Incidence of Social Security contributions: Evidence from France Preliminary results

advertisement
Incidence of Social Security contributions: Evidence
from France
Antoine Bozio, Thomas Breda and Julien Grenet∗
Preliminary results
Thursday 25th February, 2016
Abstract
In this paper, we study the earnings responses to the four largest increases in Social
security contributions (SSCs) in France over the period 1976–2010, using both a
difference-in-differences (DiD) estimation and a regression kink design (RKD) approach. We find systematic evidence of increased labour cost, i.e., the absence of
full tax shifting to workers, at the individual level, within five to six years after each
reform. Our DiD estimates point to some shifting of SSCs to employees in the form
of lower wages, with a estimated share of the tax burden between 40 percent and
50 percent. This conclusion is not borne by our RKD estimates which do not reject
full incidence on employers. We treat these estimates with caution, as preliminary
results, while arguing that the empirical strategies followed in this paper offer a
promising prospect in a literature which generally lacks convincing identification
strategies or external validity.
∗
Bozio: Paris School of Economics (PSE), antoine.bozio@ipp.eu. Breda: PSE, thomas.breda@ens.fr.
Grenet: PSE, julien.grenet@ens.fr. We thank Emmanuel Saez and participants of the Open research
area (ORA) consortium from PSE, IFS, DIW and CPB. We acknowledge financial support from Agence
nationale de la recherche (ANR) under the grant number ANR-12-ORAR-0004 under the ORA call.
1
1
Introduction
Social security contributions (SSCs) represent an important part of total tax revenues
in OECD countries (on average 25 percent of total tax revenues), and an even larger share
in countries with extended social insurance systems (40 percent of total tax revenues in
France). Surprisingly, these taxes have attracted much less attention in the public finance
literature than features of income taxation. When a large literature has been devoted
to measuring behavioural responses of income taxation in the U.S., and in most other
countries (see Saez, Slemrod & Giertz (2012) for a survey), relatively few studies have
looked at SSCs. This lack of empirical evidence is in stark contrast with the policy debate
which has largely focused on the detrimental effect of employer SSCs on employment, in
particular in Continental Europe. In a number of countries (including France, Belgium,
the Netherlands), recent policies have aimed to deliberately reduce employer contributions
for low wage earners, as a way to reduce the labour cost at the minimum wage, and
hence to reduced low-skilled unemploment. Most recent studies of SSCs have therefore
focused on the employment effects of reduced employer SSCs (see for instance Kramarz &
Philippon (2001)). Above the minimum wage, the employment impact of SSCs depends
on the ultimate incidence of these taxes: if employer SSCs are shifted to workers in the
form of lower wages, then the employment effect of SSC increases are likely to be limited.
The incidence of SSCs, originally an important issue in the public economics literature (Musgrave 1959, Atkinson & Stiglitz 1980, Kotlikoff & Summers 1987, Fullerton
& Metcalf 2002), has led to little recent work. Most of the empirical literature dates
back to the 1970s and has found relatively mixed results, based on either cross-country
or time-series evidence (Brittain 1972, Feldstein 1972, Hamermesh 1979, Holmlund 1983).
Among recent papers, using micro data and adequate identification strategies, the results
are equally scattered, from full shifting to employees (Gruber & Krueger 1991, Gruber
1994, 1997, Anderson & Meyer 1997, 2000) to full shifting to employers (Lehmann et al.
2013, Saez, Matsaganis & Tsakloglou 2012). There is, however, little sense in expecting
similar incidence of SSCs, as it is likely to depend on a number of features of the SSC
reform under consideration, the functioning of the labour market, and the type of estimation carried out. First, it is important to recall that the incidence of SSCs is bound to
differ in the short run (the day after the reform) and in the long run. Economic incidence
is expected to coincide with legal incidence in the short run, and the interesting question
2
is how long does it take for wages to adjust to changes in SSCs (if they adjust at all).
Second, the tax-benefit linkage is likely to matter for the incidence of SSCs. It has long
been recognised that the intrinsic nature of SSCs (Musgrave 1959, Summers 1989, Gruber
1997), as opposed to income taxation, is to offer benefits linked with the SSC payment.
This feature of SSCs is actually one of the main arguments in favour of institutional construction of the form of social insurance, as it is assumed that SSCs have, for this reason,
lower efficiency cost. The consequence is that SSCs with strong tax-benefit linkage are
expected to be fully shifted to employees. Unfortunately, prior research provides limited
and inconclusive empirical evidence to support this argument.
If one looks closely at the two most prominent incidence studies, which find opposite
incidence effects, it seems clear that their institutional design differs markedly: Gruber
(1997) uses the decrease in pension SSCs in Chile during the privatization of the public
pension scheme to assess the incidence of these taxes and finds convincing evidence that
the decrease in SSCs led to an equal increase in wages, which is consistent with full incidence of SSCs on employees. In that case, the tax-benefit linkage is extremely salient
– employees need to fund an increase in private pension contributions –, and the shifting
of the reduced employer SSCs can be easily compensated by firms in the form of wage
increases. On the other hand, Saez, Matsaganis & Tsakloglou (2012) take advantage of
a SSC reform in Greece, where adjacent cohorts in the labour market were treated with
marked different SSC rates over a long period of time. Using a regression discontinuity
design, the authors provide compelling evidence of full incidence of employer SSCs on
employers and, symmetrically, of full incidence of employee SSCs on workers. In the latter case, full incidence on employees would have required firms to pay equally productive
workers differently on the sole basis of their date of entry into the labour market. As
suggested by the authors, this differential treatment might have raised serious fairness issues, thereby precluding full-shifting at the individual level. Despite providing remarkably
clean incidence results, the Chilean and Greek reforms have the disadvantage of lacking
external validity for more common SSC reforms. What is still missing in the literature is
compelling evidence from standard SSC reforms, with sufficiently good internal validity
and allowing for credible out of sample conclusions.
This paper aims to present evidence on the incidence of SSCs by investigating the
effects of large SSC reforms in France over a period of 30 years, using a long panel of
3
administrative data. France has a particularly long list of different SSCs for different
risks, and different schemes. Several reforms have changed the SSC schedule over the
past four decades, affecting each time different groups, and applying to SSC that were
funding different benefits – some with very little tax-benefit linkage (for instance health
care SSCs, or family benefit SSCs), others with a very strong link (e.g., complementary
pension schemes). For the purpose of identifying the earnings responses to changes in
SSCs, we exploit the four reforms which led to largest increases in SSC marginal rates: two
of them relate to non-contributory SSCs, while the other two relate to highly contributory
SSCs. We use two complementary empirical approaches: first, we carry out difference-indifferences estimations comparing wage earners just below or just above the social security
threshold (SST), before and after each reform; second, we exploit the existence of a kink
in the relationship between pre-reform earnings and the reform-induced changes in SSCs
at the SST to estimate earnings responses to changes in marginal contribution rates using
a regression kink design (RKD). Both strategies allow us to compare changes in labour
costs to changes in net earnings, in order to assess how much of the initial increase in SSCs
increase has been shifted to employers or to employees. By carrying out the estimations
for every year after the reform, we can identify effects up to 6 or 7 years after the reform,
and hence provide evidence of incidence in the long term, if not in the very long term.
While we consider our estimates with caution and treat them as preliminary results,
we find systematic evidence of less than full shifting up to six years after the occurrence
of each reform. These findings suggest that even if employers were able to shift SSCs
to employees in the form of lower wages, this process takes a very significant amount of
time. Using a regression kink design, we cannot reject that these taxes are fully incident
on employers.
The paper is organised as follows. In section 2 we present the institutional design of
SSCs in France as well as the main reforms being studied. Section 3 describes the administrative data and the microsimulation model that we use to compute SSCs. Section 4
presents the methodology and the results are described in the following section. Section 6
concludes.
4
2
Social Security Contributions Reforms
In this section, we describe the main features of the four major Social security con-
tributions (SSCs) reforms that we exploited in this paper. Before doing so, we provide a
brief overview of the architecture of SSCs in France.
2.1
Social Security Contributions in France
SSCs are a very important part of taxation in France, as their represent approximately
40 percent of total tax revenues. These contributions fund a number of aspects of the
welfare system, notably health spending and pensions, but also family benefits, unemployment insurance and housing benefits. There is a large number of different SSCs, one
for each scheme and type of risk, for instance one for the main pension system of private
sector employee, one funding family benefits, another one funding unemployment insurance, etc. The different schemes differ according to the type of governance (State, Social
security, management by employer and employee unions) and according to the nature
of the contributory link between SSCs and benefits (pensions are contributory, health is
not). For instance, in the case of pensions for private sector employees, the main scheme
from Social security (caisse nationale assurance vieillesse, CNAV) is an defined benefit
pay-as-you go pension scheme, with an implicit contributory link (reference to average
earnings of the last 25 years), but incorporates some elements of redistribution. Alongside
this main scheme are two mandatory complementary schemes (Agirc and Arrco) which
are managed by employer and employee unions, and, like the basic pension, are financed
on a pay-as-you go basis. These schemes, however, use a point-based system which creates
a very salient contributory link between the amount of pension benefits received and the
amount paid in contributions. The SSC schedule that applies to wage earnings thus varies
according to a number of characteristics (e.g., number of employees of the firm, location
of the firm, pension scheme affiliation), and in particular it differs within the private sector according to whether wage earners are executives (i.e., affiliated to the Agirc pension
scheme) or non-executives (i.e., affiliated to the Arrco pension scheme).
