I. Introduction
Tax amnesties are a common instrument used by numerous governments to increase tax revenue
in the short term.[1] Nevertheless, there is an unsettled
debate as to whether the benefits of the short term increase in tax revenue outweigh
the possible negative
effects of tax amnesties
(Marchese, 2014; Nar, 2015; Dunning et al., 2015), with empirical evidence revealing ambiguous results (Torgler et al., 2003). Previous studies, however, did
not account for the possibility of heterogeneous effects
of tax amnesties on tax compliance. In this paper, we estimate
the differential effect
of a tax amnesty on tax compliance across two dimensions
of heterogeneity: access to public goods and
wealth. We find that the reduction in
tax compliance is stronger
for wealthier taxpayers and those with lower access to public goods, indicating that tax amnesties
have differential effects on tax compliance.
The literature identifies several advantages and disadvantages of tax amnesties. The key advantage is the short
term increase in tax revenue, while other advantages include—but are not limited to—reducing administrative costs, offering a “soft option”
for those who became
tax evaders by mistake, getting some tax evaders
back on compliance, and signaling that the tax problem will be tackled by the government. The disadvantages include
lowering tax compliance
morale, signaling a weak government, increasing
awareness of non-compliance, reducing moral
costs of behaving
dishonestly, and potentially reducing future compliance by creating anticipation of future tax amnesties (Leonard and Zeckhauser, 1987). Whether the advantages of tax amnesties outweigh the disadvantages is an unsettled debate that has warranted
a substantial amount of research to this day.
Previous empirical
evidence regarding the effect of tax amnesties
on tax revenue reveals ambiguous results,
with some studies
indicating increases in tax revenue (Alm and Beck, 1991; Das-Gupta and Mookherjee, 1995), others
indicating negligible or null effects
(Alm and Beck,
1993; Christian et al., 2002), and others indicating reductions in tax collection (Alm et al., 1990). Most of the previous literature relies on the use of time series
aggregate data analysis and the effects of tax amnesties
in developing countries
have seldom been studied
(Torgler et al., 2003). Moreover, these previous studies
did not address the possibility of heterogeneous effects of
tax amnesties on tax compliance, which might explain some differences in the
estimated effects of tax amnesties
on posterior revenue.
Tax amnesties are not the only measure
implemented by governments looking to increase tax revenue. Argentina and
Uruguay have raffled prizes to
compliant taxpayers in order to
boost tax compliance using positive incentives (Chelala and Giarrizzo, 2014). Sending
messages to tax payers
was also explored in recent literature: a field experiment conducted in the municipality of Junín found that messages
emphasizing fines and other legal
consequences of non-payment increased tax compliance in 4 percentage
points (Castro and Scartascini, 2015). Finally, a field experiment in the United
States found that shaming penalties has a large effect
on repayment of smaller debt amounts, but no effect
on larger debt amounts
(Perez-Truglia
and Troiano, 2015). Despite these alternative
policies, the most prominent measure to increase tax revenue
in the short term are tax amnesties (Bergman, 2003).[2]
In
this paper, using a five-year panel from taxpayers
in Argentina, we estimate the
differential effect of a tax amnesty on a property-based tax on tax compliance
across two dimensions of heterogeneity: access to public
goods and wealth. On one hand, we capture
wealth with a proxy variable constructed
as the ratio of property valuation to the property area
(i.e. the valuation of a squared
meter of the property).[3] On the other hand,
we capture differential access to public goods through a variable that
indicates whether a tax payer is
located at a gated community or not, since gated communities do not receive
some public services (e.g. garbage
collection and street cleaning). We measure
tax compliance as the ratio of the yearly amount paid to the total amount of
tax liabilities for the same year
(Payment-ratio). Our main result is that the tax amnesty
induced a greater
decrease in tax compliance across both dimensions of heterogeneity: wealthier taxpayers and those located at gated communities showed a stronger decline in tax
compliance after the amnesty. An increase
of one standard deviation on the value of a squared
meter implies a 1.04
percentage points differential reduction on tax compliance, while taxpayers
located at gated communities show a 4.01 percentage points
differential decrease in tax compliance.
To further argue our results, we group
another set of monthly data
into two periods (pre- and post-amnesty) and estimate
the differential effect
of the tax amnesty on the probability of having at least one unpaid
month of tax across both of our dimensions of heterogeneity.
