Download Displaced workers and the effects of outplacement and

Displaced workers and the effects of
outplacement and severance pay
Wiljan van den Berge
Master Thesis Economics
Tilburg University
No. of words: 17,149 (excl. tables and figures)
Date of defense: August 29, 2013
Supervisors:
prof. dr. ir. J.C. van Ours (Tilburg University)
dr. M. de Graaf-Zijl (Centraal Planbureau)
Third reader:
dr. M.A. van Tuijl (Tilburg University)
Displaced workers and the effects of outplacement and severance
pay
By Wiljan van den Berge
I use a dataset with social plans from firms from the Netherlands covering
8,751 displaced workers to estimate the effects of outplacement and severance
pay on a displaced worker’s probability of moving from job to job, the hazard
rate and the quality of the subsequent job. I find that both severance pay and
outplacement are not associated with a higher average probability of moving
from job to job. For those who do experience an unemployment spell, I also
find that outplacement is not associated with a higher exit rate on average.
In addition I find (i) that outplacement seems to be most effective for elderly
workers; (ii) that severance pay is associated with a slightly higher exit rate out
of unemployment on average, which likely indicates a selection effect, and (iii)
little evidence of effects of either outplacement or severance pay on subsequent
job quality.
Governments frequently provide job-search support to unemployed workers, but a different
form of support, job-to-job support, is starting to play a larger role in many job markets.
A prominent form of job-to-job support is outplacement, generally offered by private outplacement agencies to workers who are about to lose their job or who are looking for another
job. Outplacement aims to help workers who have been given notice in the transition from
job-to-job without an intervening spell of unemployment. While outplacement is rare in some
countries – e.g. France and Spain – it is very common in others, such as the Netherlands and
Belgium. Nevertheless, little is known about how these services operate and even less about
how effective they are.
In this study I attempt to answer the question of the effectiveness of outplacement by
examining its effect on unemployment duration and subsequent job quality. Since there is no
data available from outplacement agencies, I draw upon a sample of social plans agreed upon
between labor unions and firms facing mass layoffs in the Netherlands. These firms often offer
outplacement services and financial arrangements – such as severance pay – to workers who
will be laid off. These arrangements are detailed in social plans. The major advantage of using
this sample is that I have information on the entire package that workers were offered when
given notice, so that I can take into account the effects of elements other than outplacement,
most notably severance pay. Hence, this sample will allow me to estimate the effects of both
outplacement and severance pay on the unemployment duration and subsequent job quality
of workers who are laid off. In addition, since I have data on the firms involved in these plans,
I will examine whether there is a relation between several firm and workforce characteristics
and the contents of a social plan.
In section I I start with a discussion of what outplacement exactly consists of and the role
it plays within the Netherlands and some other European countries. I continue to discuss
the institutional context of terminations in the Netherlands. In section II I conceptualize the
effects of outplacement and severance pay in a simple job-search model. I also discuss the
results of some more complex models. In section III I look at the existing empirical literature
on outplacement. However, since there is very little empirical research on outplacement as
such, I mainly draw upon the literature on active labor market policies. Some of these policies,
such as job search assistance, are very close to outplacement services, so evaluations of these
2
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
3
policies can provide some insights into what we might expect of outplacement. In section IV
I present my sample of social plans, firm characteristics and the sample of displaced workers.
In section V I present my empirical strategy. My first step is a descriptive analysis of the
contents of social plans and their relation with several firm characteristics. Then I continue
to the main focus of my thesis and estimate the effects of outplacement and severance pay
on the probability of moving from job to job, the hazard rate, unemployment duration and
subsequent job quality. Finally, in section VI I discuss my results.
At the firm level I find no significant relation between firm characteristics, such as whether
a firm pays relatively high wages or the median age of the workers in a firm, and the content
of its social plans. At the individual level I find that both outplacement and severance pay are
not associated with a higher probability of moving from job to job, except for older workers.
For those who do experience an unemployment spell, I also find that outplacement is not
associated with a higher exit rate. When I restrict my sample to specific subgroups, I find
that outplacement seems to be most effective for elderly workers, with the estimates being
largest when the sample is restricted to 55 years and older. For younger workers I find no,
or even negative, effects of outplacement. For severance pay I find that it is associated with
a slightly higher exit rate out of unemployment. Other measures of severance pay, both in
levels and in relative indicators, confirm that indeed just receiving severance pay is associated
with a shorter unemployment duration, which likely indicates a selection effect. Finally, I find
little evidence of effects of either outplacement or severance pay on subsequent job quality,
except for a positive effect of outplacement on the wages of those younger than 35.
4
I.
Institutional context of outplacement
In this section I outline the institutional context of outplacement, starting with a discussion of a typical outplacement program and the role outplacement plays in labor markets in
the Netherlands and other European countries. Since the data that I use consist of social
plans and the associated displaced workers, I also discuss the institutional structure of Dutch
termination law and the role of outplacement and severance pay in the case of mass layoffs.
A.
What is outplacement?
Outplacement is a combination of services offered by private companies that help a worker
who has been given notice find a new job as quickly as possible. Outplacement is aimed at
helping people transition from their current job to a new job, without an intervening spell of
unemployment. In this respect outplacement is different from the services frequently offered
by public employment services, since they aim at helping people who are already unemployed
make the transition to a job.
An outplacement program typically consists of the following elements
• Psychological support aimed at helping the employee cope with being displaced.
• Self-evaluation aimed at understanding what the employee wants and what he or she
can do.
• Simple training, such as short (one or two day) courses on for example the use of
computer software, or in applying for a job.
• Starting to apply for a job, where the initiative is with the job-seeker, but where the
outplacement bureau offers support in job search and with applying (e.g. De Cuyper
et al. (2008)).
Note that the outplacement agency generally doesn’t try to match a job-seeker with a
specific employer and, apart from some simple courses, outplacement also doesn’t try to
retrain the employee. The goal is rather to support the displaced worker and prepare her for
the labor market, so that she is able to find a fitting job as quickly as possible.
Outplacement agencies are usually hired by firms who have to lay off some workers, be
it through individual layoffs or mass layoffs. The outplacement program typically starts a
couple of months before the worker is actually displaced, so that it can help with a smooth
transition to a new job. Nevertheless, despite that workers are often not technically displaced
yet when they are offered outplacement, I will refer to them as displaced workers, since they
most certainly will lose their current job.
While a firm may for example offer six months of outplacement to a displaced worker, this
doesn’t mean the worker can’t quit the program earlier if she finds a job. However, if she
doesn’t find a job, she is generally expected to use up the full six months.
Since I only have information on social plans drawn up in the case of mass layoffs, I restrict
my attention to the role of outplacement in mass layoffs, but this is in principle no different
from the role of outplacement in individual layoffs.
B.
Outplacement and public job-to-job support within Europe
While outplacement is widely used in the Netherlands, and is also quite common in the
United Kingdom and Germany, only Belgium has made it a systematic part of their policy.1
1 It is also common with corporate layoffs in the USA. A survey for The Wall Street Journal estimates that more
than two-thirds of 265 employers with layoffs from 2007 to 2009 used outplacement services, at an average cost of $3,589
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
5
The Netherlands
The current government, in collaboration with labor unions and employers’ unions, has
made it a priority to support people in job-to-job transitions. But the Dutch labor market
has had a relatively high level of job-to-job support for quite some time. Some of this consists
of specifically tailored programs by the government. For example in 2010 the government set
up nine experiments in different regions and sectors to provide job-to-job support (see e.g.
Borghouts-van de Pas (2012, 230-2)).
Apart from such government programs the private market for outplacement is substantial.
A 2011 survey by the trade organization for companies that deal with various types of job and
job-transition support estimates that outplacement bureaus had a gross turnover of about 35
million euros, with most of it coming from firms that hire outplacement services for supporting
their displaced workers. The survey also estimates that about 9,300 outplacement programs
were started in 2011 (OVAL, 2012).
Large firms, such as the National Railways (NS), have setup their own so-called mobility
center. These centers are aimed at helping displaced workers find a fitting job and fulfill
a role similar to private outplacement programs. Such centers are only possible for larger
firms though. The government has therefore helped to setup 33 temporary mobility centers
throughout the Netherlands in 2008 and 2009 to help smaller firms facing layoffs find new
jobs for their employees (Borghouts-van de Pas, 2012, 228-229).2
Belgium
Belgium is the only country where outplacement is a systematic part of public policy. As
of December 2007 employers must offer outplacement to every displaced worker 45 years or
older. The fine for not complying is 1,800 euros.3 In 2009 this obligation was extended to all
employees. Displaced workers are also obliged to accept the outplacement offer, unless they
are able to find a job within 14 days.
Unfortunately, there has been no treatment evaluation study into outplacement in Belgium.
Jacobs and De Cuyper (2013) are the only ones who have looked at gross effects of outplacement on the probability of finding a job. They sent out a survey to people who were displaced
in 2010 and asked whether they found a job after outplacement and whether they still had a
job in June 2012. They found that around 60% found a job after outplacement and 70% of
them still held that job in 2012, whereas 20% already went to another job. They also looked
at whether age affects the probability of finding a job and found a significant negative effect
for people aged 50 and older compared to people between 45 and 50.
Spain
Outplacement services are only officially allowed in Spain since 2010. However, before that
time some already operated as a so-called “service-sector company”. Private outplacement is
only used by large multinational firms and is mainly aimed at highly-educated males in the
age range of 35 – 45 (Borghouts-van de Pas, 2012, 206). Arellano (2007, 2009) examines the
effects of one large outplacement bureau in Spain. He looks at approximately 500 workers who
received outplacement services in the period 1998 – 2003 and compares these, via a matching
per employee (The Wall Street Journal, 2009).