The SSC tax schedule in France is similar to that observed in most other OECD
countries. The tax base is gross earnings capped at different thresholds. The reference
threshold, which is referred to as the Social Security threshold (SST) corresponds roughly
5
to the mean gross earnings, and SSCs can be defined as a function of 1, 3, 4 or 8 times the
SST. One distinctive feature of French SSCs is that the main threshold (1 SST) is lower
than in most other countries (around P70) while there are SSCs for very high level of
earnings (the highest threshold being at P99.95). Importantly for our empirical strategy,
the SSC schedule is expressed in terms of hourly wage, i.e., the SST is adapted to the
actual hours of work and duration of the job spell. This means that marginal tax rates
are unaffected by changes in hours of work – contrary to what is usually the case with
income taxation.
During the period of our study, from 1976 to 2010, a number of SSC reforms have been
carried out in France. Some of the most well-known of these changes are the reductions
in employer SSCs around the minimum wage that have been put in place since the 1990s.
In this paper, we focus on an other set of reforms, which have attracted far less attention
in the literature – we are not aware of any previous analysis. These reforms involved
large increases in employer SSC rates above the SST, affecting the top three deciles
of the earnings distribution. Figure 1 shows the evolution of marginal employer SSC
rate for different fractions of earnings, separately for non-executive workers (Panel A)
and for executives, who are covered by a mandatory pension scheme up to 8 times the
SST (Panel B). While the rates of employer SSCs applied to the fraction of earnings
below the SST have increased modestly (from 36 percent in 1976 to 38 percent in 2001),
the rates applied to the fraction of earnings above the SST have increased dramatically
over the same period (from 7 percent to 38 percent for non-executives, from 10 percent
to 38 percent for executives). These changes were driven by two set of SSC reforms:
(i) the uncapping of SSCs (previously capped at the SST) for the main schemes (health,
family, pensions); and (ii) an increase in the complementary schemes’ contribution rates
for earnings above the SST. These reforms are at the core of this paper as they led to
marked increase in SSCs for some individuals and not for others. We discuss below the
institutional setting for each of these reforms.
2.2
Uncapping of Health Care SSCs (1981 and 1983)
Health care SSCs are a set of contributions funding access to the French health care
system. It funds a public health insurance (assurance maladie) which reimburses individuals covered for the health expenses incurred from both private and public health care
6
providers. Health care SSCs can be characterised as non-contributory in the sense that
the level of insurance does not depend on the amount of contributions. There was, at the
time of the reform, a contributory link insofar as eligibility to health insurance was conditional on being covered (hence on having paid contributions in the past), but a change
in the rate of SSCs did not change the amount of benefits received.
At the onset of the scheme, health care SSCs took the form of large employer SSCs
under SST, and much smaller employee SSCs. In the early 1980s, employer health SSCs
were “uncapped” in two stages, i.e., became applicable to the full earnings instead of
just the fraction of earnings below the cap. In November 1981, marginal employer SSCs
on full earnings jumped from 4.5 percent to 8.0 percent (+3.5 ppt), while remaining at
13.45 percent for earnings below the SST. In January 1984, marginal employer SSCs
jumped further to 12.6 percent (+4.5 ppt), while they were decreased to the same level
for the fraction of earnings below the threshold (-0.85 ppt). The first panel in Table 1
presents the total changes in employer and employee SSC rates that were brought about
by this uncapping of heatlh care SSCs between the last pre-reform year and the first
post-reform year.
2.3
Uncapping of Family Benefit SSCs (1989 and 1990)
Family SSCs are a peculiar feature of the French social security system, as these SSCs
do not fund a social insurance scheme, but universal child benefits. All families with
children are entitled to these benefits, irrespective of their employment statuts, with no
link between contributions and benefits.
From the onset of the scheme, family SSCs have only taken the form of employer SSCs
capped at the SST. Over two years, in 1989 and 1990, these SSCs were uncapped, the
marginal rate below the ceiling dropping from 9 percent to 7 percent and the rate above
the SST jumping from 0 percent to 7 percent (see Table 1, second panel).
2.4
Increase in Pension Contributions for Top Executives (1991)
Complementary pensions in France are private pension schemes, managed by employer
and employee unions, without oversight from the Government or Parliament. Contributions to these schemes are, however, mandatory. Rates and benefits are determined by
union representatives. Two main schemes exist for private sector workers: Agirc for ex7
ecutives, and Arrco for non-executives (see below). Both schemes cover earnings above
the SST and work as unfunded defined contribution point-based systems. Wage earners
pay contributions (both employer and employee SSCs) which are converted from euros
into points, and are accumulated as points, before being converted into annuity pensions.
Hence there is a strong contributory link, as each euro of additional contribution leads to
increased level of pension.
From the onset of the Agirc scheme (1947), contributions were based on earnings
between the SST and 4 times the SST, which covered roughly all earnings between average
earnings and P97. In 1990, marginal employee contributions amounted to 2.34 percent,
whereas employer contributions reached 7.02 percent for earnings between 1 and 4 SST.
In 1991, Agirc’s board decided to create a new layer of contributions (and benefits) by
levying contributions of similar rates on the fraction of earnings between 4 and 8 SST.
This earnings range was referred to as “tranche C ” (as opposed to tranche B for earnings
between 1 and 4 SST, and tranche A for earnings below the SST). This increase in the
pensions contributions for top executives represented an increase of 9.36 ppt in SSC for
the marginal rate above 4 SST, of which 7 ppt for the employer side (see Table 1, third
panel).
This apparently large-scale reform was in reality the product of the poor financial
health of existing schemes, which were based on voluntary contributions for earnings
between 4 and 8 times the SST. Three different schemes (CCSBTP, IRCASUP and IRICASE) were indeed operating before 1991, and were offering their services to individuals
and to firms that were willing to cover their top employees with pension rights. It was
possible by collective agreement (at the branch level) to decide mandatory contribution to
these schemes even though it was not mandated at the economy-wide level. For instance,
the construction sector signed such an agrement and the CCSBTP scheme was covering
all employees from this sector with earnings between 4 and 8 times the SST. The reason
for the 1991 reform can be found in the financial difficulties that these schemes were facing. The reform consisted in their incorporation into the Agirc scheme in 1988, a gradual
increase in their contribution rates to the level of Agirc rates and, in 1991, a mandate
that all firms join the extended scheme. For firms previously affiliated with a voluntary
pension scheme, the change would have been small while for unaffiliated employers, the
reforms induced a large increase in employer contributions for their top executives.
8
2.5
Increase in Pension Contributions for Non-Executives (1999
to 2005)
The complementary scheme Arrco is dedicated to non-executive workers. It offers
both a complementary pension below the SST and a supplementary pension for earnings
above the SST and up to 3 times the SST. While SSC rates below the cap have only
marginally changed over time, the marginal rates applicable to earnings above the SST
increased steadily during the early 2000, from 4.5 percent in 1999 to 12.0 percent in 2005
for employer SSCs, and from 3.0 percent to 8.0 percent for employee SSCs (see Table 1,
fourth panel).
The above discussion on the contributory nature of contributions to the Agirc scheme
applies to the Arrco scheme as well, since it uses the exact same point-based system for
deriving pension benefits. The only differences with the 1991 Agirc reform are that the
1999-2000 Arrco reform and did not lead to an incorporation of other schemes and that
the level of earnings for which the contribution rate increased was much lower.
In summary, the four SSC reforms described in this section all increased SSCs above
the SST, but differ in their timing and in the contributory nature of the benefits they fund.
The first two reforms affected only employer SSCs and did not lead to proportional changes
in benefits; by contrast, the other two reforms affected both employee and employer SSCs
and increased the level of expected pension benefits for the workers concerned.
3
Data
3.1
The DADS Panel Dataset
Our primary source of data for the analysis comes from the matched employer-employee
panel DADS (Déclaration Annuelle de Données Sociales), which was constructed by the
French National Institute for Statistics (INSEE) from the compulsory fiscal declarations
made annually by all employers for each of their employees. The main purpose of these
forms is to provide the different social security schemes with the earnings information necessary to determine workers’ eligibility to benefits and to compute their levels, notably for
pension schemes. The French national statistics office, INSEE, transforms the raw DADS
9
data into user files available to researchers under restricted access1 . The panel version of
the DADS consists of a 1/25 sample of private sector employees, born in October of evennumbered years, from 1976 onwards. In 2002, the sample size was doubled to represent
1/12 of all private sector workers. The data includes roughly 1.1 million workers each
year between 1976 and 2001, and 2.2 million workers from 2002 onwards. Unfortunately,
some years of the original data sources were lost (1981, 1983 and 1990) and are therefore
missing in the Panel data.
The DADS Panel provides information about the firm (identifier, sector, size) and each
job spell (start and end date, earnings, occupation, part-time/full-time). Hours of work
are available since 1993. Importantly for our study, raw data about earnings come under
the form of “net taxable earnings” (earnings reported for income tax). This definition of
earnings is net of social security contributions, but does not recover exactly the definition
of net earnings as some flat rate contributions (CSG and CRDS) are not deductible for
the income tax, and are therefore added back to compute net taxable earnings. Gross
earnings reported by employers are available from 1993 onwards. In earlier years, they
were only estimated by the INSEE on the basis of the reported net taxable earnings.
