FIT Working Paper 14

Essi Eerola, Tuomas Kosonen, Kaisa Kotakorpi and

Teemu Lyytikäinen

Tax Compliance in the Rental

Housing Market: Evidence
from a Field Experiment
 Tax Compliance in the Rental Housing Market:
Evidence from a Field Experiment∗
† ‡ §
Essi Eerola Tuomas Kosonen Kaisa Kotakorpi
¶
Teemu Lyytikäinen

June 2023

Abstract

We study rental income tax compliance using novel third-party information and
a large-scale randomized field experiment. The third-party information combines
register data on the ownership and occupancy of apartments. The RCT sent infor-
mation letters about the usage of third-party information in tax enforcement. This
increased the propensity to report rental income and the amount of reported rental
income net of expenses. Our research design also allows us to identify members of
ownership networks and analyze spillover effects in tax enforcement between them.
We find positive reporting spillovers. We do not find evidence of real effects on asset
market transactions.
JEL: H26, H31
Keywords: Tax compliance, field experiment, rental market, spillovers, real effects
of tax enforcement.
∗
We thank the team at the Finnish Tax Administration for cooperation in conducting the field exper-
iment. We also thank participants at IIPF Congress 2018, NTA Congress 2018, MaTax Conference 2018,
Workshop on Empirical Analysis of Tax Compliance at University of Oslo, and seminar audiences at NHH
Bergen and the Ministry of Finance forcomments. The research received funding from the Academy of
Finland (grant no. 277283, 299373, 315591, 346250 and 346253). The opinions expressed in this paper
are those of the authors and do not necessarily reflect the views of the Bank of Finland. The experiment
has been pre-registered at the AEA RCT Registry, https://www.socialscienceregistry.org/trials/2575.
†
essi.eerola@bof.fi, Bank of Finland.
‡
tuomas.kosonen@vatt.fi, Finnish Centre of Excellence in Tax Systems Research (FIT), VATT Insti-
tute for Economic Research.
§
kaisa.kotakorpi@tuni.fi, FIT, Tampere University and VATT.
¶
teemu.lyytikainen@vatt.fi, FIT, VATT.
1 Introduction
Rental income from buy-to-let housing received by landlords is in most developed countries
heavily taxed and a significant source of tax revenue.1 Rental income tax is potentially
easy to evade for taxpayers and difficult to enforce for tax authorities: Ownership of
buy-to-let housing is highly dispersed with a large number of landlords each renting out a
small number of flats. Also, rental income is not subject to third-party information or tax
withholding, and therefore differs from e.g. wage earnings where third-party reporting has
been shown to be crucial for effective enforcement (Kleven et al., 2011). Currently, we have
only scant evidence on the extent of evasion from rental income, partly due to the lack of
third-party information. Further, housing constitutes a large share of household wealth,
and hence tax treatment of housing may have broad efficiency and equity implications.
In this paper we build a novel source of third-party information for rental income re-
vealing potential rental apartments, and hence potential landlords. We then investigate
tax compliance among potential landlords by a large-scale randomized field experiment
(RCT), where the Finnish Tax Administration sent information letters to a randomized
group of the potential landlords notifying them about the usage of the third-party in-
formation in tax enforcement. Our main results indicate a significant increase in tax
compliance on the extensive margin and also an increase in net rental income on the in-
tensive margin. We also find enforcement spillovers in networks formed by co-ownership of
rental apartments. Further, we do not find evidence of real effects of stricter enforcement
on asset allocations.
The innovation in constructing the third-party information is to combine information
on ownership of properties with information on residence from different administrative
registers. Owners of apartments that are occupied by someone else than the owner are
identified as potential landlords and hence potential recipients of rental income.
We build an RCT that leverages this new source of third-party information in two ways:
First, the new information enables targeting enforcement at a relevant sub-population
of taxpayers. Without this information, for example, landlords who never report any
rental income might largely go unnoticed. This type of information was not available
in tax enforcement at the time of the study. Second, our results show that the new
type of information is effective in deterring non-compliance among those taxpayers. In
detail, in the RCT a randomly selected subset of potential landlords received a letter from
the Finnish Tax Administration, notifying them of various features of rental income tax
filing and enforcement, such as the usage of third-party information revealing potential
1
In many OECD countries, rental property is the most heavily taxed type of asset (OECD, 2018).

1
landlords.
In addition to analyzing direct effects on tax reporting, our experimental design allows
for studying various types of spillovers in tax enforcement. Previous research has indi-
cated that spillovers have potentially important effects on the efficiency of the tax system
(e.g. Pomeranz (2015)) and spillovers imply that considering direct effects only would
understate the effects of enforcement. However, a systematic analysis of spillovers is often
hampered by the inability to observe the relevant network within which spillovers can be
expected to arise. In this regard, buy-to-let housing is interesting as apartments are often
co-owned by several individuals. This creates well-defined networks between owners that
we can identify with our administrative data. We can examine the tax reporting behavior
of those co-owners who themselves did not receive a treatment letter, but are in the same
network with a potential landlord who did. In addition, we examine spillovers within
social networks, in this case within the household.
Moreover, our experimental design allows a study of spatial spillovers between land-
lords within local rental markets. Estimating the magnitude of spatial spillovers is of
interest in itself. More importantly, if spatial spillovers exist they would create a bias
in our main estimates if not accounted for; see Crépon et al. (2013) for a similar RCT
design. An analysis of these various types of spillovers in a unified setting is exceptional
in the tax enforcement literature.
Further, we contribute to a broader analysis of the efficiency effects of rental income
taxation. An efficient tax creates as little behavioral responses as possible. As rental
housing is merely one among various assets individuals can invest in, applying differential
tax treatment to different assets - along with varying opportunities for tax evasion -
potentially distorts investment and savings choices. Consequently, stricter enforcement of
rental income tax could affect the portfolio choice of investors. Ultimately, rental income
taxation potentially affects rental housing supply and the tenure-type distribution in
the economy. We are able to analyze real effects on portfolio choice with our extensive
administrative data.
As our first main empirical result, we find that the treatment letters affected the
reporting behavior of potential landlords. The strongest treatment informing potential
landlords of the usage of third-party information in tax enforcement caused a marked
increase in the propensity to report rental income. Compared to previous literature,
the type of third-party information we utilize has interesting features: The information
consists of a signal of potential existence of economic activity that confers a tax liability
on the individual (rental activity), but provides no indication of the scale of this activity
(actual income). Accordingly, we find the most pronounced effect on the extensive margin

2
of compliance: the letters caused a larger fraction of potential landlords to report some
rental activity. Nevertheless, we also find clear effects on the intensive margin (i.e. the
euro-amount of income reported net of expenses).
Further, our results indicate interesting heterogeneity. Potential landlords who did not
report any rental income in the year prior to the experiment respond very strongly to the
use of third-party information. The strongest treatment letter increased the propensity
to report rental income by more than 50% in this group in the treatment year. We also
find that the effects differ by the scale of rental activity. The responses are strongest in
the case of small-scale renting. These findings are important for tax enforcement: In the
absence of third-party information, the group of landlords who never report any rental
income would likely largely go undetected. Our novel third-party information allowed
targeting enforcement measures specifically at this group of potential evaders. Further,
an enforcement strategy relying on tax audits would be particularly costly in the case of
small-scale renting. This points to a potential reason for why and where tax evasion exists
from rental income. Alternative methods to tax audits, such as the use of third-party
information in tax enforcement, are particularly attractive in this institutional setting
where small-scale activity makes up a significant share of the market.
Regarding enforcement spillovers, we find a positive spillover effect on tax reporting
within ownership networks: Those in the ownership network who did not receive a letter,
responded to a letter received by someone else within the same network. The effect
is statistically significant and the point estimate is about as large as the direct effect.
Because not all potential landlords are members of an ownership network, the total impact
of the treatment letters increases by about 20% because of this spillover effect. For spouses
we do not find any clear spillover effects. A potential explanation for the difference
between ownership network and spouse spillover effects is that there seems to have been
more tax evasion in the ownership network in the baseline, than among spouses.
Further, to analyze spatial spillovers, we use a block-design inspired by Crépon et al.
(2013), where we analyze how treatment effects spills over to other landlords not directly
treated within tightly defined geographical areas. We do not find clear evidence of enforce-
ment spillovers between landlords in local rental markets. This result provides confidence
that spatial spillovers do not cause a bias in our main results.
Finally, as for potential broader efficiency effects of rental income taxation and en-
forcement, we analyze the real effects of our experimental treatments. In particular, we
examine whether there are effects on housing market transactions and on the ownership
of other assets. We do not find clear effects on either of these outcomes.