2 Note that these shouldn’t affect my results, since I will only be looking at the years 2003 to 2007.
3 Interestingly, the average price for outplacement services has been driven down to the same level as the fine. Part of
the reason is the increased competition following the new laws, but another reason is that firms would otherwise simply
prefer to pay the fine (De Cuyper et al., 2008, 52).
6
procedure, with similar workers who did not receive outplacement services. He finds, perhaps
surprisingly, a general increase in unemployment for those in outplacement.
Arellano (2007) explains this by a reservation wage effect – workers in outplacement have
higher wage demands than those not in outplacement (also see section II). In a 2009 study
he examines the same sample and indeed finds that wages are higher for those receiving
outplacement services. Individual outplacement increases annual earnings by 33,500 euros
and group outplacement by 8,400 euros. For men the effects are larger: a 36,500 annual
increase for individual outplacement and 12,000 for group outplacement. For women the
effects are 10,300 and 5,600 respectively (Arellano, 2009).
Spain does not have a similar public outplacement service. When facing unavoidable layoffs,
unions typically aim at achieving the highest level of severance pay possible or try to have
workers leave through early retirement (Borghouts-van de Pas, 2012, 200-9).
Austria
Outplacement in Austria primarily consists of public foundations known as “labor foundations” (Arbeitsstiftungen). Private outplacement services are rarely used, and if they are, only
by large multinational companies. The first labor foundation was established in 1987 during
the large-scale privatization and scaling down of the steel industry. The idea was to help employees towards new employment, so that both employees and employers could benefit from
a more or less smooth transition to a smaller, private steel industry. At first 12 companies
joined, but it grew quite quickly to 28 firms in 1988 and 58 firms in 1998 (Winter-Ebmer,
2001). The success of this labor foundation was quickly copied, and sector-, region-, and
company-specific labor foundations arose (Borghouts-van de Pas, 2012, 160).
The main goal of labor foundations is to help displaced workers to a new job. A typical
program for a labor foundation starts with trainees receiving help with professional orientation: what are they able to do and what do they want to do. This typically takes about six
weeks. About 75% of trainees continues into training of some kind, which can range from
simple courses to full-fledged education at schools or universities. Finally, trainees are actively supported with job search. In 2009 trainees stayed in the foundation for an average of
295 days (Borghouts-van de Pas, 2012, 183). So except for training, labor foundations offer
similar programs as most outplacement agencies do.
Some studies have looked at the effectiveness of the services offered by labor foundations.
Gross effects show that about 67% of trainees are fully or mostly active in the labor market
one year after leaving a labor foundation. Winter-Ebmer (2001) evaluates the first labor
foundation for the steel industry. He finds that, on average, employment of trainees in a year
is 45 days longer than for the control group that didn’t join the labor foundation.
He also uses an instrumental variables (IV) strategy, because there could be endogeneity
problems relating to the trainees. It is likely that trainees are those people who expect to
have lower labor market prospects and that people who don’t join a foundation will typically
have better prospects. This would lead to an underestimation of the effect. And indeed,
he finds that when using the IV strategy those that followed the foundation’s services have
employment that is up to 70 to 80 days higher than for those that didn’t join the foundation.4
4 Winter-Ebmer (2001) also estimates effects on income and finds that income increases about 5 percentage points,
and about 6 to 7 percentage points with the IV models. These should be interpreted with caution though, since only
monthly income data is used and it could be that people simply work more rather than earn a higher hourly wage.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
7
United Kingdom
In the UK private outplacement is quite common for large firms. It is mostly used for
senior, high-skilled employees, since for low-skilled employees the costs are too high. One of
the reasons is that the government started offering a similar service for displaced workers in
2002, the so-called “Rapid Response Service” (RRS). The RRS was setup in addition to the
regular public employment service – the Jobcentre Plus – which offers counseling, job-search
assistance and training for the long-term unemployed (3 – 6 months). The RRS essentially
offers those that are about to be displaced a fast track into the Jobcentre services. In addition
to counseling, the Jobcentre also has the largest vacancy database in the UK, so matching
between employers and employees also plays an important role.
Unfortunately, there are no evaluations of this policy, since the government argues that it
would be too expensive to collect the data (see Borghouts-van de Pas (2012, 118-145) for
more details on the UK policy).
Sweden
Private outplacement services were prohibited in Sweden until 1993, and are still rarely
used. One of the reasons is that, as with Austria and in a lesser respect the UK, public
transition foundations take up the role of arranging job-to-job transitions. These foundations
are sector-based, with early ones already established in the 1970s. For example, in 1974
the Trade & Industry foundation for white-collar workers was established, which currently
covers approximately 700,000 employees. Since 2012, when the municipalities setup their own
transition foundation, about 75% of Swedish workers with a permanent contract are covered.
Transition foundations are public, but without a role for the government. Rather, they are
run by labor unions and employers. They provide support to displaced workers that is very
similar to that provided by private outplacement providers.
Again, due to data limitations there are no evaluations that allow us to say anything about
the causal effects of these programs. Some studies suggest that about 70 – 80% of displaced
workers find a job within a year, and unemployment is primarily a factor for those who
(re)enter the labor market (Borghouts-van de Pas, 2012, 98-102).
Now that we have some idea of what outplacement consists of and the role it plays within
various labor markets in Europe, I will focus on the Dutch institutional background for the
remainder of this section. My data derives from mass layoffs accompanied by a social plan in
the Netherlands, so it is important to consider the system of Dutch termination law.
C.
Institutional background: Dutch termination law
The Dutch labor market combines quite rigid job protection with a remarkable good performance on unemployment. Employment protection legislation (EPL) – with an average of
2.23 on the OECD scale from zero to six – is right around the OECD average of 2.25. But this
masks some strong variety in protection. Permanent workers are quite strongly protected, at
2.73, with an OECD average of 2.26. Protection of temporary workers, on the other hand,
scores a 1.42, with an average of 2.10. Mass layoff rules, which only cover permanent workers
in the Netherlands, are again tougher, at 3.00, with an average of 2.58. While the share of
Dutch workers with a permanent contract has declined, in 2008 still almost 56% of Dutch
workers held a permanent contract (compared to 65% in 2000) (OECD, 2008).
This might suggest that the Dutch labor market is quite rigid, and as a consequence should
suffer from higher unemployment rates. However, from 2002 to 2012 the average unemploy-
8
ment rate was only 4.2 percent. The EU average over this period was 8.9 percent and the
Euro area average was 9.2 percent. The Netherlands in fact had the lowest unemployment
rate within the EU and only Norway had a lower unemployment rate within Europe, at 3.5
percent (Eurostat, 2012). So let us look behind the OECD numbers and consider the system
of Dutch termination law in detail.
In the Netherlands every planned termination of an employee by an employer has to be
requested beforehand. The employer can choose between three possible routes:
1) Request the cantonal judge to end the labor agreement (art. 7:685 BW, 1639w oud
BW).
2) Request a license to terminate at the public employment service UWV (art. 6 BBA).
3) End the labor agreement by mutual consent.
If a request is rejected by the cantonal judge the employer can always ask UWV, and vice
versa. The cantonal judge and UWV routes are most common and used about just as much.
Figure 1a shows the division in terminations between UWV and the cantonal judge routes
over time.
Ending by mutual consent has become more popular as of late. For one, it allows the
employer to circumvent the rules related to mass layoffs. Such agreements are typically
associated with a high level of severance pay. Another reason is that the blameworthiness
criterion for receiving unemployment benefits has been changed in 2006. Whereas before
an employee would be blameworthy if he agreed on ending the labor agreement by mutual
consent and hence not receive unemployment benefits, this is no longer necessarily the case.
It is however unclear how often this route is used, since it is not registered as such, hence
why it is not included in Figure 1a.
There are some important differences between the cantonal judge and UWV routes:
• UWV checks beforehand whether a termination complies with all regulations, whereas
the cantonal judge determines this after the fact.
• The cantonal judge can decide on severance pay, whereas UWV can’t. However, many
of the terminations through UWV are mass layoffs and in such cases severance pay is
often part of the social plan.
• It is common, although not necessary, that UWV primarily examines terminations for
economic reasons (layoffs), whereas the cantonal judge looks at non-economic reasons
for terminations. In 2009 84% of the requests at UWV were for economic reasons, with
66% being for individual and 18% for mass layoffs. Figure 1b shows the development
in reasons for terminations via UWV route over time. We see an increase in individual
layoffs for economic reasons and a decrease in other reasons. This is mostly due to a
decrease in disability-related terminations caused by more stringent rules for disability
benefits.
Employers will typically choose the cheapest route for terminations. There are three principle components that have to be considered:
• The time before the request is granted; at UWV this typically takes about 4 – 6 weeks,
and then there is at least a one month notice period. The cantonal judge has no predetermined period of notice, this will be determined by the judge on an individual
basis.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
Figure 1. : Different routes for terminations in the Netherlands.
Number of terminations (x 1,000)
180
160
140
120
100
80
60
40
20
0
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
Year
UWV
Cantonal judge
Total
Percentage of terminations via UWV route
(a) Terminations by UWV and cantonal judge routes.
100%
90%
80%
70%
60%
50%
40%
30%
20%
10%
0%
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
Year
individual
collective
other
(b) Reasons for termination via UWV route.
Source: Jaarraportage Ontslagstatistiek 2001, 2003, 2005, 2007 and 2009.
9
10
• Severance pay; the cantonal judge can allocate severance pay. Ending by mutual consent
also typically includes a (large) severance pay allocated by the employer.
• The chance of being rejected; estimates are that about 2% of the requests at the cantonal
judge are rejected and about 8% of those at UWV (Frenk and Pfann, 2009, 117-8).5
Severance pay
In the case of individual terminations, severance pay can only be assigned by the cantonal
judge. A specific formula for determining the level of severance pay was introduced in 1997.
