Who Pays for the Payroll Tax? Quasi

Who Pays for the Payroll Tax? QuasiExperimental Evidence on the Incidence of
the Payroll Tax
Evelina Gavrilova, Floris Zoutman, Arnt-Ove Hopland,
Jarle Møen
Who Pays for the Payroll Tax? Quasi-Experimental Evidence on
the Incidence of the Payroll Tax ∗
Evelina Gavrilova, Floris Zoutman, Arnt-Ove Hopland, Jarle Møen,
NHH Norwegian School of Economics and
Norwegian Center for Taxation
Preliminary and Incomplete
August 29, 2015
Abstract
We exploit quasi-experimental variation in changes of the payroll tax rates in zones within Norway,
as well as changes in the borders of these zones, to estimate the incidence of the payroll tax. Our
data is a panel of the universe of manufacturing firms in Norway between 1996-2012. Since, the
data includes the number of hours worked at each plant we can construct the gross wage rate per
hour, allowing us to fully separate incidence from employer and employee behavioral responses.
We find that a 1 percent increase in the payroll tax reduces the gross wage rate by 0.66 percent
indicating that employees pay for 66 percent of the payroll tax on average. In large plants and in
plants with high wage rates the incidence is fully borne by the employees, while in smaller plants
with lower wages the incidence is on the employers. We further use the changes in the payroll tax in
combination with our incidence equation to instrument for changes in the total wage cost per hour.
We find evidence that firms increase their labor demand on the intensive margin. The number of
employees is unaffected.
1
Introduction
Who pays for the payroll tax? Theory suggests that the majority of the long-run economic incidence of
the payroll tax resides with employees even if the statutory incidence falls on the employer. However, by
using a quasi-experimental variation, Saez et al. (2012) shows that the economic incidence of employerpaid payroll taxes fully resides with the employer. This suggests that the payroll tax can be considered
as an additional tax on firm capital. In this paper we use Norwegian firm level data to verify whether
the incidence of employer-paid payroll taxation indeed resides with the firm, and study whether reforms
in the payroll tax elicit behavioral responses at the firm level.
In a policy dating back to 1975 Norway was split into three regions with different payroll tax rates.
The aim of the policy is to stimulate employment in the Norwegian periphery. Over the years, the
policy has seen a multitude of reforms. On one hand, the number payroll-tax zones has expanded to
seven, with a tax rate of 14.1 percent in cities in the south of Norway, and a zero rate in the sparsely
populated areas in the north. On the other hand, the borders of the various zones have shifted, and the
rates within zones have changed. Additionally, EU regulations have forced the Norwegian government
to make adjustments in the tax rates. Many of these reforms create variation in the payroll tax rate
that is plausibly exogenous to local labor market conditions. Hence, the Norwegian policy of payroll
tax differentiation offers quasi-experimental circumstances to estimate the incidence of, and behavioral
responses to, payroll taxation over a long time-frame.
A main concern in empirical tax studies is that it is often difficult to separate incidence from
behavioral responses in the data. Conceptually, incidence at the firm level arises if the wage cost per
∗
We would like to thank Research Council of Norway Grant Number 239120. All opinions expressed herein belong to
the authors.
1
hour of labor employed increases when the payroll tax increases. The behavioral response of firms to
this incidence is a decrease in the amount of effective units of labor demanded. However, we generally
only observe the total wage sum which is the product of the wage rate and the number of labor hours.
Therefore, it is usually impossible to separate between tax incidence and behavioral effects.1 Our data
is a panel with the universe of manufacturing plants in Norway between 1996 and 2012, taken from
administrative data sources. A plant record contains information on gross wage, payroll tax paid,
overall wage cost payments and overall hours worked for each plant. The main contribution of our
study is therefore that we observe hours worked at each plant, such that we can construct the wage
rate per hour worked. This allows us to estimate incidence independently from behavioral responses.
To study the incidence of the payroll tax we run a fixed-effect regression where the gross wage rate
per hour is the dependent variable, and the the net-of-payroll tax the main independent variable. We
include sector-time and plant fixed effects to filter out any unobserved heterogeneity that may bias
our results. Since the wage rate, rather than the overall wages paid, is the dependent variable this
regression allows us to estimate the incidence of the payroll tax in isolation of behavioral effects.
Furthermore, to test for behavioral effects of the payroll tax on firms we run an instrumental
variable (IV) regression between several plant-level outcome variables and the total wage cost per hour
instrumented by the payroll tax. We include all fixed effects as well as several control variables to
filter out observed and unobserved heterogeneity. This IV-approach has an important advantage over
a reduced-form OLS approach where the payroll tax is simply used as the independent variable, since
the second-stage coefficient can be interpreted as the effect of the wage cost per hour on firm behavior.