3
2 Previous literature
Our paper is related to several strands of literature.
First, to the best of our knowledge, we are the first to conduct a comprehensive analysis
of the effects of rental income tax enforcement. Two previous studies have analyzed rental
income tax compliance, but have focused on a specific question or group of taxpayers.
Wenzel and Taylor (2004) carried out an experiment with the Australian Taxation Office
concerning itemization of rental property expenses. The study shows that those landlords
who were required to itemize deductions significantly reduced deductions compared to
landlords who received an information letter only. López-Laborda et al. (2023) analyze a
warning system set up by the Spanish Tax Agency to deter tax evasion of vacation rental
income. The system targeted individuals who had advertised rental properties in the
media and informed them that the tax authority is aware of the advertizing activity. The
results indicate a 6% increase in the share of income reported as vacation rental income.
Second, our study relates to the literature on the role of third-party information in
tax enforcement, by building a novel source of third-party information that enables more
effective enforcement, and using this information in an RCT. The literature using RCTs
to study the effects of tax enforcement started with Slemrod et al. (2011) and Kleven et al.
(2011), has since expanded, and a meta-analysis is provided by Antinyan and Asatryan
(2020). Although the importance of third-party information in tax enforcement has been
acknowledged in the literature (e.g. Slemrod (2007), Kleven et al. (2011)), research utiliz-
ing randomized variation in third-party information is scarce.2 In studying responses to
an audit experiment, Harju et al. (2020) implemented randomized variation in the salience
of third-party information, albeit in a quite different context, namely international trade:
they found third-party information to be effective in detecting and deterring tax evasion
from imports of used cars.3
Third, we contribute to the literature on spillovers in tax enforcement. A few earlier
papers have studied regional enforcement spillovers between individuals in the context
of TV license fee collection (Rincke and Traxler (2011), Drago et al. (2015)), income
tax filing (Meiselman, 2018) and taxation of firms (Lediga et al. (2022)). Frimmel et al.
2
In Kleven et al. (2011), variation in third-party reporting comes from certain types of income being
subject to third-party reporting, while others (notably self-employment income) are not. Studies that
utilize quasi-experimental variation in third-party information include Pomeranz (2015), who compares
VAT declarations involving line-items that are covered by paper trail (transactions between two firms) to
line-items that are not (sales to final consumers). Naritomi (2019) compares retail transactions (where
third-party information increased due to a campaign that incentivized consumers to send in their receipts
to the authorities) and wholesale transactions (not affected by the campaign).
3
Kotakorpi et al. (2022) provide complementary evidence on how randomized third-party information
affects both tax reporting and pricing decisions in a double-auction lab experiment.

4
(2018) and Alstadsaeter et al. (2019) analyze tax evasion and avoidance spillovers within
the family but do not focus on the effects of enforcement measures. Pomeranz (2015),
Boning et al. (2018) and Brockmeyer et al. (2018) analyze enforcement spillovers in firm
networks. We contribute to this literature by analyzing tax enforcement spillovers in a
well-defined ownership network, created by the nature of the institution that we analyze.
Landlords in these networks are more closely connected than for example landlords within
a geographical area. We show that spillovers can exist in closely related networks although
they do not arise in a spatial setting or between spouses.
Fourth, we contribute to literature on real economic effects of enforcement. The effects
of tax incentives on real economic behavior in many contexts has of course been widely
documented in economic research.4 However, the potential real effects of tax enforcement
- with implications for effective tax rates - have been much less studied. Much of the
empirical work on deterrence (e.g., Kleven et al. (2011) and subsequent field experiments)
focuses on reporting responses while there is little evidence on whether enforcement affects
real economic activity. Exceptions include Kopczuk et al. (2015), who provide an analysis
of price effects of enforcement. Related to our setting, Bomare and Le Guern Herry
(2022) show how enforcement may affect allocations of off-shore wealth into different types
of assets. In general, reporting responses and real responses (i.e. responses pertaining
to market behavior) are difficult to disentangle with field data. We use data on asset
transactions of landlords to detect potential real effects on housing market transactions
or other aspects of portfolio choice.

4
In our particular context, empirical evidence suggests that differential tax treatment indeed affects
household portfolio allocation, but most studies focus on financial assets - see e.g. Alan et al. (2010) and
references therein. Fossen et al. (2020) in turn distinguish between six asset types using German data:
Private business equity, owner-occupied housing, rental property, financial assets (stocks and bonds), life
and private pension insurance, and tangible assets and show that an increase in the marginal income tax
rate has a significant negative effect on the probability of holding rental property.

5
3 Institutions and Experimental Design
3.1 Institutional Setting
3.1.1 Rental Housing Market in Finland

The functioning and size of the private rental market varies from one country to another
depending on a range of institutional details (e.g. tax treatment of rental and owner-
occupied housing, rent-setting and tenancy security regulations, prevalence of non-profit
housing sector and demand subsidies). However, there are also some key common features
across countries, including Finland. First, the private rental market is often dominated
by private landlords who typically own only a few dwellings (Whitehead et al., 2016).
Second, taxable rental income includes a complicated set of different types of deductible
cost items ranging from fixed costs to costs that vary with the size and/or rent of the
dwelling.5 Third, tenants in the private rental market tend to be more mobile than owner-
occupied households. For instance, in Finland, the annual mobility rate among renters is
around 20% and substantially higher than the mobility rate of owner-occupied households.
One third of all Finnish households live in rental housing.6 Especially the share of
private rental housing has been increasing since the early 2000’s. Currently, the private
rental market constitutes two thirds of the overall rental market.7 Roughly half of the
rental units in the private rental market are owned by households or individuals.
In the private rental market, legislation on rental agreements is very flexible. In this
respect the institutional setting in Finland is very similar to countries like the U.K. or the
U.S. Rent-setting is not subject to any restrictions, and in the case of long-term rental
agreements, the rent is typically reviewed annually. The size of annual rent increases must
be specified in the lease agreement and is typically based on the cost-of-living index. Valid
reasons for contract termination include unpaid rents, sale of the dwelling, or personal
use by the landlord.

3.1.2 Taxation of Rental Income

The Finnish tax system is a dual system combining progressive taxation of labour income
with a separate tax on all capital income. At the time of the experiment, the net rental
5
See Sommer et al. (2013), Whitehead et al. (2016) and Ball (2016), for the tax treatment of rental
income in the U.S., Germany and U.K., respectively.
6
See, e.g. http://www.oecd.org/social/affordable-housing-database/housing-market/ for information
on the tenure distribution in different countries.
7
The rest of the rental market can be characterized as social housing. In the social housing sector,
rents and tenant selection are regulated. The housing units are owned by municipalities and non-profit
organizations that are not subject to regular capital income taxation.

6
income was subject to a capital income tax rate of 30%, and 33% if taxable capital income
exceeded an annual threshold of 30,000 euros.8 The rental income tax is a non-negligible
source of tax revenue. In 2015, total reported rental income net of expenses amounted
to 1.6 billion euros. The corresponding tax revenue was more than 480 million euros (or
1.1% of central government tax revenue).
In the analysis, we focus on rental apartments owned by individuals. For the purposes
of this study, we identify likely landlords by combining register data on ownership and
flat occupancy in a manner explained in Section 3.2.1. In what follows, we refer to these
individuals as potential landlords.
Figure 1 illustrates the phenomenon under study. The figure shows the share of po-
tential landlords reporting rental income, as well as the share of potential landlords and
total tax revenue, by the number of potential rental apartments owned.

Figure 1: Rental income tax reporting by potential landlords

Notes: Figure describes reporting of rental income and the share of total tax revenue by the number of
potential rental flats owned at the end of tax year 2015. The data used in the figure contains individuals
in untreated control blocks (N = 19,208).

There are a number of interesting points to note from the figure. First, small scale
renting is highly prevalent, and significant from a tax revenue perspective: More than
80% of all potential landlords own only one potential rental apartment, and their share of
the overall rental income tax revenue was almost 60%. Second, out of those individuals
owning one potential rental flat, roughly 75% reported some rental income in tax year
8
The tax rate has slightly increased during the recent decades and became progressive in 2012.

7
2015. The figure also shows (right axis) the amount of rental income reported in 2015 by
the number of potential rental apartments owned.
As the ownership of rental units is widespread across households and small-scale rent-
ing makes up a large share of tax revenue, enforcement may be costly for tax authorities.
This underlines the importance of looking for ways to steer taxpayers to comply without
tax audits.
Turning next to the tax-filing procedure, pre-populated income tax returns are sent
out to taxpayers each year in late April. The pre-populated return contains information
on incomes that are subject to third-party reporting. The taxpayer is required to submit
a revised return to the tax authority if any income information is missing from the pre-
populated return. The taxpayer can also apply for discretionary deductions (e.g. expenses
for travel to work). The taxpayers have to submit their corrections in May; otherwise,
the original proposal is implemented.
As income from rental property was not subject to any third-party reporting at the
time of our study, all individuals with rental income had to revise the pre-populated tax
return and submit the revision to the tax authority. Rental income is reported on a sep-
arate form (see Appendix B), and income and deductible expenses have to be reported
separately. In the experiment, we implement a nudge to the regular tax filing proce-
dure, whereby treated individuals received different types of communication from the Tax
Administration shortly after receiving the pre-populated tax return.
In the absence of third-party information, tax enforcement needs to rely on self-
reported income or to audit taxpayers whose reporting behavior raises a red flag. Obvi-
ously, the red flags indicating non-compliance are difficult to implement based on reporting
behavior only if an individual never reports any rental income. The idea of our information
experiment is to create an environment where compliance is high even without large-scale
audits.