This is the so-called cantonal judge formula (kantonrechtersformule, henceforth KRF). The
formula consists of three elements that determine the level of severance pay.
A. weighted number of years of tenure with the firm.
B. gross monthly wage.
C. correction factor.
These factors are multiplied to arrive at the level of severance pay. Every year of tenure
until the age of 40 has a weight of 1, every year from 40 to 50 has a weight of 1.5 and every
year later than age 50 has a weight of 2.6
The correction factor essentially takes into account who is at fault in the termination. A
neutral factor is simply C = 1. But if the judge finds that the employee is at fault, he can
decide on a lower factor or even a factor of 0. If the employer is to blame, the factor can be
larger than 1, sometimes even going up to 2. This means severance pay has similar effects as
a layoff tax, because employers have to internalize part of the costs of unnecessary turnover.
Severance pay is taxed as regular income, with the top bracket being 52% in the Netherlands, which starts to apply at 56,000 euros. Dismissed workers have the option of setting
their severance pay aside, which will generally reduce the tax burden because it can either
be paid out over time or paid at a time when workers earn less income, so their severance
pay will not be taxed at the top marginal tax rate. The usefulness of the option depends on
the amount the worker gets. If the amount is high – exceeding around 60,000 euros – it is
typically beneficial to set the money aside. If the amount is lower, it can typically be paid
out without moving into the upper tax bracket, since, for income tax purposes, the amount
can be spread out over three years, leaving workers with only a 20,000 euro income hike for
three years, which will often keep them under the top bracket. The option of setting aside
the money is used frequently, although exact numbers are unavailable. It can be very useful
if a laid off worker finds a job quickly or has enough of a buffer to not face any liquidity
constraints (see section II).
Mass layoffs
Since 1976 there is a special law for mass layoffs (wet melding collectief ontslag). A mass
layoff is defined as laying off at least 20 employees within a period of three months for economic
5 Opponents argue that the dual system of UWV and cantonal judge routes is confusing and inefficient. However,
others claim that the system offers flexibility to the employer. One could think that, because both routes are used just
as frequently, that the employer is indifferent between the two. However, it is probably the case that the costs depend
very much on the type of termination and the type of employee and that this determines the choice of termination
(Frenk and Pfann, 2009).
6 Since 2009 these weights have been lowered, with every year until 35 weighted at 0.5, every year from 35 to 45 at
1, every year from 45 to 55 at 1.5 and every year older than 55 at 2. This doesn’t matter for my purposes, as almost all
social plans from 2009 and 2010 still use the old formula. Also, for the analysis on individual unemployment durations
I will only be using data from 2003 to 2007.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
11
reasons.7 The employer has to present advance notice to both the relevant labor unions and
UWV. This notification, apart from informing of the layoffs and the reason for them, is also
meant to examine possibilities to prevent layoffs or soften the possible consequences through
a social plan.
Employers are not free to determine who will be laid off. From 1976 to 2006 tenure was
leading in determining who would be laid off, with those who came in last being the first to
leave. However, since this provided strong protection for insiders at the expense of outsiders
(women, young people, people with foreign backgrounds), the government decided that a
different system was required. In March 2006 the reflection principle (afspiegelingsbeginsel )
became the norm. The goal is to have a good reflection of the personnel file in the case of
mass layoffs. Conditional upon similar positions, employees are divided in five age brackets
(15 - 24, 25 - 34, 35 - 44 and 55 - 64 years) and the number of employees that will be laid
off from each bracket is dependent on the number of employees the firm has in each bracket.
Finally, tenure considerations are applied to these brackets, so that those who came in last
are the first to leave in each age bracket.
Employers do have some leeway, since they can deviate from the reflection principle if the
person to be laid off has a smaller chance of finding a job than a colleague, if the person is
indispensable or if the person is seconded at a different company and that company doesn’t
want to lose him or her.
Mass layoffs should be checked in advance by UWV. However, if the involved unions confirm that the layoffs are indeed due to economic reasons, UWV will only check whether the
reflection principle is applied correctly and whether there are possibilities for replacement.
Note that UWV only checks whether all the rules of mass layoffs have been applied correctly.
The actual act of laying off the employees can still be through either cantonal judge or UWV
routes.
While the cantonal judge and UWV routes are still most popular, ending agreements by
mutual consent has become more popular because it offers employers a way around the stringent rules of mass layoffs – i.e. the reflection principle and tenure rules –, since they don’t
have to notify UWV. This flexibility comes at a cost however, since ending by mutual consent
usually involves a relatively high level of severance pay.
Social plans
In the context of mass layoffs, employers and labor unions often decide on a social plan to
deal with the consequences for displaced workers. Social plans for larger firms are typically
around twenty pages and contain the agreements on the possibilities for internal replacement,
severance pay, outplacement, training and other financial and non-financial arrangements that
compensate the employee for losing her open-ended contract.
While one might expect that these plans offer the displaced worker a choice in whether she
prefers a more generous financial arrangement or more support in the form of outplacement
or training, this is typically not the case. Severance pay is generally allocated to each worker
according to the kantonrechtersformule (KRF), while training and outplacement, if they are
offered, are usually on a voluntary basis for each displaced worker.
In deciding on the level of severance pay, the correction factor C is the main bargaining
tool. Considerations in determining the correction factor in a social plan are how much the
firm spends on other arrangements, such as outplacement, but mostly whether the firm has
a lot of money to spend, rather than who is to blame for the layoff. Obviously, if the firm
7 An economic reason can be a bad financial situation, a reorganization, ending business activities, disappearance of
some jobs because of automation or a move of some business units.
12
is facing bankruptcy, severance pay will be a lot lower than when a firm is part of a large
corporation with good prospects.
While a social plan is not legally compulsory in the case of mass layoffs, many collective
labor agreements do oblige it. Aside from these agreements, the most important reason for
employers to draw up a social plan is that employees who are laid off without a social plan can
challenge the layoff at the cantonal judge. The judge will typically allocate a higher level of
severance pay if employers have decided to lay workers off without a social plan. To prevent
this from happening, employers want to have a social plan that the labor unions agree with,
because this makes it effectively impossible for individual employees to challenge the layoff.
A judge will usually not deviate from the social plan if it was agreed upon with the unions.
Also, a social plan provides certainty for the employer regarding the costs of the layoff.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
II.
13
Theory
In this section I formalize the effects of outplacement and severance pay in a simple jobsearch model. I show that severance pay does not have an effect on the hazard rate, whereas
outplacement has an ambiguous effect. Then I continue to discuss some extensions of this
simple model in the recent literature that show that we can expect that outplacement will
increase the hazard rate, whereas severance pay tends to lower it.
A.
A simple job-search model
Let’s start with a very simple job-search model to get an idea of the effects of outplacement
and severance pay on the hazard rate and unemployment duration (see Mortensen (1986);
Rogerson, Shimer and Wright (2005); Boeri and van Ours (2008) for the elements of this
model). In the model workers search for a job in an uncertain world. During her search
a worker receives job offers according to a Poisson process with arrival rate µ. Each job
offer is a random drawing from a known wage offer distribution with cumulative distribution
function F (w). While being unemployed a worker receives benefits b. These benefits could be
unemployment benefits, but they could also refer to the value of home production or leisure.
If the worker accepts the offer, she will get the offered wage w forever. To decide whether the
worker should accept the job offer or continue searching, she must compare the present value
of remaining unemployed with the present value of accepting the current job offer. These can
be specified as follows. The flow value of accepting the current job offer is
(1)
ρVw = w
where ρ is the subjective time discount rate, Vw is the value of accepting the current job
offer and receiving w forever and w is the currently offered wage. The flow value of being
unemployed is
Z
(2)
∞
max{0, Vw − Vu } dF (w)
ρVu = b + µ
0
where b is the benefits and µ is the job offer arrival rate, which is multiplied with the expected
increase in the value of the job offer. This is 0 if the offer is rejected and Vw − Vu if the offer
is accepted.
Comparing the flow values of being employed and unemployed implies a reservation wage
wR where the worker will accept any job offer that exceeds it. Accepting is optimal if w ≥ ρVu
8 , so the reservation wage is
wR = ρVu
(3)
If we combine equations 2 and 3 we get
(4)
8I
µ
w =b+
ρ
R
Z
∞
(w − wR ) dF (w)
wR
assume that the worker will accept if she is indifferent.
14
Integrate by parts and we get:
µ
w =b+
ρ
R
(5)
Z
∞
[1 − F (w)] dw
wR
Equation 5 gives us the formulation of the reservation wage. The reservation wage is static.
The equation shows that an increase in the benefits of being unemployed, b, will increase the
reservation wage. We also see that an increase in µ will increase the reservation wage and
that an increase in ρ will decrease the reservation wage. But how does this affect the expected
unemployment duration?
The hazard rate is the rate at which workers leave unemployment. It is defined as
H = µ[1 − F (wR )]
(6)
The hazard rate is simply the rate at which job offers arrive, µ, multiplied by the probability
of accepting the offer. The probability that a worker who has been unemployed for a length
of t has still not found a job is then e−Ht .
At this point we can examine the effects of outplacement and severance pay on the hazard
rate. Outplacement can be conceptualized as having a positive effect on the job offer arrival
rate, because the worker can search more efficiently than without outplacement. This means
that outplacement will increase µ. This will have two effects on the hazard rate. By equation 6,
it will increase the hazard rate through the increase of µ. But it will also have a negative
effect on the hazard rate, since it raises the reservation wage wR . The intuitive reason is
that, with more job offers coming in, workers can be more ‘picky’ in deciding the job they
want, and this translates into a higher reservation wage. This is also clear from equation 5,
where an increase in µ will translate into a higher wR . The net effect of ∂H
∂µ is thus ambiguous
without a specification of F (w). However Van den Berg (1994) shows that for most of the
distributions that are frequently used for wage-offer distributions the sign of ∂H
∂µ is positive.