As such, the elasticity can also be useful when forecasting the effect of other policy measures that effect
the wage cost per hour, such as changes in the income-tax rate. Moreover, the assumption that the
payroll tax only affects firm behavior through its effect on the wage cost per hour is likely satisfied,
since there are no other plausible channels through which the payroll taxation could affect decisions
made by firms.
Our preliminary results indicate the incidence of payroll taxation is shared between employers and
employees. We use two types of payroll tax rates: the effective tax rate that is derived from the amount
a plant paid in payroll tax and the statutory tax rate coded with the municipality- and sector-specific
tax rates. Conceptually, the first measure of the payroll tax may be endogenous, since firms may
react to changes in the tax rate by hiring employees that are payroll-tax exempt. However, the second
measure may be too coarse, since statutory tax rates differ not only by municipality and sector, but also
by firm size (and a fair amount of discretion). Empirically, the two measures are strongly correlated,
but the correlation between the change in each of the variables is weak. This indicates that a change in
the municipal payroll tax rate only has a small effect on the actual payroll tax rate paid. Nonetheless,
both payroll tax measures reveal that employees receive about 66-77 percent of the incidence in our
central estimate.
Using effective tax rates, we find that incidence strongly depends on plant size and wage rate. In
particular, for small plants where the average wage rate lies below the median the incidence lies with
the employer, while for big plants with high wages the incidence falls completely on the employee. In
addition, small plants with high wages and big plants with low wages shift up to 80 percent of the
incidence on the employees. In the second part of our analysis we look at the viability of the payroll
tax as an instrumental variable and find that, given that most of the incidence falls on employees, it is
a weak instrument to estimate the elasticity of labor demand. Thus, reduced form results of a decrease
in the employees is rather a supply side response than a demand side one.
Our results on incidence strongly contrast the results by Saez et al. (2012), since a large part of the
incidence is shifted onto employees in our study. Saez et al. (2012) show that there is no behavioral
response to the payroll tax by using employee data and a reform in Greece where the payroll tax rate
differs by the date of entry to the labor market. Similarly, using a sample of low-income workers in
France and a simultaneous income - and payroll-tax reform, Lehmann et al. (2013) find that workers
adjust their taxable income in response to a change in the income tax, but show no behavioral response
1
Saez et al. (2012) circumvents this issue, since in their study there is no behavioral response at the employee level.
In this special case, it is possible to also calculate incidence using only overall wage payments. However, this result does
not generalize to a setting where behavioral responses are present.
2
to the payroll tax. The authors conjecture, in correspondence with Saez et al. (2012), that low-income
workers in France are sheltered from a change in the payroll tax due to wage rigidity in gross wages.
Our results partially confirm this conjecture by showing that incidence on low-income workers is indeed
a lot lower than for high income workers. However, we find that even low-income workers do face a
significant decrease in their net-wage rate when the payroll tax goes up. As such, our results do not
fully solve the puzzle of why low-income workers in France respond differently to payroll, than to
income taxation.
We are not the first to study the incidence of payroll taxation on firms.2 In Norway Carlsen
and Johansen (2005), Dyrstad (1992), Dyrstad and Johansen (2000) and Johansen (2002) include
the payroll tax rate as a control variable when explaining the wage cost per workers. Their results
indicate the burden of the payroll tax is split about equally between employer and employee. Using
a quasi-experimental reform in Sweden, Bennmarker et al. (2009) find that payroll tax cut is mostly
paid for by employers. However, by evaluating a Finnish reform Korkeamäki and Uusitalo (2009) find
that the majority of the incidence resides with the employee respectively. This finding is echoed by
earlier studies in the US (e.g. Gruber, 1994, Anderson and Meyer, 1997, Murphy, 2007) which show
that several taxes and subsidies levied at the employer level are mostly, but not entirely, shifted onto
employees. Gruber (1997) studies a payroll tax reform in Chile and finds that there the payroll tax
is entirely shifted onto employees. However, this may be the result of a simultaneous reform in the
income tax that was meant to offset the change in revenue induced by the payroll tax reform.
In each of these studies the dependent variable used is the wage per worker, instead of the wage
rate per hour. Therefor, the results may be biased by behavioral responses of the employer and/or the
employee with respect to the payroll tax. As far as we know, the only paper in the literature that does
use the wage rate per hour as the dependent variable is Johansen and Klette (1997). However, their
study was never concluded. Their preliminary results are in line with our preliminary results.