3.2 Experimental Design

3.2.1 Constructing the base population

The base population for the experiment was formed using the tax authority’s register on
flat ownership and an extract from the national population register on flat occupancy.9
We linked the information from these registers using personal identification numbers that
uniquely identify individuals across different national registers. Both data sets are based
9
We focus on flats in apartment buildings and leave out detached houses which are predominantly
owner-occupied and often located in rural areas with thin rental markets.

8
on the situation at the end of year 2015.10
In these data we classify flats that are occupied by someone else than one of the owners
as potential rental flats. The owners of these flats are classified as potential landlords.
We drop flats with more than 15 tenants and more than 5 owners. We also keep only
potential landlords who own up to 15 potential rental flats.
In order to guarantee that only one potential landlord from owners of the same flat
receives a letter, we restrict the base population as follows. First, we construct households
using information on home address and for each household include only the potential
landlord with the largest number of potential rental flats.11 This guarantees that only
one member of each household receives the treatment and this individual is the one with
most extensive ownership.
Second, multiple individuals from different households may own a share of the same
flat, which creates co-ownership networks. For each potential rental flat owned by at least
two individuals, we use ownership shares to identify the main owner and allocate the flat
to this owner. In case of equal ownership shares, the main owner is randomly chosen.12
This minimizes unintentional spillovers across experimental treatment groups, and at the
same time allows a rigorous analysis of spillovers within ownership networks.
We do not expect everybody in the base population to have received rental income in
2015. Rather, we have an imperfect indication of likely rental income. This is because
the end of year snapshot nature of the data and potential reporting lags and errors in
the registers means that the base population is based on an imperfect measure of rental
apartments owned during the year. For example, our measure does not capture rental
apartments sold before the end of the year or apartments rented out earlier during the
year but vacant at the end of the year. Further, the register of flat ownership and flat
occupancy may sometimes be updated with a lag. Finally, some of the potential rental
apartments may have been occupied free of rent (e.g. by a friend or a relative of the
owner).
10
We drop flats that have been bought in November or December 2015 because it is unlikely that in
these cases a new rental contract could have been made with a tenant before the end of 2015.
11
In case of equal number of flats, we choose the person with the biggest ownership share. In case the
ownership shares are also equal we randomly choose one person.
12
Here we use information on all potential rental apartments owned and not only apartments for which
the individual is the main owner. For example, say, that individual A and B jointly own a potential
rental flat and A is the main owner. In addition, B owns another potential rental flat alone. Then only
one of them is randomly selected into the base population.

9
3.2.2 Treatments

The treatment letters were sent out in April 2016 by the Finnish Tax Administration. We
randomized letter recipients from the whole base population, but the main analysis focuses
on dense rental markets using a block design discussed below. Without the novel third-
party information used in this study, we could not have targeted the letters to potential
landlords, and for example landlords not reporting any rental income might have gone
unnoticed in tax enforcement. The letters were sent out just after the taxpayers received
their pre-populated income tax returns for income earned in 2015. The Finnish Tax
Administration committed to the experimental design with a non-treated control group
for year 2016, but not necessarily in consequent years.
All in all, roughly 45,000 treatment letters were sent. The experiment consisted of four
different treatments: 1) Letter with a neutral reminder to file tax returns and to report
deductions and missing income, such as rental income; 2) Letter providing information
on how to file rental income; 3) Letter notifying the recipient of a general increase in
the intensity of rental income tax enforcement; 4) Letter on intensified enforcement of
rental income taxation and a mention of the use of third-party information on ownership
of dwellings. All treatment letters (2)–(4) contained also the information provided in
treatment letter (1). Group (1) can serve as an alternative comparison group for the
other treatment groups.
The enforcement measures described in letters (3) and (4) were implemented by the
Finnish Tax Administration in summer 2016. The full letters are shown in Appendix B.
Table 1 illustrates our experimental design. We used a randomized block design,
similar to the design in Crépon et al. (2013), to be able to analyze potential spatial
spillovers of the treatments. We then allocate the individually randomized four letter
groups to the blocks.
We use the following procedure. We first allocate each potential landlord in our base
population to a postcode area based on where the flats owned are located. Those owning
flats in several postcode areas are allocated to the postcode area with most flats.13 As
we wish to analyze spatial spillovers, we leave out areas that are mostly populated by
owner-occupiers living in detached houses, and select into the block design only postcode
areas with a reasonably dense rental market.14 These postcode areas (or blocks) are then
13
In case of equal number of flats in two postcode areas, we use the sum of ownership share to allocate
the owner to the postcode area. In case of equal ownership shares, we randomly allocate the owner to
one postcode area.
14
We leave out rural municipalities with less than 5,000 flats. Furthermore, we leave out postcode areas
with less than 60 flats and with on average less than five flats per building. After these restrictions, we
have 263 postcode areas.

10
randomly assigned into three treatment groups with varying intensity of treatment: i)
control blocks where no letters were sent; ii) low-intensity blocks where 24% of potential
landlords in the base population received a letter; iii) high-intensity blocks where 62% of
potential landlords received a letter.15 In addition, the share of the stronger treatment
letters (3) and (4) was higher in the high-intensity blocks. For instance, out of those
receiving a letter, roughly a third in the low-intensity blocks and a half in the high-intensity
blocks received letter (4). Finally, to obtain (direct) treatment effect estimates for the
entire population of potential landlords, some treatment letters are sent to randomly
chosen potential landlords who live outside of the areas included in the block design
(second last column in Table 1).

Table 1: Experimental design.

Low intensity High intensity

Control blocks blocks blocks Not in blocks Total

No letter 19208 21320 14995 28178 83701

Letter 1 0 1713 2502 4779 8994
Letter 2 0 1739 2383 4871 8993
Letter 3 0 1118 6476 1397 8991
Letter 4 0 2310 12863 2813 17986
Total 19208 28200 39219 42038 128665

Postcode areas 111 62 90 4200 4463

Notes: Table shows the number of letters sent to different groups of potential landlords in the base
population in the treatment and control groups as well as the number of postcode areas.

Figure 2 illustrates the block design for Helsinki, the capital city and largest municipal-
ity in our data. There are roughly 650,000 inhabitants and 80 postcode areas in Helsinki.
The postcode areas with a reasonably dense rental market (as defined above) are ran-
domly assigned to control, low-intensity or high-intensity groups (for data confidentiality
reasons, we are not able to show which ones).
15
We first form groups of postcode areas of similar size. Then within each strata, randomly assign
postcode areas to different blocks.

11
 Figure 2: An illustration of the block design for Helsinki.

Source: City of Helsinki, Map service. https://kartta.hel.fi/

4 Data and Empirical Strategy

4.1 Data
Our data covers the whole population and contains information on individual rental in-
come (gross and net), place of residence, ownership of apartments, transactions of apart-
ments, ownership and transactions of other assets, family status (id of spouse/partner
included), age and gender. Summary statistics of key variables in the data are reported
in Table A1 of Appendix A.
Given that landlords and rental markets not in the blocks are quite different from
those in the blocks, we only utilize data from the blocks in our main analysis and report
the results for individuals outside the blocks in Appendix A. This choice also allows us
to analyze spatial spillovers, and to isolate the direct treatment effects of the letters from
such spillovers.
Table 2 describes reporting of rental income before the treatment (Panel A) and after
the treatment (Panel B). The comparison of different treatment groups in Panel A shows
that the randomization has been successful as the groups are very similar in terms of the
pre-treatment propensity to report, reported gross rental income and reported net rental
income. This is to be expected by construction.

12
 Overall, a comparison of Panel A and B indicates that the propensity to report rental
income is higher after the treatment. This is true also in the ”No letter” group. Such
changes over time may be due to general developments in the rental market. One specific
reason may be related to turnover: some of those who owned a potential rental flat in
2015 may not have owned one in 2014. This would mechanically increase the propensity
to report from 2014 to 2015.
A first indication that the treatment had some effect on the propensity to report
rental income is visible in Panel B: For example, those receiving Letter (4) had a higher
propensity to report than those not receiving a letter or receiving Letter (1).

Table 2: Reporting of rental income before and after the treatment.

Reported rental Gross rental Net rental

income 1/0 income income
Mean Std. Dev. Mean Std. Dev. Mean Std. Dev.