Severance pay, on the other hand, does not have an effect on the hazard rate. To see
this, simply note that severance pay increases both the value of being unemployed and the
value of being employed by the same fixed amount. It doesn’t affect the flow values of either
employment or unemployment.
B.
Outplacement in a more complex model
There has been no theoretical research on the effects of outplacement on unemployment
duration. However, a very much related literature is the one on active labor market policies
(ALMP). These policies generally consist of a mix of monitoring, counseling, job-search assistance and training aimed at bringing unemployed who receive unemployment benefits back
to work and reducing the disincentive effects of unemployment benefits. Some elements from
ALMP are close to the services offered by outplacement programs, in particular counseling and
job search assistance. But outplacement typically already starts before the the termination
of the labor contract, whereas ALMP focus on the unemployed. In addition, outplacement is
usually only offered to displaced employees, whereas ALMP focus on all unemployed workers.
Van den Berg and Van der Klaauw (2006) present a job-search model with two search
channels, formal and informal search, and examine the effects of two elements of ALMP:
counseling and monitoring. Formal search is search through the formal channels of the employment office and via advertisements on the Internet or in newspapers. Informal search is
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
15
through a job-seeker’s network. Counseling and monitoring only directly affect the formal
channel.
Van den Berg and Van der Klaauw (2006)’s model – in contrast to the simple model above
– endogenizes search intensity: workers can decide how much effort they put into search and
how to allocate this between formal and informal search. This means that the effect of an
increase in the job offer arrival rate µ that we found earlier is magnified. A higher job offer
arrival rate (more efficient search) means that a worker will put more effort into search, since
each unit of effort is more productive. But it will also mean that workers can be even more
picky with respect to the job they take, since there are even more offers due to the increased
search intensity. This is in line with the ambiguous result we had earlier. But because we
now have two search channels, there is an additional effect. Due to an increased efficiency in
formal job search, workers will now substitute effort away from the informal search to formal
search. It turns out that this leads to an overall positive effect on the hazard rate (see Van
den Berg and Van der Klaauw (2006, 904-6) for the details).9
The implication of this model for outplacement is that, via counseling and job-search assistance, it will lead to an unambiguously positive effect on the hazard rate. It will also, via
an increased reservation wage, lead to an increase in the average quality of the future job.10
C.
Severance pay and liquidity constraints
The result obtained above – that severance pay does not have an effect on the hazard rate
– is a special case. It only applies when job seekers face no liquidity constraints, or in other
words, when they are able to perfectly smooth their consumption over their life cycle. In any
other case job seekers who receive a (higher) severance pay could use that severance pay to
maintain a longer search period, or equivalently, the (higher) severance pay would raise their
reservation wage.
This idea requires a more complex model where the reservation wage is duration-dependent
and declines as time without a job continues. Card, Chetty and Weber (2007) present a jobsearch model with liquidity constraints. They take search intensity as endogenous, but the
basic idea is the same as with an endogenous reservation wage. Both imply that job seekers
who face a binding liquidity constraint will have a higher hazard rate than those who face a
non-binding constraint, but Card et al assume that this is due to a higher search intensity
rather than a lower reservation wage. They show that if job seekers perfectly smooth their
consumption, severance pay has no effect, just as we saw above. But if job seekers face
liquidity constraints or they reduce their consumption while unemployed to maintain some
level of savings, an increase in the level of assets a job seeker has at her disposal will reduce
her search intensity. The intuitive mechanism is that severance pay initially increases the
level of assets, relieving the unemployed from their liquidity constraint. This allows workers
to be more selective in their search, but as their search time goes on their buffer will continue
to fall until she faces a binding liquidity constraint again (see Card, Chetty and Weber (2007,
1516-9) for the details).
Note that dismissed workers are in principle free to use their severance pay as they like.
Some might prefer to save their severance pay for their pension for example. These workers
don’t face liquidity constraints, for example because their partner has a well-paying job, or
9 Van den Berg and Van der Klaauw (2006) also look at monitoring. They argue that monitoring might have a
negative effect on the hazard rate since it sub-optimally diverts attention away from the informal search channel. If this
channel is used very little however, monitoring could have a positive effect. However, since monitoring plays no major
role in outplacement services, we can safely discard these effects.
10 Hujer, Thomsen and Zeiss (2006) arrive at similar conclusions with regard to the search assistance component of
ALMP, but they also include a training component that is usually absent from outplacement programs.
16
because they have a large buffer stock of savings to get through the search period. For these
workers severance pay will have no effect on their hazard rate.
The implications of this model for severance pay are that it has either none, or a negative
effect on the hazard rate. Severance pay could also increase the quality of the future job,
because it increases the reservation wage.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
III.
17
Empirical literature: Outplacement and severance pay
Only a few studies have examined the effects of outplacement, and actual causal evaluations are even rarer.11 For this reason, I cast my net a little wider and also consider ALMP
evaluations, especially focusing on those policies where job search assistance plays an important role. Finally, I go into some research on the effects of severance pay on unemployment
duration.
A.
Outplacement
Arellano (2007) examines the effects of outplacement in Spain, where it is generally only
used for those with high-earning jobs. He looks at approximately 500 workers who received
outplacement services from a large outplacement bureau in the period 1998 – 2003 and compares these, via a matching procedure, with similar workers who did not receive outplacement
services. He finds a general increase in unemployment for those in outplacement, possibly
due to a reservation wage effect. In a follow-up study he indeed finds that wages are higher
for those receiving outplacement services. Individual outplacement increases annual earnings
by 33,500 euros and group outplacement by 8,400 euros. For men the effects are larger: a
36,500 annual increase for individual outplacement and 12,000 for group outplacement. For
women the effects are 10,300 and 5,600 respectively (Arellano, 2009).
Jacobs and De Cuyper (2013) have looked at gross effects of outplacement on the probability
of finding a job in Belgium. They sent out a survey to people who were displaced in 2010 and
asked whether they found a job after outplacement and whether they still had a job in June
2012. They found that around 60% found a job after outplacement and 70% of them still
held that job in 2012, whereas 20% already went to another job. It remains unclear whether
they also would have found a job without outplacement.
B.
Active labor market policies
Empirical research on ALMP could be useful in thinking about the possible effectiveness of
outplacement, but we must keep in mind the differences between outplacement and ALMP
discussed earlier. In particular, outplacement typically only focuses on displaced workers and
tries to bridge the gap from job to job. ALMP on the other hand often focus on the long-term
unemployment. This could mean that the effects on unemployment duration of ALMP are
smaller than for outplacement because there might be important differences in skills between
people who are out of a job because they are displaced and people who were fired (Gibbons
and Katz, 1991). Also, the compulsory nature of ALMP adds a threat dimension which is
absent in outplacement. The positive effect of a given policy on unemployment duration could
be caused by the threat of the services, rather than the services themselves (Black et al., 2003;
Graversen and van Ours, 2008).
Card, Kluve and Weber (2010) present a meta-analysis of the literature on ALMP in different countries, including Europe as well as the US and Canada. They find that job-search assistance programs have overall weakly positive effects on the probability of being re-employed
or the unemployment duration. Direct public employment services are found to be less effective, whereas on-the-job training is particularly effective in the medium to long run (2 – 3
years).
Thomsen (2009) surveys the literature on job search assistance programs in nine European
countries. Overall he finds that most programs have either no or small positive effects on the
11 A major problem of course is that most of the data would have to come from private companies, who are probably
less willing to provide this data if it could show that their services have little effect.
18
re-employment probability. Some studies examine the hazard rate and they all find positive
effects of the varying programs on the hazard rates. It is hard to disentangle which elements of
the programs are most effective, but Thomsen argues that particularly programs that combine
some services, such as counseling and job search assistance, are most effective. So while the
results do suggest some small positive effects on the re-employment probability and hazard
rate of elements of ALMP also found in outplacement outplacement programs, we must be
careful in generalizing these results, as most programs also include such elements as close
monitoring of search activity and retraining.
One study specifically interesting for my purposes is Van den Berg and Van der Klaauw
(2006), who study a counseling and monitoring program in the Netherlands. This program
was aimed at people who just started collecting UI benefits between August and December
1998 in two large Dutch cities. The program ended in February 1999. The program had an
experimental setup, with about half of the inflowing unemployed being randomly assigned
to the treatment group. Only so-called Type I unemployed were assigned to the treatment
or control group. These are people who are thought to have sufficient skills to find a job
without retraining or education. This means that their labor market prospects are probably
comparable to those in outplacement programs. The treatment consisted of a meeting every
four weeks for six months, where progress on job applications and a plan for the coming four
weeks is discussed. Overall, they find no treatment effect of counseling and monitoring on
the hazard rate, although it could be that counseling has a positive effect and monitoring
negates this positive effect. They do find that counseling and monitoring has a larger positive
effect on the hazard rate for older people. In addition, they find that due to counseling and
monitoring people substitute search effort away from informal channels (Van den Berg and
Van der Klaauw, 2006, 921-6).
C.
Severance pay
Card, Chetty and Weber (2007) test the predictions of their search model that incorporates
liquidity constraints (see section II). The model predicts that severance pay will increase
search duration and possibly the quality of the new job. They use a sample of 650,992 job
losses in Austria. Severance pay in Austria depends on tenure – workers with at least three
years tenure will receive a severance pay of 2 months salary, whereas workers with less than
three years tenure will receive no severance pay.
This discontinuity lends itself for a regression-discontinuity design to estimate the causal
effects of severance pay on search duration. They find that, depending on the exact specification, severance pay reduces job-finding hazards in the first 20 weeks by 9.4% to 13.2%.