Finally, a measure of reported net earnings is available from 1993 onwards, corresponding
to earnings as paid through the pay slip, i.e., after deduction of some specific employee
contributions to restaurant vouchers or public transport cards.
3.2
Microsimulation of SSCs
Microsimulation techniques are required to compute the labour cost based on the
information that is available in the DADS Panel data. The present work relies on the
use of the TAXIPP model developed at the Institut des politiques publiques (IPP), and
in particular on the social security contribution module. The model takes as input the
changes in SSC schedule, as collected in the IPP Tax and Benefit Tables2 and computes
employee and employer SSCs, reductions in employer SSCs, flat-rate income tax (CSG
and CRDS) as well as other payroll taxes. Given the complexity of French SSCs, a number
of simplifications have been carried out: local variations in the public transport payroll
tax (versement transport) were not simulated, nor were the specific reductions in SSCs
1
We were granted access to the DADS data by the decisions of the Comité du secret statistique ME27
of 02/10/2013, ME56 of 25/06/2014 and ME91 of 25/06/2015.
2
See http://www.ipp.eu/en/tools/ipp-tax-and-benefit-tables/social-security-contributions/.
10
that were granted to firms operating in a small number of disadvantaged areas. Besides
these elements, most of the SSCs has however been carefully simulated, including local
social security schemes such as the one in place in the Alsace-Moselle region.
The main challenge in computing SSCs from the DADS data comes from the missing
information in the raw data. Three main issues must be noted. First, because gross
earnings are not available in the raw data before 1993, we decided to compute them based
on net taxable earnings using the microsimulation model. Second, SSCs are defined as
a function of hourly wage when working part-time – SST are defined for each period of
work and adjusted for hours worked. Since we do not observe hours in the DADS data
before 1993, the SSCs for part-time workers cannot be computed before 1993. Third, the
data does not provide information on the actual date at which firms decided to join the
35 hours week in the early 2000s, which determines eligibility to specific SSCs reductions.
We estimated this date based on each firm’s reported working hours3 .
4
Empirical approaches
We take advantage of the different SSCs reforms described in section 2 to identify the
earnings responses to changes in SSC rates. For all reforms, the year-to-year shifts in the
total amount of SSCs vary with base year earnings according to a well-defined schedule:
they are null below the SST (or below four times the SST in the case of the Agirc reform),
and increase linearly above it.
The most straightforward way of estimating earnings responses to changes in SSCs
rates is to compare individuals with earnings above the SST in the last pre-reform year
(treatment group) to individuals with earnings below the cap (control group), before and
after the reforms. The validity of this difference-in-differences (DiD) approach relies on
the assumption that the average earnings of treatment and control group workers would
have followed parallel trends over time, absent the reform. This assumption could be
violated, e.g., in the presence of increasing wage inequality. To address this concern, we
complement the DiD estimation with a regression kind design (RKD) estimation. Both
the uncapping of SSCs and the increase in marginal SSC rates above the SST create kinks
in the relationship between the level of earnings in the base year and subsequent changes
3
We compute the share within firms of workers declaring 35 or 39 hours, and decide to allocate firms
to either hours reference depending on the highest proportion of observed employees.
11
in average SSC rates. We exploit these kinks to locally identify the earnings responses to
changes in the tax rates.
4.1
Difference-in-Differences Estimation
To implement the difference-in-differences approach, we use four balanced panels of
workers, which are constructed separately for each of the reforms being studied. Each
sample includes workers who are observed in employment throughout a period which
starts 4 years before the reform and ends 6 to 7 years after. These time windows were
chosen to avoid contaminating the estimated earnings responses to a particular reform
with the effects of other reforms. The time periods used in the analysis are 1977–1988
for the uncapping of health care SSCs in 1981 and 1983, 1985–1996 for the uncapping of
family SSCs in 1989 and 1990, 1986–1997 for the Agirc reform of 1991 (increase in pensions
SSCs for top executives), and 1996-2008 for the Arrco reform of 2000-2005 (increase in
pensions contributions for non-executives). The samples used for the analysis of the first
two reforms include both executive and non-executive workers. By contrast, the samples
used for the analysis of the Agirc reform in 1991 and of the Arrco reform of 2000–2005
are restricted to individuals with executive and non-executive status respectively.
In each of the balanced panels, workers are assigned to the treatment and control
groups based on the their level of gross earnings relative to a predetermined earnings
threshold in the base year. The relevant threshold for the two uncapping reforms and for
the Arrco reform is the SST. In the case of the Agirc reform, the relevant threshold is
four times the SST. Individuals with earnings just below the threshold are assigned to the
treatment group whereas individuals with earnings just above are assigned to the control
group.
Because the Agirc reform only impacted a relatively small number of top executives
with earnings in the upper end of the wage distribution, we were not able to construct
control and treatment groups that would be large enough for estimation purposes while
following similar earning trends before the reform. We therefore ended up discarding this
reform from the rest of the analysis (see results below, and Figure 4).
Regarding the other reforms, the main tradeoff in selecting the treatment group is that
while expanding this group’s upper earnings threshold mechanically inflates the reforminduced variation in average SSC rates, it also increases the risk of dissimilar earnings
12
trends between the treated and controls. For our baseline analysis of the three reforms
that occurred at the SST, we assigned to the treatment group individuals whose gross
earnings in the base year were 1 to 1.6 times the SST in that year. We assigned to
the control group individuals in a smaller range of gross earnings in the base year, i.e.,
between 0.8 and 1 times the SST. This range is large enough to build a control group of
significant sample size, and going further down the earnings distribution would bring the
risk of including workers whose earnings were affected by the diffusion effects of increases
in the national minimum wage.
Normalizing to t = 1 the first year after the reform, we focus on workers observed
every year between t = −3 and t = 8 and with annual earnings between 0.8 and 1.6
times the SS cap in the base year 0 (or −1 if the base year happens to be missing from
the data–see section 3). We use the following dynamic DiD specification to estimate the
earnings responses to each of the four SSCs reforms:
log(wit ) = α +
8
X
k=−3
βk 1{t = k} + δTi +
8
X
γk (Ti × 1{t = k}) + εit ,
with γ0 = 0 (1)
k=−3
where wit denotes the annual net earnings (or, alternatively, the annual labour cost) of
worker i in year t, 1{t = y} is a year fixed effect and Ti is a treatment group dummy
that takes the value one if worker i’s earnings are between 1 and 1.6 times the SST
in the base year. The coefficients γk measure the pre- and post-reform log-differences
in earnings between treated and control workers in year k relatively to the base year.
Non statistically significant coefficients γk ’s for years k < 0 would be consistent with the
parallel pre-trends assumption. In that case, and assuming that the earnings trends would
have remained parallel for years k ≥ 1 in the absence of reform, one can interpret γk for
k ≥ 1 as “treatment effects” for each year k ≥ 1, i.e., the reform’s impact on earnings
after k years.
The model’s key identifying assumption is that absent the SSC reforms, average earnings for the treated and control groups would have followed parallel trends over time. In
light of the general pattern of rising earnings inequality during this period, this may seem
an unreasonable assumption (see for instance Bozio et al. 2016). This concern, however,
is mitigated by the fact that our identification strategy uses relatively narrow earnings
ranges around the SS cap and that the parallel trends assumption can be tested for the
13
pre-reform periods.
Equation (1) is estimated using either the log of net earnings or the log of labour cost
as the dependent variable. The employer share of the incidence of changes in SSC rates
between the base year 0 and end year y can then be recovered by restricting the sample
to workers observed in both years, and estimating the following equation using 2SLS:
log(zit ) = α + β. log(1 + SSCrit ) + δ.1{t = y} + γ.Ti + εit
(2)
where zit is worker i’s labour cost in year t, SSCrit is the worker’s average SSC rate in
t (measured as the ratio of her SSCs to her net earnings in t), and log(1 + SSCrit ) is
instrumented by the interaction between the treatment dummy Ti and the post-reform
year dummy 1{t = y}. This equation is estimated separately for all years 0 < y ≤ 8. The
first-stage equations estimate the average reform-induced variation in SSC rates, while
the coefficient β̂ in equation (2) estimates the share of this variation that was passed on
to labour costs.
4.2
Regression Kink Design
Our second approach exploits the fact that the reform-induced changes in SSCs exhibit
a kink at the SST. The empirical analysis draws on the recent contributions by Calonico
et al. (2014a), Calonico et al. (2014b), Landais (2015) and Card et al. (2015). In the
current version of the paper, we stick to a simple parametric framework. We focus on
the changes in net earnings and the changes in labour costs experienced by workers who
are observed in employment both in the base year 0 and in the post-reform year y, for
1 ≤ y ≤ 8. We assume that the underlying expected change in earnings in the absence
of reform is a continuous and smooth (i.e., differentiable) function f of initial earnings
w0i : E[wyi − w0i |w0i ] = f (w0i ), with f continuous and differentiable. This assumption
can be considered as relatively weak since there is no obvious economic reason for the
relationship between average changes in earnings and initial earnings not to be smooth.
