Majlesi, 2020; Lettau and Ludvigson, 2004). Because nominal wages take time to adjust
to inflation, it might also take time for households to adjust their consumption, savings,
investment and debt decisions, even if they become wealthier in real terms.
We study the extent to which households are aware of the wealth effects of inflation, how
awareness affects beliefs about nominal positions and own wealth, and how these beliefs feed
into debt choices, consumption plans, and actual consumption decisions using administra-
1
See also Bhamra et al., 2023; Kang and Pflueger, 2015; and Leombroni et al., 2020.

1
tive data. To study these questions, we run a large-scale randomized control trial (RCT) on
customers of a major German bank in which participants receive information on inflation-
induced erosion of either nominal assets, or nominal debt. We find participating households,
on average, are well-informed about the current inflation rate and concerned about its wealth
impact. However, whereas households are largely aware of nominal-asset erosion, they have
limited knowledge about the debt-erosion channel of surprise inflation. Consequently, re-
spondents who receive information about nominal-debt erosion form more positive beliefs
about debt and increase their estimates of their own real net wealth. Households’ changes in
beliefs have real effects: they plan to spend more, update actual consumption ex post, and
choose a higher share of debt financing in a hypothetical real-estate investment.
We implemented the RCT on several thousand bank customers in July 2022 when infla-
tion in Germany was at a 70-year high of 8.7%. We first ask questions on knowledge about
the erosion channel. Respondents also estimate the recent change in their real net wealth
and decompose their balance sheet, allowing us to calculate their net nominal position. The
subsequent information experiment builds the core of the survey. We randomly assign re-
spondents into two treatment groups and one control group. We inform all groups about
prevailing inflation. One treatment additionally discusses the erosion of savings, whereas
the other discusses the erosion of loans. Specifically, we explain that unexpected inflation
hurts savers / benefits debtors because it erodes nominal positions and calculate the change
in real value of a savings product / loan due to inflation. By comparing the treatment
groups to the control group, we can thus isolate the effect of information about the erosion
channel of inflation. Post-treatment, we elicit beliefs about nominal positions, own wealth,
and the economy. Moreover, we ask respondents about their spending plans, they perform
a hypothetical real-estate investment, and we track their actual spending using bank data.

2
 We have three main findings. First, we document asymmetric awareness of the erosion
channel. Households on average are well-informed about nominal-asset erosion. For example,
75% of respondents believe the impact of unexpected inflation on fixed-rate savings products
is very negative or negative. However, knowledge about loan erosion is more limited. Only
9% of respondents believe the impact of unexpected inflation on fixed-rate loans is very
positive and 25% believe it is positive. Limited awareness of debt erosion, including of
debtors, suggests muted short-run responses to the redistributive effects of inflation.
Limited awareness is surprising, because survey respondents are on average better edu-
cated than the average German, hold large nominal positions (e.g., 55% have outstanding
debt), and state they care about inflation and its wealth effects. Which characteristics then
predict awareness of the erosion channel? First, awareness only weakly varies by households’
net nominal position. That is, debtors, for example, are not significantly more aware of the
redistributive effects of inflation. Second, business education, wealth, and stock ownership
correlate with awareness. Third, general knowledge about inflation, such as an accurate
perception of current inflation, predicts awareness of the wealth effects of inflation.
Second, providing information on the erosion channel, in particular on debt, affects beliefs
about nominal positions and own real net wealth. We ask respondents to rank household
balance-sheet items in terms of their inflation protection. The savings-erosion treatment
group assigns a slightly worse inflation-protection rank to savings products than the control
group. Respondents who receive the loan treatment rank loans more favorably. In addition,
learning about inflation-induced debt erosion reduces general debt aversion, suggesting that
information effects extend beyond the context of inflation. Knowledge about the wealth
effects of inflation hence impacts beliefs about nominal positions.
Treated respondents also update their estimates of real net wealth. Conditioning on

3
pre-treatment estimates of wealth changes, we find respondents who receive the savings-
erosion information on average estimate their change in real net wealth over the past and
next 12 months 1.6–1.8 percentage points lower than respondents in the control group.
Learning about inflation-induced loan erosion, instead, increases wealth estimates relative to
the control group by 2.5–2.9 percentage points. This latter effect is large: it corresponds to
around 50% of the average estimate of -5.6% in the control group. Respondents also report
they were less aware of the information provided in the loan as compared to the savings
treatment, suggesting the differential treatment effects reflect asymmetric prior awareness.
We then study heterogeneity in the treatment effect on perceived and expected real net
wealth. Respondents with a positive net nominal position (net savers) drive the negative
average effect of the savings treatment, but the economic magnitude remains modest. The
positive wealth effect of the loan treatment comes from respondents with a negative net
nominal position (net debtors), who on average estimate their past- and next-12-month
change in real net wealth five percentage points higher. Furthermore, we study heterogeneity
in respondents’ ability to map the loan-treatment information into their own wealth situation.
The sensitivity of the wealth effect of the loan treatment to net nominal positions is strongest
for respondents with high cognitive abilities and strong interest in the topic of inflation
(Agarwal and Mazumder, 2013; D’Acunto et al., 2022).
Third, learning about inflation-induced nominal-debt erosion has real effects both in
survey data and in actual account-transaction data from our partner bank. We focus on
the effects of the debt-erosion treatment because only this treatment generates significant
effects on beliefs. We study households’ spending plans as well as their actual spending over
the weeks after our survey intervention compared to the previous weeks. Learning about
debt erosion leads to higher planned and actual spending, both in reduced-form and in

4
instrumental-variable estimations. The estimates map into MPCs of close to 3%, in line with
estimates from the literature on MPCs from unrealized capital gains (e.g., Chodorow-Reich,
Nenov, and Simsek, 2021; Di Maggio, Kermani, and Majlesi, 2020; Lettau and Ludvigson,
2004). Our results suggest changes in real net wealth affect households’ consumption response
to inflation, conditional on households being aware of the wealth effects of inflation.
We show another real effect of learning about debt erosion, on hypothetical debt choices.
In the survey, respondents engage in a hypothetical real-estate transaction. The transaction
encompasses choosing both the value of the property and the financing of it, using both equity
and debt. Respondents in the loan treatment choose real estate of similar value compared to
respondents in the control group. However, the two groups differ in their preferred debt/price
ratio, which is significantly higher in the loan-treatment group. Moreover, the loan treatment
tilts the mortgage choice towards fixed- rather than adjustable-rate mortgages, with longer
fixation periods for nominal interest rates. These choices likely reflect treatment-induced
shifts in beliefs about the inflation-hedging properties of long-maturity fixed debt and an
associated reduction in debt aversion.
Overall, our results increase the understanding of redistributive effects of surprise inflation
at the micro level and provide useful moments to target for models featuring heterogeneous
consumers such as Heterogeneous-Agent New Keynesian (HANK) models (e.g., Kaplan, Moll,
and Violante, 2018). Auclert (2019) shows wealth redistribution caused by surprise inflation
boosts the economy because the beneficiaries of it, net debtors, have higher MPCs than the
losers. Limited awareness of inflation-induced debt erosion suggests muted debtor responses
in the short run. Extending the HANK framework by allowing for deviations from full
information rational expectations (FIRE), for example via information frictions, would be a
fruitful avenue for future research. More generally, a positive covariance between individual

5
MPCs and exposure to economic shocks is a core amplification mechanism in HANK models
(e.g., Bilbiie, 2008; Patterson, 2023). Allowing additionally for heterogeneity in the awareness
of these shocks might help to better understand their aggregate consequences (Pfäuti and
Seyrich, 2022; Pfäuti, Seyrich, and Zinman, 2023).
We focus on a sample of German households because this setting allows us to combine
administrative data from a bank with a customized survey. This combination is unique and
enables us not only to study prior awareness of consumers about inflation-induced wealth
redistribution but also to provide causal estimates via an information provision experiment.
The existing literature that we discuss below either assumes everyone is fully aware of the
redistributive effects, or studies consumption-savings decisions in realized data that do not
allow for distinguishing frictions from limited awareness.
Nevertheless, our setting also has potential shortcomings. German households are noto-
riously concerned about inflation. Moreover, our sample consists of relatively educated and
wealthy Germans who care about and have accurate perceptions of inflation. One might
hence expect our sample is already well aware of the effects of inflation. Yet, Weber et al.
(2023) show that German households update their expectations to information about in-
flation in a similar way to consumers in the U.S. and other European countries, once they
condition on the level of inflation because household attention to inflation is endogenous
(Cavallo, Cruces, and Perez-Truglia, 2017; Maćkowiak, Matějka, and Wiederholt, 2023). In
addition, we find even mortgagors in our sample are largely unaware of nominal-debt ero-
sion. Consumers might be learning about the Fisher channel as high inflation persists, yet
Germans had been concerned about inflation even prior to its surge (Braggion et al., 2024).
We therefore argue we likely uncover a systematic failure of households in understanding
inflation-induced nominal-debt erosion.

6
 Related literature We contribute to several strands of the literature. First, we build
on work on inflation-induced debt erosion that goes back to at least Fisher (1933). Doepke
and Schneider (2006) document the net nominal positions of different economic actors in the
U.S. over time and study the redistributive effects of various inflation scenarios within and
across these groups. A key finding is young households with large debt positions benefit at
the expense of wealthy middle-aged households.2 Auclert (2019) and Pallotti (2023) study
theoretically the redistributive effects of monetary policy on spending based on differential
MPCs of winners and losers of the wealth effects of inflation. Coibion et al. (2017) document
inequality dynamics due to monetary policy, including by household net wealth. Brunner-
meier et al. (2023) and Gomes, Jermann, and Schmid (2016) analyze how unanticipated
inflation transmits to the real economy through nominal corporate debt. We are the first to
study how households perceive and adjust to the redistributive effects of inflation, using an
information provision experiment paired with administrative bank data.
Second, we contribute to a burgeoning literature on households’ response to inflation. Ev-
idence on the sensitivity of spending to inflation expectations is mixed. Some studies report
a positive effect of inflation expectations on spending, consistent with an intertemporal-
substitution motive, whereas other studies find households associate higher future inflation
with worse economic outcomes and hence spend less if researchers do not condition on eco-
nomic outlook (Kamdar (2018); see also D’Acunto, Malmendier, and Weber (2023), Jiang
et al. (2024), and Weber et al. (2022) for recent surveys of work in this area). We add to
this literature by showing that wealth effects of inflation mediate the consumption-inflation
sensitivity once households are aware of the effects. Differences across studies in how house-
holds update their real-net-wealth estimates to inflation can hence possibly also rationalize
2
Recent papers that underline the important role of the Fisher channel on the distributional impact of
the post-pandemic inflation include Del Canto et al. (2023), Ferreira et al. (2024), and Yang (2023).

7
part of the mixed results in the literature.
Beyond effects on spending, the literature studies how households seek protection from
inflation. Botsch and Malmendier (2023) find that experiencing high inflation in the 1970s
is associated with an aversion to adjustable-rate mortgages. Leombroni et al. (2020) use an
asset-pricing model to link high inflation expectations in the 1970s to a portfolio shift toward
housing. Braggion, von Meyerinck, and Schaub (2023) find investors purchase fewer stocks
when facing higher inflation during the hyperinflation in Germany in the 1920s. Schnorpfeil,
Weber, and Hackethal (2024) use an RCT to study the effects of beliefs about inflation on
portfolio choices. We add to these papers by showing that learning about inflation-induced
debt erosion has a positive effect on beliefs about debt and increases debt financing.
Finally, a recent literature in economics employs information-provision experiments to
study how consumers, who make consumption, savings, and investment decisions in the
field, perceive economic policies and phenomena, as well as their effects (e.g., Andre et al.,
2022; D’Acunto et al., 2022).3 We contribute to this literature by analyzing how households
perceive the wealth effects of inflation, how they map information about these effects into
their own economic situation in an experimental setting, and how exogenous changes in their
economic situation feed into real-world choices. A methodological novelty is explaining an
economic mechanism rather than purely providing information about a single variable, such
as a point forecast for inflation, as is common in the macro literature (e.g., Coibion, Gorod-
nichenko, and Weber, 2022). We thereby manipulate the interpretation of the consequences
of inflation instead of exogenously varying inflation expectations.

3
See Haaland et al. (2023) for a review on how to run surveys and implement information experiments.

8
2 Experimental design and data

In this section, we discuss the survey design and characteristics of our sample. Section 2.1 fo-
cuses on the survey design, emphasizing the information-provision part of the survey, whereas
Section 2.2 describes the bank data as well as the sample composition and characteristics.

2.1 Experiment

We implemented the RCT on more than 3,800 bank customers in July 2022. The survey
consists of three sections: a pre-treatment section on demographics, respondents’ balance
sheet, and economic beliefs; an information-provision section; and a post-treatment section
on economic beliefs, hypothetical and planned economic choices, and additional background
characteristics. Online Appendix B contains the survey questions translated to English.
Pre-treatment section Respondents start by answering two questions on their educa-
tional background. We then assess respondents’ marginal propensity to save, consume, and
pay down debt in the three months following a hypothetical one-time payment of e 10,000
(see, e.g., Fuster, Kaplan, and Zafar, 2020), framed as either a gain or a loss. In addition,
respondents receive a slight variation of this question asking about their hypothetical re-
sponse to noticing by chance that their own net wealth is e 10,000 higher or lower than
previously thought. We randomize the order of the two questions. The two questions allow
us to infer differential hypothetical spending responses to unexpected actual payments versus
unexpected changes in perceived net wealth with no actual cash flows involved.
Respondents then answer questions about the economy and its impact on their net wealth.
The questions include the importance of macroeconomic factors, such as inflation and GDP,
for their own wealth; the consumption response to recent changes in these factors; the percep-

9
tion of current and forecasts of future inflation; and whether holding cash, fixed-rate savings
products, stocks, real estate, and fixed-rate loans provide a hedge against an unexpected
surge in inflation. These questions aim to elicit existing knowledge about inflation-induced
erosion of nominal assets and liabilities.
Furthermore, we ask respondents to provide a decomposition of their balance sheet into
nominal assets (we mention cash, bonds, and life insurances as examples), stocks, real estate,
other assets (we mention vehicles and gold as examples), and nominal liabilities (divided into
mortgages and consumer loans).4 Respondents state the value of each balance-sheet item as
a fraction of their gross wealth. Given the cognitive burden of this question, we carefully
explain the concept of gross wealth and provide several examples. Moreover, we provide
a warning message if the sum of respondents’ assets differs from their gross wealth. The
median time survey participants spent on the balance-sheet-decomposition screen is more
than two minutes, indicating that they carefully answer the question.
Respondents also estimate the change in their real net wealth over the past 12 months.
We follow the design in the New York Fed Survey of Consumer Expectations (Armantier
et al., 2017) and first elicit a directional estimate before asking for a point prediction, on a
scale from -60% to 60%. We familiarize respondents to net and gross wealth in the preceding
balance-sheet-decomposition task, by providing a simple sample balance sheet to differentiate
between gross and net wealth and by stating their net wealth as a fraction of their gross
wealth. As we discuss in Section 2.2 below, the elicited responses appear meaningful.
Treatment section The second part of the survey contains the information provi-
sion experiment. The objective of the intervention is to generate exogenous variation in
knowledge about inflation-induced erosion of nominal assets and liabilities. To generate this
4
We follow Adam and Zhu (2015) in classifying life insurances as nominal claims, as insurance companies
in the euro area predominantly invest in nominal assets.

10
variation, we randomly split the sample into three equally-sized groups, two of which receive
information and one serves as a control group. The two treatments are similar, but one
focuses on the erosion of nominal assets, whereas the other focuses on the erosion of nominal
liabilities. The savings-erosion treatment information are as follows:

The current inflation rate in Germany is 8.7%, the highest rate in more
than 70 years. That is, goods and services priced at e100 one year ago now
cost e108.7 on average. This price increase has a relatively negative effect
on savers: the savings amount (e.g., checking account, bond, life insurance) is
unchanged nominally or lower, but worth less in real terms as a consequence of
money depreciation.
As an example, consider a e50,000 savings product with a three-year maturity
that you took out one year ago. The real value of the savings product has already
fallen sharply, and will depreciate further if inflation remains high:
e50,000 savings value one year ago ⇓ e38,800 real value today
The inflation-induced savings depreciation thus has a negative effect on the
real net wealth of savers.
Note: the numbers come from current calculations by the Universities of Chicago
and Frankfurt (calculation details).

Respondents in the loan-erosion treatment group read the following text:

The current inflation rate in Germany is 8.7%, the highest rate in more
than 70 years. That is, goods and services priced at e100 one year ago now
cost e108.7 on average. This price increase has a relatively positive effect on
borrowers: the loan amount is unchanged nominally, but worth less in real terms
as a consequence of money depreciation.
As an example, consider a e50,000 loan with a three-year maturity that you took
out one year ago. The real value of the loan has already fallen sharply, and will
depreciate further if inflation remains high:
e50,000 loan value one year ago ⇓ e38,800 real value today
The inflation-induced loan depreciation thus has a positive effect on the real
net wealth of borrowers.