They also include the eligibility for unemployment benefits and find that this decreases the
job-finding hazards by 6.4% to 9.3%. These results suggest that, in contrast to the basic
search model where workers can perfectly smooth their consumption, severance pay can have
a substantial effect on search duration, possibly even a stronger effect than unemployment
benefits.12
Uusitalo and Verho (2010) examine a policy change in Finnish unemployment insurance,
where severance pay was replaced by higher unemployment benefits. They find that the
introduction of unemployment benefits reduced the exit rate to unemployment. They also
find that most of this reduction was due to a higher daily allowance rather than the loss of
severance pay. Although these results are less precise due to small sample sizes, this suggests
12 The role of liquidity constraints is also important with unemployment benefits. Chetty (2008) estimates that around
60% of the increase in unemployment duration due to unemployment benefits is due to liquidity constraints, rather than
disincentive effects.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
19
that severance pay has a small effect on re-employment rates.
To recap, previous evaluations of the causal effects of private outplacement services are
rare. The only existing evaluation is for the Spanish market, where outplacement is aimed
specifically at the high-educated, where they find a negative effect on unemployment duration,
but a positive effect on subsequent wages, indicating an increase in the reservation wage caused
by outplacement. Studies on the gross effects of both private and public forms of outplacement
are more common, since this practice is quite widespread in Europe. The results suggest
positive effects, but it is unclear whether this is due to these programs or whether these
workers would have found a job anyway. Studies on job search assistance in ALMP suggest
either no or small positive effects on unemployment duration, but the threat effect of these
policies makes it again hard to determine whether this is in fact due to the effectiveness of
assistance. Finally, recent studies on severance pay suggest that liquidity constraints play an
important role, indicating that severance pay will generally increase unemployment duration.
20
IV.
Data description
To estimate the effects of outplacement and severance pay on unemployment duration,
I use data gathered from social plans that contain the services firms offer their displaced
workers (see section I.C). I connect these data to the firms involved in these plans and finally
use registration data from Statistics Netherlands (CBS) to gather information regarding the
displaced workers involved in these plans.
The advantage of using only data on displaced workers is that there is no relation between
unobserved characteristics of these workers, such as ability, and the fact that they were
displaced, whereas such a relation would probably exist for a sample containing all dismissed
workers (Gibbons and Katz, 1991). As discussed before, mass layoffs in the Netherlands are
primarily based on people’s age and tenure, rather than their productivity.
This advantage could be undermined by the growing popularity of ending labor agreements
by mutual consent due to relaxing of the blameworthiness criterion for unemployment benefits
(see section I.C). A disadvantage of the data is that endings by mutual consent are not
identified as such, as they are either included with all other mass layoffs or not included at
all. This could introduce selection bias, although the problem shouldn’t affect my sample too
much, since the blameworthiness criterion was only relaxed in 2006. It nevertheless remains
important that it becomes clear how often this route is used and whether it really affects
some particular group with poor labor market prospects (e.g. older workers with a low level
of education), as this could seriously undermine not only research, but also the intentions
behind the rules on mass layoffs.
The advantage of using data on social plans is that I have an idea of the comprehensive
package that people were offered when displaced. I know for example whether they received
severance pay in addition to outplacement, so that I can try to disentangle the effects of
severance pay and outplacement support on unemployment duration.
A major disadvantage is that we don’t know whether workers who were offered outplacement
support actually took up the offer. This means that I can only estimate the “intention to
treat”, rather than an average treatment effect for outplacement (see section V). Furthermore,
since I don’t know what other workers were offered when dismissed, I can only rely on variation
within my sample of social plans. This severely limits my possible control group.
My sample consists of 552 social plans from 2002 to 2010 from a database from the Ministry
of Social Affairs and Employment.13 This database only contains social plans that were
registered as collective labor agreements, whereby they apply to all displaced workers for a
firm. This means that it doesn’t contain all social plans agreed upon in this period. For
example EIM (2008) estimates that between 2004 and 2007 1,130 social plans were agreed
upon. A representative from the largest labor union in the Netherlands confirmed that each
year there are about 1,500 – 2,000 social plans. Most of these (around 1,000) are very small
and only contain minimal arrangements. Labor unions are not involved with these, as they
are agreed upon between the employer and the works council. Then there are about 350 –
500 plans where the labor union is involved, but that are not registered as collective labor
agreements. These are more substantial, but are what the representative referred to as “batch
production”. Finally we have the social plans from the database, which are more substantial
– usually around twenty pages, but sometimes over a hundred pages long – and tailored to
the situation at a specific firm.
13 Originally the database contained 672 social plans from 2002 to 2010 involving displacements. However, it proved
impossible to connect every plan with data on firm- and displaced worker-characteristics. Section A.A1 explains how
the sample is constructed and checks for significant differences between the full sample, the restricted sample from 2002
to 2010 and the restricted sample covering 2003 to 2007.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
21
Figure 2. : Venn diagram showing the sub-population covered by my dataset.
Laid off workers in NL
Laid off workers
in registration
data (CBS)
Social
plans
Sample
I connect this sample of social plans to registration data from Statistics Netherlands on the
firms involved. My first step is analyzing the relation between several firm characteristics and
the contents of social plans. For this analysis I gathered data on the employees of these firms
through registration data and aggregate these at the firm level.
However, the main focus of my thesis is on the individual-level effects of outplacement
and severance pay. I use a special dataset from Statistics Netherlands containing all workers
laid off for economic reasons for the years 2003 – 2007 for whom a request to lay off was
granted by the UWV and whose job actually ended (BEONTTAB). In addition to the pure
UWV layoffs, Statistics Netherlands also includes those who were involved in mass layoffs.
They use a statistical method to determine whether people were involved in a mass layoff. In
months with an outflow that is substantially larger than an average month, all job endings
are included, whether they went through the UWV, cantonal judge or by mutual consent.
However, mass layoffs are frequently characterized by an outflow that stretches over a longer
period – where the later months don’t have an outflow significantly above average – so not
necessarily all workers involved in a mass layoff are included (see the discussion below in
section IV.C).
Figure 2 shows the sub-population covered by my dataset of social plans and individual
displaced workers. My sample includes a subsample of all social plans in this period and
a subsample of all laid off workers in this period. The subsamples are non-random, which
means that it could introduce selection bias in my sample. In the remainder of this section I
will discuss the extent of this selection bias by comparing my sample of social plans to earlier
ones in section IV.A. In section IV.B I compare several firm characteristics of my sample
to the general population of Dutch firms. In section IV.C I present descriptive statistics on
displaced workers covered by the social plans in my sample for the period 2003 to 2007.
A.
Social plans
Table 1 details the contents of the social plans in my sample. The three most important
elements of a social plan are usually severance pay, outplacement support and arrangements
for early retirement or a way to bridge the period to (early) pension schemes. We see that
the large majority of firms offer severance pay, with most of them being related to the kantonrechtersformule (KRF). Other financial arrangements include providing displaced workers
with an incentive to leave quickly or providing workers who were not displaced with an incentive to replace a displaced worker. We see that 62% of the firms offer outplacement support
22
in their plans, with an average length of eight months. Other support includes possibilities
for getting a new job within the same firm or getting training. Finally, a substantial share of
the plans offer specific arrangements for the elderly – usually starting from around fifty-eight
years old – in the form of a financial bridge to their pension or early retirement schemes. In
2006 the government made it fiscally less attractive to quit working at an early age, so there
has been a shift towards providing more job-to-job support for the elderly in recent years.
To get a sense of the representativeness of my sample, Table 1 also compares my sample
with earlier overviews of social plans from different periods. While these are difficult to
compare, since the sample sizes and periods are different, we do see that most of the contents
are similar. I’m particularly interested in severance pay and outplacement and we can see
that for severance pay my sample is roughly in between the sample by EIM (2008) and Tros,
Rayer and Verhulp (2005). For outplacement my sample is similar to the one used by Tros,
Rayer and Verhulp (2005). Overall the samples seem to match quite well.14
B.
Description: firm characteristics
Table 2 shows some firm characteristics for the firms in my sample and compares them to
all Dutch firms and all Dutch firms with more than 20 employees, as only layoffs involving 20
or more employees are considered mass layoffs. The sample size is somewhat smaller than in
Table 1 because I was unable to find enough workers for all firms (see section A.A1).
Two things stand out in the table. First, the firms are relatively large, with about half
of the firms in my sample having more than 250 employees, whereas in the Netherlands the
large majority of firms have only one or two employees. Second, almost half of the firms
in my sample are in manufacturing, whereas this only holds for around 5% of the firms in
the Netherlands and 16% of the firms with more than twenty employees. It is clear that my
sample is not representative of firms in the Netherlands.
This could mean that it could be difficult to generalize any conclusions about the effects of
outplacement and severance pay to the total population of laid off workers in the Netherlands.
One reason why this might not be such a problem however, is that firms with mass layoffs and
an accompanying social plan are usually quite large, so that the total population of workers
involved in social plans could in fact be relatively similar. For example, EIM (2010) finds
in a recent study (not reported in Table 1) that in their sample of small and medium-sized
companies in the Netherlands, only 9% offered outplacement to laid off workers in 2010.
My sample in Table 1 on the other hand shows that 62% of the examined social plans offer
outplacement and that this is quite similar to what other studies of social plans found.
C.
Description: individual characteristics
Table 3 shows summary statistics for my sample of 8,751 displaced workers. The dataset on
displaced workers is linked – through information on the job people were dismissed from – to
my sample of social plans. I was able to find information on dismissed workers for 148 social
plans. This means that for 146 social plans no individuals could be found. There could be
several reasons for why some firms with a social plan in the period 2003 – 2007 are not found
in the database of dismissed workers. For one, it could be that workers were laid off already
before 2003 or only after 2007. Second, it could be that ultimately no one had to be laid off,
since a firm was able to internally replace everyone. Finally, it could be that the employer
14 A clear difference between my sample and the others is the difference in percentages of training between EIM and
my sample. This is because I’ve only included training aimed at finding an external job, whereas EIM also includes
training aimed at getting another job within the organization. See section A.A2 for an elaborate description of all the
variables.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
23
Table 1—: Percentage of social plans in my sample containing specific arrangements compared
with those of plans used in other recent studies.