Under this assumption, a discontinuity in the slope of the relationship between changes in
earnings and initial earnings at the SST identifies the incidence of SSCs. This is precisely
because the extra SSC burden induced by the reform is zero up to the SST, and increases
linearly above it.
14
This identification result can be easily derived in the the case where the expected
change in (real) earnings around the SST is zero in the absence of reform and the SST
itself is constant over time (in real terms).4 Denote τly and τry the marginal SSC rates
below and above the SST in year y. The reforms we analyze are such that τly ≈ τl0 and
τry >> τr0 for all y > 0. The reform-induced increase in the amount of SSCs, which we
denote ∆SSC, is then a kinked function of earnings in year 0:

 ∆SSC ≈ 0
if w0i ≤ SST
 ∆SSC = (τ − τ ) × (w − SST ) if w > SST
ry
r0
0i
0i
To estimate whether net earnings or labour costs variations between year 0 and year
y exhibit a kink at the SST, we estimate the change in the slope of the conditional
expectation function of the outcome given the assignment variable at the kink by running
parametric polynomial models of the form:
E(∆wiy |wi0 ) = α0 +
P
X
βp (w0i − SST0 )p + γp (w0i − SST0 )p .Di
(3)
p=1
where |w0i − SST0 | < h, ∆wiy is the change in earnings between year 0 and year y, wi0
is the assignment variable (base year earnings), Di = 1{w0i > SST0 } is an indicator for
being above the SST in the base year, h is the bandwidth size, and the change in the
slope of the conditional expectation function is given by γ1 .
The incidence of SSC on labour cost is identified locally at the kink and can be
estimated using a fuzzy RKD estimator. We implement this approach by estimating the
following equation using 2SLS:
E(∆ziy |wi0 ) = α0 + β.∆SSCiy +
P
X
p
γp (w0i − SST0 ) +
p=1
P
X
δp (w0i − SST0 )p .Di
(4)
p=2
where |w0i − SST0 | < h, ∆ziy is the change in worker i’s labour cost between the baseline
year and year y, and ∆SSCiy (the change in worker i’s SSCs) is instrumented by γ1 (w0i −
SST0 ).Di . The estimation is performed separately for all years 1 ≤ y ≤ 8. The first-stage
equation estimates the kink (τry − τr0 ) in SSCs induced by the reform at the SST, while
the coefficient β̂ in equation (4) estimates the extent to which this increase in marginal
4
In the appendix, we discuss the more general case where real earnings and the SST grow at different
rates.
15
SSC rates was passed on to labour costs.
4.3
Behavioral Responses vs Incidence
What we measure using our two empirical strategies is in effect earnings response
to changes in SSCs. Earnings response to change in taxation cannot be interpreted as
incidence as they mix a response in terms of wage rate (incidence) and a potential response
in hours worked (behavioural responses). It is common in the elasticity of taxable income
literature (Saez, Slemrod & Giertz 2012) to assume full incidence on employees, in order
to derive estimates of behavioural response from earnings response. In the case of income
taxation, nominally borne by employees, the assumption seems innocuous (see however
Bingley & Lanot 2002, Kubik 2004, for dissenting evidence) but it is certainly not valid
when considering impact of SSCs. On the contrary, one assume away any hours response,
and interpret earnings responses as incidence effect (i.e., wage rate changes).
In the current version of this paper, this is mostly the assumption we make to interpret
our results as incidence effects. We do not observe hours worked before 1993, during the
period when most of our reforms have been carried out. We can therefore only measure
total earnings response to change in SSCs. We must stress however that, in our empirical
analysis, we only use wage earners working full-time and during the entire year, which
is likely to limit the amount of behavioural response we capture in our estimates. It
is possible to argue that other margins of behavioural responses are at play (namely
unobserved effort), but in that latter case they would be undistinguishable from incidence
effect even with hours observed.
One final argument can be discussed in the size of the expected effect of behavioural
responses: an increase in SSCs should lead to a reduction in hours of work, hence a reduction in the total earnings. If there is behavioural responses in our estimates, they should
bias our analysis into finding more incidence on employee than in reality (confounding
hours response with incidence effects).
5
Results
We present below the main summary of our results using the two empirical approaches
described in the previous section.
16
5.1
Graphical Evidence from DiD
Graphical evidence on the earnings responses to the SSC reforms is presented in Figures 2 to 5. For each of the four SSC reforms, the figures compare the evolution of average
net earnings (left panel) and of average labour cost (right panel) between the treatment
and control groups around the reform years. All earnings measures are indexed to 100 in
the base year, i.e., the year immediately preceding the start of the reform.
When considering the first two uncapping reforms (Figures 2 and 3), we find that
treatment and control groups have a very similar evolution of net earnings while labour
cost diverge markedly immediately after the reform. This pattern suggests that, in the
short term, employer SSCs are incident on employer – which is what one would expect.
We are able to follow the evolution of earnings up to 6 years after the reform (before
the next reform) and we do not find evidence of full convergence in terms of labour cost.
Reassuring for our DiD approach is the fact that the pre-reform trend align very well for
the control and treated groups.
Figure 4 presents similar evidence for the increase in SSC rates for executives with
earnings above 4 times the SST. At these very high earnings levels, our treated (earnings
between 4 and 6 times the SST) and control groups (earnings between 2.5 and 4 times
the SST) are very small (respectively 815 and 173 observations) and we find strongly
diverging patterns in terms of both net earnings and labour cost. Wage earners observed
before the reform above 4 SST tend to see their earnings decrease post-reform, while they
were increasing at faster rate before the reform. This signals to us the likelihood of mean
reversion, making this reform unsuitable for our empirical strategies. As a result, we leave
this reform aside in what follows.
Figure 5 shows the treatment-control comparison for the fourth SSC reform, which
affected non-executives above the SST. The pattern here is similar to that observed for
the two uncapping reforms, with a short-term increase in labour cost after the reform
that does not seem to converge over the period during which we follow individuals. One
should however point to the pre-reform trends that do not seem totally aligned, with
treated groups apparently growing at faster rate than control groups.
17
5.2
Regression Analysis from DiD
We now present the results from the dynamic DiD regressions, which we carried out
separately for each of the three reforms.
For the first reform, i.e., the uncapping of health care SSC, the regression results are
shown graphically in Figures 6 and 7 and the corresponding coefficients can be found in
Panel A of Table 2. The results suggest that after the reform, the increase in employer
SSCs lead to an increase in labour cost. After 6 years, the impact on the labour cost
is still positive and statistically different from 0. We find evidence of partial shifting,
as net earnings appear to have declined after the reform. The pre-reform estimates are
not statistically different from 0 and are therefore reassuring regarding the possibility of
pre-reform differential trends. The 2SLS estimates yield a statistically significant point
estimate of the share of SSC born by employer 6 years after the reform of 47.5 percent,
with a relatively large standard error.
Evidence from the second reform, i.e., the uncapping of family SSC, can be found in
Figures 8 and 9, with the corresponding coefficients reported in Panel B of Table 2. The
results are qualitatively similar those found for the first reform – an increase in labour
cost and less than partial shifting 5 years after the reform – but the results are more
imprecise, and the pre-reform estimates show evidence of differential earnings trends.
The 2SLS estimates yield a point estimate of the share of SSC born by employer 6 years
after the reform of 60.2 percent, down from an estimated share of 1.3 immediately after
the reform’s implementation. Again, standard errors are large, but we cannot reject
statistical difference from 0.
Evidence from the fourth reform, i.e., the increase in SSC for non-executives between
2000 and 2005, is shown in Figures 10 and 11, with the corresponding coefficients reported
in Panel C of Table 2. Since the reform was very gradual, we need to look at T+6
to see it fully in place. Similar to the previous reform, we cannot reject pre-reform
differential trends. We find no impact on net earnings and almost full incidence on
employers. The 2SLS estimates yield a point estimate of the share of SSC born by
employer of 103.5 percent in the last year we look at, i.e. 2 years after the full installment
of the reform, which can be arguably be considered a short term impact.
18
5.3
Regression Kink Designs
We now turn to the analysis using the RKD approach. We start by presenting graphical
evidence of the existence of kinks around the SST in the earnings distribution.
Fig 12 plots for each reform the average change in net earnings (blue triangles) and
the average change in labour cost (red circles) against earnings in the last pre-reform base
year. The graphs in the left panel show the changes 4 years after the reform, while the
graphs in the right panels show the changes after 8 years. The kink in labour cost changes
at the SST after 4 years materialises the short-term incidence of an increase in employer
SSCs (which mechanically increase labour cost for those with earnings above the SST).
After 8 years, we still observe kinks in the labour cost for all three reforms.
Table 3 presents the coefficient estimates from the RKD, using a local linear estimator
with a bandwidth of 0.2 times the STT. We find very significant impacts of increases in
SSCs on labour cost using this approach, with estimated employer shares of incidence
systematically above 1 for the first two reforms, and only slightly below one for the third
one. The estimated employer share of incidence do not decrease with time, so that we do
not find any evidence of partial shifting even in the final period of our estimation.