11
 Note: the numbers come from current calculations by the Universities of Chicago
and Frankfurt (calculation details).

Respondents assigned to the control group receive only the first two sentences on infla-
tion. We provide this information, instead of having a fully passive control group, because
information on current inflation being historically high can have a confounding effect on
post-treatment beliefs and choices, for example via effects on income expectations or labor-
supply decisions. By comparing respondents in the treatment groups with respondents in
the control group, we can identify the effect of learning about the inflation-induced erosion
of nominal assets and liabilities and absorb any effect that the information about the current
inflation level might have.
The real-value numbers in the treatments come from a simple present-value calculation.
Real changes in the positions’ present value occur because future nominal cash flows are
discounted at higher interest rates. Discount rates increase linearly over the course of one
year, because inflation in 2021–22 increased gradually. We disregard any reallocation or
reinvestment after the products’ maturity. We thereby assume exposure to nominal erosion
ends at maturity, or that inflation suddenly drops back to its baseline level.5 At the end of
the treatment text, respondents can click on a button to see calculation details.6
Post-treatment section Following the information intervention, we first elicit 12-
month-ahead expectations for real-estate prices, the unemployment rate, interest rates, and
respondent’s income. If the treatments affect some of these expectations, it might be through
these expectations that respondents alter their economic choices. For example, Coibion et al.
(2024) show that people’s macroeconomic outlook causally affects their consumption. We

5
This assumption is similar to the lower-bound scenario by Doepke and Schneider (2006), who calculate
wealth effects of inflation assuming that households switch to inflation-indexed securities once their nominal
positions expire.
6
19% of respondents click on the button; they do not differ in their survey responses.

12
investigate this possible alternative channel below.
We then ask survey participants about planned spending on nondurables and durables.
Following Roth and Wohlfart (2020), respondents state their consumption plans for multiple
nondurables categories over the next four weeks relative to the previous four weeks. Response
options, on a five-point scale, range from “much less” to “much more.” The short time window
mitigates concerns that changes in inflation or other economic news confound the responses.
On durables spending, we ask whether respondents plan to make major purchases over the
next 12 months, such as buying a car or an apartment. If they do, we ask them to state
the amount they plan to spend. These self-reported spending plans are useful because (i)
we cannot observe actual account transactions for all respondents, (ii) we can distinguish
between nondurables and durables spending, and (iii) we can corroborate treatment effects
on actual spending with spending plans.
We again elicit respondents’ perception about the past-12-month change in their real net
wealth, as well as their expectation for the next 12 months. This time, respondents directly
provide a point estimate, with answer options between -60% and 60%. We elicit posteriors
for real net wealth changes in a different format compared to priors to mitigate concerns of
survey fatigue and demand effects, following Coibion, Gorodnichenko, and Weber (2022).7
We are therefore able to study the instantaneous revision in own-wealth perceptions and
expectations in response to the provided information.
The final belief questions relate to nominal assets and debt. We elicit beliefs about the
relative wealth protection that savings products, stocks, real estate, and fixed-rate loans
provide. In addition, we ask about debt aversion (Almenberg et al., 2021). Moreover, re-
spondents engage in a hypothetical real-estate transaction. They choose a preferred purchase

7
De Quidt, Haushofer, and Roth (2018) show that demand effects tend to be small in settings like ours.

13
price and mode of financing, with up to e 500,000 equity and a mortgage of up to e 500,000.
Respondents also choose whether they prefer an adjustable- or a fixed-rate mortgage, and if
they choose the latter, the length of the fixation period.
The last section of the survey elicits respondents’ risk tolerance, money illusion (Shafir,
Diamond, and Tversky, 1997), and the value of their nominal positions. After completion,
we ask respondents how interesting they found the survey and they can leave comments.

2.2 Data

Survey administration We run the survey experiment in partnership with a large German
bank. The bank offers both retail-banking and brokerage services. In July 2022, the bank
sent out a short email to around 215,000 customers, inviting them to participate in a survey
on inflation administered by Goethe University Frankfurt. The survey was in the field for
two weeks. After the first week, the bank sent out a reminder email, informing customers
they have one week left to complete the survey. During the field period, inflation in Germany
was at 8.7%, the highest rate in more than 70 years.
Overall, 3,846 bank customers complete the survey. 45% of respondents complete the
survey within a day after the bank sent the invitation email, and 43% of the responses
came in within a day after the reminder email. The overall response rate is 1.8%, which is
comparable to other surveys of the bank, and the median response time is 18.3 minutes.8
Administrative bank data The partnering bank provides us with data on customer
demographics and, importantly, categorized account transactions. Demographic information
include age, gender, marital and employment status, and zip code. Account transactions

8
When compared to bank customers who receive a survey invitation but do not participate, respondents
are more likely to be male, use the partnering bank’s online banking, have outstanding loans with the bank,
have a securities portfolio with the bank, and have higher income.

14
come from the bank’s personal-financial-management (PFM) tool, which registers customers’
in- and outflows and classifies them into more than 50 categories.9
Broad spending categories include expenses on living (e.g., groceries and clothing); hous-
ing (e.g., rent and furniture); leisure (e.g., restaurants and events); mobility (e.g., cars and
fuel); health (e.g., pharmacies and hospitals); occupation and education (e.g., office supplies
and tuition fees); insurances, loans, and investments; cash withdrawals; credit cards; and
online shopping. Income categories include salaries, other forms of regular income (e.g.,
pension and rental income), additional sources of income (e.g., tax refunds and children’s
allowances), and capital income (e.g., dividends).
Around one third of transactions are uncategorized. These transactions usually con-
stitute payments to a party unknown to the categorization algorithm, such as peer-to-peer
transactions or transfers between accounts. We omit uncategorized transactions for our main
consumption measures, because we do not know whether these transactions reflect consump-
tion. In a robustness test, we incorporate uncategorized transactions identified by the bank
as likely representing consumption to our spending measures.
We consider three measures of consumption. Total spending subtracts from all cate-
gorized outflows those related to investments, insurances, and loans. We also split total
spending into discretionary and nondiscretionary spending, following D’Acunto, Rossi, and
Weber (2023). Discretionary spending includes categories such as clothing, leisure, cash
withdrawals, and online shopping. Nondiscretionary spending is the difference between to-
tal and discretionary spending. Appendix Table A1 details all categories as well as the
subcomponents of our different measures of consumption.

9
The bank defines categories based on the classification used by the German Federal Statistical Office.
The tool is similar to personal-finance apps such as Mint; an important difference, however, is that the tool
is embedded in the bank’s online-banking environment.

15
 Sample selection Together with the bank, we select the survey sample based on two
criteria. First, because of the importance of observing actual consumption choices, all bank
customers with an activated PFM tool and who had received any cash inflows over the past
six months receive a survey invitation. We randomly incentivize survey participation with
either a voucher, or participation in a lottery. All payoffs are in the form of online-shopping
vouchers.10 Second, bank customers who do not satisfy the above criteria but have an
outstanding mortgage or consumer loan at the partnering bank receive a survey invitation,
which comes with the lottery incentive (15% of the selected sample). Given their outstanding
debt, these customers should be particularly exposed to inflation-induced debt erosion.
Given noise in survey data (D’Acunto, Fuster, and Weber, 2021), we take two steps to
filter respondents. First, we omit 563 respondents who make wrong or implausible entries in
the balance-sheet-decomposition task: those who enter a negative share of a balance-sheet
item relative to gross wealth; a share greater than one, that is, the value of a single balance-
sheet item exceeds gross wealth; a share of nominal assets equal to zero; or a value of an
outstanding mortgage, consumer loan, or the sum of the two that is equal to gross wealth.11
Second, we drop respondents who take less than seven minutes or more than 120 minutes to
complete the survey, roughly corresponding to the 1.5th and 98.5th percentiles. After these
two steps, 3,190 individuals remain in the baseline sample. The results are robust to not
performing the two screening steps.
Our sample to analyze treatment effects on spending using bank data comprises 2,671
bank customers. We arrive at this smaller sample because some survey participants with

10
Survey responses do not systematically differ by type of participation incentive.
11
Most respondents (10%) that we screen out enter mortgage and/or consumer-loan values that sum up
to the value of their gross wealth. Because we ask respondents for the sum of assets to equal 100% of gross
wealth, some respondents might have mistakenly assumed that the sum of all liabilities should also equal
100% gross wealth. We are unable to calculate the net nominal position of these survey respondents.

16
outstanding debt at the bank have not activated the PFM tool, and hence we cannot observe
their account transactions. Moreover, we want to ensure that our bank-customer sample
actively uses the observed bank accounts. We therefore require the sample to receive at least
e 100 as average regular income per month on observable accounts.12
Sample characteristics Table 1 reports summary statistics for our sample, with basic
demographics in the top panel. 45% of respondents are female and the average age is 48
years. The educational level is relatively high, with 48% having completed college, around
20% of them in business. We benchmark our sample against the most recent (2017) wave
of the Bundesbank’s Panel on Household Finances (PHF), which is a representative survey
of German households’ finances and comprises the German data for the ECB’s Household
Finance and Consumption Survey. In the PHF, the share of the population with completed
college education is 29%. 72% of the sample is employed (54% in the PHF).
Regarding household finances, respondents’ average gross wealth is e 386,000 (e 238,000
in the PHF). Average nominal assets are 43% (37% in the PHF) and mean nominal debt is
17% of gross wealth (10% in the PHF). The average net nominal position is hence 26% of
gross wealth. We follow Auclert (2019) and Doepke and Schneider (2006) by defining the net
nominal position as all nominal assets minus all nominal liabilities. Similar to Auclert (2019),
we exclude indirect nominal positions, which originate from investment intermediaries and
the ownership of firms, to reduce complexity for respondents. We operationalize this measure
by taking the sum of cash and fixed-rate savings products minus the sum of mortgage and
consumer loans.13 Respondents estimate the value of each balance-sheet position relative to
their gross wealth, so the net nominal position is naturally scaled by gross wealth. 54% of
12
The consumption results do not rely on the cutoff choice.
13
We do not distinguish between fixed- and variable-rate loans in the balance-sheet-decomposition task
to reduce complexity and because around 90% of loans in Germany have a fixed rate. Our bank partner
only offers fixed-rate loans (though mortgagors in the prolongation period can have a variable rate).

17
respondents own stocks and 59% own real estate (20% and 44% in the PHF, respectively).
The middle panel of Table 1 presents statistics on average monthly spending and income.
We calculate the individual-level average based on the six months preceding the survey. Total
spending amounts to e 1,844 for the average respondent, with around 50% of it being dis-
cretionary spending. Average regular income, constituting of, among other things, salaries,
pensions, children’s allowances, and rental income, is e 2,862. Gross individual income in
the PHF is e 2,274 on average.14
The bottom panel of Table 1 reports statistics for respondents’ beliefs, elicited prior to
the information provision. The average perceived inflation rate is 8.8%, close to the actual
inflation rate of 8.7%. 75% of respondents have a perception error of at most 1.5 percent-
age points, likely because they actively acquire information about inflation when it is high
(Cavallo, Cruces, and Perez-Truglia, 2017; Weber et al., 2023). Respondents expect inflation
to remain high, with a mean one-year forecast of 10.4% and a five-year forecast of 10.7%.
Cross-sectional dispersion increases with the forecast horizon. The estimated average change
in the respondents’ real net wealth over the past 12 months is -7.5%. Many respondents state
that inflation matters for their own wealth, and relatively more so than GDP growth and
the level of interest rates. Overall, respondents on average are relatively well-educated and
wealthy, have accurate inflation perceptions, and are subjectively (given their beliefs) and
objectively (given their large nominal positions) strongly exposed to changes in inflation.
Integrity of randomization Appendix Table A2 reports a balancing table for the two
treatment groups and the control group. Our sample is largely well-balanced across groups
for a wide array of demographic characteristics and perceptions and expectations. The three

14
In our bank data, we observe net flows, such as salaries net of taxes. We restrict income to be at
least e 100 to ensure some account usage; nonetheless, our respondents can receive income on accounts
unobservable to us. We include the same restriction in the PHF for consistency.

18
groups are also statistically indistinguishable from each other in terms of actual average
spending and income. Overall, only a few imbalances occur, such as for gender and wealth.
To address the slight imbalances, we include a set of control variables in all specifications.
Reliability of survey responses We try to alleviate concerns about inaccurate or
untruthful reporting in our survey (Stantcheva, 2023). First, we attempt to mitigate in-
accuracies in responses by carefully explaining the concepts we aim to elicit, tracking time
spent on the questions, and filtering the sample. In addition, we test for reliability of survey
responses using administrative data provided by the bank partner (Dutz et al., 2022). We
focus on the reliability of reported outstanding debt because (i) these data are particularly
important for our analysis, (ii) the balance-sheet-decomposition task is demanding, and (iii)
holding debt in Germany may be stigmatized (e.g., D’Acunto, Schnorpfeil, and Weber, 2022).
Appendix Figure A1 shows the association between outstanding debt reported in the
survey and at the partnering bank in the month of survey participation. For debt reported in
the survey, we take the mid point of the debt-balance range respondents select. Outstanding
debt at the partnering bank increases monotonically with survey-reported debt. Importantly,
virtually no survey respondent has outstanding debt with the partnering bank when stating
in the survey they have zero debt. The debt level we observe in the bank data is generally
lower than what survey participants report, indicating that other lenders originated some
of the debt and underscoring the importance of using surveys to elicit the overall household
balance-sheet positions. The correlation between survey-reported and bank debt is 0.49.

19
3 Prior knowledge about the wealth effects of inflation

Level of knowledge We start by documenting respondents’ prior knowledge of the wealth

effects of inflation. Figure 1 shows a large share of respondents appears aware of inflation-
induced erosion of nominal assets. Panel A displays the distribution of beliefs about how a
surprise increase in inflation affects nominal positions. 88% of respondents believe that the
impact on cash on hand is “very negative” or “negative,” whereas that number is 75% for
fixed-interest savings products. Panel B of Figure 1 reports respondents’ relative ranking
of nominal assets, stocks, real estate, and nominal loans in terms of wealth protection pro-
vided against unexpected inflation.15 70% of respondents ascribe relatively poor inflation
protection (rank 3 or 4) to nominal assets.16
Pre-existing knowledge about the erosion of nominal liabilities appears more limited,
however. 25% of respondents believe that the impact of unexpected inflation on those with
a fixed-rate loan is “rather positive,” and only 9% believe it is “very positive” (top panel).
Similarly, 62% assign a relatively poor inflation-protection rank, of 3 or 4, to fixed-rate loans
(bottom panel). Overall, the findings indicate asymmetric knowledge about the erosion of
nominal assets versus nominal debt. Incomplete knowledge is necessary for the information
intervention to plausibly have scope to affect beliefs and hence choices. We therefore expect
stronger responses to the loan-erosion than to the savings-erosion treatment.
Heterogeneity in knowledge According to models of endogenous information acqui-
sition, awareness of the wealth effects of inflation might vary with exposure to the effects
(e.g., Kindermann et al., 2021; Maćkowiak, Matějka, and Wiederholt, 2023). Households

15
We show statistics for the control group only, because we ask this question post-treatment. We discuss
how the treatments affect responses to this question below.
16
The majority of respondents assign the best inflation protection (rank 1) to real estate, consistent with
?, who show that inflation protection is a key motivation for homeownership.

20
with larger nominal positions might hence exhibit greater ex-ante awareness of the Fisher
channel. Figure 2 documents how prior knowledge about the wealth effects of inflation varies
by respondents’ net nominal position. Consistent with greater awareness among those who
are more exposed, respondents with a negative net nominal position have a more positive
perception about the impact of unexpected inflation on fixed-rate loans (left scale), but the
magnitudes are small.17 Beliefs about the impact on savings products do not vary by net
nominal position (also left scale).18 Small differences in knowledge by net nominal position
do not translate into differences in respondents’ self-reported consumption response to the
recent increase in inflation (right scale).19
Which other individual characteristics explain knowledge about the wealth effects of in-
flation? Table 2 shows results of regressing beliefs about the impact of unexpected inflation
on balance-sheet items, including nominal assets and debt.20 Two results stand out. First,
business education, higher wealth, and stockholdings predict knowledge about the inflation-
induced erosion of nominal assets (Columns 1–2) and nominal debt (Column 5). Second,
general knowledge about inflation correlates with awareness about the wealth effects of in-
flation. Especially respondents who consider inflation to be important for their wealth, have
an accurate perception of current inflation, or expect inflation to be lower in five years than
they perceive it today appear more knowledgeable. Appendix Table A3 shows results for a
more comprehensive set of correlates.

17
We abstain from analyzing differences in the inflation-protection ranking (bottom panel of Figure 1) by
net nominal position because of the smaller sample size coming from a restriction to the control group.
18
Appendix Table A3 confirms these patterns in a multivariate analysis, controlling for a wide array of
demographic characteristics, perceptions and expectations, as well as wealth.
19
Consistent with limited variation in prior knowledge by net nominal position, Appendix Figure A2 shows
that knowledge about the information provided in the treatments does not vary by net nominal position. If
anything, knowledge about savings erosion is lower for respondents with a positive net nominal position.
20
The number of observations is lower than the 3,190 of the baseline sample because we trim the 1% tails
of perceived and expected inflation, elicited as point estimates.

21
4 The effects of information treatments on beliefs

In this section, we investigate how providing information about the wealth effects of inflation
in the survey experiment shifts beliefs of our participants. To characterize average effects of
the information interventions, we estimate variants of the following equation:

2
X
posteriori = const + βj I{i ∈ treat j} + γ prior rnwi + controlsi + errori , (1)
j=1

where posteriori is a post-treatment measure of beliefs of respondent i. I{i ∈ treat j}

indicates whether respondent i received treatment j. The omitted category is the control
group, so coefficients {βj }2j=1 can be interpreted as being relative to the control group.
Because we randomize the treatments, controls help with the precision of the estimates but
have no material effect on the point estimates. prior rnwi refers to estimates of past real-
net-wealth change. We control for two wealth estimates because the elicitation occurs in
two steps, first with a directional and then with a point estimate. Other controlsi include
a quadratic polynomial in the respondent’s age, risk tolerance measured on a 1–5 ordinal
scale, the log of gross wealth (1% tails winsorized), and a rich set of dummy variables for
a respondent’s gender, marital status, educational level (equal to one for college degree or
higher), a business degree, employment status, debt and stock holdings, net nominal position
relative to gross wealth (<-50%, -50% to <-25%, etc.), accuracy of perceived current inflation
(<1.5 pp deviation from actual rate), expectation that inflation will be lower in five years (1%
tails of inflation estimates trimmed), whether survey participation is based on the voucher
or lottery incentive, and whether participation is after the reminder email sent by the bank.