Results from this exercise yield a similar finding: wealthier taxpayers and those located at gated
communities show a differential increase in the probability of having at least one unpaid month. An increase
in one standard deviation in the value of a squared meter implies an 11.3% stronger
increase in the probability of having at least one unpaid month, while
tax payers located at gated communities show an increase
25% higher in the
probability of having at least one un paid month.
Our results
indicate important heterogeneous effects of tax amnesties across
both of our dimensions of heterogeneity (access
to public goods and wealth). These
findings might explain some ambiguity on previous
estimates of the effects of tax amnesties by
indicating that heterogeneity of taxpayers induces differential responses in tax compliance. Furthermore, our findings also bear an important policy implication: heterogeneous effects on tax compliance
should be considered when performing
the cost benefit analysis of undertaking a tax amnesty, since future tax compliance is sensitive to the
composition of the population regarding wealth and access to public goods.
The
rest of the paper is structured as follows. Section 2 describes the tax amnesty
we study. Section 3 describes the
data. Section 4 presents the empirical strategy and results. Section 5 presents
some robustness checks. Section 6 concludes.
II. The Natural Experiment
The property-based tax in Pilar is computed
by considering the property linear frontage extent (meters), the
property valuation and, mainly, the amount of indirect and direct
services received by the property
from the municipality. The tax name
is Tributo
de Mantenimiento de la Via Publica
y Servicios Generales (property-based tax, PBT
henceforth) and property owners are billed every month.[4]
Even
though the tax frequency payment is monthly, the municipality allows taxpayers to pay a yearly
basis with various discounts according to the months paid in advance.[5] Nevertheless,
less than 10% of taxpayers incur in
this payment method. Also, all
individuals who do not have any debt on the PBT, get a 10% discount on
it. Tax payers have approximately 10 days to pay from the moment they receive the bill to the first due date.[6] In case of payment delays, a cumulative compound monthly interest rate of
3% is applied to the outstanding liabilities. Most of all tax payers pay their duties personally at the municipality or through other private offices (e.g.
banks and private tax collection
agencies) instead of using automatic payment
methods.[7] At the
same time war veterans, religious institutions, volunteer firemen, and social security
recipients (retirees) receiving
the minimum pension
are exempted from the tax.
Pilar
is a municipality in the Province of Buenos Aires and is 35 miles away from Autonomous
City of Buenos Aires (Capital City of Argentina). The municipality is known as
one of the richest in the country since the 30% of its properties belong to
luxurious gated communities (62 in total,
see Figure 1). The average income
in Pilar is much higher than the
average in the country
(AR$13,520 vs AR$9,825)[8] and, according to the last
2010 Argentine census, it is
also the 9th most populated city in the Province of Buenos Aires.[9] Pilar is divided into 11 districts
(Figure 2). Districts are large and tax compliance, public services’ provision, as well as the property-built
area and the property valuation vary across districts
as a reflection of heterogeneous characteristics of the underlying taxpayer population. Over the last years in Pilar the PBT revenue has decreased as in many other municipalities in
Argentina.
Pilar
is not an exception to the downward trend
in tax collection accompanied with measures
to increase revenue.
In 2014, the municipality carried
out a tax amnesty in order
to revert the shortfall in public revenues in
recent years. The tax amnesty initially was
supposed to last three months (July, August
and September) but was extended
during the rest of the year (October, November
and December). The policy consisted in a reduction
of the surcharge fees of PBT debt, depending in the payment plan incurred by the taxpayer. There were three
payment plans.
The first plan
offered a 100%
reduction in penalty
fees if the debt
was paid entirely within
one to three
monthly payments. The second payment
plan was a 50% reduction if the total payment was done within two to
six monthly payments. And, the third
payment plan
was a 20% discount if paid within
seven to twelve monthly
payments. It is worth noting
that the incentives to incur in the first payment
plan are stronger (total penalty forgiven), since the main reason to carry out a tax amnesty in municipalities like Pilar that have income problems, is to increase short-term
revenue.[10]
The tax amnesty was widely
announced to the population through
a publicity campaign in local newspapers and radios.
Also, public spaces were used as platforms to inform the population (through
signs, information campaigns, and pamphlets), and stands were set in strategic points of the city and
business centers where tax payers could
consult their debt and simulate the amount of monthly
fees according to different payment plans.
Any person [11] could participate in the tax amnesty without
the need to get a lawyer or an accountant. The only requirement was to approach the municipality and sign
the agreement.