The estimates provided by the RKD are larger than DiD estimates, the differences
being always statistically significant for the first reform, and also significant seven or
eight years after the second one. The strong stability over time of the RKD estimates also
contrast with the slight decline of the employer share of incidence that we found for the
second reform using the DiD approach. The comparison should not been pushed forward
however, as we acknowledge that our RKD identification is still in progress, and may
currently be compromised by three issues. First, if for structural reasons, workers above
the kink have larger increases in earnings than those below, this might be captured by
the kink in our simple linear specification. Using a larger order polynomial to fit the kink
may help taking into account more flexibly counterfactual increases in earnings. Second,
some workers just above the kink before the reform may be caught-up by the SST and
see their average SSC rate increase for this reason rather than because of the reform.
This may contaminate our identification at the kink (see appendix). Third, the standard
errors provided in Table 3 are certainly too small and need to be corrected (Calonico et al.
2014a, Card et al. 2015). We are currently in the process of dealing with these issues.
19
6
Conclusion
Using two empirical strategies, difference-in-differences and regression kink design, we
have provided evidence on the incidence of three large changes in social security contributions that took place in France over the 1980s and 1990s. These changes consisted mainly
in an increase in employer SSCs for high wage earners on the portion of their earnings
above the main social security threshold.
While we consider our estimates with caution and treat them as preliminary results,
we have found systematically evidence of less than full shifting up to 6 years after the
occurrence of the reform. These findings suggest that even if employers are able to shift
these contributions to employees in the form of lower wages, this process takes a very
significant amount of time. Using a regression kink design, we cannot reject that these
taxes are fully incident on employers. Our results are however unprecise, as large standard
errors are attached to our point estimates.
If these results were to hold, they would not imply an absence of any shifting of
employer SSCs onto employees, but more precisely they would rule out a particular mechanism of tax incidence, namely shifting at the individual level of SSC in the form or lower
wages. Shifting towards employees could happen through a number of other mechanisms
that we are not able to test: firms could shift these taxes to employees at the firm level,
reducing all wages beyond those of the workers directly affected by an increase in employer
SSCs; alternatively the incidence could happen at market level through firm destruction
and creation, or through changes in the extensive margins of employment (lower hiring,
higher dismissal of those affected).
There are reasons to think that at least some of those mechanisms are at play in the
long run. Indeed, the long trend during this period of skill-biased technological change,
which led to a significant increase in labour cost inequality, provides suggestive evidence
that the long-term incidence of employer SSCs is on employees Bozio et al. (2016). The
discrepancy between our results at the micro level, based on seemingly convincing identification strategies, and this indirect evidence at the macro-level deserves to be investigated
further. We plan to extend the analysis carried out in this version of the paper by looking
at how employment probabilities were affected by these SSC reforms. This may indicate
that workers who became more costly because of the reforms have been replaced in the
long-run by equally skilled but less paid ones.
20
References
Anderson, P. M. & Meyer, B. D. (1997), ‘The effects of firm specific taxes and government
mandates with an application to the U.S. unemployment insurance program’, Journal
of Public Economics 65(2), 119–145.
Anderson, P. M. & Meyer, B. D. (2000), ‘The effects of the unemployment insurance
payroll tax on wages, employment, claims and denials’, Journal of Public Economics
78(1-2), 81–106.
Atkinson, A. B. & Stiglitz, J. E. (1980), Lectures in Public Economics, McGraw-Hill.
Bingley, P. & Lanot, G. (2002), ‘The incidence of income tax on wages and labour supply’,
Journal of Public Economics 83(2), 173–194.
Bozio, A., Breda, T. & Guillot, M. (2016), ‘Taxes and technological determinants of wage
inequalities: France 1976–2010’, Working paper .
Brittain, J. (1972), The Payroll Tax for Social Security, Brookings Institution.
Calonico, S., Cattaneo, M. D. & Titiunik, R. (2014a), ‘Robust data-driven inference in
the regression-discontinuity design’, Stata Journal 14(4), 909–946.
Calonico, S., Cattaneo, M. D. & Titiunik, R. (2014b), ‘Robust Nonparametric Confidence
Intervals for Regression-Discontinuity Designs’, Econometrica 82(6), 2295–2326.
Card, D., Lee, D. S., Pei, Z. & Weber, A. (2015), ‘Inference on Causal Effects in a
Generalized Regression Kink Design’, Econometrica 83(6), 2453–2483.
Feldstein, M. S. (1972), ‘The incidence of the social security payroll tax: Comment’, The
American Economic Review 62(4), 735–738.
Fullerton, D. & Metcalf, G. (2002), Tax incidence, in A. Auerbach & M. Feldstein, eds,
‘Handbook of public economics’, Vol. 4, Elsevier/North Holland.
Gruber, J. (1994), ‘The Incidence of Mandated Maternity Benefits’, American Economic
Review 84(3), 622–41.
Gruber, J. (1997), ‘The Incidence of Payroll Taxation: Evidence from Chile’, Journal of
Labor Economics 15(3), S72–101.
21
Gruber, J. & Krueger, A. B. (1991), The Incidence of Mandated Employer-Provided
Insurance: Lessons from Workers’ Compensation Insurance, in ‘Tax Policy and the
Economy, Volume 5’, NBER Chapters, National Bureau of Economic Research, Inc,
pp. 111–144.
Hamermesh, D. (1979), ‘New estimates of the incidence of the payroll tax’, Southern
Economic Journal 45, 1208–1219.
Holmlund, B. (1983), ‘ Payroll Taxes and Wage Inflation: The Swedish Experience’,
Scandinavian Journal of Economics 85(1), 1–15.
Kotlikoff, J., L. & Summers, L. H. (1987), Tax Incidence, in A. J. Auerbach & M. Feldstein, eds, ‘Handbook of Public Economics’, Vol. 2, Elsevier, pp. 1043–1092.
Kramarz, F. & Philippon, T. (2001), ‘The impact of differential payroll tax subsidies on
minimum wage employment’, Journal of Public Economics 82(1), 115–146.
Kubik, J. D. (2004), ‘The incidence of personal income taxation: evidence from the tax
reform act of 1986’, Journal of Public Economics 88(7-8), 1567–1588.
Landais, C. (2015), ‘Assessing the Welfare Effects of Unemployment Benefits Using the
Regression Kink Design’, American Economic Journal: Economic Policy 7(4), 243–78.
Lehmann, E., Marical, F. & Rioux, L. (2013), ‘Labor income responds differently to
income-tax and payroll-tax reforms’, Journal of Public Economics 99, 66–84.
Musgrave, R. (1959), The Theory of Public Finance, McGraw-Hill.
Saez, E., Matsaganis, M. & Tsakloglou, P. (2012), ‘Earnings Determination and Taxes:
Evidence From a Cohort-Based Payroll Tax Reform in Greece’, The Quarterly Journal
of Economics 127(1), 493–533.
Saez, E., Slemrod, J. & Giertz, S. H. (2012), ‘The Elasticity of Taxable Income with
Respect to Marginal Tax Rates: A Critical Review’, Journal of Economic Literature
50(1), 3–50.
Summers, L. H. (1989), ‘Some Simple Economics of Mandated Benefits’, American Economic Review 79(2), 177–83.
22
Appendix: Increase in the SST versus Increase in the
Marginal SSC rate above the SST
In this appendix, we derive the relationship between base year earnings and the reforminduced variation in SSCs, in the general case where the SST in year 0 (SST0 ) and in
year y = 1 (SST1 ) is allowed to differ, and where w1 = w0 + f (w0 ) with f an increasing,
continuous and differentiable function. For expositional purposes, we ignore expectations
(or individual shocks).
We have in year 0:
(
SSC0 = τl0 w0
if w0 ≤ SST0
SSC0 = τl0 SST0 + τr0 (w0 − SST0 ) if w0 > SST0
Analysis of our data reveals that for all the reforms under study, the SST grows either
faster or roughly at the same pace as earnings around it. We therefore assume that there
exists w̄0 > SST0 such that:
(
w1 = w0 + f (w0 ) ≤ SST1 if w0 ≤ w̄0
w1 = w0 + f (w0 ) > SST1 if w0 > w̄0
w̄0 is the earnings level in year 0 for which earnings in year 1 are expected to be exactly
at the SST :
w̄1 = w̄0 + f (w̄0 ) = SST1
We can then express employer SSCs in year 1 as a function of base year earnings w0 :


 SSC1 = τl1 (w0 + f (w0 ))


textSSC 1 = τl1 SST1 + τr1 (w0 + f (w0 ) − SST1 ) if w0 > w̄0
From these equations, we


- If w0 ≤ SST0 :





- If SST0 < w0 ≤ w̄0 :





 - If w0 > w̄0 :












if w0 ≤ w̄0
derive the variation in SSCs between year 0 and year 1:
∆SSC = (τl1 − τl0 )w0 + τl1 f (w0 ) = g(w0 )
∆SSC = g(w0 ) + (τl0 − τr0 )(w0 − SST0 )
∆SSC = g(w0 ) + (τl0 − τr0 )(w0 − SST0 )
−(τl1 − τr1 )(w0 + f (w0 ) − SST1 )
= g(w0 ) + (τl0 − τr0 )(SST1 − SST0 )
+ {(τr1 − τr0 ) − (τl1 − τl0 )} (w0 − SST1 )
−(τl1 − τr1 )f (w0 )
The above calculations make clear that in the general case, there can be two kinks
induced by the reform:
1. A first kink at SST0 of slope (τl0 − τr0 ), which comes from the fact that workers
23
above the SST in year 0 are “caught up” by the SST and therefore see their average
SSCs increase, because the higher marginal SSC rate below the cap now applies to
their entire earnings.