22
4.1 Effects on beliefs about nominal positions

Table 3 presents the results of estimating Equation 1 with and without controls. In Columns
1–4, the dependent variable captures beliefs about relative inflation protection provided
by nominal assets and debt. The survey question reads: “With which of the following
financial instruments would you expect the most positive real-net-wealth impact in times
of unexpectedly high inflation?” Respondents rank nominal assets, stocks, real estate, and
nominal liabilities. Responses are hence on a 1–4 ordinal scale, which we reverse for simplicity
so that a higher number indicates a better hedge against unexpectedly high inflation.21 In
Columns 5–6, the dependent variable proxies debt aversion, based on agreement with the
statement “I am uncomfortable with taking on debt.” Five possible responses range from
“completely disagree” to “completely agree.” We standardize the dependent variables.
Provision of information about the wealth effects of inflation impacts beliefs about nom-
inal positions. Respondents who learn about inflation-induced erosion of nominal savings
attach a significantly poorer inflation-protection ranking to nominal assets (Column 1–2).
Provision of the loan-erosion treatment has a significantly positive effect on the ranking of
nominal debt (Columns 3–4). We do not find strong evidence of cross-learning, that is, re-
spondents who learn about the erosion of nominal assets due to unexpectedly high inflation
do not infer high inflation also erodes nominal debt, and vice versa.22 Moreover, learning
about inflation-induced loan erosion reduces individuals’ debt aversion, indicating that ef-
fects extend beyond the direct inflation context (Columns 5–6). Overall, knowledge about
the wealth effects of inflation impacts beliefs about nominal assets and debt, opening up the
possibility to affect economic choices, which we discuss in Section 5.2.

21
The reduced sample size is due to the fact that respondents were not required to respond to the question.
22
A statistically weakly significant coefficient on the loan-treatment indicator in Columns 1–2 is partially
mechanical, compensating for the treatment-induced effect on the ranking of nominal debt.

23
4.2 Effects on perceived and expected real net wealth

Average effects Table 4 quantifies the treatment effects on the perceived and expected
change in real net wealth, again based on estimating Equation 1. Importantly, we control
for the pre-treatment estimate of the past-12-month change in real net wealth. In Columns
1–6, results are from OLS estimations. The savings treatment has an insignificantly negative
effect on the perceived wealth change over the past 12 months, the expectation over the next
12 months, and the sum of the two changes. The loan treatment, instead, significantly in-
creases perceived and expected real net wealth. In Column 7, results are from Huber-robust
regressions of perceived and expected wealth changes. Huber regressions allow us to system-
atically control for outliers and influential observations (see, e.g., Coibion, Gorodnichenko,
and Ropele, 2019), which is useful because of rounding, heaping, mean reversion, and survey
noise. We find that estimates are quite similar to the ones using OLS, with coefficients being
more precisely estimated. The wealth effect is economically sizable, corresponding to nearly
50% of the average estimate of -5.6% to -6.3% in the control group. In estimations below
that involve perceptions and expectations of wealth changes, we focus on Huber regressions
to mitigate the impact of extreme observations and for increased precision.23
In addition to quantitative wealth-change estimates, we analyze treatment effects on
qualitative wealth perceptions. Appendix Table A4 shows survey respondents in the loan-
treatment group have significantly more positive perceptions about their wealth in the con-
text of unexpected inflation than survey participants in the control group. Specifically, the
loan treatment induces more positive responses to a question on the effect of inflation on
the respondents’ real net wealth over the last twelve months (Columns 1–2). The loan

23
The variation in the number of observations across specifications is because (i) Huber regressions weight
observations, with particularly influential observations receiving a weight of zero, and (ii) regressions with
controls (Columns 2, 4, 6, and 7) include trimmed inflation perceptions (1% tails).

24
treatment also increases agreement to the statement that with their current balance sheet,
respondents are well-prepared for times of high inflation (Columns 4–5). The effects of the
savings treatment are insignificant throughout.
To sum up, the loan-erosion treatment leads to higher perceived and expected real net
wealth, whereas the effect of the savings-erosion treatment is limited. Success of the informa-
tion provision in shaping wealth estimates likely reflects respondents’ limited prior knowledge
about inflation-induced erosion of nominal debt. Differential prior knowledge about the ero-
sion of nominal assets versus debt also shows up in respondents’ stated knowledge about
the treatment information they receive. Appendix Figure A2 shows that respondents report
less awareness of the loan- than the savings-erosion information. Our findings imply that
consequences of the redistributive effects of unexpected inflation are likely muted in the short
run in light of limited awareness of inflation-induced erosion of nominal debt.
The information provision on nominal-debt erosion thus induces significant exogenous
variation in perceptions and expectations of real-net-wealth changes. As a result, this treat-
ment can serve as an instrument to help us identify how and whether wealth effects of
inflation due to nominal-debt erosion affect spending. Information on nominal-savings ero-
sion, instead, does not shift beliefs sufficiently to serve as a strong first stage. In Section 5.1,
we thus focus on the effects of the loan treatment on spending, through changes in perceived
and expected real net wealth.
Heterogeneity by net nominal position Do treatment effects on real-net-wealth
estimates vary with net nominal exposure? Table 5 reports results of regressing changes in
real net wealth on a treatment indicator interacted with respondents’ net nominal position.
Estimates of wealth changes again focus on the past 12 months (Columns 1–2), the next 12
months (Columns 3–4), and the sum of the two (Columns 5–6). The effect of the savings

25
treatment is insignificant for respondents with a negative net nominal position and respon-
dents with a positive net nominal position have a significantly lower wealth estimate than
those with a negative net nominal position. Respondents in the loan-treatment group with
a negative net nominal position relative to respondents in the control group with a negative
net nominal position report significantly higher changes in real net wealth. In our preferred
specification on perceived past and expected future wealth changes with the full set of con-
trols (Column 6), the difference between the two groups amounts to five percentage points.
This effect nearly fully offsets the reported average change in real net wealth of -5.8% of
respondents with a negative net nominal position in the control group.
Figure 3 graphically illustrates treatment effects by net nominal position. Based on a
regression of changes in real net wealth over the past and next 12 months, we plot coefficients
and 95% confidence bounds on the interaction of a treatment indicator and respondents’ net
nominal position. We group the net nominal position into bins of less than -50% of gross
wealth, -50% to less than 0%, 0% to 50%, and more than 50%. The estimated change in real
net wealth of respondents with a net nominal position of less than -50% to gross wealth is
7.1 percentage points higher in the loan-treatment group than in the control group, whose
average estimate is -1.2% (Panel A). Moreover, the treatment effect monotonically falls as
the net nominal position increases. The pattern is comparable albeit much weaker for the
savings-erosion treatment: respondents with a net nominal position of more than 50% of
gross wealth have a 2.9 percentage points lower estimate of real-net-wealth changes than
similar respondents in the control group.24 Appendix Figure A3 similarly shows treatment
effects on wealth estimates, but decomposes the net nominal position into nominal debt and
savings as a fraction of gross wealth. We find the loan treatment has a large, significant
24
Positive coefficients for those with a negative net nominal position are unlikely to reflect cross-learning,
because Table 3 shows that the savings-erosion treatment does not impact beliefs about nominal debt.

26
effect on wealth estimates for highly levered respondents, whereas the savings-treatment
effect increases only modestly with the fraction of wealth held in nominal savings.
Two additional pieces of evidence corroborate the mediating role of net nominal exposure
for the effects of the loan treatment. First, Appendix Table A4 documents respondents in
the loan-treatment group with a negative instead of a positive net nominal position express
a more positive qualitative perception about the effect of the recent rise in inflation on their
real net wealth (Column 3). Similarly, agreement to the statement that with their balance
sheet, respondents are well-positioned to cope with high inflation is stronger for those with
a negative net nominal position (Column 6). In both cases, however, differences by net
nominal position are not statistically significant.
Second, Appendix Figure A4 shows the cross-sectional raw-data relationship between
respondents’ prior estimate about the past-12-month change in real net wealth and posterior
estimates about the past- and future-12-month change in real net wealth.25 In Panel A,
we restrict the sample to respondents with a positive net nominal position. The panel
reveals that, among net savers, those who receive the savings treatment tend to have lower
posterior estimates of wealth changes. In Panel B, the sample includes only respondents with
a negative net nominal position. Among net debtors, posterior wealth-change estimates are
higher in the loan-treatment than in the control group. Moreover, respondents with extreme
prior real-net-wealth estimates do not appear to drive the effect of the treatment.
Comprehension of treatment information Incorporating information about inflation-
induced nominal-debt erosion into estimates of changes in real net wealth is challenging. Two
features of our study likely help respondents to map the treatment information into their
own wealth situation. First, respondents are well-educated on average, interested in inflation,
25
We do not elicit the expected 12-month change in real net wealth pre-treatment to avoid survey fatigue
when asking the same question twice (Stantcheva, 2023).

27
and take their time to go through the survey. Second, we carefully introduce concepts such
as changes in real terms and net wealth, with respondents providing plausible responses on
questions involving these concepts. Nonetheless, variation in respondents’ ability to incorpo-
rate the treatment information into their estimates of real net wealth may exist. Specifically,
differential treatment reactions may reflect heterogeneity in cognitive abilities and interest
in the topic of inflation (Agarwal and Mazumder, 2013; D’Acunto et al., 2022).
We analyze heterogeneity in respondents’ ability to map the loan-treatment information
into their own wealth situation in Figure 4, displaying results of regressing real-net-wealth
changes on the loan-treatment indicator for different subsamples. In Panel A, we limit the
sample to respondents with a negative net nominal position. Importantly, we additionally
split the sample by education and beliefs about inflation as proxies for cognitive abilities and
interest in inflation. We argue respondents who completed higher education, expect inflation
to be lower in five years, have a relatively accurate perception of current inflation (deviation
from the actual rate of at most 0.5 percentage points), or consider inflation to be relatively
important should be better able to incorporate the loan-treatment information into beliefs
about their own real net wealth. We indeed find treatment effects on real net wealth of highly
educated respondents or those with general knowledge or interest in inflation are relatively
high and significant (6.9–8.3 pp) compared to those without these characteristics (2.1–3.6
pp).26 Importantly, on the subsample of respondents with a positive net nominal position
(Panel B), the loan-treatment effects on real net wealth are close to zero, independent of
the subsample we consider. These findings suggest that respondents with greater ability
and with a higher relative interest in the topic are better able to map the implications of
inflation-induced nominal-debt erosion to their own net nominal exposure.

26
Appendix Table A5 displays the full regression output.

28
5 Wealth effects of inflation and economic choices

In this section, we study the real effect of learning about the wealth effects of inflation on
economic choices. In Section 5.1, we link exogenous variation in respondents’ knowledge of
the effect to planned and actual spending. In Section 5.2, we study how knowledge feeds
into debt financing in a hypothetical real-estate investment.

5.1 Effects on spending

Spending data Do the changes in perceived and expected real net wealth generated by the
loan-erosion information treatment have any effect on spending decisions? We investigate
this question using two data sets. First, in the survey, we elicit planned changes in spending
on a wide category of nondurable goods and services over the next four weeks relative to
the previous four weeks, following Roth and Wohlfart (2020). We elicit these plans via a
five-point Likert scale, ranging from “much less” to “much more.” Respondents additionally
state whether they plan to buy big-ticket items over the next 12 months, as in Coibion,
Gorodnichenko, and Weber (2022). Second, the partnering bank provides us with categorized
transaction-level account data. We use the bank data to study actual changes in spending
in the 30–90 days after survey participation relative to the 30–90 days before.
The two data sets are complementary to each other. All respondents report spending
plans, allowing us to test for effects of the information treatment on intended spending
on both nondurable goods and services as well as larger durable goods. The survey data
are useful because the administrative bank data (i) do not cover all respondents, (ii) do
not necessarily comprise all spending decisions, as respondents may spend from accounts
unobservable to us, and (iii) cannot fully distinguish between nondurable and large/durable

29
spending because big spending categories are cash withdrawals, credit cards, and online
shopping, which can entail both. However, the bank data provide a useful check on whether
respondents actually follow through with their spending plans.
We first verify that spending plans and actual spending are positively related. Naturally,
the correlation will be substantially below one because of survey noise, measurement error,
spending in accounts at other banks, other news after our survey intervention that affect
actual spending, and the fact that we elicit planned spending via a Likert scale. Appendix
Table A6 shows results of regressing actual spending changes on planned changes in non-
durable (Panel A) and large/durable spending (Panel B). We consider actual changes in
spending over the 30 days (Columns 1, 4, and 7), 60 days (Columns 2, 5, and 8), and 90
days (Columns 3, 6, and 9) after survey participation relative to before. Planned changes in
nondurable spending are the average across categories such as groceries and restaurants. On
durable spending, we take the sum across categories such as real estate and cars (the average
sum is around one; that is, respondents plan to purchase one large/durable item over the
next 12 months). We standardize both survey-based measures.
In regressions of actual total and discretionary spending, coefficients on planned spend-
ing tend to be significantly positive. Moreover, the association strengthens with the length
of the event window, even for the four-week plan on nondurable goods and services as the
independent variable, suggesting that changes in planned spending only gradually trans-
late into actual changes in spending. In terms of economic magnitude, a one-standard-
deviation increase in planned nondurable spending corresponds to e 37 (60 days) and e 98
(90 days) higher actual discretionary spending, and with e 59 (60 days) and e 88 (90 days)
for large/durable items. Overall, these results indicate that spending plans are consistent
with actual measures of spending, which is comforting given that self-reported spending is

30
the only type of information available for all respondents.
Planned spending Table 6 links the loan-erosion information treatment, through its
effect on perceived and expected changes in real net wealth, to planned spending. We run
regressions of spending plans for each nondurable category (Columns 1–4) and, because of
its low frequency, the sum of planned purchases on durable goods (Column 5). To make
nondurable and durable spending comparable, we standardize both measures. As a first
step, we show that spending plans are significantly positively correlated with respondents’
post-treatment estimates of changes in their real net wealth over the past and next 12
months (Panel A).27 This result indicates that respondents who perceive and expect a more
positive change in their real net wealth plan to reduce spending by less, as we find that the
control group on average intends to reduce spending on nondurables over the next four weeks.
Groceries is the single nondurable-spending category not significantly correlated with wealth
changes. Respondents on average do not plan to alter spending on groceries as a function of
their real-net-wealth change, which likely reflects its largely non-discretionary nature.
In Panel B, we observe a positive reduced-form effect of the loan-erosion treatment on
planned spending. Households in the loan-treatment group expect to spend significantly
more on restaurants (Column 2) and leisure (Column 3) than comparable respondents in
the control group. Economically, the treatment increases planned spending by 11% of a
standard deviation. The effect on planned spending on clothing (Column 4) and durable
items (Column 5) is positive but statistically insignificant. As expected, planned spending
on groceries does not vary across groups (Column 1).
27
We include all respondents in the estimation sample. We account for this choice by adding a treatment
indicator to the list of controls. Results are similar when restricting the sample to the control group.
Moreover, we abstain from controlling for the pre-treatment estimate of the past-12-month change in real
net wealth, because we do not want to relate spending plans to perceived and expected changes in real
net wealth that arise as respondents go through the survey. When we add the pre-treatment estimate as a
control, the coefficient on the post-treatment wealth-change estimate is still significantly positive but weaker.