Several previous studies have considered the role of payroll taxes in determining labor demand and
employment. Similar to our results, Korkeamäki and Uusitalo (2009) and Bohm and Lind (1993) find
no effects on employment following payroll tax reform, consistent with the incidence being borne mostly
by the employee. In contrast, Bennmarker et al. (2009) look at employment effects after a payroll tax
cut in Sweden. Initially they do not find employment effects, when evaluating the incidence. However,
once they account for firm numbers, they find positive effects on firm entry and generally positive effect
on employment, while the incidence becomes fully borne by the employer. We are not aware of studies
that have evaluated other margins of behavioral responses of firms.
The remainder of the article is organized as follows. The next section provides a background on
the Norwegian payroll tax and a description of the data that we use. The third section provides a
theoretical framework and the identification strategy that follows from it. Section four presents results
on the incidence, an extension on heterogeneity and behavioral effects. The fifth section concludes.
2
Background
2.1
Norwegian Payroll Tax
Our empirical approach will exploit the reform-induced variation in payroll tax rates in Norway to
estimate the economic incidence of payroll taxation. The payroll tax was introduced in Norway in
1967 in order to finance the social security system. Starting in 1975, the tax rates were differentiated
across regions based on which municipality the employee lived in, with the intention of stimulating
employment in rural areas. In the early years, the country was divided into three zones, where the
lowest rate applied to the most rural areas. The policy aim of the differentiated payroll tax is to
stimulate employment in the Norwegian periphery, and the regionally differentiated payroll tax has
been reformed in several ways over the years. The borders of the various zones have shifted, the degree
of differentiation has increased through introduction of extra tax zones and the rates within zones
have changed. EU regulations regarding the reduced rates as subsidies have forced the Norwegian
government to make further adjustments.
2
See section 3 in Bennmarker et al. (2009) for a more detailed overview of all papers in the literature.
3
The differentiated rates has been a popular policy tool, as it is considered to be precise and involves
low administrative costs. The difference in rates increased steadily in time. In 1990 there were five
zones with rates of 2.2 to 16.7 percent. This development continued throughout the 90s, and in 2003
the five zones had rates of 14.1 (zone 1), 10.6 (zone 2), 6.4 (zone 3), 5.1 (zone 4), and 0 percent (zone
5).
Over the years, the system has been extended, and at the time of writing there are seven zones
(1, 1a, 2, 3, 4, 4a, 5), as presented in figure 1. In figure 2 we show the tax rates for these zones. The
payroll tax rate is highest in cities in the south of Norway (14.1%), and zero in the sparsely populated
areas of North Troms and Finnmark. The tax generally decreases from the south to the north, with
the exception being that the big cities face a higher tax rates than their immediate surroundings.
The EFTA Surveillance Authority (ESA) defines the differentiated tax rates as subsidies, which in
general are not allowed according to European Economic Area (EEA) regulations. Consequently, the
Norwegian system for payroll taxes had to be changed in 2004 in order to adapt to these regulations.
Zone 2 were given the full rate immediately, while the rates in zones 3 and 4 were increased gradually
(until 2007). Norway was temporarily allowed to keep the zero rate that applied for the most rural
regions (zone 5). Further, industries exempt from the EEA agreement (agriculture and fisheries) were
also allowed to keep the differentiated rates. Moreover, Norway was allowed to keep the differentiated
rates as long as the total support to a firm does not exceed “trivial support”, defined as 100,000 Euro
over a three year period. For firms where the tax subsidy constituted less than 33,333 Euro per year,
the new system therefore involved no real changes. Firms where the tax subsidy in the old system
exceeded 33,333 Euro per year, from 2004 paid the low rate until reaching the threshold and the full
rate onwards. About 55 percent of the man years and 90 percent of the firms were fully compensated.
This has led to a “smoothing” of the tax rates as can be seen when comparing figures 2 and 3.
Figure 1: Payroll Tax Zones in Norway
Source: Statistics Norway
2.2
Data Description
The data used in this study come from the Annual Manufacturing Census of Statistics Norway. This
panel covers all firms and plants in Norway in the manufacturing sector, from 1996 to 2012. A firm is
said to have a plant in a municipality, if it has at least one production facility within a municipality.
4
Figure 2: Statutory Payroll Tax Rates per Zone
Figure 3: Effective Payroll Tax Rates per Zone
5
Table 1: Summary Statistics
Mean
SD
Min
Max
Total Gross Wages
Employees
Hours
Wage Rate
Wage Cost
Effective Payroll Tax
Statutory Payroll Tax
7429.25
21.63
34783.72
0.20
0.23
0.13
0.13
26920.36
66.01
120084.24
1.50
1.73
0.04
0.03
0.10
1
0.50
0.00
0.00
0.00
0.00
1456394.00
3753
26230738
512.46
593.36
1.00
0.14
Plants:
28045
Obs.:
203 626
Notes: Total gross wages, the wage rate and the wage costs are denominated
in 1000 Norwegian Krones.