Panel A: Before treatment (tax year 2014)

No letter 0.724 0.447 6725 8320 3731 5005
Letter 1 0.730 0.444 6600 7870 3655 4833
Letter 2 0.731 0.443 6845 8642 3775 5220
Letter 3 0.728 0.445 6698 8127 3777 5110
Letter 4 0.733 0.443 6722 9301 3752 4912

Panel B: After treatment (tax year 2015)

No letter 0.772 0.419 7397 8513 4063 5119
Letter 1 0.792 0.406 7311 8223 4021 4944
Letter 2 0.801 0.399 7545 8431 4197 5189
Letter 3 0.803 0.398 7525 8358 4240 5206
Letter 4 0.815 0.388 7498 8040 4194 4904

Notes: Table shows rental income reporting before the treatment (tax year 2014) and after the treatment
(tax year 2015) in the treatment groups.

4.2 Empirical strategy

Since individuals are randomly assigned to treatment groups we could estimate average
treatment effects consistently through simple OLS regression of outcomes on treatment
indicators. However, we can use the panel property of the data to improve the reliability
of the estimation through a Differences-in-Differences strategy which compares outcomes

13
in treatment groups over time. This strategy controls for possible idiosyncratic differences
across treatment groups, and thus reduces the noise in the estimates. We estimate the
following model:

X X
yit = αt + βjt Letterj + γkt Blockk + µi + it (1)
j k

where yit is the outcome for individual i at time t. We consider the effects of the different
treatment letters (Letterj ) separately, and include dummies for high or low intensity
blocks (Blockk ). The coefficients βjt then identify the effects of the different letters in
different years compared to the base year before the treatment. The γkt identify the effects
of being in a high or low-intensity block (over and above the direct effect of receiving a
letter), relative to the control block, and provide a measure of regional spillovers. To
estimate spillover effects within ownership networks and/or between spouses, equation
(1) is estimated on a sample of individuals consisting of co-owners or spouses of letter
recipients. We control for general changes in outcomes through time dummies. In addition
we include individual fixed effects and control for the number of all flats owned to improve
precision. We cluster standard errors at the postcode level - the same level at which we
randomize geographic treatment blocks.16 We also control for the enforcement measures
associated with the experiment but we cannot disclose the estimates.
We first estimate equation (1) with data from two years before the treatment and
two years after the treatment and present the findings in the form of event-study type
graphs of regression coefficients, where we pool the stronger treatment letters (Letter 2-
4) into one. We illustrate the impact of these letters by plotting the coefficient on the
letters relative to the group not receiving a letter. We then estimate the model with data
form one year before and after the reform, and report the results in tables where we also
perform formal tests for the difference of each of the stronger treatment letters to Letter
1.
We analyze heterogeneous responses first visually by plotting the estimates of model
(1) for sub-samples based on previous reporting status and the number of potential rental
apartments owned. In order to disentangle the impacts of different characteristics on the
responses to the treatments, we estimate a model which includes interactions of letter*year
dummies with a dummy for not reporting rental income in tax year 2014, a dummy for
16
We trim the data by excluding individuals with extreme values for the following variables: net rental
income, spouse’s net rental income, number of flats, number of sold flats and number of bought flats. We
drop the individual from the sample in all years if the value of any of these variables is above the 99.5
percentile or below the 0.5 percentile in any year during the study period. This reduces the sample size
by 5.9% and decreases standard errors considerably.

14
owning only one potential rental flat, as well as dummies for age below 40 and gender.
This model uses data from two years (tax years 2014 and 2015) and also controls for the
interaction of the characteristics and year.

5 Results
5.1 Direct reporting effects
In this subsection we analyze the direct effects of the treatments on tax reporting by let-
ter recipients. According to the deterrence model of tax evasion (Allingham and Sandmo
(1972)), the extent of evasion depends crucially on the probability that evasion is detected.
Recent literature has emphasized that the detection probability is particularly high for
income items that are subject to third-party reporting (e.g. Kleven et al. (2011)). Our
experimental treatments (3) and (4) are designed to affect the recipient landlords’ percep-
tion of the detection probability - treatment (3) through providing a signal of a general
increase in enforcement intensity, and treatment (4) additionally through notifying land-
lords about the use of third-party information in tax enforcement. Also letter (2) might
increase the perceived detection probability, because it is a letter from the Tax Authority
mentioning how rental income should be reported. We therefore expect the experimental
treatments to affect landlords’ reporting behavior, in particular if there has been non-
compliance in the baseline.
Recall that treatment letters were sent out in spring 2016, just after the taxpayers
received their pre-populated tax returns for income received in 2015. Therefore tax year
2015 is the first treatment year, and any effects observed in that year are pure reporting
effects as any initial real effects are ruled out by the timing of the experiment. Any
effects observed in subsequent years may incorporate both reporting and real responses.
Moreover, the experimental design of having treatment groups and the control group not
facing any treatment applies only to tax year 2015. Thus, we cannot reliably estimate the
impact of the treatments for reporting in tax year 2016 as the control group could have
been partially treated then.
Figure 3 shows the development of the propensity to report rental income (left panel)
and the reported net rental income (right panel).17 Figures 4 and 5 show the same
outcomes for different subgroups of interest: In Figure 4 we divide the base population
into those who reported rental income in tax year 2014, that is one year before the
17
Figure A1 and Table A2 in Appendix A show the corresponding results for potential landlords in our
base population but outside the treatment blocks.

15
treatment, and to those who did not report any rental income. In Figure 5, in turn, we
divide the base population according to the number of potential rental flats owned.
The figures show that the treatment letters affected rental income reporting of potential
landlords, both on the extensive margin (propensity to report) and on intensive margin
(the amount of net rental income). The effects appear to be concentrated on those who
reported no rental income in the previous year, and among small-scale renters (individuals
with only one potential rental flat). Despite of the randomized experiment there appears
to be pre-trends especially for reporting of rental income. This could be due to the block
design where the number of blocks is limited. Nevertheless, the pre-trends do not appear
in all outcomes, and even when they do, the effects are clearly distinguishable from any
possible pre-trends in most cases.
The effects appear to be short lived, but again that could be because we cannot reliably
rely on the experimental design for tax year 2016; we discuss this more extensively at the
end of this subsection.

16
 Figure 3: Reporting of rental income in letter groups 2-4

Notes: Figures show regression coefficients on treatment letter by year dummies (ref. no letter and year
2014). Controls include individual fixed effects, dummies for the number of apartments, treatment block
by year dummies and additional enforcement measures related with the treatments. Vertical lines indicate
95% confidence intervals (clustering at postcode level).

17
Figure 4: Reporting of rental income in letter groups 2-4 - by reporting status before treatment

Reporting of rental income 0/1 Reporting of rental income 0/1

No reported rental income in 2014 Reported rental income in 2014
.1

.1
-.04 -.02 0 .02 .04 .06 .08

-.04 -.02 0 .02 .04 .06 .08

Difference to no letter group

Difference to no letter group

2

2
2 2
2
2

2013 2014 2015 2016 2013 2014 2015 2016

Tax year Tax year

Net rental income Net rental income

No reported rental income in 2014 Reported rental income in 2014
400

400
Difference to no letter group

Difference to no letter group

200

200

2
2
2
2 2
0

2
-200

-200

2013 2014 2015 2016 2013 2014 2015 2016

Tax year Tax year

Letters 2-4

Notes: Figures show regression coefficients on treatment letter by year dummies (ref. no letter and year
2014). Controls include individual fixed effects, dummies for the number of apartments, treatment block
by year dummies and additional enforcement measures related with the treatments. Vertical lines indicate
95% confidence intervals (clustering at postcode level).

18
Figure 5: Reporting of rental income in letter groups 2-4 - by number of apartments owned
before treatment

Reporting of rental income 0/1 Reporting of rental income 0/1

1 potential rental flat 2 or more potential rental flats

.04
.04

2
Difference to no letter group

Difference to no letter group

.02
.02

0
2
0

2
2
2

-.02
-.02

2013 2014 2015 2016 2013 2014 2015 2016

Tax year Tax year

Net rental income Net rental income

1 potential rental flat 2 or more potential rental flats
1000
400
Difference to no letter group

Difference to no letter group

200

500

2
2 2 2 2
0

0
-200

-500

2013 2014 2015 2016 2013 2014 2015 2016

Tax year Tax year

Letters 2-4

Notes: Figures show regression coefficients on treatment letter by year dummies (ref. no letter and year
2014). Controls include individual fixed effects, dummies for the number of apartments, treatment block
by year dummies and additional enforcement measures related with the treatments. Vertical lines indicate
95% confidence intervals (clustering at postcode level).