31
 We then examine how treatment-induced knowledge about nominal-debt erosion caused
by unexpected inflation affects planned spending through its impact on own real-net-wealth
perceptions and expectations. We estimate the following specification:

spendi = β posterior rnwi + γ prior rnwi + controlsi + errori . (2)

spendi measures either planned spending on nondurable or durable goods and services, or
actual spending. posterior rnwi is the sum of the post-treatment estimate of past- and
next-12-month change in real net wealth. We instrument this variable using Equation 1. By
using the loan-erosion treatment as a source of exogenous variation in perceived and expected
wealth, the instrumental-variables approach can resolve possible endogeneity. prior rnwi
again refers to the pre-treatment directional and point estimate of the past-12-month change
in real net wealth. controlsi are similar to Equation 1. We run Huber-robust regression in
the first stage and OLS in the second stage when studying planned spending. Following
Coibion, Gorodnichenko, and Weber (2022), we use Huber regression in the first stage and
a jackknife approach in the second stage to control for outliers and influential observations
in both stages when analyzing actual spending.
Panel C of Table 6 reports the results from estimating Equation 2. The Kleibergen-Paap
F-statistic for the first stage is 10.3, which is non-homoskedasticity robust in settings like
ours with a single endogenous regressor (Andrews, Stock, and Sun, 2019). The coefficient on
posterior real-net-wealth changes is significantly positive for the planned change in spending
on restaurants (Column 2) and leisure (Column 3). A one-percentage-point increase in real
net wealth causes planned spending to go up by around 4% of a standard deviation.28 The
28
The economic magnitude of the coefficient on real net wealth is substantially above the magnitudes
shown in Panel A. The difference possibly in part reflects that in the absence of news, wealth estimates
are sticky, and households are unlikely to make large adjustments to their short-term spending plans as a

32
coefficient is positive but insignificant for clothing (Column 4) and durables (Column 5).
Again, exogenous variation in changes in wealth does not causally affect planned spending
on groceries (Column 1). The results indicate that knowledge about the wealth effects of
inflation transmits to spending through its effect on perceived and expected real net wealth.
Actual spending We now turn to the treatment effects on actual spending. In Ta-
ble 7, we report effects on total spending, as well as separately on nondiscretionary and
discretionary spending. Panel A presents reduced-form evidence of a significantly positive
effect of the loan-erosion information on total and discretionary spending. Comparing the
loan-treatment group with the control group in the 60 days following survey participation,
relative to the 60 days prior, the treatment group increases total spending by e 187 and dis-
cretionary spending by e 121. Economically, the effect on discretionary spending constitutes
a 6% increase in spending over the average measured spending in the 60 days pre-treatment.
The magnitudes are roughly similar over a 90-day window. Effects are weaker over a 30-day
horizon, likely because it takes time to adjust spending.
Panel B reports results from an IV regression based on Equation 2. The Kleibergen-
Paap F-statistic varies between 7.4 and 10.9, rejecting that the worst-case bias of two-stage
least squares exceeds 15–20% of the worst-case error of OLS. We find a positive relation
between instrumented posterior real-net-wealth perceptions and expectations and total and
discretionary spending. Effects are strongest when considering spending over the 60 days
post-treatment relative to pre-treatment: a one-percentage-point increase in real net wealth
raises total spending by e 92 and discretionary spending by e 70. Again, the effect on
nondiscretionary spending is close to zero and insignificant. The patterns are hence similar
to the reduced-form evidence (Panel A), albeit statistical significance is more sparse.

consequence. Once households receive news about real net wealth instead, for example due to our information
intervention, the updating of perceptions and expectations can cause larger changes in spending plans.

33
 To gauge the economic magnitudes, we translate the average treatment effects into MPCs
using back-of-the-envelope calculations. The loan-erosion treatment increases perceived and
expected real net wealth by 2.5 percentage points on average (Table 4). Average net wealth
of our respondents is e 326,000, corresponding to an increase of e 8,150. Hence, with an
average treatment effect on total spending of e 187 over 60 days (Panel A of Table 7), the
estimates translate into an MPC of 2.3%. We calculate a comparable MPC of 2.8% based on
the IV regression (Panel B of Table 7), as a one-percentage-point increase in real net wealth
(e 3,260) induces e 92 more total spending. The magnitudes are consistent with estimates
from the literature on the stock-market-wealth effect on consumption.29
Appendix Table A7 illustrates the robustness of the treatment effects on actual spending.
In Panel A, we estimate treatment effects using OLS instead of Huber-robust regressions.
Economic magnitudes remain non-trivial but estimates become a bit noisier. Panel B shows
estimates when fixing the baseline time window to 60 days before the survey intervention.
Treatment effects are stronger. In Panel C, we add uncategorized spending to our measures of
total and discretionary spending. We restrict transactions to uncategorized non-integer out-
flows of below e 100, which according to the bank are more likely to constitute consumption
(rather than, e.g., peer-to-peer transfers). Treatment effects are slightly larger on average.
Results are also robust to additionally controlling for changes in regular income, such as
salary, pension, and children’s allowances (Panel D), which we do because inflation-induced
wealth changes might alter the demand for leisure (Doepke and Schneider, 2006). Similarly,
in Panel E, we add other post-treatment expectations to our standard set of controls. These
additional controls include expectations about house prices, unemployment, own economic

29
Chodorow-Reich, Nenov, and Simsek (2021) calculate an MPC of three percent in a quantitative model,
a magnitude similar to the central scenario in the review by Poterba (2000). Di Maggio, Kermani, and
Majlesi (2020) report an MPC of below three percent for the top ten percent of the wealth distribution.

34
situation, and interest rates, and could in principle confound the loan-treatment effect on
spending.30 The loan-treatment coefficients are slightly larger with the added controls.
Real-net-wealth change as the underlying mechanism We interpret our results as
reflecting how knowledge about the wealth effects of inflation, through own real-net-wealth
changes, impacts spending. Our approach, based on Equation 2, yields a “total” estimate
of how exogenous variation in these changes ultimately affects spending. However, the loan-
erosion information provision could alter other perceptions and expectations, which in turn
could shape spending choices. For example, households, realizing that debt erosion harms
banks’ profits, might expect increases in interest rates. Table 8 reports insignificant loan-
treatment effects on a set of other expectations: house prices, unemployment, own income,
general optimism, and interest rates.31 This result suggests knowledge about the wealth
effects of inflation indeed affects spending through perceived and expected wealth.
Our results suggest households spend out of wealth gains, likely because of increased
confidence about their household balance sheet. Indeed, respondents state that their hypo-
thetical MPC out of such unrealized gains is smaller but close to their MPC out of realized
gains (Appendix Figure A5). 51% of respondents have a positive MPC out of realized gains
and 43% spend out of unrealized gains.32 Households that receive the loan-erosion informa-
tion report more positive beliefs about debt, both in the context of inflation and generally,
than those in the control group (Table 3). Treated respondents also agree more with the
statement that with their balance sheet they are well-positioned for high inflation (Appendix
Table A4). Higher household spending might thus reflect greater perceived financial security

30
As the loan treatment might alter these expectations, we sacrifice econometric rigor by estimating such
a model (Angrist and Pischke, 2009).
31
The point estimate on expected house prices is marginally significant but negative and therefore is
unlikely to drive the positive effect of the loan-erosion treatment on consumption.
32
10% of respondents would spend less, a fraction similar to Fuster, Kaplan, and Zafar (2020).

35
and confidence about own wealth. This channel is consistent with Fisher (1933) who posits
changes in real debt feed into economic activity through consumer confidence. More gener-
ally, this channel supports the view that fluctuations in consumer sentiment causally affect
consumption (e.g., Angeletos and La’O, 2013).
Moreover, our respondents on average accumulate savings over time (Table 1), similar
to other samples of German households (e.g., Bachmann et al., 2021). These authors also
discuss based on representative data from the Deutsche Bundesbank that most Germans are
financially unconstrained. Hence, our survey participants can likely increase spending when
they feel wealthier in real terms by drawing down on their existing stock of savings.

5.2 Effects on debt financing

Can more positive beliefs about nominal debt induced by the loan-erosion information provi-
sion affect debt choices? Table 9 documents effects of the loan-erosion treatment on choices
in a hypothetical real-estate investment task. Respondents purchase real estate with up
to e 500,000 in equity and e 500,000 in debt. We also elicit preferences on the mortgage
type: adjustable (ARM) versus fixed rate (FRM) and, conditional on selection of a FRM,
the length of the fixed-rate period.
Panel A reports reduced-form evidence that survey participants in the loan-treatment
and control group are similar in their preferred home value on average. However, the two
groups differ significantly in their preferred financing structure. Information about inflation-
induced debt erosion increases the average mortgage size by nearly e 18,000. This increase
leads to a three-percentage-points higher debt/price ratio, relative to an average of 49%.
Moreover, the treatment group prefers a FRM over an ARM.
In Panel B, we analyze the effect of the loan-erosion treatment through the perceived

36
inflation protection of debt. We therefore instrument beliefs about the relative inflation
protection provided by nominal debt using the loan-erosion-treatment dummy (Columns
3–4 of Table 3 report the first stage). The Kleibergen-Paap F-statistic for the first stage
is between 15 and 20. Treatment-induced increases in the perceived inflation protection
of nominal debt reduce the use of equity, increase the use of debt, and hence lead to a
higher debt/price ratio of the transaction. A one-unit increase in the perceived inflation
protection of debt (ranging from 1 to 4) increases this ratio by 16 percentage points. These
results indicate that knowledge about inflation-induced erosion of nominal liabilities affects
hypothetical financing choices through its positive impact on beliefs about debt.

6 Implications for theory

Our findings have implications for the HANK literature that focuses on the transmission
of economic policy, while at the same time being consistent with micro data on the wealth
distribution and the composition of wealth (e.g., Bayer, Born, and Luetticke, 2023; Kaplan,
Moll, and Violante, 2018). A key insight of this literature is that the impact of macroe-
conomic shocks on aggregate consumption depends on the covariance between individual
MPCs and exposure to the shock (e.g., Bilbiie, 2008; Patterson, 2023). Consider an expan-
sionary monetary policy that increases inflation. In HANK models, redistribution caused by
the Fisher channel amplifies the effect of monetary policy on aggregate consumption because
the beneficiaries, households with a negative net nominal position, have higher MPCs than
the losers (e.g., Auclert, 2019; Pallotti, 2023). This result hinges on the assumption that all
households have full information rational expectations; that is, they are aware of the Fisher
channel, and adjust their consumption accordingly.

37
 We argue that taking into account heterogeneity in awareness of the effects of macroe-
conomic shocks is important to assess their aggregate consequences (D’Acunto et al., 2023).
One key finding of our paper is that households with large nominal-debt positions, who typ-
ically have higher MPCs, are largely unaware of inflation-induced debt erosion. Based on
our findings, it would be fruitful to extend the HANK framework to allow for information
frictions and other deviations from FIRE. One possible avenue could be following the mod-
eling strategy of Auclert, Rognlie, and Straub (2020) who introduce information rigidity to
jointly get large impact MPCs at the micro level and humped-shaped macro responses.
We sketch how asymmetric awareness of the Fisher channel might alter its short-run
effects. Suppose for example that nominal prices unexpectedly rise by 10%. A nominal
saver with N N P = e100k experiences a real wealth effect of −e10k. Conversely, a nominal
debtor with N N P = −e100k gains the equivalent of e10k. Assume a one-year MPC of
0.3 for the nominal saver and of 0.5 for the nominal debtor. Under FIRE, the saver then
reduces consumption by e3k, whereas the debtor spends e5k more. Aggregate consumption
increases by e2k. Limited awareness of debt erosion alters this calculation and the effects
on aggregate demand. Assume only 50% of borrowers know about debt erosion, whereas
everyone knows about savings erosion, roughly consistent with our evidence. This assump-
tion cuts the increase in consumption of net debtors to e2.5k, leading to a fall in aggregate
consumption by e0.5k. While highly stylized, this simple example illustrates the possible
importance of heterogeneity in awareness for the effects of macroeconomic shocks and for
the redistributive effects of surprise inflation on aggregate demand in particular.
Two additional points are noteworthy. First, as we show in Appendix Figure A5, MPCs
out of balance-sheet revaluations akin to the Fisher channel may be lower than MPCs out
of realized gains. If lower MPCs out of the wealth effects of inflation compress the MPC gap

38
between debtors and savers, effects on consumption due to redistribution might be muted.
Second, limited awareness adds an intertemporal-substitution component: even unaware
debtors may raise consumption once they repay fixed nominal obligations with higher future
nominal income. High impact MPCs, as in FIRE, hence frontload consumption that might
occur later. One implication of frontloading is that inflation can feed on itself and trigger
inflationary spirals, as in Pallotti (2023).

7 Conclusion

We causally study the wealth effects of inflation in an information-provision experiment on

customers of a large German bank during a historic inflation episode. On average, households
are well-informed about prevailing inflation and are highly concerned about the impact of
inflation on their wealth. Yet, they know surprisingly little about the reduction in the real
value of nominal debt due to surprise inflation. Once we inform respondents about the debt-
erosion channel, they increase estimates of their real net wealth and causally increase their
planned and actual consumption. Moreover, treated respondents differ in leverage choices
in a hypothetical real-estate transaction. Our results document the redistributive nature of
surprise inflation across households and provide causal estimates for how individuals adjust
behavior following inflation-induced redistribution of wealth.
Our findings inform a recent class of HANK models. A core amplification mechanism
in these models comes from a positive covariance between exposure to economic shocks
and propensities to spend. Applied to the Fisher channel, amplification occurs because the
beneficiaries of unexpected inflation, households with a negative net nominal position, have
higher MPCs than the losers (Auclert, 2019). We show the beneficiaries are largely unaware

39
of inflation-induced debt erosion, which suggests effects of redistribution might be limited in
the short run. Based on our findings, it would be interesting to extend the HANK framework
to consider how shock exposure varies with shock awareness.
On the policy side, understanding optimal monetary policy in general and the implica-
tions for the optimal inflation target in particular remain largely unexplored in a framework
that deviates from FIRE (Coibion, Gorodnichenko, and Wieland, 2012; Dávila and Schaab,
2023). Moreover, our results suggest suboptimal household debt choices given a large frac-
tion of our survey population was initially uninformed about inflation-induced debt erosion.
In addition, wealthier, more educated survey participants, which are overrepresented in
our sample relative to the German population, are better informed about inflation-induced
nominal-position erosion, which might raise concerns about possible redistribution from parts
of the population that are less well informed to those that are better informed (D’Acunto
et al., 2023). This concern could possibly be counteracted through information campaigns
and robo-advise (D’Acunto and Rossi, 2023).

40
References
Adam, K. and J. Zhu (2015). Price-level changes and the redistribution of nominal wealth
across the euro area. Journal of the European Economic Association 14 (4), 871–906.

Agarwal, S. and B. Mazumder (2013). Cognitive abilities and household financial decision
making. American Economic Journal: Applied Economics 5 (1), 193–207.

Almenberg, J., A. Lusardi, J. Säve-Söderbergh, and R. Vestman (2021). Attitudes towards

debt and debt behavior. The Scandinavian Journal of Economics 123 (3), 780–809.

Andre, P., C. Pizzinelli, C. Roth, and J. Wohlfart (2022). Subjective models of the macroe-
conomy: evidence from experts and representative samples. The Review of Economic
Studies 89 (6), 2958–2991.

Andrews, I., J. H. Stock, and L. Sun (2019). Weak instruments in instrumental variables
regression: Theory and practice. Annual Review of Economics 11 (1), 727–753.

Angeletos, G.-M. and J. La’O (2013). Sentiments. Econometrica 81 (2), 739–779.

Angrist, J. D. and J.-S. Pischke (2009). Mostly Harmless Econometrics. Princeton University
Press.

Armantier, O., G. Topa, W. Van der Klaauw, and B. Zafar (2017). An overview of the survey
of consumer expectations. Economic Policy Review (23-2), 51–72.

Auclert, A. (2019). Monetary policy and the redistribution channel. American Economic
Review 109 (6), 2333–67.

Auclert, A., M. Rognlie, and L. Straub (2020). Micro jumps, macro humps: Monetary policy
and business cycles in an estimated hank model. Technical report.

Bachmann, R., B. Born, O. Goldfayn-Frank, G. Kocharkov, R. Luetticke, and M. Weber

(2021). A temporary vat cut as unconventional fiscal policy. Working Paper .

Bayer, C., B. Born, and R. Luetticke (2023). Shocks, frictions, and inequality in us business
cycles. American Economic Review (forthcoming).

Bhamra, H. S., C. Dorion, A. Jeanneret, and M. Weber (2023). High inflation: Low default
risk and low equity valuations. The Review of Financial Studies 36 (3), 1192–1252.

Bilbiie, F. O. (2008). Limited asset markets participation, monetary policy and (inverted)
aggregate demand logic. Journal of Economic Theory 140 (1), 162–196.

41
Botsch, M. J. and U. Malmendier (2023). The long shadows of the great inflation: Evidence
from residential mortgages. Working Paper .

Braggion, F., F. von Meyerinck, and N. Schaub (2023). Inflation and individual investors’ be-
havior: Evidence from the german hyperinflation. The Review of Financial Studies 36 (12),
5012–5045.

Braggion, F., F. von Meyerinck, N. Schaub, and M. Weber (2024). The long-term effects of
inflation on inflation expectations. Working Paper .

Brunnermeier, M., S. Correia, S. Luck, E. Verner, and T. Zimmermann (2023). The debt-
inflation channel of the german hyperinflation. Working Paper .

Cavallo, A., G. Cruces, and R. Perez-Truglia (2017). Inflation expectations, learning, and
supermarket prices: Evidence from survey experiments. American Economic Journal:
Macroeconomics 9 (3), 1–35.

Chodorow-Reich, G., P. T. Nenov, and A. Simsek (2021). Stock market wealth and the real
economy: A local labor market approach. American Economic Review 111 (5), 1613–1657.

Cohen, R. B., C. Polk, and T. Vuolteenaho (2005). Money illusion in the stock market: The
modigliani-cohn hypothesis. The Quarterly Journal of Economics 120 (2), 639–668.

Coibion, O., D. Georgarakos, Y. Gorodnichenko, G. Kenny, and M. Weber (2024). The effect
of macroeconomic uncertainty on household spending. American Economic Review 114 (3),
645–677.

Coibion, O., Y. Gorodnichenko, L. Kueng, and J. Silvia (2017). Innocent bystanders? mon-
etary policy and inequality. Journal of Monetary Economics 88, 70–89.

Coibion, O., Y. Gorodnichenko, and T. Ropele (2019). Inflation expectations and firm
decisions: New causal evidence. The Quarterly Journal of Economics 135 (1), 165–219.

Coibion, O., Y. Gorodnichenko, and M. Weber (2022). Monetary policy communications and
their effects on household inflation expectations. Journal of Political Economy 130 (6),
1537–1584.

Coibion, O., Y. Gorodnichenko, and J. Wieland (2012). The optimal inflation rate in new
keynesian models: should central banks raise their inflation targets in light of the zero
lower bound? Review of Economic Studies 79 (4), 1371–1406.