Moreover, if a production facility moves from to a new municipality it is considered a new plant.
For the purposes of our analysis we consider plants with multiple employees, in order to capture the
elasticity of labor demand. In our sample the biggest sectors are newspaper publishing, shipyards,
diverse machinery, furniture, metal treatment, mining and food production. The data that we extract
is a panel that contains in total 28 thousand plants.
The main variables are described in table 1. The Total Gross Wage is the sum of wages for all
employees in the plant, as is the variable Hours. The relatively high standard deviation for the first
three variables hints that the plants in the data are very diverse. The Wage Rate is defined as the ratio
between gross wages and the total hours worked at the plant. Thus, the variable presents the average
gross wage rate per hour per plant with mean value NOK200. The Wage Cost is the product of the
wage rate and the Effective Payroll Tax, representing total average wage cost rate per hour per plant.
Given that the data includes the payroll tax expenses actually incurred by the plant, we calculate
the Effective Payroll Tax by dividing total payroll expenses by the gross wage rate. Data on Statutory
Payroll Tax Rates has been collected by Statistics Norway on the basis of administrative data sources.
The mean and median of the two variables coincide and they have a correlation coefficient of .73.
However, the correlation coefficient between the difference of the variables is close to zero. This
indicates that effective payroll taxes are not very strongly affected by changes in the statutory rate.
This could be because firms respond to changes in the payroll tax by hiring or firing payroll-tax-exempt
employees, but it likely also has to do with exemptions given to certain sectors and firms when the
payroll tax system was reformed.
In figures 2 and 3 we observe the statutory and effective tax rates. Most notably, the main difference
between the figures is in the period 2004-2006 when an increase in the tax rate in zones 2, 3 and 4 was
offset by the support scheme outlined in the previous subsection.
3
3.1
Theory
Model
To illustrate our empirical analysis we first construct a simple theoretical model that allows us to
consider the theoretical impact of a change in the payroll tax rate on the labor market. We assume
a neo-classical labor market where both plants and employees are price takers, and as such, treat the
wage rate as given. We consider both cases where wages are flexible, and where they are fixed. In
correspondence with the Norwegian institutional setting we assume all payroll taxes are paid for by
the employer. We first describe labor demand by plants, and supply by workers in a setting with
payroll taxation. Afterwards, we turn to the market equilibrium and show the theoretical predictions
for incidence and behavioral responses. Finally, we discuss our empirical methodology.
6
logw
1
D
logw0
LD
logl1
LD
logl0
loglj
Figure 4: Plant Labor Demand
3.1.1
Plants
In our theoretical analysis we assume that each plant maximizes its own profits. The production
function of each plant j is assumed to depend on capital and labor employed by the plant:
yj = F (kj , lj ) ,
where F (.) is a twice differentiable function which is assumed to be increasing and concave in both
arguments. Plants are assumed to make zero profit in equilibrium. We assume labor is homogeneous,
and an effective unit of labor is priced at overall wage cost rate wP , which the plant takes as a given,
where wP includes all cost of hiring a unit of labor. It follows that labor demand of plant j is implicitly
given by equating the wage cost rate to the marginal product of labor:
wP = Fl (kj , lj ) ,
(1)
The total labor cost paid for a unit of effective labor supply consists of two parts: the gross wage rate
wj received by workers and the payroll tax τj paid over the gross wage by plant j. The relationship
between labor cost and the gross wage rate is therefore given by:
wP = (1 + τj ) wj .
Inserting this equation in the labor demand equation we arrive at:
wj =
Fl (kj , lj )
.
1 + τj
To arrive at an estimatable we assume labor-demand takes the following log-linear form:
log ljD = f (kj ) − εD (log wj − log (1 + τj )) .
where εD is the labor demand elasticity, and f (kj ) some function of capital. Labor demand is
drawn in figure 4, together with equilibrium wage rate w0 . All quantities are drawn in logs such that
labor demand becomes a straight line under our assumptions with slope coefficient −1/εD . For wage
rate w0 equilibrium plant labor demand is given by l0j .
Now consider a payroll tax reform, where the payroll tax is increased. We assume for simplicity
that plant capital is unaffected by the reform, such that we can write f (kj ) = αD , where αD is a
constant. This assumption implies that a payroll tax rise increases the relative price of labor with
respect to capital. Considering this assumption as a first approximation, it can be easily seen from the
7
logW
LS
logw0
logw1
LD
LD
logL1 logL0
logLj
Figure 5: Market Equilibrium
labor demand equation that an increase in the payroll tax rate, holding constant plant capital, shifts
labor demand to the left. If market wages are rigid the new labor demand is given by the intersection
between the labor demand line, and the old wage rate. However, if wages are flexible the equilibrium
wage rate will also be affected.