2. A second kink at w̄0 where the marginal SSC rate above the SST (τr1 ) starts to
replace the higher marginal rate τl1 prevailing below the SST.
As shown in the last expression above, the difference in the slopes of SSC variations
below SST0 and above w̄0 captures three effects: (i) the variation in marginal SSCs
(τr1 − τr0 ) above the SST which is high due to the reform, (ii) the variation in marginal
SSCs (τl1 − τl0 ) below the SST (which is empirically always very small), and (iii) an
“increase in earnings” effect capturing the fact that the SSC marginal rate applying to
this increase is the above-kink rate rather than the below-kink rate.
In practice, both kinks appear to be too close to each other to allow fitting both. We
therefore opted for the estimation of a single kink at SST0 for changes in earnings or
labour costs between year 0 and year y. We chose to fit the kink at SST0 rather than at
w̄0 to avoid relying on additional assumptions to calculate w̄0 . As w̄0 is close to SST0 for
most reforms and most years 1 ≤ y ≤ 8, taking SST0 as the single kink in our analysis
should not make a strong difference. To ensure that our estimates of the kink capture the
effect of the reform and not only the fact that some workers are caught up by the SST,
we impose a relatively large bandwidth of 0.2 times the SST in the base year on each side
of the kink in our preferred empirical specifications. We therefore estimate the kink using
workers with earnings between 0.8 and 1.2 times the SST in the base year.
Note that the “increase in earnings” effect implies that the second kink is not linear
in w0 as long as f (w0 ) > 0. In principle, the second kink should be fitted using year 1
earnings w1 = w0 + f (w0 ) instead of year 0 earnings. However, as year 1 earnings may
be endogenous to the reform (e.g., include behavioral responses), we decided to use w0 as
described in the previous section. Since earnings growth in real terms is usually not very
large, this should not lead to large differences.
The use of a high-order polynomial to capture the kink at SST0 might help solving the
two issues mentioned above. First it will naturally capture the fact that the extra SSC
paid above the kink do not always vary linearly. Second, it will more flexibly incorporate
the change in slope induced by the second kink5 . The concerns put forward in this section
should nevertheless be kept in mind when analyzing the results.
5
Another route would have been to estimate f (w0 ) below SST0 only (parametrically, e.g. with a
polynomial fit), then to calculate a counterfactual w˜1 above SST0 by applying the parametrized f to
w0 > SST0 , and finally to estimate the kink using w˜1 at the right of w¯0 as explained in this section. We
have not followed this route at that stage.
24
Figure 1: Marginal Employer SSC Rates, Private Sector (1976–2010)
(a) Non-executives
45%
40%
35%
30%
25%
20%
15%
10%
Under SST
5%
1-3 SST (non-executives)
1976
1977
1978
1979
1980
1981
1982
1983
1984
1985
1986
1987
1988
1989
1990
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
2010
0%
(b) Executives
45%
40%
35%
30%
25%
20%
15%
Under SST
1-4 SST (executives)
10%
4-8 SST (executives)
5%
2010
2009
2008
2007
2006
2005
2004
2003
2002
2001
2000
1999
1998
1997
1996
1995
1994
1993
1992
1991
1990
1989
1988
1987
1986
1985
1984
1983
1982
1981
1980
1979
1978
1977
1976
0%
Notes: Marginal tax rates are here expressed as a percentage of gross earnings, as they are legislated. These rates are applied to
different fraction of earnings, defined with respect to the Social Security threshold (SST).
Sources: IPP tax and benefit tables (April 2015) ; TAXIPP 0.4.
25
Figure 2: Graphical Evidence on Earnings Responses to the Uncapping of Health Care
SSCs (1981 and 1983)
(a) Net Earnings
(b) Labour Cost
Notes: The figure shows the evolution of average real net earnings (left panel) and of average real labour cost (right panel) between
1977 and 1988 for groups that were affected differently by the uncapping of health care SSCs in 1981 and 1983. The figure is based
on a balanced panel of individuals who are observed throughout the period. The vertical lines denote the reform years, and earnings
levels in the last pre-reform year (1980) are normalized to 100 in all groups. The treatment-control assignment is based on the
reform-induced variation in average SSC rates. The treatment group includes individuals whose gross earnings in 1980 were 1 to
1.6 times the SS earnings cap that year. These workers experienced an increase in their average SSC rate due to the reform. The
control group includes individuals whose gross earnings in 1980 were 0.8 to 1 times the SS cap that year. These individuals did not
experience a change in their average SSC rate due to the reform.
Sources: DADS Panel 2010; TAXIPP 0.4.
Figure 3: Graphical Evidence on Earnings Responses to the Uncapping of Family SSCs
(1989 and 1990)
(a) Net Earnings
(b) Labour Cost
Notes: The figure shows the evolution of average real net earnings (left panel) and of average real labour cost (right panel) between
1985 and 1996 for groups that were affected differently by the uncapping of family SSCs in 1989 and 1990. The figure is based on
a balanced panel of individuals who are observed throughout the period. The vertical lines denote the reform years, and earnings
levels in the last pre-reform year (1988) are normalized to 100 in all groups. The treatment-control assignment is based on the
reform-induced variation in average SSC rates. The treatment group includes individuals whose gross earnings in 1988 were 1 to
1.6 times the SS earnings cap that year. These workers experienced an increase in their average SSC rate due to the reform. The
control group includes individuals whose gross earnings in 1988 were 0.8 to 1 times the SS cap that year. These individuals did not
experience a change in their average SSC rate due to the reform.
Sources: DADS Panel 2010; TAXIPP 0.4.
26
Figure 4: Graphical Evidence on Earnings Responses to the Increase in Pension Contributions for Top Executives (1991)
(a) Net Earnings
(b) Labour Cost
Notes: The figure shows the evolution of average real net earnings (left panel) and of average real labour cost (right panel) between
1986 and 1997 for groups that were affected differently by the increase in the pension SSCs of executives in 1991. The figure is based
on a balanced panel of executives who are observed throughout the period. The vertical line denotes the reform year, and earnings
levels in the last available pre-reform year (1989) are normalized to 100 in all groups. The treatment-control assignment is based on
the reform-induced variation in average SSC rates. The treatment group includes executives whose gross earnings in 1989 were 4 to
6 times the SS earnings cap that year. These individuals experienced an increase in their average SSC rate due to the reform. The
control group includes executives whose gross earnings in 1989 were 2.5 to 4 times the SS cap that year. These individuals did not
experience a change in their average SSC rate due to the reform.
Sources: DADS Panel 2010; TAXIPP 0.4.
Figure 5: Graphical Evidence on Earnings Responses to the Increase in Pension Contributions for Non-Executives (2000–2005)
(a) Net Earnings
(b) Labour Cost
Notes: The figure shows the evolution of average real net earnings (left panel) and of average real labour cost (right panel) between
1996 and 2008 for groups that were affected differently by the increase in the pension SSCs for non-executives between 2000 and 2005.
The figure is based on a balanced panel of non-executives who are observed throughout the period. The vertical lines denote the first
and final reform years, and earnings levels in the last pre-reform year (1999) are normalized to 100 in all groups. The treatment-control
assignment is based on the reform-induced variation in average SSC rates. The treatment group includes non-executives whose gross
earnings in 1999 were 1 to 1.6 times the SS earnings cap that year. These individuals experienced an increase in their average SSC
rate due to the reform. The control group includes non-executives whose gross earnings in 1999 were 0.8 to 1 times the SS cap that
year. These individuals did not experience a change in their average SSC rate due to the reform.
Sources: DADS Panel 2010; TAXIPP 0.4.
27
Figure 6: Uncapping of Health Care SSCs: Log Differences in Net Earnings and Labour
Cost between Treatment and Control Groups
(a) Log Difference in Net Earnings
(b) Log Difference in Labour Cost
Notes: The figure shows the estimated log differences in real net earnings (left panel) and in real labour cost (right panel) between
groups of individuals who were affected differently by the uncapping of health care SSCs in 1981 and 1983. The difference-in-difference
estimation is performed on a balanced panel of individuals who are observed throughout the period 1977–1988. The treatment-control
assignment is based on the reform-induced variation in average SSC rates. The treatment group includes individuals whose gross
earnings in 1980 were in the range of 1 to 1.6 times the SS earnings cap that year. These workers experienced an increase in their
average SSC rate due to the reform. The control group includes individuals whose gross earnings in 1980 were in the range of 0.8 to 1
times the SS cap that year. These individuals did not experience a change in their average SSC rate due to the reform. The vertical
line denotes the last pre-reform year (1980). The solid lines show the parameter estimates on the interaction between the treatment
group and year dummies, which are normalized to zero in 1980. The dashed lines show the 95 percent confidence intervals around
the estimates.
Sources: DADS Panel 2010; TAXIPP 0.4.