Turning to the magnitude of the effects, Table 3 shows the estimated effects of the
different treatment letters (equation (1)), on the reporting of rental income for tax year

19
2015 (Panel A) and on the net rental income (Panel B). The table shows separately the
effects of all four letters (βj coefficients from equation 1). The first column shows the
results for all potential landlords.
The first column shows that all letters caused a statistically significant increase in the
propensity to report. Letter (2) providing information on reporting procedures and re-
quirements on rental income increased compliance, which suggests that outright mistakes
may play a role in non-compliance. Letter (4), the strongest treatment, which notified po-
tential landlords of the use of third-party information in tax enforcement, had the largest
effect. The effect of letter (4) is to increase the compliance rate by about 3.2%-points,
which amounts to a relative effect of 4.0% compared to the baseline compliance rate of
73.7%.18
Compared to previous literature, the type of third-party information that we utilise has
interesting features: The information consists of a signal of potential existence of economic
activity that confers a tax liability on the individual (rental activity), but provides no
indication of the scale of this activity (actual income levels). Accordingly, the effects
that we find are strongest on the extensive margin of reporting. Nevertheless, there are
also significant effects on the amount of rental income reported. Interestingly, we find a
positive effect on net rental income. This is in contrast with findings in some previous
studies where taxpayers have scaled up deductible costs to offset the effect of specific
third-party information on the amount of taxable income (Slemrod et al. 2017; Carrillo
et al. 2017).
Turning to the analysis of heterogeneous responses in the other columns of Table 3,
we report the interaction of landlord characteristics and letters from a regression that
includes all interactions at the same time. We study heterogeneous responses by whether
the individual reported any rental income in tax year 2014 or not, by the number of
potential rental flats owned before the treatment, as well as by age and gender.
The second column shows the results for potential landlords who did not report any
rental income in tax year 2014 compared to those who did report. While some of these
individuals may not have owned or rented out a flat in the previous year, this is a subgroup
where non-compliance appears more likely. Indeed, the baseline compliance rate (at the
extensive margin) in the control block in tax year 2015 in this subgroup is only about 15%.
The effects of the treatment letters on the propensity to report are now much stronger
than in the first column. Given the low baseline compliance rate in this subgroup, the
relative effect on compliance is very large. Receiving Letter 4 causes an over 50% increase
18
In Appendix A, Table A3 shows the estimation results for reporting behavior in 2015 separately for
reported gross rental income and deductions.

20
in the propensity to report rental income in this group. This finding suggests that non-
compliance was especially prevalent in this group, which is also a group that would have
been difficult to detect in tax enforcement in the absence of our third-party information.
The result also highlights that findings from studies where interventions are targeted at
likely evaders (e.g. Bott et al. 2020) may not generalize to the average taxpayer.
The third column shows that the effects are larger for potential landlords who own
only one potential rental flat. This is a group where traditional enforcement measures
may have a relatively poor cost-benefit-ratio due to dispersed, small-scale activity.
The fourth column shows that Letters 1 and 2 have stronger effects on those younger
than 40, indicating that reminders and instructions are more effective for less experienced
tax payers. In the fifth column we find that the effects of letters on the propensity to
report do not seem to vary by gender.

21
 Table 3: Effects of treatment letters.

No reported One potential

rental income rental flat
in 2014 in 2014 Age below 40 Female
All *Letter *Letter *Letter *Letter

Panel A: Reporting rental income (0/1)

Letter 1 0.0125** 0.0252 0.0209 0.0389*** -0.00613
[0.00598] [0.0156] [0.0128] [0.0146] [0.00916]
Letter 2 0.0250*** 0.0455*** 0.0213* 0.0471*** -0.00536
[0.00617] [0.0152] [0.0111] [0.0146] [0.00985]
Letter 3 0.0195*** 0.0512*** 0.0145* -0.00667 -0.00815
[0.00410] [0.0106] [0.00766] [0.0124] [0.00771]
Letter 4 0.0323*** 0.0795*** 0.0142** 0.00208 -0.00855
[0.00395] [0.0101] [0.00637] [0.00922] [0.00627]

Baseline mean 0.737 0.164 0.729 0.682 0.756

Panel B: Net rental income

Letter 1 20.68 -54.57 -77.43 145.4 -255.6**
[48.97] [118.5] [256.1] [115.5] [103.0]
Letter 2 87.11* 110.7 -10.27 126.6 33.88
[52.34] [124.3] [235.4] [120.4] [109.4]
Letter 3 126.7*** -2.013 -269.9 64.41 -162.3**
[41.09] [78.12] [169.5] [104.2] [75.10]
Letter 4 96.71*** 91.60 165.1 42.29 -61.49
[33.12] [73.62] [121.3] [69.90] [55.50]

Baseline mean 3778 514 2995 3359 3734

Notes: The first column shows the overall effects of treatment letters (ref. no letter) using data from tax
years 2014 (before treatment) and 2015 (after treatment). Sample size is 162,632. Columns 2-5 show the
interaction of characteristics of potential landlords and letters from a model where all interactions are
included at the same time. Controls include individual fixed effects, treatment block by year dummies,
dummies for the number of apartments owned and additional enforcement measures related to the treat-
ment letters. In the interaction model landlord characteristics interacted with year dummies. Standard
errors clustered at postcode area level (263 clusters) are in brackets. Significance is denoted by asterisks:
* p < 0.1, ** p < 0.05, *** p < 0.01. Bold font indicates significant difference at 5% level relative to
Letter 1.

Finally, we we already noted, the positive reporting responses observed for tax year
2015 fade out already in 2016, as was shown in Figure 3. Appendix Table A4 reports

22
DiD estimates for 2016. However, we cannot draw strong conclusions based on this
finding, as the experimental design applies only to tax year 2015. Thus, we do not
have exact information about the developments in enforcement policy beyond the initial
implementation of the experiment and the associated measures. In particular, it is possible
that some of the control group has been subject to more intensive enforcement regarding
reporting of taxes for tax year 2016 (which took place in spring 2017). However, it is
important to note that this potential concern does not contaminate our results regarding
the effects on asset market transactions in 2016, analyzed in subsection 5.3, because
these transactions took place within 2016, before any tax enforcement measures were
implemented in 2017 for tax year 2016.

5.2 Spillover effects

We have constructed the base population for the study in such a way that we can ex-
amine potential spillovers within ownership networks of individuals who own flats to-
gether. Co-ownership is a natural type of close relationship where information sharing
and hence spillovers may occur. We also examine spillovers within the household (co-
owning spouses). Further, the randomized block design similar to Crépon et al. (2013)
enables us to analyze spillover effects from intensified enforcement across landlords within
local rental markets.
Figure 6 and Figure 7 first show the spillover effect on reporting behavior within
ownership networks and within households. The figures point towards significant reporting
spillovers among co-owners, but not within the household.

23
 Figure 6: Reporting spillovers in ownership network

Notes: Data includes people who owned apartments together with people in our main estimation sample
(spouses excluded). Figures show regression coefficients on co-owner’s treatment letter by year dummies
(ref. no letter and year 2014). Controls include individual fixed effects, dummies for the number of
apartments and additional enforcement measures related with the treatments. Vertical lines indicate
95% confidence intervals (clustering at postcode level).

24
 Figure 7: Reporting spillovers within household

Notes: Figures show regression coefficients on treatment letter by year dummies (ref. no letter and year
2014). Controls include individual fixed effects, dummies for the number of apartments, treatment block
by year dummies and additional enforcement measures related with the treatments. Vertical lines indicate
95% confidence intervals (clustering at postcode level).

Table 4 shows the proportions of people potentially exposed to different treatments

through their co-owners.

25
 Table 4: Treatment spillovers in ownership network.

Proportion

Co-owner received:
No letter 0.6406
Letter 1 0.0484
Letter 2 0.0485
Letter 3 0.0925
Letter 4 0.1736

Notes: Data includes people who owned apartments together with people in our main estimation sample
(spouses excluded). Table shows the proportions of people potentially exposed to different treatments
through their co-owners.

Table 5 quantifies the effects of the different treatment letters on the reporting behavior
of co-owners and spouses of letter recipients (in 2015). In the first and second columns
we look at individuals who owned apartments together with individuals in our main
estimation sample (excluding spouses). In the third and fourth columns we look at spouses
of the individuals in our main estimation sample. We find significant and positive spillover
effects for co-owners, but not for spouses.
Finding no spillovers within the household may relate simply to the fact that in our
data, spouses have a higher reporting rate in the baseline than other co-owners (60 % vs.
38 %), which translates into a larger reaction in the latter group.
To quantify the overall significance of spillovers, it should be noted that while the point
estimates of the spillover effects on co-owners are of the same order of magnitude as the
direct effects of the letters, not all landlords are part of a co-ownership network. Taking
this into account, our results indicate that the size of the spillover effect (measured by
the number of additional rental income tax reports arising due to the spillover) amounts
to 20% of direct effect.