3.1.2
Labor Market Equilibrium
We assume that all labor is created equal. That is, labor supply (demand) of different workers (plants)
are perfect substitutes. Aggregate labor demand is therefore given by the sum of plant-specific labor
demand, and labor supply is given by the sum of plant labor supply:
M
X
LD =
ljD ,
j=1
N
X
LS =
liS .
i=1
The equilibrium labor demand and supply is depicted in figure 5. Initially equilibrium employment is
given by L0 and the equilibrium wage rate by w0 . A payroll tax reform where the payroll-tax increases
will shift labor demand down. Consider first the case were wages are flexible. If wages are fully flexible
the reform will result in a new equilibrium with wage rate w1 and equilibrium employment L1 . The
change in the wage rate will also have its repercussions in individual labor supply and plant-specific
labor demand by the fact that the equilibrium wage shifts from w0 to w1 . How much the wage rate
changes crucially depends on the relative size of the labor demand, and the labor supply elasticity.
Generally, the larger the labor demand elasticity relative to the labor supply elasticity, the more able
are plants to shift the burden of the payroll tax onto employees by lowering their wage rate.
In the case were wages are fully rigid, statutory and economic incidence of the payroll tax will
always coincide as can be seen in figure 5. In particular, when the wage rate remains at w0 firms are
by definition unable to shift any of the burden on to employees. Instead an increase in the payroll tax
will lead to excess supply of workers, and hence, involuntary unemployment.
8
3.2
3.2.1
Empirical Strategy
Estimating the incidence
The discussion above suggests a simple framework in order to estimate the incidence of the payroll
tax. In particular, incidence is a measure of how well firms are able to shift the burden of taxation
onto their employees. This can be captured in the following regression equation:
log wjkst = β log (1 + τjkst ) + αst + ηj + γXkt + εjkst ,
(2)
where s denotes the sector in which the plant is active, and k the municipality in which it resides. β is
the incidence-parameter, αst are time-sector fixed effects, ηj are plant-fixed effects, and Xkt are a set
of municipality control variables.
If β = 0 this indicates that the gross wage rate is unaffected by changes in the payroll tax. Hence,
the burden of taxation is fully on the employees. This could be the case if wages are flexible and labor
demand is inelastic, or if wages are fully rigid. On the other hand, if β = −1 this indicates that the
full burden of the payroll tax resides with employees. In our model, this could be the result if wages
are flexible, and labor demand is infinitely elastic or labor supply is inelastic.
The sector-time fixed effects, αst are meant to capture unobserved time-variant variation in the
economic conditions for each sector. These could for example include trends in preferenes for goods
produced by each sector. ηj is meant to capture unobserved heterogeneity between plants. Since plants
in our sample do not change municipality these fixed effects also capture unobserved heterogeneity
between municipalities. Finally, control variables Xkt are meant to capture time-variant heterogeneity
between municipalities such as local changes in unemployment. Note that we cannot fully saturate our
model by including municipality-time fixed effects, since the main variation in the payroll tax occurs
at the municipal level.
Moreover, suppose that the payroll tax rate is exogenous to firm behavior after controlling for the
fixed effects and control variables described above. In that case, including time-variant plant-level
control variables could bias our estimate for the incidence parameter β. To see this assume that the
model above fully describes plant behavior. In addition, assume that in regression equation (2) we
could control for equilibrium labor demand. From figure 5 it is clear that part of the effect of payroll
taxation comes through its effect on labor demand. As such, the coefficient on firm labor demand will
absorb part of the effect of payroll taxation on the gross wage rate, whereas the coefficient β will now
only measure the direct effect of the payroll tax on the gross wage rate. However, we are interested
in the overall effect of payroll taxation on the gross wage rate. Hence to measure this, we should not
control for any plant-level variable that may conceivably be affected by the payroll tax. Therefore, we
choose not to include plant-level control variables in our regression.
In equation (2) our measure for the wage rate per unit of work is the wage rate per hour. As
discussed above, in all previous studies the dependent variable is the wage rate per worker. However,
an increase in the payroll tax rate might induce a worker to work less, if the payroll tax affects his net
wage rate. Similarly, an increase in the wage cost per hour may induce the firm to hire the worker for
fewer hours. Both behavioral responses bias the incidence parameter β downwards. Hence, using this
approach we overestimate the incidence of payroll taxation on workers. As such, we get a much better
estimate of the pure incidence of the payroll tax when we use the wage rate per hour instead of the
wage rate per worker.3
3.2.2
Estimating Behavioral Responses of Plants
With our empirical framework we are able to determine the labor demand elasticity provided firms
bear some of the incidence of payroll taxation. To see this, note that labor demand depends on the
wage cost rate, wF as described in the theoretical model above. As such, if the payroll tax affects the
wage cost rate, an exogenous shock in the payroll tax rate allows us to determine the effect of a change
3
Even the wage rate per hour may also be affected by behavioral responses, since workers may decide to exert less
effort during each hour worked. However, here the behavioral responses of the firm run in the opposite direction, since
an increase in the wage cost rate induces them to force employees to work harder.