D’Acunto, F., A. Fuster, and M. Weber (2021). Diverse policy committees can reach under-
represented groups. Technical report.

42
D’Acunto, F., D. Hoang, M. Paloviita, and M. Weber (2022). IQ, expectations and choice.
The Review of Economic Studies (forthcoming).

D’Acunto, F., D. Hoang, M. Paloviita, and M. Weber (2023). Human frictions in the trans-
mission of economic policy. Working Paper .

D’Acunto, F., D. Hoang, and M. Weber (2022). Managing households’ expectations with
unconventional policies. The Review of Financial Studies 35 (4), 1597–1642.

D’Acunto, F., U. Malmendier, and M. Weber (2023). What do the data tell us about inflation
expectations? In Handbook of Economic Expectations, pp. 133–161.

D’Acunto, F. and A. G. Rossi (2023). Robo-advice: Transforming households into rational

economic agents. Annual Review of Financial Economics 15.

D’Acunto, F., A. G. Rossi, and M. Weber (2023). Crowdsourcing financial information to

change spending behavior. Journal of Financial Economics.

D’Acunto, F., P. Schnorpfeil, and M. Weber (2022). Big brother is watching you (even if
he’s dead): Surveillance and long-run conformity. Working Paper .

Dávila, E. and A. Schaab (2023). Optimal monetary policy with heterogeneous agents:
Discretion, commitment, and timeless policy. Technical report.

De Quidt, J., J. Haushofer, and C. Roth (2018). Measuring and bounding experimenter
demand. American Economic Review 108 (11), 3266–3302.

Del Canto, F., J. Grigsby, E. Qian, and C. Walsh (2023). Are inflationary shocks regressive?
a feasible set approach. Working Paper .

Di Maggio, M., A. Kermani, and K. Majlesi (2020). Stock market returns and consumption.
The Journal of Finance 75 (6), 3175–3219.

Doepke, M. and M. Schneider (2006). Inflation and the redistribution of nominal wealth.
Journal of Political Economy 114 (6), 1069–1097.

Dutz, D., I. Huitfeldt, S. Lacouture, M. Mogstad, A. Torgovitsky, and W. van Dijk (2022).
Selection in surveys: Using randomized incentives to detect and account for nonresponse
bias. Working Paper .

Ferreira, C., J. M. Leiva, G. Nuño, A. Ortiz, T. Rodrigo, and S. Vazquez (2024). The
heterogeneous impact of inflation on households’ balance sheets. Working Paper .

Fisher, I. (1933). The debt-deflation theory of great depressions. Econometrica 1 (4), 337.

43
Fuster, A., G. Kaplan, and B. Zafar (2020). What would you do with $500? spending
responses to gains, losses, news, and loans. The Review of Economic Studies 88 (4), 1760–
1795.
Gomes, J., U. Jermann, and L. Schmid (2016). Sticky leverage. American Economic Re-
view 106 (12), 3800–3828.
Haaland, I., C. Roth, and J. Wohlfart (2023). Designing information provision experiments.
Journal of economic literature 61 (1), 3–40.
Jiang, J. H., R. Kamdar, K. Lu, and D. Puzzello (2024). How do households respond to
expected inflation? an investigation of transmission mechanisms. Working Paper .
Kamdar, R. (2018). The inattentive consumer: Sentiment and expectations. Working Paper .
Kang, J. and C. E. Pflueger (2015). Inflation risk in corporate bonds. The Journal of
Finance 70 (1), 115–162.
Kaplan, G., B. Moll, and G. L. Violante (2018). Monetary policy according to hank. Amer-
ican Economic Review 108 (3), 697–743.
Kindermann, F., J. Le Blanc, M. Piazzesi, and M. Schneider (2021). Learning about housing
cost: Survey evidence from the german house price boom. Technical report.
Leombroni, M., M. Piazzesi, M. Schneider, and C. Rogers (2020). Inflation and the price of
real assets. Technical report.
Lettau, M. and S. C. Ludvigson (2004). Understanding trend and cycle in asset values:
Reevaluating the wealth effect on consumption. American Economic Review 94 (1), 276–
299.
Maćkowiak, B., F. Matějka, and M. Wiederholt (2023). Rational inattention: A review.
Journal of Economic Literature 61 (1), 226–273.
Modigliani, F. and R. A. Cohn (1979). Inflation, rational valuation and the market. Financial
Analysts Journal .
Pallotti, F. (2023). The fisher channel according to hank: Unexpected inflation and the
missing recession. Working Paper .
Pallotti, F., G. Paz-Pardo, J. Slacalek, O. Tristani, and G. L. Violante (2024). Who bears
the costs of inflation? euro area households and the 2021–2022 shock. Working Paper .
Patterson, C. (2023). The matching multiplier and the amplification of recessions. American
Economic Review 113 (4), 982–1012.

44
Pfäuti, O. and F. Seyrich (2022). A behavioral heterogeneous agent new keynesian model.

Pfäuti, O., F. Seyrich, and J. Zinman (2023). Heterogeneity in what? cognitive skills, beliefs
and the liquid wealth distribution.

Poterba, J. M. (2000). Stock market wealth and consumption. Journal of Economic Per-
spectives 14 (2), 99–118.

Roth, C. and J. Wohlfart (2020). How do expectations about the macroeconomy affect
personal expectations and behavior? The Review of Economics and Statistics 102 (4),
731–748.

Schnorpfeil, P., M. Weber, and A. Hackethal (2024). Inflation and trading.

Shafir, E., P. Diamond, and A. Tversky (1997). Money illusion. The Quarterly Journal of
Economics 112 (2), 341–374.

Stantcheva, S. (2023). How to run surveys: A guide to creating your own identifying variation
and revealing the invisible. Annual Review of Economics 15 (1).

Weber, M., B. Candia, T. Ropele, R. Lluberas, S. Frache, B. H. Meyer, S. Kumar, Y. Gorod-

nichenko, D. Georgarakos, O. Coibion, et al. (2023). Tell me something i don’t already
know: Learning in low and high-inflation settings. Technical report.

Weber, M., F. D’Acunto, Y. Gorodnichenko, and O. Coibion (2022). The subjective infla-
tion expectations of households and firms: Measurement, determinants, and implications.
Journal of Economic Perspectives 36 (3), 157–184.

Yang, Y. (2023). Redistributive inflation and optimal monetary policy. Working Paper .

45
Figures and Tables

Figure 1: Prior knowledge about the wealth effects of inflation

Panel A. Impact of unexpected inflation on nominal positions

50
% of respondents
40
30
20
10
0

e
no
tiv

iv

tiv
iti
at
ga

si
er

os
eg

po
th
ne

rp
rn

ei

ry
he
ry

N
he

Ve
Ve

at
at

R
R

Impact of unexpected inflation increase

Cash on hand Fixed income Fixed-rate loans

Panel B. Wealth protection by nominal positions during unexpected inflation

60
% of respondents
40
20
0

st

st
or

Be
k

k
an

an
W

Wealth protection when unexpected increase in inflation

Savings Stocks Real estate FR loans

Notes: The figures show the distribution of beliefs about how financial instruments fare during an
unexpected increase in inflation. The top panel shows the distribution of responses to “What do you
think, is an unexpected increase in inflation positive or negative for owners of the following financial
products?” For fixed income, we list savings accounts, bonds, and life insurances as examples. The
bottom panel shows the distribution of responses to “With which of the following financial products
would you expect the highest real-net-wealth return during unexpectedly high inflation? Please
state “1” for the product with the best inflation protection, “2” for [...]” We plot responses of the
control group only, as we pose this question post-treatment. We list examples of savings products.

46
 Figure 2: Prior knowledge about the wealth effects of inflation by NNP

e
or
m

Consumption over past 12 months

h
uc
Impact of unexpected inflation

M
s

e
po

or
rm
ry
Ve

he
at
R
os
rp

ge
an
he

ch
at
R

o
N
r
no

s
es
er

rl
th

he
ei
N

at
R
eg
rn

ss
he

le
at

h
uc
R

M
g
ne
ry
Ve

%
%

5%

0%

5%
50

75
25

,0

,2

,5

,7
<-

>=
5%
,-

5%

0%
0%

[-2

[0

[2

[5
[-5

Net nominal position / gross wealth

FR loans (left scale) Savings (left)

Effect of inflation (right)

Notes: Referring to the left vertical axis, we plot average responses by respondents’ net nominal
position to the question “What do you think, is an unexpected increase in inflation positive or
negative for owners of the following financial products?” FR loans refers to fixed-rate loans. Savings
are the average of cash on hand and fixed income. For fixed income, we list savings accounts, bonds,
and life insurances as examples. Referring to the right vertical axis, we plot average responses by
respondents’ net nominal position to the question “Have you consumed rather more or less as a
consequence of changes in the following factors over the past 12 months, such as abstained from a
purchase or purchased something extra?” One of the factors we ask about is the rate of inflation.

47
 Figure 3: Treatment effect on changes in real net wealth by NNP
Panel A. Loan-erosion treatment

15%
Real-net-wealth change (past + future 12 months)
-10% -5% 0% 5% 10%

<-50% -50% to <0% 0% to 50% >50%

Net nominal position / gross wealth

Panel B. Savings-erosion treatment

15%
Real-net-wealth change (past + future 12 months)
-10% -5% 0% 5% 10%

<-50% -50% to <0% 0% to 50% >50%

Net nominal position / gross wealth

Notes: The figures report coefficients and 95% confidence bounds from regressions of respondents’
post-treatment perceptions and expectations of changes in real net wealth over the sum of the past
and next 12 months. We regress these wealth changes on a loan-treatment (Panel A) or savings-
treatment indicator (Panel B), interacted with respondents’ net nominal position as a fraction of
their gross wealth. We describe the treatments in detail in Section 2.1. Regressions include the
standard set of controls. Results are from Huber-robust regressions to control for outliers and
influential observations. Standard errors are robust to heteroscedasticity.

48
 Figure 4: Heterogeneity in loan-treatment effect on real net wealth
Panel A. Subsample of households with negative net nominal position

15%
Real-net-wealth change (past + future 12 months)
-5% 0% 5% 10%

Education high Inflation down 5 yrs Inflation right Inflation important

No Yes

Panel B. Subsample of households with positive net nominal position

15%
Real-net-wealth change (past + future 12 months)
-5% 0% 5% 10%

Education high Inflation down Inflation right Inflation important

No Yes

Notes: The figures report coefficients and 95% confidence bounds on the loan-treatment indicator
from regressions of respondents’ post-treatment estimates of changes in real net wealth over the
past and next 12 months. We describe the treatment in detail in Section 2.1. In Panel A, these
regressions are limited to households with a negative NNP; in Panel B, regressions are limited to
households with a positive NNP. Moreover, we split the sample by another respondent characteristic:
Education high, which equals one for completed higher education and zero otherwise; Inflation down
5 yrs, which is one if the rate of inflation expected in five years is below currently perceived inflation;
Inflation right, which is one if currently perceived inflation deviates by at most 0.5 pp from actual
inflation; and Inflation important, which is one if respondents perceive inflation to be substantially
more important for their own wealth than the average of GDP growth, stock prices, and interest
rates. Regressions include the standard set of controls. Results are from Huber-robust regressions
to control for outliers and influential observations. Standard errors are robust to heteroscedasticity.

49
 Table 1: Descriptive statistics

Data sources: PHF Bank sample

Statistics: Mean Mean SD P25 P50 P75
Demographic characteristics
Female (0/1) 0.50 0.45 0.50 0.00 0.00 1.00
Age (years) 52.97 48.08 15.08 36.00 48.00 60.00
University completed (0/1) 0.29 0.48 0.50 0.00 0.00 1.00
Business at university (0/1) 0.10 0.30 0.00 0.00 0.00
Employed (0/1) 0.54 0.72 0.45 0.00 1.00 1.00
Gross wealth (e k) 238.13 385.57 659.37 25.00 120.83 463.02
Nominal assets / gross wealth (%) 37.21 42.67 33.66 10.00 30.00 79.00
Nominal debt / gross wealth (%) 10.12 16.78 22.88 0.00 5.00 30.00
Homeowner (0/1) 0.44 0.59 0.49 0.00 1.00 1.00
Stockholdings (0/1) 0.20 0.54 0.50 0.00 1.00 1.00
Spending and income
Total spending (avg./month) 1,844 1,202 1,001 1,604 2,385
Nondiscretionary spending (avg./month) 818 584 397 702 1,106
Discretionary spending (avg./month) 978 763 432 785 1,329
Regular income (avg./month) 2,274 2,862 2,002 1,457 2,381 3,826
Perceptions and expectations
Inflation rate today (%) 8.78 6.24 7.00 7.90 8.00
Inflation rate in 12 months (%) 10.39 9.80 6.00 8.50 10.00
Inflation rate in five years (%) 10.67 15.07 3.00 5.00 10.50
Real-net-wealth change past 12 months (%) -7.45 14.91 -14.00 -6.00 0.00
Inflation important for own wealth (0–4) 2.37 1.02 2.00 2.00 3.00
GDP growth important for own wealth (0–4) 1.73 1.06 1.00 2.00 2.00
Interest rates important for own wealth (0–4) 1.34 1.14 0.00 1.00 2.00
Notes: This table reports summary statistics for respondents’ characteristics (survey and bank
data), spending and income (bank data), and perceptions and expectations (survey data). We
present the variables’ mean, standard deviation (SD), 25th percentile (P25), median (P50), and
75th percentile (P75). We compare our respondents to a representative German sample from the
2017 wave of the Bundesbank’s Panel on Household Finances. In our sample, we winsorize the 1%
tails of gross wealth. Spending and income measures are monthly averages spanning the six months
preceding the survey, and are winsorized at the 97.5th percentile. The belief variables reported in
the table refer to priors elicited before the treatment. Priors on inflation and own real net wealth
are point estimates, with the 1% tails trimmed. The baseline number of observations is 3,190.

50
Table 2: Correlates of beliefs about balance-sheet effects of unexpected inflation

Dependent variable: Unexpected inflation increase positive or negative for...

cash fixed income stocks real estate FR loans
(1) (2) (3) (4) (5)
Female (0/1) −0.049 0.001 −0.125*** −0.038 −0.116***
(0.035) (0.036) (0.036) (0.036) (0.035)
Age group
>35 to 45 years 0.000 −0.059 −0.150*** −0.208*** −0.024
(0.057) (0.057) (0.056) (0.057) (0.053)
>45 to 55 years −0.003 −0.128** −0.135** −0.131** −0.055
(0.058) (0.058) (0.058) (0.058) (0.055)
>55 to 65 years −0.022 −0.255*** −0.216*** −0.107* −0.085
(0.060) (0.059) (0.060) (0.060) (0.057)
>65 years 0.078 −0.280*** −0.096 −0.059 −0.072
(0.064) (0.061) (0.063) (0.061) (0.061)
University completed (0/1) −0.144*** 0.046 0.087** −0.014 0.051
(0.041) (0.040) (0.040) (0.039) (0.039)
Business at university (0/1) −0.119** −0.035 −0.013 0.053 0.214***
(0.052) (0.066) (0.060) (0.060) (0.062)
Log gross wealth −0.084*** −0.036*** 0.016 0.079*** 0.100***
(0.012) (0.012) (0.012) (0.012) (0.012)
Stockholdings (0/1) −0.114*** −0.021 −0.029 −0.007 0.135***
(0.040) (0.041) (0.042) (0.041) (0.039)
Homeowner (0/1) 0.141*** 0.101** 0.000 0.071 −0.052
(0.046) (0.045) (0.045) (0.045) (0.042)
Inflation relatively important −0.152*** −0.054 0.006 0.108*** 0.074**
(0.037) (0.038) (0.038) (0.038) (0.037)
Accurate inflation perception (0/1) −0.151*** −0.064 0.040 0.085* 0.023
(0.046) (0.043) (0.043) (0.044) (0.040)
Inflation lower in five years (0/1) −0.131*** 0.063 0.133*** 0.100** 0.250***
(0.042) (0.040) (0.042) (0.041) (0.039)

Observations 3,134 3,134 3,134 3,134 3,134

R-squared 0.06 0.02 0.02 0.04 0.09
Notes: This table reports estimates of regressions of beliefs about the impact of an unexpected
inflation increase on various balance-sheet items, measured on an ordinal scale from 0 (“very neg-
ative”) to 4 (“very positive”). We standardize the outcome variables using the mean and standard
deviation of our sample. For fixed income, we list savings accounts, bonds, and life insurances as
examples. FR loans refers to fixed-rate loans. Inflation relatively important measures respondents’
beliefs about the importance of inflation for own wealth relative to the average importance of GDP
growth, interest rates, and stock prices. Accurate inflation perception is a dummy equal to one if
respondents’ estimate of current inflation is at most 1.5 pp off actual inflation. Inflation lower in
five years indicates the expectation that inflation will be lower in five years than it is perceived
today. Robust standard errors are in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

51
 Table 3: Treatment effects on beliefs about nominal positions

Dependent variable: Inflation-protection ranking

Nominal assets Nominal debt Debt aversion
(1) (2) (3) (4) (5) (6)
Treat: savings erosion −0.124*** −0.126*** 0.051 0.056 0.037 0.046
(0.045) (0.043) (0.044) (0.044) (0.043) (0.041)
Treat: loan erosion −0.080* −0.080* 0.200*** 0.190*** −0.119*** −0.123***
(0.045) (0.044) (0.045) (0.046) (0.044) (0.042)