26
 Table 5: Spillovers in ownership network and household.

Co-owners Spouses
Reporting rental Net Rental Reporting rental Net Rental
Dep. Var. income (0/1) income income (0/1) income

Letter 1 0.0092 125.0* 0.00367 -29.76

[0.0128] [68.53] [0.00789] [47.21]
Letter 2 0.0104 -49.43 -0.0113* 24.54
[0.0133] [47.89] [0.00649] [44.91]
Letter 3 0.0145 40.67 0.00677 66.88
[0.0101] [52.70] [0.00593] [41.24]
Letter 4 0.0250*** -14.86 0.00852** 37.78
[0.00776] [40.79] [0.00419] [26.35]

Baseline mean 0.377 1315.7 0.604 2167.8

N 39220 39220 69766 69766

Notes: Table shows DiD estimates for the effects of treatment letters (ref. no letter) using data from tax
years 2014 (before treatment) and 2015 (after treatment). Co-owners in columns 1 and 2 include people
who owned apartments together with people in our main estimation sample (spouses excluded). Controls
include individual fixed effects, treatment block by year dummies, dummies for the number of apartments
owned and additional enforcement measures related to the treatment letters. Standard errors clustered
at postcode area level (263 clusters) are in brackets. Significance is denoted by asterisks: * p < 0.1, **
p < 0.05, *** p < 0.01. Bold font indicates significant difference at 5% level relative to letter 1.

Finally, we utilize the block design to analyze enforcement spillovers between landlords
in local rental markets. As Figure 8 shows, we find no evidence of local reporting spillovers.
Nevertheless, utilizing a block design that allows us to examine and control for potential
spatial spillovers, has the important benefit that we can be confident that the estimates
of the direct effects are not biased by potential spatial spillovers.

27
 Figure 8: Reporting spillovers in the local rental market

Notes: Figures show regression coefficients on treatment area by year dummies (ref. control area and
year 2014). Controls include individual fixed effects, dummies for the number of apartments, treatment
letter by year dummies and additional enforcement measures related with the treatments. Vertical lines
indicate 95% confidence intervals (clustering at postcode level).

5.3 Real effects

We next analyze potential real effects of the experiment in 2016. Stricter enforcement
increases an evader’s effective tax rate from rental income, and therefore reduces the
perceived after-tax profitability of investing in rental housing. Stricter enforcement may
then lead to a real effect whereby affected individuals reduce the number of apartments
held, and may invest in other assets instead.19
19
Possibilities for tax evasion from different assets likely differ in the baseline. While the taxation
of rental income was based on self-assessment, tax is deducted automatically from many other types of
capital income such as income from dividends, mutual funds or savings accounts

28
 Figure 9 shows the effect of the treatment letters on housing market transactions. In
this case, the first treatment year is 2016, since the letters were sent in spring 2016. The
figure indicates that the letters did not affect housing transactions of potential landlords.
The result is perhaps not surprising: The effect on tax reporting is on average relatively
small, leading to modest increases in effective tax rates. As rental housing is a lumpy
investment, effects of relatively small changes in effective tax rates may in the end be
difficult to detect. Nevertheless, we regard this analysis as providing a novel example of
how new types of administrative data may be used to identify potential real effects of
tax enforcement. Further, an analysis of trade in other assets (listed stocks and shares
in mutual funds)20 , documented in Table A5 of Appendix A, does not reveal significant
effects on portfolio choice overall.21
20
The data on transactions of listed stocks and shares in mutual funds is available to us only from year
2016. This means that we cannot estimate DiD type models and estimate a cross-sectional OLS model
comparing letter groups in 2016.
21
The analysis of trade in other assets complements the analysis of housing market transactions: smaller
reductions in investment in rental housing - such as lower expenses on improvements - may show up in
amounts invested in other assets.

29
 Figure 9: Housing transactions in letter groups 2-4

Notes: Figures show regression coefficients on treatment letter by year dummies (ref. no letter and year
2015). Controls include individual fixed effects, treatment block by year dummies and additional enforce-
ment measures related with the treatments. Vertical lines indicate 95% confidence intervals (clustering
at postcode level).

30
6 Conclusions
In this paper we built a new source of third-party information that can be used in the
enforcement of rental income taxation, not previously subject to third-party reporting, by
combining information from different administrative registers. We carried out an RCT to
study the effectiveness of this information in tax enforcement. The information allowed
to target enforcement measures at a relevant group of taxpayers, whose rental activity
might otherwise go unnoticed. Notifying taxpayers of the use of this information in tax
enforcement increased reporting of rental income both at the extensive and intensive
margin. The reactions are very large for potential landlords who had not previously
reported any rental activity, increasing the reporting rate compared to the control group
by 50%.
We also provided a comprehensive analysis of tax enforcement spillovers taking into
account informational spillovers within ownership and social networks, as well as regional
spillovers within local rental markets. Analyzing spillovers is crucial because taking into
account direct effects only may understate the effects of enforcement. Typically, the anal-
ysis of spillovers is difficult because networks where spillovers may be expected to arise
may not be well-defined. Rental housing is typically co-owned which creates well-defined
networks between owners. Taking into account spillovers within ownership networks in-
creases the estimated effects of enforcement by 20 %. We also analyze spillovers within
the household, but do not find evidence of spillovers in this context. Further, we ex-
amine regional spillovers of the treatments within local rental markets, but do not find
evidence of regional spillovers. These findings are important in showing that spillovers
may be significant within well-defined networks even if they do not arise in other settings.
Co-ownership may be important in many other contexts and a similar approach can be
applied to analyze enforcement spillovers for example within networks of firm owners.
Our third contribution was to provide an example of analyzing potential real effects
on tax enforcement on market allocations, using administrative data on asset market
transactions. We did not find clear effects on housing market transactions, nor on the
ownership of other assets. Real effects of tax enforcement are understudied in previous
literature, and this type of an analysis hence provides an important avenue for further
research also in other tax enforcement contexts.
As for the policy implications of our findings, the first and primary implication of our
results is to highlight the possibilities of more effective tax enforcement through creat-
ing novel third-party information by combining information from different administrative
registers. A natural conjecture is to systematically go through income items not cur-

31
rently subject to third-party reporting, and assess whether relevant information may be
created by novel combinations of register data. For example, the information we used
relates to ownership and occupancy of apartments, while information on rent levels was
not available.
Our results may also provide several more nuanced lessons for enforcement policy.
First, our results suggest that an effective enforcement strategy might combine the use
of two types of information: third-party information that provides a signal of potential
income, as well as information on previous tax reporting behaviour (in our case no report
in the previous year). Second, the tax authorities might consider publicizing at least the
type of information used in tax enforcement, in contrast to the tendency of many tax
authorities to rather hide this information; c.f. a more general discussion on this point in
Slemrod (2009).Third, information spillovers significantly strengthen the effectiveness of
enforcement measures and affect the associated cost-benefit calculus as well as the optimal
targeting of enforcement.

32
References
Alan, S., Atalay, K., Crossley, T. F., Jeon, S.-H., 2010. New evidence on taxes and
portfolio choice. Journal of Public Economics 94 (11), 813 – 823.
URL http://www.sciencedirect.com/science/article/pii/S0047272710001192

Allingham, M., Sandmo, A., 1972. Income Tax Evasion: A Theoretical Analysis. Journal
of Public Economics 1 (3-4), 323–338.

Alstadsaeter, A., Kopczuk, W., Telle, K., 2019. Social networks and tax avoidance: Evi-
dence from a well-defined norwegian tax shelter. International Tax and Public Finance
forthcoming.

Antinyan, A., Asatryan, Z., 2020. Nudging for tax compliance: A meta-analysis. CESifo
Working Paper 8500.

Ball, M., 2016. Investor costs and returns in the english private rented sector. Working
Papers in Real Estate and Planning 02/16, University of Reading.

Bomare, J., Le Guern Herry, S., 2022. Will we ever be able to track offshore wealth? evi-
dence from the offshore real estate market in the uk. Sciences Po Economics Discussion
Paper 2022-10.

Boning, W., Guyton, J., Hodge, II, R. H., Slemrod, J., Troiano, U., 2018. Heard it through
the grapevine: Direct and network effects of a tax enforcement field experiment. NBER
Working Paper 24305.

Bott, K. M., Cappelen, A. W., Sørensen, E. , Thungodden, B., 2020. You’ve got mail: A
randomized field experiment on tax evasion. Management Science 66, 2801–2819.

Brockmeyer, A., Hernandez, M., Kettle, S., Smith, S., 2018. Casting a wider tax net:
Experimental evidence from Costa Rica. Mimeo, World Bank.

Carrillo, P., Pomeranz, D., Singhal, M., 2017. Dodging the taxman: Firm misreporting
and limits to tax enforcement. American Economic Journal: Applied Economics 9,
144–164.