9
in wages on labor demand. To find the exact size of the effect we adopt a two-step approach. First,
we identify the effect of payroll-taxation on the wage cost rate. Second, using the payroll-tax as an
instrument, we estimate the effect of a change in the wage rate on labor demand. Mathematically, this
leads us to estimate the following set of equations:
P
P
log wjkst
= β P log (1 + τjkst ) + αst
+ ηjP + γ P Xkt + εPjkst ,
D
log ljkst
D
= −ε log ŵjkst +
I
αst
+
ηjI
I
+ γ Xkt +
εIjkst ,
(3)
(4)
where ŵjkst is the predicted outcome of the wage-rate from first-stage equation (3), and each of the other
variables was already described above. Note that since wP = (1 + τj ) wj , it follows that β P = β + 1.
D
In the second equation, εD is the labor-demand elasticity. We measure ljkst
both in the number of
workers, and in the number of hours at each plant.
This approach offers important advantages over a reduced-form equation where the payroll tax
rate is directly regressed on labor demand, as is common in the literature (see e.g.Bennmarker et al.
(2009)). To see this note that in the reduced-form approach if labor demand does not respond to the
payroll tax this may be indicative of one of two facts. First, the labor demand elasticity may be zero.
Second, it may be that all incidence of the payroll tax resides with employees, for example because of
infinite labor demand elasticity. That is, these two extreme cases cannot be separated by using the
reduced-form.
However, in our approach if the employer does not bear the incidence of the payroll tax this will
appear as a weak-instrument problem in the first stage. On the other hand the case where the labor
demand elasticity is zero, but incidence is non-zero, will result in a non-significant estimate for εD .
4
4.1
Results
Incidence
In table 2 we present the main empirical results from the estimation of equation (2). The independent
variable of interest, the payroll tax rate, is represented by two measures. First, we observe the effective
payroll tax paid by the plant, which allows us to recover the individual tax rate that the plant faces.
Second, we observe the payroll tax rate at the municipality level where the plant is located. As
discussed before, the first measure may be endogenous to firm behavior, while the second does not
capture all variation in actual statutory rates. In order to capture labor demand effects we concentrate
on results from the use of the first measure.
The dependent variable in columns 1 and 6 in table 2 is the logged wage rate per hour. We observe
that depending on the measure, as the payroll tax rate increases by 1 percent the wage rate declines
by .66 to .77 percent, which effectively puts most of the burden on the tax with the employee.
From column 2 to 5 we explore behavioral effects of the payroll tax at the plant level. We observe
that an increase in the tax has a weak effect on the total gross wages, and similarly, no effect on the
total amount of hours worked at the plant. Yet, there is also evidence of a slight increase in the hours
per employee and a decrease in the number of employees per plant.
4.2
Extension
In table 3 we separate the sample into several parts in order to uncover heterogeneous effects in the
sharing of the burden of the payroll tax. The subsample in the first column is characterized by being
composed of firms with a small number of employees and wages below the median wage, by different
zones. We conjecture that the wage rate could be also interpreted as a proxy for whether a worker is
high skilled or low-skilled. The last two lines in the table present the mean values of the wage rate and
number of employees for the sample. We observe a coefficient that is close to 0, meaning that in that
in such plants employers bear the burden of the payroll tax.
In columns 2 and 3 we observe a coefficient that is higher than the one in table 2. In plants with a
few high-skilled employees or many low-skilled employees, the incidence lies mostly with the employees.
In column 4 we observe that large plants with high-skilled employees fully shift the incidence onto the
10
Table 2: The Effect of the Payroll Tax
Effective Payroll Tax
(1)
(2)
(4)
(5)
(6)
Total Wage
(3)
Hours per
Employee
Wage Rate
Employees
Hours
Wage Rate
-0.666***
(0.154)
-0.481*
(0.270)
0.417**
(0.206)
-0.233**
(0.111)
0.185
(0.257)
Statutory Payroll Tax
Observations
R-squared
-0.772***
(0.176)
203,515
0.528
203,515
0.902
203,515
0.505
203,515
0.927
203,515
0.893
203,515
0.528
Notes: The dependent variable in each regression is noted at the top of the column. All variables are in logs.