III. Data
Exploiting
the natural experiment of tax amnesties, we aim to identify its differential
impact on tax compliance across two dimensions of heterogeneity: access to
public goods and wealth. To answer this question, we use a data set on tax compliance
provided by the Municipality of Pilar. The data set has information about all Partidas in the municipality from 2012
to 2016 (about 700 thousand observations).
Each property
in Pilar has a unique
identification number called
Partida. The holder of the Partida
is the property-owner and is not necessarily the one who pays the PBT. As
in most cases, if not all, we assume
that the individual who pays for the tax associated to
a property’s Partida is the one who lives in the property
(tenant, owner, householder, etc.).[12]
Our unit of observation is the
number of Partida associated to
individual tax payers.[13]
We excluded private companies
(e.g. local industry and retailers) and social
organizations from the sample since payment
decision for those taxpayers do
not consider the public services
received by the property or any other issue related to it, so
including those tax payers could bias the results.
The name of the holder of the Partida and the individual who pays the
tax associated to the Partida was not provided for confidentiality reasons. Nevertheless, since the
instrument exploited for identification only varies at the Partida level,
this is not a nuisance for our econometric analysis. For each Partida, we collected administrative information on tax liabilities, tax arrears, property
valuation, property
area and property
built-up area, Partidas’
category, use of the property,
neighborhood identification, and access type to the property (i.e. street, avenue, or highway). Particularly, from 2012 to June 2016[14] we have the yearly amount of tax liability
for each Partida and the total amount
paid corresponding to it. For 2014
and thereafter, for those months where the PBT was paid, we have a variable that identifies the exact date on which
the payment was made. This variable
allows us to identify
which months were paid and which were not for each Partida in this period.
The
Partida’ category refers to the
amount of public services received by the
property. Among public services
there are two kinds: direct ones
(e.g. garbage collection and street cleaning
services) and indirect ones
(e.g. maintenance of public spaces
and recreational and leisure activities).[15] Properties classified as category 1 receive all
indirect and direct
services; properties classified as category 2 receive all indirect
services and only 3 (out of 4) direct services
and; properties classified as category 3 receive all indirect services
and only 2 (out
of 4) direct services. Properties in gated communities in Pilar are considered as category 3 since they are private neighborhoods that do not receive some public services. Finally,
the use of each property is a variable
that identifies the principal or representative use of the property
(e.g. unifamiliar house, sumptuous house, hospital, hotel, etc.).
For
each Partida we calculated the yearly
Payment-ratio as the ratio of the amount paid over total tax liability billed.
This is our main outcome since it represents how tax compliance behavior varies
through the years. This ratio allows us to study how tax amnesties affect tax
compliance behavior across two dimensions of heterogeneity: access to public
goods and wealth. The level of public goods provision is captured by a dummy
variable that indicates whether the individual is located at a gated community
(since gated communities receive less public services) and as a proxy of income
we calculated the valuation per square meter of each Partida (as the ratio between valuation and property area).
Finally, we also analyze how the probability of having at least one unpaid month
is affected by the tax amnesty. In this sense, we collapse our data into two
periods (pre- and post- amnesty) and create a dummy variable equals 1 if the
tax payer had at least one month unpaid, and 0 otherwise.
By
June 2016, there were 149.470 Partidas in the municipality. During 2012 and
2016, the average taxpayer in Pilar pays AR$81 every month (62% of the average
billed amount) and, only the
24% of the taxpayers pay the PBT in
term. The average payment rate is
46.7%, like the average in most municipalities in the Province
of Buenos Aires.
Finally, the average
property valuation
is 298,514 and, the average
property area and property built up
area is 2,285
and 75 squared
meters correspondingly (For descriptive statistics see table1).
IV. Empirical Strategy and Results
Our purpose
is to estimate the differential impact of the tax amnesty
on tax compliance across two dimensions
of heterogeneity: access to public goods (captured
by living in a gated community) and a
proxy for wealth (constructed as the value of
a squared meter of the property). We measure
tax compliance as the ratio of the yearly amount paid to the total amount
of tax liabilities for the same year (Payment-ratio). Formally, we estimate the following regression model:
(1)
where Yit denotes the Payment-ratio of Partida i at time t, αi is
the Partida fixed effect, µt
is a time period fixed effect common to all Partidas, and εit is
the usual error term. The variable Iit has both time and cross-sectional variability.