Crépon, B., Duflo, E., Gurgand, M., Rathelot, R., Zamora, P., 2013. Do labor market
policies have displacement effects? evidence from a clustered randomized experiment.
Quarterly Journal of Economics 128.

33
Drago, F., Mengel, F., Traxler, C., 2015. Compliance behaviour within networks: Evi-
dence from a field experiment. IZA Discussion Paper 9443.

Fossen, F., Rees, R., Rostam-Afschar, D., Steiner, V., 2020. The effects of income tax-
ation on entrepreneurial investment: A puzzle? International Tax Public Finance 27,
1321–1363.

Frimmel, W., Halla, M., Paetzold, J., 2018. The intergenerational causal effect of tax
evasion: evidence from the commuter tax allowance in austria. Journal of the European
Economic Association https://doi.org/10.1093/jeea/jvy033.

Harju, J., Kosonen, T., Slemrod, J., 2020. Missing miles: Evasion responses to car taxes.
Journal of Public Economics 181 (C).
URL https://ideas.repec.org/a/eee/pubeco/v181y2020ics0047272719301707.html

Kleven, H., Knudsen, M. B., Kreiner, C. T., Pedersen, S., Saez, E., 2011. Unwilling or
unable to cheat? evidence from a tax audit experiment in Denmark. Econometrica 79,
561–692.

Kopczuk, W., Marion, J., Muehlegger, E., Slemrod, J., 2015. Does tax-collection invari-
ance hold? evasion and the pass-through of state diesel taxes. American Economic
Journal: Economic Policy 7, 177–208.

Kotakorpi, K., Nurminen, T., Miettinen, T., Metsälampi, S., 2022. Bearing the burden -
implications of tax reporting institutions and image concerns on evasion and incidence.
FIT Working Paper No. 3.

Lediga, C., Riedel, N., Strohmaier, K., 2022. Tax enforcement spillovers - evidence from
business audits in south africa. Working paper, University of Münster.

López-Laborda, J., Vallés-Giménez, J., Zarate-Marco, A., 2023. Fighting vacation rental
tax evasion through warnings to potential evaders. Real Estate Economics , forthcom-
ing.

Meiselman, B. S., 2018. Ghostbusting in detroit: Evidence on nonfilers in a controlled

field experiment. Journal of Public Economics 158, 180–193.

Naritomi, J., September 2019. Consumers as tax auditors. American Economic Review
109 (9), 3031–72.
URL https://www.aeaweb.org/articles?id=10.1257/aer.20160658

34
OECD, 2018. Taxation of household savings. OECD Tax Policy Studies 25.

Pomeranz, D., 2015. No taxation without information: Deterrence and self-enforcement

in the value added tax. American Economic Review 105 (8), 2539–2569.

Rincke, J., Traxler, C., 2011. Enforcement spillovers. Review of Economics and Statistics
93 (4), 1224–1234.

Slemrod, J.and Collins, B., Hoopes, J. L., Reck, D., Sebastiani, M., 2017. Does credit-
card information reporing improve small-business tax compliance? Journal of Public
Economics 149, 1–19.

Slemrod, J., 2007. Cheating ourselves: The economics of tax evasion. Journal of Economic
Perspectives 21, 25–48.

Slemrod, J., Blumenthal, M., Christian, C., 2011. Taxpayer response to an increased
probability of audit: evidence from a controlled experiment in minnesota. Journal of
Public Economics 79 (3), 455–483.

Sommer, K., Sullivan, P., Verbrugge, R., 2013. The equilibrium effect of fundamentals on
houseprices and rents. Journal of Monetary Economics 60, 854–870.

Wenzel, M., Taylor, N., 2004. An experimental evaluation of tax-reporting schedules: a

case of evidence-based tax administration. Journal of Public Economics 88 (12), 2785–
2799.

Whitehead, C., Scanlon, K., Monk, S., Tang, C., Haffner, M., Lunde, J., Lund Andersen,
M., Voigtländer, M., 2016. Understanding the role of private renting: A four-country
case study. Boligokomisk Videncenter.

35
Appendix

A Additional tables and figures

Table A1: Summary statistics for key variables 2014-2016.

1st 99th
Obs Mean Std.Dev. Median percentile percentile

Reported rental income (0/1) 243948 0.763 0.425 1 0 1

Gross rental income 243948 7298 8708 5640 0 41470
Net rental income 243948 4032 5109 2840 -1142 24722.32
Spouse reported rental income (0/1) 148089 0.442 0.497 0 0 1
Spouse’s gross rental income 148089 3005 5411 0 0 24346
Spouse’s net rental income 148089 1626 3018 0 -528 14261
Owned apartemnts 243948 2.097 1.482 2 0 8
Potential rental apartments in 2015 243948 1.215 0.623 1 1 4

Notes: Table shows summary statistics for individuals in our control and treatment blocks and their
spouses for tax years 2014-2016.

36
Figure A1: Reporting of rental income in letter groups - areas not included in the block design

Notes: Figures show regression coefficients on treatment letter by year dummies (ref. no letter and
year 2014). Controls include individual fixed effects, dummies for the number of apartments and addi-
tional enforcement measures related with the treatments. Vertical lines indicate 95% confidence intervals
(clustering at postcode level).

37
 Table A2: Effects of treatment letter in areas outside the block design.

Reporting rental Net rental

Dep. Var. income (0/1) income

Letter 1 0.0240*** 51.09

[0.00548] [41.93]
Letter 2 0.0442*** 68.72
[0.00559] [44.67]
Letter 3 0.0223** 3.805
[0.0107] [75.46]
Letter 4 0.0308*** -57.35
[0.00763] [51.32]

Baseline mean 0.705 2837.4

N 79586 79586

Notes: Table shows DiD estimates for the effects of treatment letters (ref. no letter) using data from
tax years 2014 (before treatment) and 2015 (after treatment). Controls include individual fixed effects,
dummies for the number of apartments owned and additional enforcement measures related to the treat-
ment letters. Standard errors clustered at postcode area level (263 clusters) are in brackets. Significance
is denoted by asterisks: * p < 0.1, ** p < 0.05, *** p < 0.01. Bold font indicates significant difference at
5% level relative to letter 1.

38
 Table A3: Effects of treatment letters on gross and net rental income and deductions.

Gross rental income Deductions Net rental income

Letter 1 34.48 13.8 20.68

[70.31] [52.94] [48.97]
Letter 2 42.4 -44.7 87.11*
[80.99] [58.34] [52.34]
Letter 3 132.8** 6.137 126.7***
[53.00] [38.91] [41.09]
Letter 4 95.75 -0.963 96.71***
[67.65] [56.39] [33.12]

Baseline mean 6888 3109.8 3778.1

N 162632 162632 162632

Notes: Table shows DiD estimates for the effects of treatment letters (ref. no letter) using data from
tax years 2014 (before treatment) and 2015 (after treatment). Controls include individual fixed effects,
dummies for the number of apartments owned and additional enforcement measures related to the treat-
ment letters. Standard errors clustered at postcode area level (263 clusters) are in brackets. Significance
is denoted by asterisks: * p < 0.1, ** p < 0.05, *** p < 0.01. Bold font indicates significant difference at
5% level relative to letter 1.

39
 Table A4: Effects of treatment letters two years after.

Reporting rental Net Rental

Dep. Var. income (0/1) income

Letter 1 0.00503 42.95

[0.00628] [101.3]
Letter 2 0.00676 -152.3*
[0.00549] [87.25]
Letter 3 -0.00478 -171.6**
[0.00454] [66.33]
Letter 4 0.00911** -75.23
[0.00444] [69.49]

Baseline mean 0.757 7549

N 162632 162632

Notes: Table shows DiD estimates for the effects of treatment letters (ref. no letter) using data from tax
years 2014 (before treatment) and 2016 (after treatment). Controls include individual fixed effects, treat-
ment block by year dummies, dummies for the number of apartments owned and additional enforcement
measures related to the treatment letters. Standard errors clustered at postcode area level (263 clusters)
are in brackets. Significance is denoted by asterisks: * p < 0.1, ** p < 0.05, *** p < 0.01. Bold font
indicates significant difference at 5% level relative to letter 1.