Effective Payroll Tax is transformed as log(1 + P ayroll T ax Rate). Regressions include plant × firm and year
× sector fixed effects. Standard errors are clustered at the sector-payroll tax zone level. Asterisks denote: ∗ ∗ ∗p <
0.01, ∗ ∗ p < 0.05, ∗p < 0.1.
employee. Likely, each of these 4 types of plants experiences a different type of hiring process and the
lack of burden on employees from the first group of plant might point to a peculiar hiring friction.
Table 3: Heterogeneous Effects of the Payroll Tax on the Wage Rate
(1)
(2)
(3)
Wage Rate
(4)
Effective Payroll Tax
-0.467*
(0.270)
-0.792***
(0.187)
-0.737**
(0.301)
-1.063***
(0.221)
Observations
R-squared
Small firm
Low wage
92,725
0.467
x
x
98,657
0.423
x
-
3,283
0.738
x
8,537
0.651
-
Wage Rate
Employees
0.150
9.433
0.243
13.12
0.162
151.2
0.350
201.3
Notes: The dependent variable in all regressions is the wage rate. All
variables are in logs. Effective Payroll Tax is transformed as log(1 +
P ayroll T ax Rate). Regressions include plant × firm and year × sector fixed effects. The row “Small firm” denotes with an “x” whenever a plant
employs a small number of employees. “Low wage” denotes with an “x” whenever a plant has a wage rate below the median wage rate for the tax zone.
The last two rows summarize the means of the underlying sample of each
column. Standard errors are clustered at the sector-payroll tax zone level.
Asterisks denote: ∗ ∗ ∗p < 0.01, ∗ ∗ p < 0.05, ∗p < 0.1.
4.3
Behavioral Effects
In table 4 we are looking for behavioral responses of plants from an increase in the wage cost per hour
induced by an increase in the payroll tax. In the first column we present results from the first stage,
in the second and third we present results from the reduced form on the behavioural responses of the
firm, reproduced from table 2. In the last two columns we present results from the second stage.
Consistent with the results in the first subsection we see in column 1 that 33 percent of the wage
cost is absorbed by the employer. In the remaining columns we observe that the payroll tax does not
11
affect the hours worked at the plant and it affects weakly the number of employees, indicating that the
labor demand elasticity is around 0 in Norway.
Table 4: Results from IV Regressions on Firm Decisions
(1)
First Stage
Wage Cost
Effective Payroll Tax
0.334**
(0.154)
(2)
(3)
Reduced Form
Employees Hours
-0.233**
(0.111)
203,515
0.534
203,515
0.927
(5)
Second Stage
Employees Hours
0.185
(0.257)
Wage Cost
Observations
R-squared
F-statistic
(4)
203,515
0.893
-0.696*
(0.405)
0.552
(0.860)
203,515
0.898
4.716
203,515
0.850
4.716
Notes: The first column present the results from the first stage of the IV regression, with
OLS standard errors and F-statistic. The other columns present results from the second stage,
where the F-statistic from the bottom is the Kleinbergen-Paap F-statistic with standard errors
clustered at the sector-tax zone level. The dependent variable in each regression is noted at
the top of the column. Effective Payroll Tax is transformed as log(1 + P ayroll T ax Rate).
Regressions include plant × firm and sector × year fixed effects. All variables are in logs.
Asterisks denote: ∗ ∗ ∗p < 0.01, ∗ ∗ p < 0.05, ∗p < 0.1.
4.4
Robustness Checks
In table 5 we show several robustness checks to the main result on the incidence of the payroll tax.
Column 1 replicates this result, while column 2 shows the same point estimate with an OLS standard
error. This reveals that there is significant positive correlation between plants within the sector-zone
clusters.
In column 3 we weight the estimate by the number of employees at the plant in order to make
a stab at an average effect at the employee level. We find that the incidence is fully borne by the
employee, in contrast to the finding of ??. In column 4 we present the estimates in first differences,
showing a similar estimate of full shifting of the burden to the employees.
5
Conclusion
In this paper we analyze Norwegian manufacturing plants and we show that the incidence of the payroll
tax is divided between the employers and the employees. We exploit variation across time and regions
and we measure the incidence in two ways. First, we use the effective payroll tax that each plant has
faced, as evident from the payroll tax costs incurred. We find that employees pay for 66 percent of
the tax. Second, we use the statutory tax rate defined at the level of the municipality in which the
plant is located. We find that employees pay for 77 percent of the payroll tax. Additionally, when
considering heterogeneous effects we find that in small plants with low-skilled employees the burden
seems to be borne by the employers, while in small plant with high-skilled employees or in big plants,
the burden is almost entirely borne by the employees. We find no responses at the level of the firm,
which is unsurprising given that most of the incidence is borne by the employees.