The time variability is captured by a dummy variable that takes the value of one
from the year 2014 onwards (Amnestyt). The cross-sectional variability is
captured by any of our two dimensions of heterogeneity (Hi): access to public goods
(captured with a dummy variable Gated − Community that takes the value of one if
Partida i is located at a gated community) and property valuation (measured as the
value of a squared meter of property). The parameter of interest is β, which captures
the differential effect of the tax amnesty on tax compliance across each of our
dimensions of heterogeneity.
The difference-in-differences model
assumes that the change in the Payment-ratio in those
taxpayers that receive
more public services
is an unbiased estimate of the counter-factual. While we cannot directly test this assumption, we can test whether time trends in the two groups of taxpayers were the same in the pre-tax amnesty
period in our full sample. If time trends are the same in the pre-tax amnesty
period, then it is likely that they would have been the same in the post- tax amnesty
period in the absence of the tax amnesty. To test
the hypothesis that the pre-tax amnesty time trends are not different in the two groups, we estimate a model like the one in (1) using pre-tax amnesty
data for all the tax payers in the sample,
but we replace the Tax Amnesty interaction with a linear
trend and an interaction term between the linear trend
and the Gated −
Community dummy. Column (1) in
Table 2 shows that
the interaction term
is not significant, thus validating our difference-in-
differences identification strategy. Column (2) reports
the estimates of the same regression
but interacting the time trend with the valuation
of a squared meter. Again, we find
that the interaction term is not statistically significant.
The usual
assumption in econometrics is that observations are independent. In our case, however, there might be potential correlation between
observations for the same Partida.
Thus, in every regression we cluster
standard errors at the Partida level.
Results are robust to clustering standard errors at the neighborhood level or
the zone level.
Table 3 reports Ordinary Least Squares (OLS) estimates of Equation (1).
Column (1) reports the estimate
of the differential effect of the tax amnesty for Partidas
located in gated communities. The estimated coefficient indicates that the tax amnesty
reduced the Payment-
ratio of Partidas in gated
communities by 4.01 percentage points
relative to those Partidas outside
gated communities. The effect is statistically significant at the one percent
level. Column (2) reports
the estimate of the differential effect of the tax amnesty
regarding the property valuation. Since
the Valuation variable
is normalized by its
standard deviation, the
estimated coefficient indicates that an increase of one standard deviation in
the value of a squared meter is
associated to a reduction of 1.03 percentage points in the Payment- ratio. This effect is, too, statistically
significant at the one percent level. Finally,
Column (3) reports the estimates of including interaction terms for both dimensions of heterogeneity. Estimates are lower
in magnitude but
otherwise results remain
virtually unchanged: the tax
amnesty differentially reduced
the Payment-ratio for gated
communities in 3.65
percentage points, while an increase of one standard deviation in the value of a squared meter differentially reduces
the Payment-ratio post-amnesty in 0.7 percentage points. Both coefficients are statistically
significant at the one percent level. These results indicate important heterogeneous effects
of tax amnesties on tax compliance: i) agents who perceive less benefits
from public goods differentially reduce their Payment-ratio following the tax amnesty, and ii) wealthier agents also
differentially decrease their Payment-ratio after the tax amnesty.
To further argue our results,
we exploit our monthly data on whether
each Partida paid its PBT on
each month. We collapse our data into
two periods: pre- and post-amnesty. We then create a dummy variable
that takes the value of one if
Partida i has so far not paid at
least one month of its corresponding PBT and zero otherwise. We then estimate Equation as a linear probability model to estimate the
differential effect of the tax amnesty on the probability of having at least
one unpaid month of PBT across our two dimensions
of heterogeneity. Table 4 reports
OLS estimates of this exercise. Column (1) shows that the probability of having
at least one unpaid month increases in 12.06 percentage points in the post-amnesty period, and the effect is 2.92 percentage points greater for Partidas in
gated communities. This result is statistically significant at the 1 percent level and implies a differential effect almost 25%
greater for Partidas in gated
communities. Column (2) reports the differential effect of the tax amnesty regarding
our proxy for wealth, the property valuation.