40
 Table A5: Effects of treatment letters on investment portfolio.

Apartments Apartments Buy shares Sell shares Buy shares Sell shares
bought sold 0/1 0/1 EUR EUR

Letter 1 -0.00975 0.000835 -0.00385 -0.00370 -437.1 -255.1

[0.00592] [0.00586] [0.00566] [0.00708] [431.8] [544.3]
Letter 2 0.00888 0.0048 -0.00351 -0.00350 311.2 306.3
[0.00600] [0.00724] [0.00571] [0.00712] [493.4] [607.1]
Letter 3 0.00264 0.000978 -0.00485 -0.00301 -457.2 360.2
[0.00525] [0.00595] [0.00458] [0.00572] [346.3] [475.2]
Letter 4 -0.00188 0.00258 -0.000843 -0.00460 -191.1 486.5
[0.00381] [0.00320] [0.00361] [0.00448] [282.7] [382.3]
Baseline
mean 0.0714 0.0742 0.131 0.227 4388.6 7767.5
N 159968 159968 80787 80787 80787 80787

Notes: Columns 1-2 show DiD estimates for the effects of treatment letters (ref. no letter) using data
from tax years 2015 (before treatment) and 2016 (after treatment). Controls include individual fixed
effects, dummies for the number of apartments owned and additional enforcement measures related to
the treatment letters. Columns 3-6 show OLS estimates of letter dummies using data from 2016 on
transactions of stocks in listed companies and shares in mutual funds. Standard errors clustered at
postcode area level (263 clusters) are in brackets. Significance is denoted by asterisks: * p < 0.1, **
p < 0.05, *** p < 0.01. Bold font indicates significant difference at 5% level relative to letter 1.

41
B Rental income form and treatment letters
Letter 1

VERO SKATT

NOTICE

Finnish Tax Administration PO Box 325 FI-

00052 Vero, Finland

Ref.

Check your pre-completed tax return

You have received a pre-completed tax return containing information on your

earnings and deductions in 2015. Review the tax return with care. If the information
is correct and nothing is missing, you need not do anything. If the information is
incorrect, or some pieces of information are missing, you must correct or
supplement the tax return. Information to be supplemented may include rental
income, travel expenses between your home and place of work, or tax credit for
household expenses, for example.

You can supplement and correct the information in the pre-completed tax return in
the Tax return online service (vero.fi/veroilmoitus). The service will remain open
until the tax return deadline indicated on your tax return. If you supplement your tax
return online, you need not use the tax return form or its appendix forms.

If you use a paper form to submit your tax return by regular post, you must also
send the required appendix forms. For example, you must use form 7H to
announce your rental income from a unit in a housing company and form 14A to
get your tax credit for household expenses. The required appendix forms are listed
in the instructions on how to complete the tax return. Don’t forget to enter the
required pieces of information in the correct part of the tax return form in addition to
the appendix forms.

For more information, please visit vero.fi/henkilöasiakkaat > Veroilmoitus

(Individual taxpayers > Tax return) or call the service number specified in your pre-
completed tax return.

Finnish Tax Administration

VATT_1 1.2016 vero.fi

Letter 2

VERO SKATT
NOTICE
Finnish Tax Administration PO Box 325 FI-
00052 Vero, Finland

Ref.

Check your pre-completed tax return

You have received a pre-completed tax return containing information on your earnings and
deductions in 2015. Review the tax return with care. If the information is correct and nothing
is missing, you need not do anything. If the information is incorrect, or some pieces of
information are missing, you must correct or supplement the tax return. Information to be
supplemented may include rental income, travel expenses between your home and place of
work, or tax credit for household expenses, for example.

If you received rental income in 2015, announce the rental income and related expenses.
The most common expenses to be deducted from rental income include maintenance
charges, annual repair costs and real estate tax. If you received rental income from several
sources (such as a unit in a housing company and a summer home), you must separately
announce the income and expenses of each property. If you own a unit in a housing
company with another person, you must only announce the share of rental income and
expenses corresponding to your share of ownership. Calculate the amount of taxable rental
income by deducting the expenses from the rental income.

Example of calculating rental income

The taxpayer owns one unit in a housing company, which they rented out for the entire year
of 2015, with the rent being EUR 1,000 per month. The taxpayer/landlord paid a
maintenance charge of EUR 250 per month. Other expenses related to the renting of the
apartment totalled EUR 1,500. The taxable rental income is the difference between the
rental income and expenses, or 12 x EUR 1,000 - 12 x EUR 250 - EUR 1,500 = EUR 7,500.
Hence, the taxable rental income is EUR 7,500.

You can supplement and correct the information in the pre-completed tax return in the Tax
return online service (vero.fi/veroilmoitus). The service will remain open until the tax return
deadline indicated on your tax return. If you supplement your tax return online, you need not
use the tax return form or its appendix forms.

If you use a paper form to submit your tax return by regular post, you must also send the
required appendix forms. For example, you must use form 7H to announce your rental
income from a unit in a housing company and form 14A to get your tax credit for household
expenses. The required appendix forms are listed in the instructions on how to complete the
tax return. Don’t forget to enter the required pieces of information in the correct part of the
tax return form in addition to the appendix forms.

For more information, please visit vero.fi/henkilöasiakkaat > Veroilmoitus (Individual

taxpayers > Tax return) or call the service number specified in your pre-completed tax
return.

Finnish Tax Administration

VATT_2 1.2016 vero.fi

Letter 3

VERO SKATT NOTICE

Finnish Tax Administration

PO Box 325
FI-00052 Vero, Finland

Ref.

Check your pre-completed tax return

You have received a pre-completed tax return containing information on your

earnings and deductions in 2015. Review the tax return with care. If the information
is correct and nothing is missing, you need not do anything. If the information is
incorrect, or some pieces of information are missing, you must correct or
supplement the tax return. Information to be supplemented may include rental
income, travel expenses between your home and place of work, or tax credit for
household expenses, for example.

The Finnish Tax Administration is boosting the monitoring of tax to be paid

for rental income. Hence, additional information on rental income and related
expenses will be requested more often than before. The additional information
is needed for the Tax Administration to verify that the rental income and expenses
specified in your tax return are correct.

If you received rental income in 2015, you must announce all rental income you
received and related expenses. If necessary, the Tax Administration can request
receipts or other additional information on your rental income and expenses. If we
need additional information on your rental income, you will receive a request to
supplement your tax return after the tax return deadline. Do not enclose your
receipts with your tax return, however; the Tax Administration will separately
request them if necessary.

You can supplement and correct the information in the pre-completed tax return in
the Tax return online service (vero.fi/veroilmoitus). The service will remain open
until the tax return deadline indicated on your tax return. If you supplement your tax
return online, you need not use the tax return form or its appendix forms.

If you use a paper form to submit your tax return by regular post, you must also
send the required appendix forms. For example, you must use form 7H to
announce your rental income from a unit in a housing company and form 14A to
get your tax credit for household expenses. The required appendix forms are listed
in the instructions on how to complete the tax return. Don’t forget to enter the
required pieces of information in the correct part of the tax return form in addition to
the appendix forms.

For more information, please visit vero.fi/henkilöasiakkaat > Veroilmoitus

(Individual taxpayers > Tax return) or call the service number specified in your pre-
completed tax return.

Finnish Tax Administration

VATT_3 1.2016 vero.fi

Letter 4

VERO SKATT NOTICE

Finnish Tax Administration

PO Box 325
FI-00052 Vero, Finland

Ref.

Check your pre-completed tax return

You have received a pre-completed tax return containing information on your earnings and
deductions in 2015. Review the tax return with care. If the information is correct and nothing
is missing, you need not do anything. If the information is incorrect, or some pieces of
information are missing, you must correct or supplement the tax return. Information to be
supplemented may include rental income, travel expenses between your home and place of
work, or tax credit for household expenses, for example.

The Finnish Tax Administration is boosting the monitoring of tax to be paid for rental
income. Hence, additional information on rental income and related expenses will be
requested more often than before.
The additional information is needed for the Tax Administration to verify that the rental
income and expenses specified in your tax return are correct.

The rental income information for 2015 will be compared to information on landlords’
property ownership more comprehensively than before. Special attention will be paid to
tax returns where the rental income information is not consistent with the property ownership
information. According to the information available to the Tax Administration, you own at
least one unit in a housing company, and the apartment may have been rented out in 2015.

If you received rental income in 2015, you must announce all rental income you received
and related expenses. If necessary, the Tax Administration can request receipts or other
additional information on your rental income and expenses. If we need additional information
on your rental income, you will receive a request to supplement your tax return after the tax
return deadline. Do not enclose your receipts with your tax return, however; the Tax
Administration will separately request them if necessary.

You can supplement and correct the information in the pre-completed tax return in the Tax
return online service (vero.fi/veroilmoitus). The service will remain open until the tax return
deadline indicated on your tax return. If you supplement your tax return online, you need not
use the tax return form or its appendix forms.

If you use a paper form to submit your tax return by regular post, you must also send the
required appendix forms. For example, you must use form 7H to announce your rental
income from a unit in a housing company and form 14A to get your tax credit for household
expenses. The required appendix forms are listed in the instructions on how to complete the
tax return. Don’t forget to enter the required pieces of information in the correct part of the
tax return form in addition to the appendix forms.

For more information, please visit vero.fi/henkilöasiakkaat > Veroilmoitus (Individual

taxpayers > Tax return) or call the service number specified in your pre-completed tax
return.

Finnish Tax Administration

VATT_4 1.2016 vero.fi