Figure 7: Uncapping of Health Care SSCs: Estimated Employer Share of Incidence
(a) Log Difference in SSC Rates (First Stage) (b) Employer Share of Incidence (2SLS)
Notes: The figure shows the estimated share of the increase in SSCs that was borne by employers following the uncapping of health
care SSCs in 1981 and 1983, over a period up to 8 years after the reform. The estimation is performed on a balanced panel of
individuals who are observed throughout the period 1977–1988 and uses a difference-in-difference estimator comparing groups of
individuals who were affected differently by the reform. The treatment-control assignment is based on the reform-induced variation
in average SSC rates. The treatment group includes individuals whose gross earnings in 1980 were in the range of 1 to 1.6 times the
SS earnings cap that year. These workers experienced an increase in their average SSC rate due to the reform. The control group
includes individuals whose gross earnings in 1980 were in the range of 0.8 to 1 times the SS cap that year. These individuals did not
experience a change in their average SSC rate due to the reform. The first stage estimates (reform’s impact on log(1+SSC rate)) are
shown in the left panel and are obtained from the interaction between the treatment group and year dummies, which is normalized
to zero in 1980. The estimated employer share of the incidence is shown in the right panel and is obtained from a 2SLS regression of
log(labour cost) on log(1+SSC rate) where log(1+SSC rate) is instrumented with the interaction between the treatment group and
year dummies. The solid lines show the estimates while the dashed lines show the 95 percent confidence intervals.
Sources: DADS Panel 2010; TAXIPP 0.4.
28
Figure 8: Uncapping of Family SSCs: Log Differences in Net Earnings and Labour Cost
between Treatment and Control Groups
(a) Log Difference in Net Earnings
(b) Log Difference in Labour Cost
Notes: The figure shows the estimated log differences in real net earnings (left panel) and in real labour cost (right panel) between
groups of individuals who were affected differently by the uncapping of family SSCs in 1989 and 1990. The difference-in-difference
estimation is performed on a balanced panel of individuals who are observed throughout the period 1985–1996. The treatment-control
assignment is based on the reform-induced variation in average SSC rates. The treatment group includes individuals whose gross
earnings in 1980 were in the range of 1 to 1.6 times the SS earnings cap that year. These workers experienced an increase in their
average SSC rate due to the reform. The control group includes individuals whose gross earnings in 1980 were in the range of 0.8 to 1
times the SS cap that year. These individuals did not experience a change in their average SSC rate due to the reform. The vertical
line denotes the last pre-reform year (1988). The solid lines show the parameter estimates on the interaction between the treatment
group and year dummies, which are normalized to zero in 1988. The dashed lines show the 95 percent confidence intervals around
the estimates.
Sources: DADS Panel 2010; TAXIPP 0.4.
Figure 9: Uncapping of Family SSCs: Estimated Employer Share of Incidence
(a) Log Difference in SSC Rates (First Stage) (b) Employer Share of Incidence (2SLS)
Notes: The figure shows the estimated share of the increase in SSCs that was borne by employers following the uncapping of family
SSCs in 1989 and 1990, over a period up to 8 years after the reform. The estimation is performed on a balanced panel of individuals
who are observed throughout the period 1985–1996 and uses a difference-in-difference estimator comparing groups of individuals who
were affected differently by the reform. The treatment-control assignment is based on the reform-induced variation in average SSC
rates. The treatment group includes individuals whose gross earnings in 1988 were in the range of 1 to 1.6 times the SS earnings cap
that year. These workers experienced an increase in their average SSC rate due to the reform. The control group includes individuals
whose gross earnings in 1988 were in the range of 0.8 to 1 times the SS cap that year. These individuals did not experience a change in
their average SSC rate due to the reform. The first stage estimates (reform’s impact on log(1+SSC rate)) are shown in the left panel
and are obtained from the interaction between the treatment group and year dummies, normalized to zero in 1988. The estimated
employer share of the incidence is shown in the right panel and is obtained from a 2SLS regression of log(labour cost) on log(1+SSC
rate) where log(1+SSC rate) is instrumented with the interaction between the treatment group and year dummies. The solid lines
show the estimates while the dashed lines show the 95 percent confidence intervals.
Sources: DADS Panel 2010; TAXIPP 0.4.
29
Figure 10: Increase in Pensions Contributions for non-Executives (2000–2005): Log Differences in Net Earnings and Labour Cost between Treatment and Control Groups
(a) Log Difference in Net Earnings
(b) Log Difference in Labour Cost
Notes: The figure shows the estimated dynamic effects of the increase in pensions SSCs for non executives between
2000 and 2005 on net earnings and labour costs, based on a difference-in-differences estimation that compares two
groups of non executives: those with earnings in 1980 between 0.8 and 1.6 of the SS earnings cap (treatment group)
and those with earnings between 0.8 and 1.0 of the SST (control group). The solid lines show the parameter
estimates on the interaction between the treatment group and post-reform year dummies using a strongly balanced
panel of executives. The dashed lines show the 95 percent confidence intervals around the estimates.
Sources: DADS Panel 2010; TAXIPP 0.4.
Figure 11: Increase in Pensions Contributions for non-Executives (2000–2005): Estimated
Employer Share of Incidence
(a) Log Difference in SSC Rates (First Stage) (b) Employer Share of Incidence (2SLS)
Notes: The figure shows the estimated share of the increase in SSCs that was borne by employers following the increase in pensions
SSCs for non-executives between 2000 and 2005, over a period up to 8 years after the reform. The estimation is performed on
a balanced panel of individuals who are observed throughout the period 1996–2007 and uses a difference-in-difference estimator
comparing groups of individuals who were affected differently by the reform. The treatment-control assignment is based on the
reform-induced variation in average SSC rates. The treatment group includes individuals whose gross earnings in 1999 were in the
range of 1 to 1.6 times the SS earnings cap that year. These workers experienced an increase in their average SSC rate due to the
reform. The control group includes individuals whose gross earnings in 1999 were in the range of 0.8 to 1 times the SS cap that
year. These individuals did not experience a change in their average SSC rate due to the reform. The first stage estimates (reform’s
impact on log(1+SSC rate)) are shown in the left panel and are obtained from the interactions between of treatment group and year
dummies, normalized to zero in 1999. The estimated employer share of the incidence is shown in the right panel and is obtained
from a 2SLS regression of log(labour cost) on log(1+SSC rate) where log(1+SSC rate) is instrumented with the interaction between
the treatment group and year dummies. The solid lines show the estimates while the dashed lines show the 95 percent confidence
intervals.
Sources: DADS Panel 2010; TAXIPP 0.4.
30
Figure 12: SSCs Reforms: Change in Net Earnings and Labour Cost around the Social
Security Threshold
(a) Uncapping of Health Care SSCs (1981 and 1983)
Base Year + 8: 1988 vs. 1980
Base Year + 4: 1984 vs. 1980
(b) Uncapping of Family SSCs (1989 and 1990)
Base Year + 4: 1992 vs. 1988
Base Year + 8: 1996 vs. 1988
(c) Increase in Pensions SSCs for Non-Executives (2000–2005)
Base Year + 5: 2004 vs. 1999
Base Year + 8: 2007 vs. 1999
Notes: The figure examines the earnings responses to SSCs reforms by plotting average changes in real net earnings (blue triangles)
and average changes in real labour cost (red circles) against base year gross earnings relative to the Social security earnings cap
(denoted by vertical line), using bins of width 0.04. Three SSCs reforms are considered (i) the uncapping of health care SSCs in
1981 and 1983 (Panel A); (ii) the uncapping of family SSCs in 1989 and 1990 (Panel B); and (iii) the increase in pensions SSCs for
non-executives between 2000 and 2005. Changes in earnings are measured in absolute terms relatively to the pre-reform base year,
based on the sample of workers with earnings in the range of 0.8 to 1.6 times the SS cap in the base year and observed with earnings
in the end year. Panel A shows changes in earnings four years and eight years after the first uncapping of health care SSCs, by
comparing executives’ and non executives’ earnings in 1980 to their post-reform levels in 1984 and 1988. Panel B shows changes in
earnings four years and eight years after the first uncapping of family SSCs, by comparing executives’ and non executives’ earnnings
in 1988 to their post-reform levels in 1992 and 1996. Panel C shows changes in earnings five years and eight years after the first
increase in pensions SSCs for non-executives, by comparing executives’ earnnings in 1999 to their post-reform levels in 2004 and 2007.
Sources: DADS Panel 2010; TAXIPP 0.4.
31
Table 1: Change in Marginal SSC Rates Before and After each Reform
Employer SSCs
Employee SSCs
Reform 1: Uncapping of health care SSCs (1981 and 1983)
Under SST 1 to 3 SST
1980
1984
Difference
38.1
39.0
0.9
10.2
19.7
9.6
Difference
Under SST
1 to 3 SST
Difference
−28.0
−19.3
8.7
12.8
15.2
2.4
8.1
9.7
1.6
−4.7
−5.5
−0.8
Difference
Under SST
1 to 3 SST
Difference
−19.0
−8.0
11.0
17.0
17.3
0.3
10.9
11.3
0.4
−6.1
−6.0
0.1
Reform 2: Uncapping of family SSCs (1989 and 1990)
Under SST 1 to 3 SST
1988
1991
Difference
39.2
36.3
−2.9
20.2
28.4
8.2
Reform 3: Increase in contributory pension SSCs – executives only (1991)
1988
1991
Difference
1 to 4 SST
4 to 8 SST
Difference
1 to 4 SST
4 to 8 SST
Difference
22.7
31.3
8.5
12.6
27.9
15.3
−10.1
−3.4
6.8
11.2
11.7
0.5
6.0
8.8
2.8
−5.2
−2.9
2.3
Reform 4: increase in contributory pension SSCs – non-executives only (2000–2005)
Under SST 1 to 3 SST
1999
2005
Difference
38.9
39.1
0.2
30.8
38.5
7.7
Difference
Under SST
1 to 3 SST
Difference
−8.1
−0.6
7.5
13.4
13.6
0.2
7.5
12.2
4.7
−6.0
−1.5
4.5
Notes: Marginal tax rates are here expressed as a percentage of gross earnings, as they are legislated. These rates are applied
to different fraction of earnings, defined with respect to the Social Security threshold (SST). For reforms 1 and 2, we represent
only the non-executive rate, the difference being similar for the executives.