The estimate indicates that a one standard deviation increase in the value of a squared meter increases the
probability of having at least one unpaid month in 1.38 percentage points,
and this is statistically significant at the 5% level. Finally, Column
(3) includes interactions for both dimensions of heterogeneity. As before, estimates are lower
in magnitude but otherwise results remain virtually unchanged: the tax amnesty differentially increased the probability of having at least one unpaid month in gated communities by 2.37 percentage points, and an increase in one standard
deviation in the squared meter
value of the property increases the probability in 1.21 percentage
points. The estimates for gated communities and valuation interactions are significant at the one and five
percent level, respectively. These results
provide further evidence
in favor of the differential effect
of tax amnesties on tax compliance across wealth and access to public goods.
V. Robustness Checks
To argue that our findings
are driven by the tax amnesty
of 2014, and serving as another
traditional way to address
the possibility of differential pre-treatment trends, we propose the following specification (see Perez-Truglia, 2015):
(2)
where Amnestyt13 is a
“fake” treatment indicator that occurs just before the actual tax amnesty (i.e.
a dummy variable that equals 1 in the year 2013 and 0 otherwise). If the Payment-ratio changed sharply in 2014, we should expect and
(the false interaction). The results are shown in Table
5. Column (1) reports the estimates considering the first dimension of heterogeneity Gated −
Community and, as expected, β is positive and statistically significant while
. Column
(2) reports the
estimates of the same regression but considering the second dimension of heterogeneity (valuation of a squared
meter), the results are the
same as before.
Our
main results suggest that tax amnesties reduce tax compliance for wealthier
taxpayers and those with low access
to public services.
To further address the validity
of these results, we use a monthly
dataset available
for a subset of the observations and we re-estimate
equation (1). For January 2014 to
December 2015, we have the monthly
amount of tax liability and the total amount
paid corresponding to each month for a subset of the sample. These two variables enable
us to calculate the Payment-ratio as the ratio
of the amount paid over total tax liability billed. As before,
the time variability is captured by a dummy variable that takes the value of
1 from July 2014 onwards (Amnestyt) and, the
cross-sectional variability is captured by each
dimension of heterogeneity (access to public
goods and property
valuation). We perform the same
analysis as in section 4 by estimating
equation (1) using our monthly
data available for a subset of the observations. First, we test the common
trend assumption. As shown in Table 6, the interaction terms
for the time trend with both
dimensions of heterogeneity are not statistically significant. Table 7 reports OLS estimates
of Equation (1) using monthly
data. Column (1) reports the estimate of the monthly
differential effect of the tax amnesty for Partidas
located in gated communities. The tax amnesty differentially reduced
the payment ratio in 0.44 for Partidas
in gated communities. Column (2) reports the differential effect of the tax amnesty
regarding our proxy for wealth,
the valuation of a squared
meter of property. The estimated coefficient indicates that an increase
of one standard deviation in the value of a squared
meter implies a differential decrease
of 0.15 percentage points
in the payment ratio, and the effect is statistically significant at the 10% level.
Finally, Column
(3) includes interactions for both dimension of heterogeneity. As before, estimates are lower in magnitude, but results remain virtually unchanged. Together, these results offer
further evidence of the differential effect of tax amnesties on tax compliance across wealth and access to public goods.
Lastly, to argue that the estimates are driven by the tax amnesty
(and serving as an additional way to address the possibility of differential pre-treatment trends), we generate
a “fake” treatment indicator that occurs just before the actual tax amnesty
(i.e. a dummy variable that equals 1 in the two months prior the amnesty
and 0 otherwise). Table 8
report the estimates of this exercises. As shown in Column (2), the false
interaction is not statistically significant, suggesting that results are in
fact driven by the amnesty.
VI. Conclusion
In
this paper we estimated the
differential effect of tax amnesties on tax compliance across two dimensions of heterogeneity: access to public
goods (captured by being located at a gated community) and wealth
(captured by the value of a squared meter of property). Our main result is that the
tax amnesty induced a differential decrease in tax compliance on wealthier
taxpayers and those who have lower access
to public goods.
We also find that the
tax amnesty induced
a differential increase
in the probability of not paying at least one month for wealthier taxpayers and those located at gated communities.
Our
results might explain some ambiguity on previous evidence of the effects of tax
amnesties by indicating that
heterogeneity of taxpayers
induces differential responses on tax compliance. Our findings also bear an important policy
implication: heterogeneous effects on tax compliance need to be considered when performing the cost-benefit analysis of undertaking a tax amnesty,
since posterior tax compliance is sensitive to the composition of the population regarding
wealth and access to public goods.