Contents of social
plans (%)
Van den
Berge
Boos et
al (2001)
Tros et
al (2005)
EIM
(2008)
Financial
Severance pay
77
83
C=1
C<1
C>1
Unrelated to KRF
40
35
13
12
31
30
12
27
Incentive to leave
Supplement to UB
Incentive to replace
45
18
24
40
37
62
42
67
64
31
43
44
68
74
7.6
3 - 30
3,742
8.1
9
3 - 24
67
9
24
21
78
17
29
26
78
18
65
34
30
Job-to-job
Outplacement
Duration (avg months)
Duration (min - max)
Budget (avg e)
Internal replacement
Mobility centre
Training
Income suppletion
47
31
Arrangements for
the elderly
Bridge to pension
More job-to-job support
N
Year
38
11
552
2002 - 2010
45
105
2000
145
2004 - 2005
198
2004 - 2007
Source: Sample of social plans gathered from database of Ministry of Social Affairs
and Employment and the studies of Boos, Nagelkerke and Serail (2001); Tros, Rayer
and Verhulp (2005); EIM (2008).
24
Table 2—: Firm characteristics
Firm characteristics (% of
all firms)
Sample
Netherlands (all
firms)
Netherlands (firms
with > 20
employees)
Firm size (no. of workers)
1
2
3-5
5 - 10
10 - 20
20 - 50
50 - 100
100 - 150
150 - 200
200 - 250
250 - 500
500 - 1,000
1,000 - 2,000
> 2,000
1.0
6.3
16.0
12.9
9.5
5.5
20.4
13.5
6.7
8.1
68.6
13.3
7.3
5.2
2.7
1.7
0.6
0.2
0.1
0.1
0.1
0.1
0.0
0.0
57.5
20.5
7.3
3.8
2.4
4.5
2.5
1.2
0.8
Manufacturing
Wholesale and retail trade
Other specialized business services
Transport and storage
Information and communication
Financial institutions
Health and social work activities
Others
46.3
10.9
8.6
5.1
3.6
3.6
3.4
18.5
4.5
17.2
18.0
2.7
4.7
6.1
4.6
42.2
15.8
20.5
8.3
5.9
3.7
1.6
7.1
37.1
N
505
1,124,405
32,955
Sectors
Source: Own calculations based on registration data from Statistics Netherlands (CBS).
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
25
took the cantonal judge route and didn’t have an outflow in a particular month which was
substantially larger than normal, and hence wasn’t registered by Statistics Netherlands. This
could raise some concerns about selection bias. However, Table A1 in the Appendix shows
that there are hardly any differences between the 148 social plans in the final sample and the
sample of 552 social plans for 2002 - 2010, suggesting that I still have a sample which is quite
representative of social plans for larger firms in the Netherlands.
My sample with displaced workers is linked to registration data to obtain information
on personal, household and firm characteristics.15 Unemployment duration is defined as the
difference between the start date of the new job and the end date of the previous job, measured
in days. I have job data until December 31, 2010, and displaced workers are followed until
a maximum of five years after their layoff (so three years for those laid off at the end of
2007). This should lead to relatively small censoring problems. Still, for 19.5% of the sample
I was unable to find a new job. This means that either they are still searching for a job or
they left the labor market. I assume that everyone who hasn’t found a job at age 65 leaves
the labor market at that point for reasons related to pension or health. This accounts for
2.7%. Through another dataset I was able to identify that some unemployed workers started
freelancing. These account for 1.2%. For the final 15.6% the search duration is set at their
respective maximum, so 1,827 days (five years) for those laid off before January 1, 2005 and
shorter periods for those laid off after that date. All workers for whom I couldn’t find a new
job are treated as censored, either at the point where they start freelancing, where they turn
65 or after December 31, 2010.
Particularly relevant in the context of outplacement is the number of people who find a job
that starts right after they lost their previous job. I’ve excluded people who had another job
more than two months before their current job ended (1,792 persons) and people who went
to work for the same company after losing their current job, for example through a take-over
or merger (4 persons). The result is that 25.7% of the full sample have an unemployment
spell of zero, and hence directly move from job to job.16
Table 3 shows that the comparison groups for outplacement and severance pay differ significantly in many respects. The group that received an offer of outplacement is substantially
larger than the group who did not receive such an offer and less likely to be female. They
have longer tenure and tend to be older. We also see large differences in sectors. Workers in
sectors with many relatively highly-educated jobs, such as specialized business services (e.g.
consulting) and information and communication tend to receive more offers of outplacement,
whereas workers in wholesale and retail trade receive fewer. In addition, larger firms are more
likely to offer outplacement.
The group who receives severance pay is also much larger than the group who doesn’t
receive severance pay. Mean unemployment spells vary widely, but suggest a shorter spell
length for those who received no severance pay, in line with expectations. In terms of personal
characteristics, the groups differ substantially, however. Those who received severance pay
are much more likely to be older and longer tenured. Their average wage is also almost twice
as high as for those who don’t receive severance pay. As with outplacement, sector seems
to matter a lot. Those in manufacturing, specialized business services and health-related
industries tend to receive severance pay, whereas those who don’t receive severance pay are
much more likely to be in retail and trade.
15 Unfortunately I have no information on education, since this type of information is only gathered through surveys
in the Netherlands. Only about one third of the sample has data on education and this would have restricted my sample
size too much.
16 If I extend the period where people can have two jobs to three or four months, this only adds 110 or 170 workers
to the sample. This indicates that most of those excluded are part-time workers who have held two or more jobs for a
long time.
26
Figure 3. : Kaplan-Meier survival curves for outplacement and severance pay.
Source: Own calculations based on registration data from Statistics Netherlands and the social plans in the database of
the Ministry of Social Affairs and Employment.
The differences suggest that those who don’t receive severance pay are typically those on the
bottom end of the labor market: young people who earn relatively little in jobs that require a
low level of education. For outplacement the differences are much less pronounced, but tend
to go in the same direction. These differences between comparison groups also make it difficult
to draw conclusions about differences in outcomes, as their personal characteristics are likely
also related to unobserved characteristics, such as education, ability and motivation.17
Unemployment duration
The two graphs in Figure 3 show the nonparametric Kaplan-Meier curves for outplacement
and severance pay. These curves show the outflow from unemployment over the duration of
the unemployment spell as it is found in the data. The figure suggests that those who received
an offer of outplacement typically experienced slightly longer spells than those who did not
and the same holds for those who received severance pay. The differences in survival curves
are small however and could be due to differences in personal characteristics.
17 Also note that the average wage in the first new job is substantially lower than the average wage in the previous
job. This is a common finding for displaced workers, see e.g. Jacobsen, LaLonde and Sullivan (1993).
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
27
Table 3—: Summary statistics of displaced workers.
Individual characteristics
(%)
Mean unemployment spell (days)
SD unemployment spell (days)
Full sample
Outplacement No outoffered
placement
offered
Severance
pay
No
severance
pay
478
667
500
684
418
616
490
672
436
647
54.8
25.7
19.5
54.3
25.3
20.4
56.3
26.8
17.0
53.5
26.0
20.5
59.4
24.7
15.8
33.1
26.6
23.6
16.6
150
2,369
1,809
37.6
78.9
54.5
82.2
32.9
26.0
23.2
18.0
155
2,447
1,886
35.4
78.4
53.3
81.7
33.8
28.4
24.8
12.9
135
2,157
1,607
43.5
80.0
49.4
83.4
24.4
29.9
27.4
18.3
168
2,631
1,974
33.8
80.4
53.6
82.4
64.4
15.0
10.1
10.5
86
1,431
1,253
50.9
73.4
57.6
81.2
28.2
2.4
32.0
13.1
11.9
9.7
3.1
26.7
2.2
27.6
16.2
15.3
9.5
2.5
32.4
3.1
44.1
4.6
2.7
10.3
2.8
32.6
2.7
18.5
16.7
15.0
11.1
3.4
12.8
1.3
80.5
0.0
0.9
4.5
0.0
< 50
50 to 99
100 to 149
150 to 199
200 to 249
250 to 499
500 to 999
1,000 to 1,999
More than 2,000
4.7
6.2
5.0
4.8
2.0
10.0
11.8
11.4
44.2
5.6
4.4
4.1
6.3
2.1
8.4
10.6
12.8
45.8
1.7
11.4
7.4
0.7
1.7
14.3
15.1
7.7
39.8
5.4
6.1
6.1
5.7
2.1
10.7
14.7
14.7
34.6
1.8
6.8
1.3
1.4
1.6
7.4
1.7
0.0
78.1
N
8,751
6,398
2,353
6,835
1,916
Exit route (%)
New job after unemployment spell
Job to job
Censored
Personal characteristics (%)
Age: < 35
35 – 45
45 – 55
≥ 55
Avg. tenure (months)
Avg. monthly wage in lost job (e)
Avg. monthly wage in new job (e)
Female
Partner
Children
Dutch
Sector (%)
Manufacturing
Construction
Wholesale and retail trade
Information and communication
Other specialized business services
Health and social work activities
Other
Firm size (nr. of employees, %)
Notes: I control for 13 sectors in the analyses, but many of these only include a very small percentage of workers,
so they are not reported here.
Source: Own calculations based on registration data from Statistics Netherlands (CBS).
28
V.