Controls N Y N Y N Y
Observations 2,977 2,928 2,977 2,928 3,190 3,134
R-squared 0.00 0.11 0.01 0.04 0.00 0.11
Notes: This table reports regression estimates of beliefs about nominal positions in each treatment
group relative to the control group. To elicit the beliefs about inflation protection (Columns 1–4), we
ask: “With which of the following financial products would you expect the highest real-net-wealth
return during unexpectedly high inflation?” Respondents rank nominal assets, stocks, real estate,
and nominal debt relative to each other, with higher numbers indicating better performance. For
nominal assets, we list cash on hand, savings accounts, bonds, and life insurances as examples. In
Columns 5–6, the dependent variable measures agreement with the statement “I am uncomfortable
with taking on debt.” Five possible responses range from “completely disagree” to “completely
agree.” We standardize the outcome variables using the mean and standard deviation of the sample.
A detailed description of the treatments is in Section 2.1. The list of controls is in Section 4.
Observations are lower in Columns 1–4 because respondents may skip this question and lower in
odd-numbered columns because we trim the 1% tails of inflation beliefs that serve as controls.
Robust standard errors are in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

52
 Table 4: Treatment effect on changes in real net wealth

DV: Change in real net wealth

Last 12 months Next 12 months Last + next 12 months
(1) (2) (3) (4) (5) (6) (7)
Treat: savings −0.932 −0.807 −0.835 −0.981 −1.767 −1.788 −1.606**
(0.648) (0.646) (0.769) (0.773) (1.217) (1.222) (0.781)
Treat: loan 1.490** 1.743*** 1.260 1.204 2.749** 2.947** 2.495***
(0.676) (0.666) (0.769) (0.761) (1.222) (1.207) (0.787)

Controls prior Y Y Y Y Y Y Y
Controls demo N Y N Y N Y Y
Robust reg N N N N N N Y
Avg. Y CG -2.51 -2.55 -3.11 -3.01 -5.62 -5.56 -6.33
Observations 3,190 3,134 3,190 3,134 3,190 3,134 3,099
R-squared 0.17 0.19 0.10 0.13 0.18 0.20 0.40
Notes: This table reports the perception and expectation of changes in the real net wealth in each
treatment group relative to the control group. Columns 1 and 2 consider real-net-wealth changes
estimated over the past 12 months, Columns 3 and 4 refer to changes estimated over the next 12
months, and Columns 5–7 show estimates of changes over the last plus next 12 months. We describe
the treatments in detail in Section 2.1. The list of controls is in Section 4. Robust reg indicates
whether the estimates are from ordinary least squares or Huber-robust regressions to control for
outliers and influential observations. Observations are lower in specifications with demographic
controls because we trim the 1% tails of inflation beliefs that serve as controls. Robust standard
errors are in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

53
 Table 5: Treatment effect on changes in real net wealth by NNP

Dependent variable: Change in real net wealth

Last 12 months Next 12 months Last + Next 12 months
(1) (2) (3) (4) (5) (6)
Treat: savings 0.283 0.559 −0.092 0.517 0.711 1.632
(0.748) (0.730) (0.956) (0.953) (1.523) (1.508)
Treat: loan 1.155 1.658** 2.254** 2.895*** 3.859** 5.017***
(0.769) (0.752) (1.001) (0.984) (1.559) (1.528)
NNP≥0 0.971 1.208** 0.502 0.999 1.388 2.280*
(0.606) (0.603) (0.804) (0.820) (1.222) (1.241)
Treat: savings × NNP≥0 −2.278*** −2.529*** −0.764 −1.673 −3.204* −4.447**
(0.879) (0.859) (1.126) (1.125) (1.773) (1.762)
Treat: loan × NNP≥0 −0.942 −1.513* −1.500 −1.880 −2.606 −3.546**
(0.900) (0.880) (1.167) (1.154) (1.808) (1.784)

Controls for prior estimate Y Y Y Y Y Y

Controls for demographics N Y N Y N Y
Avg. Y omitted group -2.52 -2.74 -3.15 -3.22 -5.56 -5.82
Observations 3,127 3,059 3,165 3,112 3,158 3,101
R-squared 0.46 0.48 0.24 0.26 0.38 0.39
Notes: This table reports the perception and expectation of changes in the real net wealth in each
treatment group relative to the control group. Columns 1 and 2 consider changes estimated over
the past 12 months, Columns 3 and 4 refer to changes estimated over the next 12 months, and
Columns 5 and 6 show estimates of changes over the past plus next 12 months. We describe the
treatments in detail in Section 2.1. NNP≥0 equals one if the respondent has a non-negative net
nominal position and is zero otherwise. Controls for prior estimate include a directional and a point
estimate of past-12-month changes in real net wealth. Controls for demographics include a quadratic
polynomial in the respondent’s age, risk tolerance, the log of gross wealth, and dummy variables
for the respondent’s gender, marital status, educational level, business-related university degree,
employment status, beliefs about inflation (trimmed at 1% tails), voucher instead of lottery survey-
participation incentive, and whether participation follows a reminder email sent to encourage survey
participation. Results are from Huber-robust regressions to control for outliers and influential
observations. Robust standard errors are in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

54
 Table 6: Changes in real net wealth and planned spending

Dependent variable: Planned spending

Groceries Restaurants Leisure Clothing Durables
(1) (2) (3) (4) (5)
Panel A. Endogenous regression

RNW change 0.001 0.004*** 0.004*** 0.004*** 0.003***

(0.001) (0.001) (0.001) (0.001) (0.001)
Controls for prior RNW N N N N N
Observations 3,134 3,134 3,134 3,134 3,134
R-squared 0.04 0.16 0.13 0.10 0.08

Panel B. Reduced form

Treat: loan erosion −0.007 0.110*** 0.108** 0.042 0.069

(0.043) (0.041) (0.042) (0.042) (0.043)
Controls for prior RNW Y Y Y Y Y
Observations 2,088 2,088 2,088 2,088 2,088
R-squared 0.04 0.16 0.12 0.10 0.09

Panel C. Instrumental variable

RNW change −0.010 0.037* 0.041** 0.015 0.027

(0.017) (0.019) (0.020) (0.017) (0.019)
Controls for prior RNW Y Y Y Y Y
Observations 2,065 2,065 2,065 2,065 2,065
1st stage F-stat 10.31 10.31 10.31 10.31 10.31

Controls for demographics Y Y Y Y Y

Notes: This table reports estimates from regressions of planned spending. In Columns 1–4, we
measure changes in planned spending on various nondurables over the next four weeks relative to
the last four weeks on a five-point scale, ranging from “much less” to “much more.” In Column
5, the dependent variable is the number of durable items expected to purchase over the next 12
months. We standardize the outcome variables. RNW change is the post-treatment point estimate
of last- and next-12-month changes in own real net wealth. In Panels B and C, we omit the savings-
erosion treatment group. In Panel C, we instrument RNW change using the loan-erosion treatment.
The list of controls is in Section 4. Results are from OLS regressions (Huber robust in first stage
in Panel C). Robust standard errors are in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

55
 Table 7: Changes in real net wealth and actual spending

DV: Total Nondiscretionary Discretionary

Window: 30 60 90 30 60 90 30 60 90
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Panel A. Reduced form

T: loan 65.6* 186.5*** 166.9** −4.8 21.7 38.2 40.8* 121.0*** 132.8**
(36.4) (59.6) (78.4) (15.8) (24.3) (33.1) (24.7) (40.8) (55.8)
N 1,465 1,513 1,477 1,431 1,414 1,405 1,451 1,488 1,497
R2 0.03 0.03 0.02 0.03 0.02 0.03 0.03 0.04 0.03

Panel B. Instrumental variable

RNW 29.8 92.4** 42.6 0.9 7.8 18.0 33.2* 70.3** 55.8
(24.1) (44.5) (46.2) (8.7) (13.8) (20.3) (19.4) (34.2) (38.7)
N 1,460 1,469 1,452 1,451 1,441 1,429 1,447 1,465 1,458
F-stat 8.49 8.83 8.12 10.92 10.51 9.01 8.00 8.18 7.35

Controls Y Y Y Y Y Y Y Y Y
Avg. Y -267.0 -308.4 -22.8 -92.6 -46.6 58.7 -147.2 -222.9 -240.3
Notes: This table reports estimates from regressions of actual spending. We study total (Columns
1–3), nondiscretionary (Columns 4–6), and discretionary (Columns 7–9) spending, as described in
Section 2.2. We compare individual-level aggregate spending in the 30, 60, and 90 days following
survey participation relative to the same time window pre-participation respectively. RNW is
the post-treatment point estimate of the sum of last- and next-12-month changes in own real net
wealth. We instrument this estimate using the loan-erosion information treatment. We omit the
savings-erosion treatment group from the estimations. Section 2.1 entails detailed descriptions of
the treatments. All regressions include the standard set of controls, described in Section 4. Results
in Panel A are from Huber-robust regressions to control for outliers and influential observations. In
Panel B, we use Huber regressions in the first stage and a jackknife procedure in the second stage.
Robust standard errors are in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

56
 Table 8: Treatment effects on other expectations

Dependent variable: 12-month expectation Change in interest rates

House prices Unemployment Own income Optimism One year Five years
(1) (2) (3) (4) (5) (6)
Treat: loan erosion −0.078* −0.063 −0.040 −0.005 −0.036 −0.274
(0.043) (0.044) (0.042) (0.042) (0.101) (0.183)

Controls Y Y Y Y Y Y
Avg. Y CG 0.00 0.00 0.00 0.00 1.47 2.48
Observations 2,088 2,088 2,088 2,088 2,067 2,065
R-squared 0.09 0.04 0.11 0.11 0.04 0.11
Notes: This table reports expectations of the loan-erosion treatment group relative to the control
group. In Columns 1–3, we study 12-month-ahead expectations using a five-point scale, ranging
from “much lower” to “much higher.” The dependent variable in Column 4 measures agreement, on
a five-point scale ranging from “completely disagree” to “completely agree,” to the statement “I am
optimistic about the future.” We standardize the outcome variables in Columns 1–4. In Columns
5–6, we compare point estimates of expected interest rates relative to the perceived current interest
rate (we trim the 1% tails of the outcome variable). Section 2.1 entails detailed descriptions of the
treatments. The list of controls is in Section 4. Results are from OLS regressions. Robust standard
errors are in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

57
 Table 9: Perceived inflation protection of debt and real-estate investment

Dependent variable: Price (e k) Equity (e k) Debt (e k) Debt/price FRM

(1) (2) (3) (4) (5)
Panel A. Reduced form

Treat: loan erosion 5.474 −9.829 17.770*** 0.031*** 0.073*

(7.574) (6.254) (6.202) (0.009) (0.039)
Observations 2,082 2,088 2,088 2,082 2,088
R-squared 0.19 0.11 0.12 0.06 0.06

Panel B. Instrumental variable

Debt as inflation hedge 0.067 −67.138* 83.528** 0.160** 0.321

(44.590) (36.228) (40.753) (0.065) (0.268)
Observations 1,914 1,940 1,933 1,913 1,910
1st stage F-stat 17.70 20.29 16.89 18.46 15.12

Controls Y Y Y Y Y
Avg. Y control group 542.79 280.25 260.48 0.48 2.22
Notes: This table reports estimates from regressions of features of a hypothetical real-estate trans-
action. Column 1 is on the preferred hypothetical purchase price, capped at e 1,000,000; Column
2 considers the equity stake, capped at e 500,000; Column 3 refers to the size of the mortgage, also
capped at e 500,000; Column 4 features the debt-to-price ratio; and Column 5 is on an ordinal-scale
variable that takes on 0 if respondents choose an adjustable-rate mortgage, 1 in case of a mortgage
with a five-year fixed-rate period, 2 for a 10-year fixed-rate period, 3 for a 20-year fixed-rate period,
and 4 for a 30-year fixed-rate period. Section 2.1 entails detailed descriptions of the treatments. To
elicit beliefs about Debt as inflation hedge, we ask: “With which of the following financial products
would you expect the highest real-net-wealth return during unexpectedly high inflation?” Respon-
dents rank nominal assets, stocks, real estate, and nominal debt relative to each other, with higher
numbers indicating higher relative returns. We instrument the perceived inflation protection of
nominal debt using the loan-erosion information treatment. The list of controls is in Section 4. In
Panel A, results are from Huber-robust regressions to control for outliers and influential observa-
tions. In Panel B, we use OLS in the first stage and a jackknife procedure in the second stage.
Robust standard errors are in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

58
 Online Appendix:

Households’ Response to the Wealth Effects of Inflation

Philip Schnorpfeil, Michael Weber, and Andreas Hackethal

Not for Publication

A Appendix figures and tables

Figure A1: Association between debt reported in survey and data from bank
200k 150k
Outstanding debt with bank
50k 100k
0

0 2.5k 7.5k 17.5k 37.5k 62.5k 87.5k 125k 175k 275k 425k 600k
Mid point of survey-reported range of outstanding debt

Notes: This figure documents respondents’ average debt balance with our bank partner in the month
of survey participation for each mid point of outstanding-debt bandwidth reported by respondents
in the survey. The debt balance at the bank includes consumer and mortgage debt. In the survey,
households select the value of their outstanding debt, if any, from various ranges we provide. We
take the respective mid point of the ranges we ask about.

1
 Figure A2: Knowledge about information provided by treatment and NNP
Know about information provided

lly
Fu
4
3
2
al
l
at

5%
%

5%

0%

5%
ot

50

25

,0

7
N

,2

,5

,7
<-

>=
5%
,-

5%

0%
0%

[-2

[0

[2

[5
[-5

Net nominal position / gross wealth

Savings erosion Loan erosion

Notes: The figure shows respondents’ stated knowledge about the treatment information. We plot
averages by treatment (savings erosion vs. debt erosion) and net nominal position as a fraction of
gross wealth. Naturally, the control group does not receive this question.

2
Figure A3: Treatment effect on changes in real net wealth by nominal position
Panel A. Loan-erosion treatment

15%
Real-net-wealth change (past + future 12 months)
-5% 0% 5% 10%

0% >0% to 20% >20% to 40% >40%

Debt / gross wealth

Panel B. Savings-erosion treatment

5%
Real-net-wealth change (past + future 12 months)
-15% -10% -5% 0%

<=20% >20% to 40% >40% to 70% >70%

Savings / gross wealth

Notes: The figures report coefficients and 95% confidence bounds from regressions of respondents’
post-treatment estimates of the change in their real net wealth over the past and next 12 months.
In Panel A, we regress the wealth changes on the loan-treatment indicator, interacted with respon-
dents’ debt as a fraction of gross wealth. In Panel B, we regress real-net-wealth changes on the
interaction of the savings-treatment indicator and respondents’ nominal savings as a fraction of
gross wealth. We describe the treatments in detail in Section 2.1. Regressions include the standard
set of controls. Results are from Huber-robust regressions to control for outliers and influential
observations. Standard errors are robust to heteroscedasticity.

3
Figure A4: Treatment effects on beliefs about RNW changes by prior and NNP
Panel A: Effect of savings-erosion treatment on positive-NNP subjects

Posterior RNW change (past + future 12 months)

-30 -20 -10 0 10 20

-40 -30 -20 -10 0 10 20

Prior RNW change (past 12 months)

Control group Savings-treatment group

Panel B: Effect of loan-erosion treatment on negative-NNP subjects

Posterior RNW change (past + future 12 months)
-30 -20 -10 0 10 20

-40 -30 -20 -10 0 10 20

Prior RNW change (past 12 months)

Control group Loan-treatment group

Notes: This figure reports binscatter plots of respondents’ post-treatment beliefs about changes in
their real net wealth over the last 12 months plus next 12 months, conditional on pre-treatment
estimates of past-12-month RNW changes. In Panel A, the sample comprises respondents with a
positive net nominal position. In Panel B, the sample contains respondents with a negative NNP.

4
Figure A5: Marginal propensity to consume out of realized vs. unrealized gain

50
40
% of respondents
30
20
10
0

<0 0 (0, 0.25) [0.25, 0.5) [0.5, 0.75) [0.75, 1]

Realized gain Unrealized gain

Notes: This figure shows a histogram of MPCs for two gain scenarios. On realized gains, we pose the
following question: “Please consider a hypothetical situation where you unexpectedly receive a one-
time payment of e 10,000 today. How would you change your spending behavior over the next three
months as a consequence of that?” On the unrealized-gain scenario, we ask: “Please consider that
today you carefully calculate your total wealth, from which you subtract the value of outstanding
loans. You are surprised to realize that you own e 10,000 more than previously believed. How
would you change your spending behavior over the next three months as a consequence of that?”
Respondents use a slider to indicate a spending response between e -10,000 and e 10,000.

5
 Table A1: Consumption categories

Total Discretionary Nondiscretionary

spending spending spending

(1) (2) (3)

Living
Groceries Yes Yes
Clothing Yes Yes
Multimedia Yes Yes
Hairdresser/wellness Yes Yes
Drugstore Yes Yes
Pets Yes Yes
Cafeteria Yes Yes
Gifts Yes Yes
Other living expenses Yes Yes

Housing
Rent Yes Yes
Energy and water Yes Yes
Furniture / home accessories Yes Yes
Housing fees Yes Yes
Domestic help Yes Yes
Property taxes Yes Yes
Renovations Yes Yes
Other housing expenses Yes Yes

Leisure
Restaurants/cafes/bars Yes Yes
Events/tickets Yes Yes
Sports/fitness Yes Yes
Hobbies/clubs/associations Yes Yes
Vacation / travel expenses Yes Yes
Books/music/movies/apps Yes Yes
Electronics/computers/games Yes Yes
Subscriptions Yes Yes
Other leisure expenses Yes Yes

Transportation Yes Yes

continued

6
 Table A1 continued

Total Discretionary Nondiscretionary

spending spending spending

(1) (2) (3)

Health
Pharmacy Yes Yes
Physician Yes Yes
Glasses / contact lenses Yes Yes
Hospital Yes Yes
Other health expenses Yes Yes

Children
Activities and toys Yes Yes
Children’s clothing Yes Yes
Childcare Yes Yes
School fees Yes Yes
Other expenses on children Yes Yes

Career
Office supplies / teaching material Yes Yes
Business travel Yes Yes
Tuition Yes Yes
Continuing education Yes Yes
Other career-related expenses Yes Yes

Other outflows
Donations Yes Yes
Cash withdrawals Yes Yes
Credit card Yes Yes
Online purchases Yes Yes
Notes: This table reports spending categories coming from the partner bank’s PFM tool, and how
we assign them to the three spending measures. Details on these measures are in Section 2.2.