It is very surprising that there exist so few recent studies of the payroll tax, given its importance as
a collector of revenue. It is of interest to obtain knowledge about whether a policy of high payroll taxes
is a sensible response to globalization, or if the absence of a payroll tax like in Denmark is the better
response. Knowledge about how the payroll tax affects firm behavior is also vital because it concerns
the efficient use of capital. To make a comparison, in Sweden, the payroll tax is 31.4 percent. Even
12
Table 5: Robustness Checks
(1)
(2)
(3)
Wage Rate
Effective Payroll Tax
-0.666***
(0.154)
-0.666***
(0.053)
(4)
(5)
D.Wage Rate
Wage Rate
-1.055***
(0.148)
First Difference in Eff. P. Tax
-0.984***
(0.161)
Statutory Payroll Tax
-0.772***
(0.176)
Constant
Observations
R-squared
Modification
0.043***
(0.001)
203,515
0.528
Baseline
203,515
0.528
OLS errors
203,515
0.667
Weighted
171,433
0.068
Differences
203,515
0.528
Alternative
Notes: The dependent variable in each regression is the wage rate. In column 4 the dependent variable is the first
difference in the wage rate. In column 3 regressors are weighted by the number of employees in the firm. In column
4 the baseline is repeated in first difference with sector-year fixed effects. Effective Payroll Tax is transformed
as log(1 + P ayroll T ax Rate). Regressions include plant × firm and sector × year fixed effects. All variables
are in logs. Standard errors clustered at the sector-tax zone level except in column 2, where the errors are OLS.
Asterisks denote: ∗ ∗ ∗p < 0.01, ∗ ∗ p < 0.05, ∗p < 0.1.
if we take into account that it encompasses the social security contribution, this is still 10 percentage
points above the Norwegian payroll rate and highly above Denmark. The difference in payroll taxes
should be compared to the corporate tax rates in these countries. The current rate in Sweden is 22%,
in Norway 27%, and Denmark 24.5. The Danish rate will be reduced to 22% in 2016. Sweden has
announced that it will reduce its corporate tax rate even further. Where globalization seems to lead
to tax avoidance, the payroll tax could be an alternative instrument to tax firms. While this may be
the case for Greece and France, based on the cases of Saez et al. (2012) and Lehmann et al. (2013), it
seems that in Norway firms are adept at shifting the burden away to employees and not one solution
applies to all.
References
Anderson, Patricia M, and Bruce D Meyer (1997) ‘The effects of firm specific taxes and government
mandates with an application to the us unemployment insurance program.’ Journal of Public Economics 65(2), 119–145
Bennmarker, Helge, Erik Mellander, and Björn Öckert (2009) ‘Do regional payroll tax reductions boost
employment?’ Labour Economics 16(5), 480–489
Bohm, Peter, and Hans Lind (1993) ‘Policy evaluation quality: A quasi-experimental study of regional
employment subsidies in sweden.’ Regional Science and Urban Economics 23(1), 51–65
Carlsen, Fredrik, and Kåre Johansen (2005) ‘Regional wages and subjective measures of employment
opportunities.’ The Journal of Socio-Economics 34(3), 377–400
Dyrstad, Jan Morten, and Kare Johansen (2000) ‘Regional wage responses to unemployment and profitability: empirical evidence from norwegian manufacturing industries.’ Oxford Bulletin of Economics
and Statistics 62(1), 101–117
Dyrstad, JM (1992) ‘Redusert arbeidsgiveravgift reduserer arbeidsledigheten.’ Arbeidsledighet-hvordan
redusere den. Oslo: Universitetsforlaget
13
Gruber, Jonathan (1994) ‘The incidence of mandated maternity benefits.’ American Economic Review
84(3), 622–41
(1997) ‘The incidence of payroll taxation: Evidence from chile.’ Journal of Labor Economics
15(S3), S72–S101
Johansen, Frode, and Tor Jakob Klette (1997) ‘Wage and employment effects of payroll taxes and
investment subsidies.’ mimeo
Johansen, Kåre (2002) ‘Regional wage curves empirical evidence from norway.’ Technical Report,
Department of Economics, Norwegian University of Science and Technology
Korkeamäki, Ossi, and Roope Uusitalo (2009) ‘Employment and wage effects of a payroll-tax cutevidence from a regional experiment.’ International Tax and Public Finance 16(6), 753–772
Lehmann, Etienne, François Marical, and Laurence Rioux (2013) ‘Labor income responds differently
to income-tax and payroll-tax reforms.’ Journal of Public Economics 99(0), 66 – 84
Murphy, Kevin J (2007) ‘The impact of unemployment insurance taxes on wages.’ Labour Economics
14(3), 457–484
Saez, Emmanuel, Manos Matsaganis, and Panos Tsakloglou (2012) ‘Earnings determination and taxes:
Evidence from a cohort-based payroll tax reform in Greece.’ Quarterly Journal of Economics
14