Estimation strategy
My first step is to estimate whether some firms are more likely to offer outplacement or
a relatively high level of severance pay in their social plans. Since employees, through their
labor unions, have a say in the contents of a social plan, it is interesting to see the results of
this bargaining process for different types of firms. This analysis will provide an informative
description of what firms and employees currently agree on. For example, do firms with more
older workers offer more outplacement or more severance pay?
My second step is the main focus of my thesis. I estimate the effects of outplacement and
severance pay on the probability of moving from job to job, the hazard rate for those who
experience an unemployment spell, the (log) wage in the new job and subsequent employment
history. Henceforth I will simply refer to the offer of outplacement or severance pay as the
assignment of treatment, indicated by Ti = {1, 0}. While estimations of the hazard will tell
us something about the effect of treatment on job finding rates and unemployment duration,
estimations of the subsequent wage and employment history tell us something about job
quality.
A.
Specification of hazard model
A limitation of the data is that I can’t identify who actually took up outplacement. The
social plan only indicates that someone received an offer of outplacement, but since these
offers are almost always on a voluntary basis, we don’t know whether the individual actually
took up that offer. This means that I can only estimate the effect of the offer, also known
as the “intention to treat”. However, because these type of programs always have to cope
with attrition, the “intention to treat” is of intrinsic interest, aside from giving an idea of
the average treatment effect. People do change their behavior in response to these services
and some people decide to join, whereas others don’t. The intention to treat gives us the
average effect on the behavior of all people who were offered treatment (Heckman, LaLonde
and Smith, 1999). This limitation doesn’t hold for severance pay, but unfortunately exact
numbers of the amount of severance pay are unavailable, so I will only examine the impact
of receiving severance pay.
To estimate the hazard (job finding) rate θi I use both a proportional hazards (PH) model
and a mixed proportional hazards (MPH) model, where I allow for unobserved heterogeneity
(Van den Berg, 2001). The employment hazard θ for individual i with characteristics Xi at
unemployment duration t in the MPH is
(7)
θ(t|Xi , ηi ) = λ(t) · exp(X0i β + Ti γ) · ηi .
where λ(t) is the time-varying baseline hazard function. Duration dependence is modeled
as a piecewise constant function of elapsed duration, specified as:

(8)
λ(t) = exp 
J
X

λj I(tj−1 ≤ t < tj )
j=1
I have a potentially long time period of five years of unemployment, but the transitions
towards employment decline rapidly after the first year. Based on eyeballing the Kaplan-Meier
estimates in Figure 3 I estimate the baseline hazard with J = 10 intervals. The first half of
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
29
the first year is estimated with constant hazards for each month (t1 = 30 . . . t5 = 150).18 For
the second half of the first year I assume constant quarterly hazards (t6 = 180, t7 = 270) and
for the second year I assume constant hazards for each half of the year (t8 = 360, t9 = 540).
Finally, I assume a constant interval after the first two years (t10 ≥ 720).19
The level of the baseline hazard is allowed to differ multiplicatively between individuals, as
indicated by the term exp(X0i β + Ti γ). All variables in Xi are time-invariant and measured at
the moment of layoff.20 Ti γ gives the individual effect of treatment on the hazard rate. Using
the exp(·) function ensures that the hazard rate is non-negative for all X0i β. In the MPH
model I allow for unobserved heterogeneity with ηi > 0, which I assume to be distributed as
gamma with mean 1 and variance σ 2 .21 Unobserved heterogeneity could be important, since
I can’t include variables such as educational levels and motivation.
I use the same models to estimate the effects of treatment on the duration of the first job. This
is an indication of both job and match quality. The interpretation of the model is reversed
though, since the hazard rate is in this case defined as the exit rate out of employment.
B.
Specification of the wage model
In addition to job duration, I also look at the effect of treatment on the wage in the new
job. As a baseline specification I use a simple (log) wage model to estimate the effect of
treatment on job quality:
(9)
(log)wi = α + X0i β + Ti γi + i .
where Xi is a vector of control variables containing the same controls as in the hazard model
specification. γi identifies the effect of interest and i is an individual-specific error term. A
problem for this simple model could be that some people don’t find a job, which means that
their wage is censored. If censoring is non-random, which it most likely is, this could lead
to selection bias. To account for this I also apply the Heckman correction (Heckman, 1979).
This approach takes two steps.
First, estimate the probability of finding a job using a probit model with all observations.
The binary choice model for having a job is
∗
y1i = 1{y1i
> 0}
with a latent index
∗
y1i
= X01i β1 + 1i
If the latent index exceeds 0, individual i decides to have a job. The estimates from the latent
index can be used to calculate the inverse Mills’ ratio
λ̂ =
18 For
φ(X01i β1 )
Φ(X01i β1 )
convenience, I assume months of 30 days.
have tried specifications with more intervals, but the resulting baseline hazard is similar.
20 These variables are age, sex, tenure and log monthly wage in the previous job, whether the workers has children
and a partner, ethnicity, sector, firm size in terms of number of workers and year and quarter of inflow. See Table 3.
21 In the PH model η = 1.
i
19 I
30
where φ is the probability density function of the standard normal distribution and Φ is its
cumulative distribution function.
Second, use λ̂ to estimate the model
(10)
(log)wi = α + X02i β2 + Ti γi + σ12 λ̂vi
using only the uncensored observations, where σ12 = cov(i , vi ). The error terms are assumed
to be jointly normal.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
31
Table 4—: Parameter estimates for probit model for the choice of offering outplacement and
severance pay on the firm level.
Median age
High wage relative to sector
Fraction of females
N
Outplacement
Severance pay
Early retirement
-0.0242
(0.0164)
-0.1490
(0.332)
-0.4180
(0.420)
477
-0.0127
(0.0196)
0.2410
(0.385)
0.1265
(0.506)
450
0.0512∗∗
(0.01752)
0.4009
(0.3220)
0.3073
(0.4601)
469
Notes: Robust standard errors in parentheses. Controls for sector are included. Significance levels: † : 10% ∗
: 5% ∗∗ : 1%.
Source: Own calculations using registration data from Statistics Netherlands on workers in companies with a
social plan, 6 months before the social plan goes into effect.
VI.
A.
Results
Firm characteristics and social plans
My first step is a descriptive analysis of the relation between several firm characteristics and
social plans. This should give us some idea of what firms and their workers, represented by
labor unions, agree on. For example, do firms with more older workers offer more outplacement? Or do firms who already pay a higher wage offer more generous arrangements when
they lay off their workers?22 The characteristics I use are median age, whether a firm pays a
high wage relative to its sector and the fraction of females in a firm. The idea is that these
factors are general enough so as to influence behavior at the firm level. I also control for year,
firm size and sector. Table 4 shows the results. We see that all indicators are insignificant,
which means that there is no apparent relation between firm characteristics and the offer of
outplacement or severance pay when controlling for sector.
To further examine this relation I divide my sample in four groups according to the generosity of the package they offered. In ascending order of generosity, the groups are
1) Firms who offer neither severance pay nor outplacement;
2) Firms who offer either low severance pay or outplacement;
3) Firms who offer only high severance pay or low severance pay and outplacement;
4) Firms who offer a high level of severance pay and outplacement.
I then ran an ordered logit model with the same set of variables as above. Figure 4 shows
the predicted probabilities for the combination of severance pay and outplacement offered
by firms in their social plan for the years 2002 to 2010. It distinguishes between firms that
pay high and firms that pay low wages relative to their sector (i.e. is their median wage
higher than the median wage in their sector) and stratifies by the median age in a firm. The
figure shows that firms who pay high wages relative to their sectors generally also offer more
generous packages when having to layoff workers than firms who pay relatively low wage
levels. Within wage levels, firms with more older workers tend to offer slightly less generous
packages. These results suggest some endogeneity at the firm level, where firms adapt their
social plan to the overall workforce. To further check for this I ran a probit analysis on
22 Technically, I aggregate at the level of the social plan, not at the level of an individual firm. Sometimes plans cover
multiple firms and in those cases I aggregate over the workers in multiple firms.
32
Figure 4. : Predicted probabilities to choose a specific combination of outplacement and
severance pay for firms who pay relatively low or relatively high wages, stratified by median
age of the workforce in a firm.
0%
20%
40%
60%
80%
100%
35
Low
wage
firms
40
45
50
35
High
wage
firms
40
45
50
Only low severance pay or only outplacement.
Low severance pay and outplacement or only high severance pay.
High severance pay and outplacement.
Source: Own calculations based on registration data from CBS and the social plans in the database of the Ministry of
Social Affairs and Employment.
whether firms offer an early retirement program and indeed found a significant, but small,
positive effect of median age (see Table 4)
B.
Individual effects: probability of moving from job to job
I use a probit model to estimate the probability of moving to another job right after losing
the current job. Since employees have to be given at least two months notice of a pending
layoff, they have time to search for another job. Outplacement services are usually already
offered during this period, so they could positively affect the probability of moving from
job-to-job. The prospect of receiving severance pay could negatively affect this probability,
because it will give workers more time to search. Finally, it is likely that the best workers
will find a job quicker, so an indicator of worker quality such as wage is expected to have a
positive effect.
Table 5 reports the results.23 It shows that, contrary to expectations, outplacement doesn’t
have a significant effect on the probability of moving from job to job. One reason could be
that the workers who do directly get another job are the best workers and their benefits from
an offer of outplacement are very small. The coefficient for severance pay shows that it has
no effect on the probability of moving from job to job. Severance pay would only have an
effect if it relieves liquidity constraints of workers, but for people who directly move through
to another job, this effect would be negligible, unless they have no buffer whatsoever. It
is therefore expected that these workers, like many Dutch workers, simply set aside their
severance pay and take it up at a later period when they earn less and hence have to pay less
tax (e.g. after retirement). I interpret wage as an indicator of unmeasured variables such as
education and ability and indeed it shows the expected significant positive effect.