7
 Table A2: Balancedness across treatment arms

Control Treat: p-value Treat: p-value p-value

group savings (1)=(2) loan (1)=(4) (2)=(4)
(1) (2) (3) (4) (5) (6)
Demographic characteristics
Female (0/1) 0.48 0.45 0.20 0.42 0.01*** 0.14
Age (years) 48.34 47.82 0.43 48.06 0.66 0.71
University completed (0/1) 0.51 0.47 0.07* 0.47 0.10* 0.88
Business at university (0/1) 0.10 0.10 0.90 0.11 0.57 0.65
Employed (0/1) 0.71 0.71 0.96 0.73 0.32 0.34
Gross wealth (e k) 407.25 401.13 0.84 348.09 0.03** 0.06*
Nominal assets / gross wealth (%) 43.09 41.55 0.29 43.37 0.85 0.21
Nominal debt / gross wealth (%) 17.06 15.94 0.25 17.35 0.77 0.16
Homeowner (0/1) 0.60 0.60 0.99 0.57 0.25 0.25
Stockholdings (0/1) 0.54 0.55 0.41 0.53 0.91 0.35
Spending and income
Total spending (avg./month) 1,870 1,840 0.63 1,808 0.32 0.60
Nondiscretionary spending (avg./month) 837 799 0.21 813 0.42 0.64
Discretionary spending (avg./month) 985 994 0.81 949 0.36 0.25
Regular income (avg./month) 2,847 2,899 0.62 2,813 0.74 0.40
Perceptions and expectations
Inflation rate today (%) 8.62 8.87 0.34 8.85 0.40 0.93
Inflation rate in 12 months (%) 10.22 10.55 0.44 10.41 0.65 0.76
Inflation rate in five years (%) 10.50 10.77 0.67 10.73 0.72 0.95
Real-net-wealth change past 12 months (%) -7.61 -7.12 0.44 -7.63 0.97 0.43
Inflation important for own wealth (0–4) 2.39 2.31 0.09* 2.42 0.52 0.02**
GDP growth important for own wealth (0–4) 1.76 1.73 0.41 1.69 0.11 0.42
Interest rates important for own wealth (0–4) 1.40 1.29 0.03** 1.33 0.19 0.41
Notes: This table shows means for different observable characteristics of respondents in each treat-
ment arm (Columns 1, 2, and 4). We provide a check of balance of means across arms in Columns
3, 5, and 6. We winsorize the 1% tails of gross-wealth estimates. Spending and income measures
are monthly averages spanning the six months preceding the survey, and are winsorized at the
97.5th percentile. The belief variables reported in the table refer to priors elicited before the in-
formation treatment. Priors on inflation and own real net wealth are point forecasts, with the 1%
tails trimmed. The baseline number of observations is 3,190.

8
Table A3: Full set of correlates of beliefs about balance-sheet effects of inflation

Dependent variable: Unexpected inflation increase positive or negative for...

cash fixed income stocks real estate FR loans

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

Female (0/1) −0.026 0.013 −0.122*** −0.043 −0.143***

(0.038) (0.039) (0.039) (0.039) (0.037)
Age group
>35 to 45 years −0.011 −0.055 −0.101 −0.136** 0.039
(0.061) (0.063) (0.062) (0.063) (0.057)
>45 to 55 years −0.021 −0.115* −0.086 −0.042 0.028
(0.064) (0.064) (0.064) (0.065) (0.061)
>55 to 65 years −0.054 −0.226*** −0.165** −0.025 0.020
(0.067) (0.066) (0.067) (0.067) (0.063)
>65 years 0.050 −0.231*** −0.085 0.018 −0.001
(0.093) (0.088) (0.091) (0.087) (0.088)
East Germany (0/1) 0.016 −0.006 0.001 0.021 −0.037
(0.045) (0.046) (0.047) (0.044) (0.044)
University completed (0/1) −0.129*** 0.050 0.077* −0.023 0.049
(0.042) (0.042) (0.041) (0.040) (0.040)
Business at university (0/1) −0.117** −0.046 −0.033 0.051 0.200***
(0.053) (0.067) (0.060) (0.060) (0.062)
Married (0/1) −0.010 0.038 0.000 0.038 −0.007
(0.038) (0.040) (0.040) (0.039) (0.038)
Retired (0/1) 0.010 −0.014 0.022 −0.043 −0.039
(0.082) (0.080) (0.085) (0.076) (0.081)
Student (0/1) −0.051 0.009 0.039 0.198** 0.088
(0.096) (0.088) (0.091) (0.092) (0.085)
Employee (0/1) 0.010 0.012 −0.003 −0.029 −0.117***
(0.042) (0.047) (0.045) (0.045) (0.043)
Craftsman (0/1) 0.191** −0.096 −0.185** −0.221** −0.197***
(0.086) (0.084) (0.088) (0.087) (0.076)
Risk tolerance 0.001 0.031 0.060*** 0.019 0.050***
(0.020) (0.020) (0.019) (0.019) (0.019)
Money illusion (0/1) 0.040 0.004 −0.037 0.030 −0.163***
(0.040) (0.041) (0.040) (0.039) (0.040)
Inflation relatively important −0.148*** −0.055 0.009 0.096** 0.075**
(0.038) (0.039) (0.039) (0.038) (0.037)
Accurate inflation perception (0/1) −0.150*** −0.070 0.035 0.087** 0.017
(0.046) (0.044) (0.043) (0.044) (0.040)
Inflation lower in five years (0/1) −0.109** 0.064 0.125*** 0.084** 0.218***
continued

9
 Table A3 continued

Dependent variable: Unexpected inflation increase positive or negative for...

cash fixed income stocks real estate FR loans

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

(0.043) (0.041) (0.042) (0.042) (0.039)

Log gross wealth −0.082*** −0.047*** 0.014 0.081*** 0.096***
(0.013) (0.013) (0.013) (0.013) (0.012)
Stockholdings (0/1) −0.134*** −0.045 −0.065 −0.020 0.116***
(0.042) (0.043) (0.044) (0.043) (0.041)
Homeowner (0/1) 0.155*** 0.039 0.036 0.073 −0.085*
(0.054) (0.055) (0.056) (0.054) (0.049)
Net nominal position / gross wealth
-50% to <-25% 0.026 −0.079 −0.074 −0.020 −0.115
(0.096) (0.115) (0.107) (0.107) (0.106)
-25% to <0% 0.102 0.027 0.002 −0.200** −0.144
(0.087) (0.102) (0.094) (0.096) (0.094)
0% to <25% 0.131 −0.107 0.065 −0.064 −0.248***
(0.082) (0.097) (0.091) (0.092) (0.090)
25% to <50% 0.150* −0.081 0.011 −0.087 −0.237**
(0.091) (0.105) (0.098) (0.099) (0.097)
50% to <75% 0.205** −0.152 0.001 −0.181* −0.260**
(0.099) (0.109) (0.105) (0.106) (0.101)
≥75% 0.070 −0.171 0.107 −0.015 −0.160
(0.100) (0.112) (0.107) (0.107) (0.103)
Participation following reminder (0/1) 0.041 0.006 0.018 −0.071** −0.010
(0.035) (0.036) (0.036) (0.035) (0.034)
Participation incentive voucher (0/1) −0.032 0.015 0.047 −0.002 −0.008
(0.037) (0.038) (0.038) (0.037) (0.036)
Observations 3,091 3,091 3,091 3,091 3,091
R-squared 0.07 0.02 0.03 0.05 0.11

Notes: This table reports regression estimates of beliefs about the impact of unexpected inflation
on various balance-sheet items, measured on an ordinal scale from 0 (“very negative”) to 4 (“very
positive”). We standardize the outcome variables. For fixed income, we list savings accounts, bonds,
and life insurances as examples. FR loans refers to fixed-rate loans. Inflation relatively important
measures respondents’ beliefs about the importance of inflation for own wealth relative to the
average importance of GDP growth, interest rates, and stock prices. Accurate inflation perception
is a dummy equal to one if respondents’ estimate of current inflation is at most 1.5 pp off actual
inflation. Inflation lower in five years indicates the expectation that inflation will be lower in five
years than it is perceived today. Participation following reminder indicates survey participation
after the bank sent a reminder email to the selected respondents. Participation incentive voucher
equals one if the respondent received a guaranteed voucher (rather than a lottery ticket) for survey
participation. Robust standard errors are in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

10
 Table A4: Treatment effects on beliefs about own wealth

Dependent variable: Inflation positive effect on RNW Strong balance sheet when inflation
(1) (2) (3) (4) (5) (6)
Treat: savings erosion −0.012 −0.019 −0.025 0.042 0.026 0.116
(0.043) (0.043) (0.088) (0.044) (0.041) (0.077)
Treat: loan erosion 0.129*** 0.142*** 0.232** 0.071 0.093** 0.141*
(0.044) (0.044) (0.096) (0.043) (0.040) (0.075)
NNP≥0 −0.028 −0.031
(0.071) (0.068)
Treat: savings erosion × NNP≥0 0.008 −0.112
(0.101) (0.091)
Treat: loan erosion × NNP≥0 −0.118 −0.064
(0.108) (0.089)

Controls N Y Y N Y Y
Observations 3,190 3,134 3,134 3,190 3,134 3,134
R-squared 0.00 0.03 0.03 0.00 0.18 0.17
Notes: This table reports beliefs about own wealth in the context of inflation of respondents in
each treatment group relative to those in the control group. Columns 1–3 consider beliefs about
the impact of inflation on real net wealth. The question reads: “What has been the impact of the
following factors on the change of your real net wealth over the past 12 months?” Five response
options range from “very negative” to “very positive.” From the response to inflation we subtract
the average response to the other factors we ask about (Ukraine conflict, COVID-19, climate change,
economic growth). Columns 4–6 refer to agreement to the statement that the respondent’s balance
sheet is well-suited for high inflation. Five response options range from “completely disagree” to
“completely agree.” We standardize all outcome variables. Section 2.1 entails detailed descriptions
of the treatments. NNP≥0 is a dummy equal to one if the respondent has a non-negative net
nominal position and zero otherwise. The list of controls is in Section 4. Results are from OLS
regressions. Robust standard errors are in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

11
 Table A5: Heterogeneity in loan-treatment effect on change in RNW

Dependent variable: Real-net-wealth change over last + next 12 months

Characteristic: Education high Inflation down in 5 Inflation accurate Inflation important
(1) (2) (3) (4)
Panel A. Subset with negative NNP and characteristic satisfied

Treat: loan erosion 7.852*** 6.850*** 7.630*** 8.302***

(1.877) (1.699) (1.851) (2.785)
Observations 270 353 270 199
R-squared 0.45 0.42 0.57 0.46

Panel B. Subset with negative NNP and characteristic not satisfied

Treat: loan erosion 2.115 3.317 2.836 3.615*

(2.619) (3.956) (2.507) (1.872)
Observations 255 166 249 327
R-squared 0.45 0.35 0.36 0.40

Panel C. Subset with positive NNP and characteristic satisfied

Treat: loan erosion −0.532 2.101** 0.290 2.520

(1.111) (0.999) (1.227) (1.544)
Observations 751 927 735 667
R-squared 0.47 0.43 0.46 0.39

Panel D. Subset with positive NNP and characteristic not satisfied

Treat: loan erosion 3.655** −0.858 2.130 0.586

(1.503) (1.923) (1.388) (1.185)
Observations 831 621 819 914
R-squared 0.30 0.31 0.34 0.32

Controls Y Y Y Y
Avg. Y control group -5.62 -5.62 -5.62 -5.62
Notes: This table reports estimates of regressions of RNW changes on various subsamples based
on respondent characteristics: Education high, which equals one for completed higher education
and zero otherwise; Inflation down in 5, which is one if the rate of inflation expected in five
years is below currently perceived inflation; Inflation accurate, which is one if currently perceived
inflation deviates by at most 0.5 pp from actual inflation; and Inflation important, which is one
if respondents perceive inflation to be substantially more important for their own wealth than the
average of GDP growth, stock prices, and interest rates. In Panels A and C, we run regressions
on the subsample of respondents who satisfy a characteristic, respectively. In Panels B and D,
we consider the subsamples of respondents not satisfying a characteristic. Moreover, we split the
sample based on respondents’ net nominal position (negative in Panels A and B, non-negative in
Panels C and D). Section 2.1 entails detailed descriptions of the treatment. The list of controls
is in Section 4. Results are from Huber-robust regressions to control for outliers and influential
observations. Standard errors are robust to heteroscedasticity. * p < 0.1, ** p < 0.05, *** p < 0.01.

12
 Table A6: Correlation between planned and actual changes in spending

Dep. var.: Total Nondiscretionary Discretionary

Window: 30 60 90 30 60 90 30 60 90
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Panel A. Planned nondurables spending

Nondurables 44.6** 62.4** 111.9*** 18.2** 18.2 9.4 17.6 36.9* 97.6***
(18.2) (30.6) (41.2) (8.4) (12.6) (17.1) (12.4) (20.5) (30.1)
N 1,466 1,514 1,472 1,432 1,413 1,407 1,455 1,489 1,493
R2 0.03 0.03 0.02 0.04 0.02 0.03 0.03 0.04 0.04

Panel B. Planned durables spending

Durables 17.9 76.7** 54.7 6.4 5.0 −5.7 4.8 58.8*** 88.3***
(18.6) (30.0) (40.9) (8.0) (12.3) (16.2) (12.5) (21.3) (29.2)
N 1,465 1,514 1,476 1,431 1,414 1,405 1,453 1,490 1,494
R2 0.03 0.03 0.02 0.03 0.02 0.03 0.03 0.04 0.04

Controls Y Y Y Y Y Y Y Y Y
Avg. Y -217.2 -174.5 91.8 -91.8 -37.2 84.1 -130.2 -124.4 -153.5
Notes: This table reports estimates from regressions of actual spending on planned spending.
We study total (Columns 1–3), nondiscretionary (Columns 4–6), and discretionary (Columns 7–9)
spending, as described in Section 2.2. We compare individual-level aggregate spending in the 30,
60, and 90 days following survey participation relative to the same time window pre-participation,
respectively. In Panel A, Nondurables is the average planned change in spending across nondurables
categories (groceries, restaurants, leisure, clothing) over the next four weeks relative to the past
four weeks. Five response options range from “much less” to “much more.” In Panel B, Durables is
the sum across indicators of planned purchases of large/durable items (real estate, vehicles, large
household items, large vacations, luxury goods, others) over the next 12 months. We standardize
the two measures of planned spending. The list of controls is in Section 4. Results are from Huber-
robust regressions to control for outliers and influential observations. Robust standard errors are
in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.

13
 Table A7: Robustness of treatment effects on actual spending

Dep. var.: Total Nondiscretionary Discretionary

Window: 30 60 90 30 60 90 30 60 90
(1) (2) (3) (4) (5) (6) (7) (8) (9)
Panel A. OLS instead of Huber-robust regressions

T: loan 117.9* 264.1** 136.2 44.3 85.9* 48.8 56.2 177.6** 116.1
(68.2) (102.5) (136.8) (31.2) (51.6) (72.0) (51.7) (74.6) (96.3)
N 1,514 1,514 1,514 1,514 1,514 1,514 1,514 1,514 1,514
R2 0.02 0.02 0.02 0.02 0.02 0.02 0.02 0.03 0.03

Panel B. Fixed pre-treatment baseline of 60 days

T: loan 180.2*** 186.5*** 132.5* 34.3 21.7 39.5 72.0* 121.0*** 185.8***
(68.2) (59.6) (79.1) (31.7) (24.3) (38.4) (42.5) (40.8) (54.3)
N 1,469 1,513 1,482 1,464 1,414 1,456 1,460 1,488 1,473
R2 0.07 0.03 0.04 0.07 0.02 0.04 0.05 0.04 0.05

Panel C. Adding uncategorized spending to total and discretionary spending

T: loan 66.1* 190.6*** 177.8** −4.8 21.7 38.2 39.3 130.9*** 141.1**
(37.0) (60.6) (79.7) (15.8) (24.3) (33.1) (25.6) (41.6) (56.4)
N 1,466 1,513 1,483 1,431 1,414 1,405 1,455 1,489 1,493
R2 0.03 0.03 0.02 0.03 0.02 0.03 0.03 0.04 0.03

Panel D. Adding change in income as control

T: loan 69.6* 171.8*** 162.3** −4.0 21.1 34.4 42.0* 117.9*** 129.3**
(37.0) (59.9) (76.6) (15.8) (24.0) (32.9) (25.1) (41.1) (55.5)
N 1,466 1,512 1,480 1,435 1,411 1,404 1,455 1,491 1,498
R2 0.03 0.04 0.03 0.04 0.03 0.03 0.03 0.04 0.04

Panel E. Adding controls for other expectations

T: loan 73.4** 198.8*** 166.0** −1.5 22.8 42.4 44.4* 130.5*** 131.8**
(36.8) (60.2) (79.5) (16.0) (24.7) (33.7) (25.5) (41.3) (55.8)
N 1,448 1,494 1,459 1,414 1,402 1,390 1,437 1,467 1,471
R2 0.04 0.03 0.02 0.04 0.03 0.03 0.03 0.04 0.04

Controls Y Y Y Y Y Y Y Y Y
Avg. Y -267.0 -308.4 -22.8 -92.6 -46.6 58.7 -147.2 -222.9 -240.3
Notes: This table shows estimates from regressions of actual spending. We study total (Columns
1–3), nondiscretionary (Columns 4–6), and discretionary (Columns 7–9) spending, as described
in Section 2.2. We compare total individual-level spending in the 30, 60, and 90 days following
survey participation relative to the same time window pre-participation respectively. In Panel A,
we use OLS rather than Huber-robust regressions. In Panel B, we vary the post-treatment time
window but hold fixed the pre-participation window at 60 days. In Panel C, we add uncategorized
non-integer outflows worth less than e 100 to total and discretionary spending. In Panel D, we add
regular income (such as salary, pension, children’s allowances) to the list of controls. In Panel E,
we add to the list of controls general optimism and the 12-months-ahead expectation about house
prices, unemployment, own income, and interest rates. Section 2.1 entails detailed descriptions of
the treatment. The list of controls is in Section 4. Results are from Huber-robust regressions (other
than in Panel A). Robust standard errors are in parentheses. * p < 0.1, ** p < 0.05, *** p < 0.01.
14
B Experimental instructions

This appendix provides the survey instructions translated from German into English. We use
green text in parentheses to highlight aspects of the survey design. We show non-numerical
response options to the questions using a), b), c), and so forth.