Sources: IPP tax and benefit tables (April 2015); TAXIPP 0.4.
32
Table 2: Dynamic Effects of SSCs Reforms on Earnings: Difference-in-Differences Estimates
First stage
Model:
Dependent Variable:
Estimated Employer
Share of Incidence
(2SLS)
Reduced Form
Log(1+SSC rate)
Log(net earnings)
Log(labour cost)
Log(labour cost)
(1)
(2)
(3)
(4)
Panel A. Uncapping of Health Care SSCs (1981 and 1983)
Base Year (T0 ): 1980
T0 +2
T0 +3
T0 +4
T0 +5
T0 +6
T0 +7
T0 +8
Number of observations
0.0138***
(0.0004)
n/a
n/a
0.0193***
(0.0004)
0.0202***
(0.0004)
0.0205***
(0.0004)
0.0195***
(0.0004)
0.0184***
(0.0004)
−0.0022
(0.0031)
n/a
n/a
−0.0100***
(0.0031)
−0.0081**
(0.0031)
−0.0073**
(0.0031)
−0.0094***
(0.0031)
−0.0097***
(0.0031)
0.0117***
(0.0029)
n/a
n/a
0.0093***
(0.0029)
0.0122***
(0.0029)
0.0132***
(0.0029)
0.0102***
(0.0029)
0.0087***
(0.0029)
0.844***
(0.224)
n/a
n/a
0.481***
(0.156)
0.602***
(0.150)
0.643***
(0.149)
0.519***
(0.155)
0.475***
(0.164)
338,700
338,700
338,700
338,700
Panel B. Uncapping of Family SSCs (1989 and 1990)
Base Year (T0 ): 1988
T0 +2
T0 +3
T0 +4
T0 +5
T0 +6
T0 +7
T0 +8
Number of observations
n/a
n/a
0.0118***
(0.0005)
0.0120***
(0.0005)
0.0118***
(0.0005)
0.0126***
(0.0005)
0.0137***
(0.0005)
0.0155***
(0.0005)
n/a
n/a
0.0031
(0.0029)
0.0047
(0.0029)
0.0043
(0.0029)
−0.0006
(0.0029)
−0.0041
(0.0029)
−0.0062**
(0.0029)
n/a
n/a
0.0150***
(0.0029)
0.0167***
(0.0029)
0.0162***
(0.0029)
0.0121***
(0.0029)
0.0096***
(0.0029)
0.0093***
(0.0029)
n/a
n/a
1.268***
(0.252)
1.389***
(0.250)
1.368***
(0.254)
0.955***
(0.233)
0.701***
(0.213)
0.602***
(0.188)
318,681
318,681
318,681
318,681
Panel C. Increase in Pensions Contributions for non-Executives (2000–2005)
Base Year (T0 ): 1999
T0 +2
T0 +3
T0 +4
T0 +5
T0 +6
T0 +7
T0 +8
Number of observations
0.0031***
(0.0006)
0.0059***
(0.0006)
0.0045***
(0.0005)
0.0052***
(0.0006)
0.0061***
(0.0005)
0.0066***
(0.0006)
0.0065***
(0.0006)
0.0028
(0.0028)
0.0053*
(0.0028)
0.0031
(0.0028)
0.0048*
(0.0028)
0.0000
(0.0028)
0.0020
(0.0028)
0.0002
(0.0028)
0.0059**
(0.0028)
0.0112***
(0.0028)
0.0076***
(0.0028)
0.0100***
(0.0028)
0.0061**
(0.0028)
0.0087***
(0.0028)
0.0067**
(0.0028)
1.886**
(0.898)
1.912***
(0.482)
1.688***
(0.624)
1.924***
(0.543)
1.001**
(0.455)
1.306***
(0.419)
1.035**
(0.427)
290,563
290,563
290,563
290,563
Notes: Each panel corresponds to a different SSCs reform: (i) the uncapping of health care SSCs in 1981 and 1983 (Panel A);
(ii) the uncapping of family SScs in 1989 and 1990 (Panel B); and (iii) the increase in pensions SScs for non-executives between
2000 and 2005 (Panel C). Estimates are obtained from a difference-in-difference specification that compares individuals whose
earnings in the pre-reform year were in the range of 1 to 1.6 times the SS cap (treatment group) with individuals whose earnings
in the pre-reform year were in the range of 0.8 to 1 times the SS cap (control group). Each panel uses a balanced sample of
individuals who are observed throughout the relevant 11-year period of study (1977–1988 for Panel A; 1985–1996 for Panel B;
1996-2007 for Panel C). The first stage and reduced form estimates are obtained from the interaction between the treatment
group and year dummies. The estimated employer share of the incidence is shown in the right panel and is obtained from a
2SLS regression of log(labour cost) on log(1+SSC rate) where log(1+SSC rate) is instrumented with the interaction between
the treatment group and year dummies. Standard errors are in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
Sources: DADS Panel 2010; TAXIPP 0.4.
33
Table 3: Dynamic Effects of SSCs Reforms on Earnings: Regression Kink Estimates
Model:
Dependent Variable:
First stage
Estimated Employer
Share of Incidence
(2SLS)
Reduced Form
∆ SSCs
∆ Net earnings
∆ Labour cost
∆ Labour cost
(1)
(2)
(3)
(4)
Panel A. Uncapping of Health Care SSCs (1981 and 1983)
Base Year (T0 ): 1980
T0 +4
T0 +5
T0 +6
T0 +7
T0 +8
4,910***
(396)
5,909***
(349)
6,439***
(533)
6,814***
(441)
7,466***
(486)
288
(1,251)
1,392**
(675)
1,321
(1,651)
2,403***
(854)
3,712***
(928)
5,198***
(1,618)
7,301***
(1,012)
7,760***
(2,152)
9,218***
(1,279)
11,178***
(1,399)
1.059***
(0.251)
1.236***
(0.101)
1.205***
(0.241)
1.353***
(0.104)
1.497***
(0.0940)
Panel B. Uncapping of Family SSCs (1989 and 1990)
Base Year (T0 ): 1988
T0 +4
T0 +5
T0 +6
T0 +7
T0 +8
6,277***
(448)
5,603***
(526)
6,021***
(626)
6,777***
(647)
6,920***
(716)
2,408***
(731)
1,000
(804)
1,156
(892)
1,642*
(934)
1,373
(998)
8,686***
(1,173)
6,603***
(1,319)
7,177***
(1,505)
8,419***
(1,568)
8,294***
(1,701)
Panel C. Increase in Pensions Contributions for non-Executives (2000–2005)
Base Year (T0 ): 1999
3,795***
−411.5
3,384**
T0 +5
(666)
(831)
(1,491)
4,591***
−616.6
3,975*
T0 +6
(864)
(1,255)
(2,104)
4,466***
−506.9
3,959**
T0 +7
(838)
(1,041)
(1,873)
4,057***
−1,022
3,035
T0 +8
(958)
(1,214)
(2,164)
1.384***
(0.0899)
1.179***
(0.127)
1.192***
(0.129)
1.242***
(0.116)
1.198***
(0.125)
0.892**
(0.238)
0.866***
(0.298)
0.886***
(0.254)
0.748**
(0.358)
Notes: Each panel corresponds to a different SSCs reform: (i) the uncapping of health care SSCs in 1981 and 1983 (Panel A);
(ii) the uncapping of family SSCs in 1989 and 1990 (Panel B); and (iii) the increase in pensions SSCs for non-executives between
2000 and 2005 (Panel C). The first stage and reduced form estimates are obtained from a local linear unconstrained regression
kink design (RKD) estimator, i.e., without imposing continuity in the conditional expectation of earnings at the SST. The
assignment variable is base year gross earnings relative to the SST. The dependent variables are the absolute change in real
SSCs (column 1), the absolute change in real net earnings (column 2) and the absolute change in real labour cost (column 3)
between the end year and the base pre-reform year. The bandwidth for the RK estimate is 0.2. Each row corresponds to a
different regression using the sample of individuals with earnings in the range of 0.8 to 1.2 times the SST in the base year
and observed with earnings in the end year. The estimated employer share of the incidence is shown in the right panel and is
obtained from a fuzzy RKD estimator which uses the estimated kink in ∆ SSCs (first stage) to scale the estimated kink in ∆
labour cost (reduced form). Standard errors are in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
Sources: DADS Panel 2010; TAXIPP 0.4.
34
Download