B.1 Welcome screen

Welcome to this survey by Goethe University Frankfurt!
The survey will take about 10 minutes. Your responses will be aggregated and only be used for
scientific research. If you feel you are unfamiliar with some of the survey topics, that is fine. We
just ask you for your best guess. Since we are interested in your unfiltered opinion, please refrain
from using external sources (e.g., Google).
In return for your completed participation, you will receive an Amazon voucher worth e 10 [if
voucher version] / get the chance to win one of 50 Amazon vouchers worth e 50 [if lottery
version]. To receive the voucher [if voucher version] / To receive the voucher in case of a win [IF
lottery version], you can enter your e-mail address at the end of the survey. Your email address
will only be used to send you the voucher.
Please do not use the “Back” button in your browser, as this may require restarting the survey.
Do you have questions? Please contact us at umfrage@finance.uni-frankfurt.de.

B.2 Pre-treatment section

Q1: What is your highest level of educational attainment?
a) Currently pursuing apprenticeship or studies (no degree yet)
b) Finished apprenticeship at training college or company
c) Finished apprenticeship at technical school or university of cooperative education
d) Obtained Bachelor, graduated from university of applied sciences or technical college
e) Diploma or Master, finished teacher training
f) Finished doctorate studies
g) Other professional degree
h) No degree (and not currently pursuing apprenticeship or studies)

[Ask if Q1=d), e), or f)]

Q1.1: In what field of study did you obtain your highest educational attainment?

15
a) Economics and Business Administration
b) Computer Science
c) Law
d) Medicine or Psychology
e) Engineering
f) Other field of study

[Randomized order of Q2 and Q3]

[Randomized assignment to either gain frame, or loss frame in Q2 and Q3]
Q2: Please imagine that you unexpectedly receive [if gain frame] / have to make [if loss frame] a
one-time payment of e 10,000 today. How would this change your spending over the next
three months?
Please click and drag the sliders below. [Range is from e -10,000 to e 10,000, in e 100 increments]
How much more/less would you save?
How much more/less would you spend?
How much higher/lower would your outstanding loans be?

Q3: Please imagine that today you carefully calculate the total value of your assets and subtract
from it the value of possible outstanding loans. You realize to your own surprise that you have
e 10,000 more [if gain frame] / less [if loss frame] than previously assumed. How would this
change your spending over the next 3 months?
Please click and drag the sliders below. [Range is from e -10,000 to e 10,000, in e 100 increments]
How much more/less would you save?
How much more/less would you spend?
How much higher/lower would your outstanding loans be?

Q4: To what extent does your wealth situation depend on the following factors?
Response options: Not at all – Very little – Somewhat – Strongly – Very strongly
[Factors presented in randomized order]
Inflation rate
Economic growth
Stock prices
Interest rates

Q5: Have you consumed rather more or less as a consequence of changes in the following factors
over the past 12 months, such as abstained from a purchase or purchased something extra?

16
Response options: Much less – Rather less – No change – Rather more – Much more
[Factors presented in randomized order]
Inflation rate
Economic growth
Stock prices
Interest rates

Q6: What do you think is the current rate of inflation in Germany, and what will it be in the
future?
Note: the rate of inflation is the percentage change in overall prices in the economy in the last 12
months, most commonly measured by the Consumer Price Index. A falling price level is commonly
known as “deflation”.
If you think there was deflation, please enter a negative value. You may enter up to one decimal
point.
Current rate of inflation: %
Expected rate of inflation in one year: %
Expected rate of inflation in five years: %

Q6.1: How certain are you about your responses?

Response options: 1 (“Not at all certain”) – 2 – 3 – 4 – 5 (“Very certain”)
Current rate of inflation
Expected rate of inflation in one year
Expected rate of inflation in five years

Q7: What is the composition of your household’s gross wealth? Gross wealth includes all assets,
without deducting outstanding loans. For example, if you bought property financed by a loan, only
the current value of the property is part of gross wealth, without deducting the outstanding loan
amount.
Example: a household has e 20 of cash, e 80 of real estate, and e 40 of loans. Gross wealth is
therefore e 100. Cash constitutes 20% and real estate 80% of gross wealth.
Please enter “0” if a category does not apply to you. The sum of your entries should equal 100%.
Cash and fixed-rate savings (e.g., saving accounts, bonds, life insurances): %
Stocks: %
Real estate (owner-occupied or investment): %
Other (e.g. vehicles, gold): %
Total (the sum should equal 100): [Sum of values above] %

17
[If the total of the asset entries does not equal 100%, the following message is displayed: “Please
make sure that the sum of your entries equals 100%.”]
How large are outstanding loans as a share of your gross wealth? Example: a household has e 100
of gross wealth and e 40 of loans. Loans have a 40% share of gross wealth.
Enter a “0” if you do not have any outstanding debt.
Mortgages: %
Consumer loans: %
[As soon as the respondent makes a loan entry, the following text is displayed.]
Your outstanding loans make up [sum of share of mortgages and consumer loans]% of your gross
wealth. Thus, your net wealth is [1 - (sum of share of mortgages and consumer loans)]% of your
gross wealth.

Q8: For some of the following questions, we kindly ask you to provide information about real-
value changes. These changes take inflation into account. For example, an investor who earns
a 5% annual return would achieve a real return of -2% with 7% inflation. Thus, despite having a
positive nominal return, this investor experiences a loss in purchasing power.
Now, please think about the net wealth of your household. This is your gross wealth minus
your outstanding loans. How has the real value of your net wealth changed in the past twelve
months?
Real value gain: Net worth has increased by more than inflation.
Net wealth has increased similar to inflation.
Real value loss: Inflation has increased by more than net wealth.

[If a) or c) in Q8, ask the following question]

Q8.1: Please estimate how positively [if a)] / negatively [if c)] your net wealth has changed in
real terms.
Please click and drag the slider below.
Positive real change in % [if a)] / Negative real change in % [if c)]. [Slider ranging from 1% to 60%]

Q9: What do you think, is an unexpected increase in inflation rather positive or negative for
owners of the following financial instruments?
Response options: Very negative – Rather negative – Neither nor – Rather positive – Very positive
Cash on hand
Fixed-rate products (e.g., time deposits, bonds, life insurances)
Stocks
Real estate
Fixed-rate loans

18
Q9.1: How certain are you about your responses?
Response options: 1 (“Not at all certain”) – 2 – 3 – 4 – 5 (“Very certain”)

B.3 Treatment section

[Control group]
The current inflation rate in Germany is 8.7%. This is the highest rate in more than 70 years.
That is, goods and services priced at e 100 one year ago now cost e 108.70 on average.
[Savings-erosion treatment group]
The current inflation rate in Germany is 8.7%, the highest rate in more than 70 years. That
is, goods and services priced at e100 one year ago now cost e108.7 on average. This price increase
has a relatively negative effect on savers: the savings amount (e.g., checking account, bond,
life insurance) is unchanged nominally or lower, but worth less in real terms as a consequence of
money depreciation.

As an example, consider a e50,000 savings product with a three-year maturity that you took out
one year ago. The real value of the savings product has already fallen sharply, and will depreciate
further if inflation remains high: e50,000 savings value one year ago ⇓ e38,800 real value
today
The inflation-induced savings depreciation thus has a negative effect on the real net wealth
of savers.
Note: the numbers come from current calculations by the Universities of Chicago and Frankfurt
(calculation details).
[Loan-erosion treatment group]
The current inflation rate in Germany is 8.7%, the highest rate in more than 70 years. That
is, goods and services priced at e100 one year ago now cost e108.7 on average. This price increase
has a relatively positive effect on borrowers: the loan amount is unchanged nominally, but
worth less in real terms as a consequence of money depreciation.
As an example, consider a e50,000 loan with a three-year maturity that you took out one year ago.
The real value of the loan has already fallen sharply, and will depreciate further if inflation remains
high: e50,000 loan value one year ago ⇓ e38,800 real value today
The inflation-induced loan depreciation thus has a positive effect on the real net wealth of
borrowers.
Note: the numbers come from current calculations by the Universities of Chicago and Frankfurt
(calculation details).
[Text presented to respondents who click on “calculation details” button]
Calculation details: “To determine the present value of a fixed-rate savings product [if savings
treatment] / loan [if loan treatment], we calculate the present value of future interest payments and

19
principal repayments (discounted cash flow). The present value calculation involves discounting
each payment using a discount rate. This step allows to make comparable payments that occur at
different points in time. The discount rate takes into account the inflation expectation prevailing
at the respective time periods.
In our calculation, the discount rate is set at 3% one year ago (1% + 2% inflation expectation)
and then increases linearly to 9.7% today (1% + 8.7% inflation expectation). This increase in the
discount rate significantly reduces the present value of future payments. The inflation data used
is based on the Harmonized Consumer Price Index from the German Federal Statistical Office. [If
savings treatment] Additionally, the calculation assumes that the savings product every month pays
interest at an annualized rate of 0.3%. [If loan treatment] Additionally, the calculation assumes a
nominal interest rate of 3%, an annual repayment rate of 3%, and monthly payments.”

[Treatment groups only]

Q10: Have you known about the information on the impact of inflation on loans [if loan treatment]
/ savings [if savings treatment]?
Response options: 1 (“Completely unknown”) – 2 – 3 – 4 – 5 (“Completely known”)

B.4 Post-treatment section

Q11: What do you expect the following economic factors to be in twelve months?
Response options: Much lower – Rather lower – No change – Rather higher – Much higher
Stock prices
Real-estate prices
Unemployment rate
Own net income

Q12: What do you estimate to be the interest rate for a newly contracted loan with a five-year
maturity and fixed interest rate currently? And what interest rate do you expect for the future?
Please enter up to one decimal place.
Current annual interest rate: %
Annual interest rate in one year: %
Annual interest rate in five years: %

Q13: Compared to the past four weeks, do you plan to spend more or less in the next four
weeks on the following:
Response options: Much less – Somewhat less – Similar amount – Somewhat more – Much more
Groceries

20
Eating out at restaurants
Leisure activities (e.g., movies/theater, vacations, hobbies)
Clothing, shoes

Q14: Do you plan to purchase any of the following products or take a major vacation within the
next twelve months?
Multiple choices are possible.
a) House or apartment
b) Car or other vehicle
c) Household item or electronic device (e.g., refrigerator, sofa, mobile phone)
d) Major vacation
e) Luxury item (e.g., watch, jewelry)
f) Other
g) None of the above

[If Q14 is not answered with g), ask Q14.1 in the next screen.]
Q14.1: How much do you plan to spend on the house or apartment [if a)], the car or other vehicle
[if b)], the household item or electronic device [if c)], the major vacation [if d)], the luxury item [if
e)], other major products [if f)]?
a) e %
b) I do not know or prefer not to answer

Q15: We would like to ask you again about the net wealth of your household. There is no right
or wrong answer here. Please estimate the real-value change of your net wealth in the past
twelve months, as well as the expected change in the next twelve months.
Please click and drag the sliders below. [Sliders ranging from -60% to 60%]
Real change of net wealth in the past twelve months
Expected real change of net wealth in the next twelve months

Q15.1: How certain are you about your estimates?

Response options: 1 (“Not at all certain”) – 2 – 3 – 4 – 5 (“Very certain”)
Real change of net wealth in the past twelve months
Expected real change of net wealth in the next twelve months

Q16: In your opinion, what impact did the following factors have on the real value of your net
wealth in the past twelve months?
Response options: Very negative – Rather negative – None – Rather positive – Very positive

21
[Randomized order]
Inflation
Ukraine conflict
COVID-19
Climate change and policies
Economic development

Q17: With which of the following financial products would you expect the highest real-net-
wealth return during unexpectedly high inflation?
Please assign a “1” to the financial product with the best inflation protection, a “2” to the second-
best protection, a “3” to the third-best, and a “4” to the fourth-best inflation protection.
Savings products (e.g., cash, savings account, bonds, life insurance)
Stocks
Real estate
Fixed-rate loans

Q18: Imagine you want to buy property. You can use up to e500,000 in borrowed capital (bank
loan) and e500,000 in equity (savings) for the purchase. How much money would you use to buy
the property, and how would you finance it?
Please click and drag the sliders below.
Purchase price [Slider ranging from e0 to e1,000,000]
Equity used [Slider ranging from e0 to e500,000]
You would finance the property using e[purchase price - equity used] of borrowed capital.
Fixed-rate period [Response options: “None” – “5 years” – “10 years” – “20 years” – “30 years”]
Note: the mortgage term describes the period for which the agreed interest rate in the loan contract
is fixed.

Q19: To what extent do you agree with the statements below?

Response options: “Strongly disagree” – “Somewhat disagree” – “Neither agree nor disagree” –
“Somewhat agree” – “Strongly agree”
[Randomized order]
a) One should use savings to buy something for themselves.
b) Taking on loans makes me uncomfortable.
c) I am concerned about high inflation.
d) With my balance sheet, I am well-equipped for times of high inflation.
e) I am optimistic about the future.

22
f) I expect to work more in the coming months than in the previous months.
g) Unexpected high inflation leads to a redistribution of wealth from savers to borrowers.

Q20: You are now nearing the end of the survey. We just have a few more questions
about you.
When making personal savings or investment decisions, how would you generally describe your risk
tolerance?
Response options: 1 (“Not at all willing to take risks”) – 2 – 3 – 4 – 5 (“Very risk tolerant”)

Q21: Imagine that your income and expenses are about 10% higher than they were a year ago. At
that time, you were planning to sell a high-quality watch but did not get around to it due to lack
of time. The price of this watch has since increased from e10,000 to e11,000. Would you now be
more inclined to sell your watch compared to a year ago?
a) Yes, more inclined to sell
b) No, more inclined to keep
c) Unchanged

Q22: How would you estimate the combined value of the cash holdings and interest-bearing assets
(such as savings accounts, bonds, life insurance) of your household?
[Dropdown menu]
a) 0 to under e2,500
b) e2,500 to under e5,000
c) e5,000 to under e10,000
d) e10,000 to under e25,000
e) e25,000 to under e50,000
f) e50,000 to under e75,000
g) e75,000 to under e100,000
h) e100,000 to under e150,000
i) e150,000 and above

Q23: How would you estimate the outstanding value of all the loans of your household?
[Dropdown menu]
a) e0 (no loans)
b) e1 to under e5,000
c) e5,000 to under e10,000
d) e10,000 to under e25,000

23
e) e25,000 to under e50,000
f) e50,000 to under e75,000
g) e75,000 to under e100,000
h) e100,000 to under e150,000
i) e150,000 to under e200,000
j) e200,000 to under e350,000
k) e350,000 to under e500,000
l) e500,000 and above

Q24: How interesting did you find this survey?

Response options: 1 (“Not at all interesting”) – 2 – 3 – 4 – 5 (“Very interesting”)

Q25: Do you have any suggestions or feedback regarding our survey? Please share them here
(optional).
[Text field]

Q26: Thank you for participating in our survey!

As a token of appreciation for your participation, you will receive an Amazon voucher worth e10
[if voucher] or a chance to win one of 50 Amazon vouchers worth e50 [if lottery]. To receive the
voucher, simply confirm that you would like to be contacted by us for the purpose of voucher
delivery and provide your email address in the next step.
a) Yes, I would like to receive the voucher [if voucher] / participate in the voucher lottery [if lottery]
b) No, I do not want to receive the voucher [if voucher] / do not want to participate in the voucher
lottery [if lottery]

[If Q26 is answered with a), present Q26.1 in the same screen]
Q26.1: Please provide your email address for voucher delivery:
Enter email address:
Confirm email address:

[Closing text below]

Thank you once again for your participation! Your answers have been saved. You may now close
this window in your browser.

24