The results also show that older workers, relative to those younger than 35 years, have a
significantly smaller probability of moving from job to job. The effect is largest for those aged
23 The missing parameter estimates for this model and the following analyses at the individual level can be found in
the Appendix in section A.A3.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
33
Table 5—: Parameter estimates for probit model on the probability of moving directly from
job to job.
Full model
Outplacement
Severance pay
0.0401
(0.0409)
0.0159
(0.0556)
Individual characteristics
Age: 35 – 45 years
Age: 45 – 55 years
Age: ≥ 55 years
ln(wage) in previous job
Tenure (months) / 10
Female
Partner
Children
Dutch
Constant
N
-0.2393∗∗
(0.0430)
-0.5175∗∗
(0.0505)
-1.3292∗∗
(0.0809)
0.1705∗∗
(0.0262)
-0.0118∗∗
(0.0019)
-0.1712∗∗
(0.0358)
0.0922∗
(0.0408)
0.0379
(0.0347)
0.1973∗∗
(0.0415)
-1.7675∗∗
(0.2407)
8,751
Notes: Robust standard errors in parentheses. Controls for year and
quarter of inflow, sector and firm size are included. Omitted categories:
Age: < 35 years. Significance levels: † : 10% ∗ : 5% ∗∗ : 1%.
Source: Own calculations using registration data from Statistics
Netherlands on displaced workers.
55 years or older (I will return to this effect below). All other personal characteristics have the
expected signs, with having dependents in the form of children and a partner positive effect
on the probability of moving from job to job, while being female implies a lower probability.
While these coefficients are interesting, they don’t yet tell us much about the actual probability of moving from job to job. Simulations show that the average male, aged 45 - 55 with
a partner and earning the median wage has a 23% chance of moving from job to job, whereas
for a similar female the probability is 16%.
Whereas there might be no average effects of outplacement or severance pay, there could be
effects for different subgroups. To check for this I have run the same model for different age
groups. Table 6 reports the estimates. The baseline estimates are included for comparison.
The results indicate that both outplacement and severance pay have no effect for those younger
than 55 years. On the other hand, for those older than 55 years both outplacement and
severance pay have a significant positive effect. However, this is relative to the very small
baseline probability for those older than 55. Without outplacement or severance pay, those
older than 55 have a 1.6% chance of moving from job to job, whereas with outplacement
this increases to 3.3% and with severance pay to 3.4%.24 So while their chances improve by
about 50%, the actual impact is still very small. Figure 5 shows the predicted probabilities
24 I
will return to this result on severance pay below, where I argue that it is likely the result of selection bias.
34
Table 6—: Parameter estimates for probit model on the probability of moving directly from
job to job for different age groups.
Outplacement
Severance pay
N
Baseline
< 35 years
35 – 45 years
45 – 55 years
≥ 55 years
0.0401
(0.0409)
0.0159
(0.0556)
8,751
-0.0070
(0.0838)
-0.0518
(0.1000)
2,900
-0.0452
(0.0728)
0.0489
(0.1053)
2,331
-0.0555
(0.0819)
0.1796
(0.1291)
2,068
0.3053†
(0.1654)
0.6377∗
(0.2477)
1,452
Notes: Robust standard errors in parentheses. Controls for year and quarter of inflow, sector and firm size are included.
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%.
Source: Own calculations using registration data from Statistics Netherlands on displaced workers.
Probability of moving from job to job
Figure 5. : Predicted probabilities of moving from job to job for males and females with
outplacement in different age categories.
0.4
0.3
0.2
0.1
0
< 35
35 - 45
45 - 55
≥ 55
Age
Outplacement
No Outplacement
Source: Own calculations based on registration data from CBS. The predicted probabilities are calculated at the
population averages of Xi for each age group based on the estimates in Table 6.
for different age group on the basis of the estimates in Table 6.25
C.
Unemployment duration: hazard estimates
The non-parametric Kaplan-Meier curves in Figure 3 show some indication of how outplacement and severance pay affect unemployment duration. However, to get a better sense
of the effects on both exit rates and unemployment duration and to account for duration
dependence, we need to model these effects.
Table 7 shows the results from a PH and an MPH model on the exit rates out of unemployment. There are two major effects from allowing for unobserved heterogeneity in the
MPH model. First, the coefficients on most individual characteristics are larger. Second,
the coefficients on duration dependence are smaller. These effects are expected, because the
individuals with the best chances of leaving unemployment (due to unobserved variables such
as ability) will leave first, leaving a population that is less likely to exit unemployment. The
PH model without unobserved heterogeneity doesn’t account for this, so the effect will go in
25 I have also run the same model for two wage groups: less than and more or equal to the in-sample median wage of
2,360 euros. In line with the baseline specification, I found no significant effects of either outplacement or severance pay
for most groups.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
35
the duration dependence parameters. The MPH model, on the other hand, separates this
“weeding out” effect from genuine duration dependence (Lancaster, 1979).
When examining the coefficients from the MPH model, we see that outplacement has no
significant effect on the exit rates. This is quite similar to what we saw in the Kaplan-Meier
estimates, which suggested a very small negative effect of outplacement. The coefficient on
severance pay provides no evidence for the idea that these unemployed workers face liquidity
constraints. Severance pay has a small positive effect on the hazard rate, although only
significant at the 10% level. A positive effect of severance pay is contrary to what we would
expect on the basis of theory or what previous studies have shown, so it likely indicates a
selection effect.
As we saw in Table 3, those who receive no severance pay are younger, earn less and
predominantly work in sectors that require less education, such as retail and trade. While I
control for most of these characteristics, they could be related to other characteristics that are
not controlled for, such as education, ability and motivation. If workers who receive severance
pay have a higher level of education, ability or motivation, this could mean they will find a
job sooner than those who receive no severance pay. This could hold even if severance pay
does in fact increase unemployment duration, which makes the estimate difficult to interpret
beyond that it indicates a selection effect.
All personal characteristics point to the expected effects. Age, relative to those younger
than 35, has a strong negative effect on the hazard rate. One likely reason for the strong
effects of age on both unemployment duration and the probability of moving from job to job
we saw above is that employers simply prefer to hire younger workers, for example because of
perceived productivity differences or because investments in them will have a larger potential
payoff. These are all valid economic reasons. However, part of the effect is likely due to
unobserved variables and selection effects. One such unobserved variable affecting the results
on age are early retirement schemes, which are quite common in social plans. I’ve tried to
control for these, but due to the multitude of different schemes, starting at different ages and
different possible effects, it is very difficult to interpret the results.26 Another reason is that
older workers will likely receive longer unemployment benefits.27 A third reason could be
that older workers on average have a lower level of education and tend to work more often in
declining occupations (Bosch and ter Weel, 2013).
Somewhat surprisingly, wage in the previous job, a proxy for worker quality, doesn’t seem to
have an effect on the hazard. Tenure has a small negative effect, which indicates that workers
who have worked at one firm for a long time find it more difficult to find a job. There could
be several reasons for this. First, they might have more firm-specific skills, making it harder
to find an appropriate match. Second, they could have received a larger amount of severance
pay, because this is partly related to tenure. Third, they have less or outdated experience on
the labor market, making it more difficult to look for appropriate vacancies and go through
the application process. Being female has a significant negative effect on the hazard, whereas
having children positively affects the hazard.
The estimates for duration dependence indicate that the exit rate declines as time progresses, which could for example be because workers who have been unemployed for some
26 In estimates where I include a variable for early retirement schemes in social plans I found mostly positive effects on
the hazard rate. This again likely indicates a selection effect, possibly relating to a correlation between the generosity
of a social plan and an employers’ other personnel-related policies.
27 Unemployment benefits last between 3 and 38 months, depending on how long a person has worked in her life. After
the expiration of regular unemployment benefits, most workers will go into means-tested welfare. However, workers who
were laid off after 50 are entitled to a different welfare system, which doesn’t take into account any property they might
have (including savings, such as severance pay). This could also lengthen their unemployment durations compared to
younger workers. Unfortunately I don’t have data on unemployment benefits or welfare.
36
time have a harder time to find a job, or that they have given up looking for a job, for
example via early retirement schemes.
While the hazard rates are important parameters, as they are implied by job search theory,
it is also relevant to look at the implied unemployment durations. Using the hazard rates,
we can estimate the median unemployment duration for different groups. For the median
male, between 45 and 55 years old and earning the median wage, the median unemployment
duration is about 450 days. For the median female, between 45 and 55 years old and earning
the median wage for females, the median unemployment duration is about 560 days.
While there might be no average effects of outplacement, it could be that, just as above,
some groups do experience significant effects. Table 8 reports the estimates of the same
MPH model as in Table 7 for several subgroups. The baseline estimates are included for
comparison. There are some interesting observations. First, the differences between low wage
and high wage workers and males and females are relatively small, whereas the differences
between age groups are quite substantial. For those younger than 35 almost all estimates
are insignificant, indicating that they simply find it relatively easy to find a new job. The
estimates for outplacement are particularly striking, with a negative effect for the age group
between 35 and 55 and a large significant effect for those older than 55.28 The estimates for
severance pay confirm the idea of a selection effect playing a role, as for those older than 55
the estimate is substantial and significant.
Figure 6 shows the cumulative job-finding probability for the full estimate and the different
age groups. I obtain this probability by first calculating the probability of finding a job for
each individual based on the estimates in Table 8, which is given by
Z
(11)
1 − Ŝp (t|Xi ) = exp(−
t
θˆ1 (z | Xi ) dz)
0
Where Ŝp is the estimated survivor function with p = outplacement, severance pay or no
treatment and t = 30, 60, . . . , 1140, 1200. I obtain the probabilities for e.g. outplacement by
imposing outplacement = 1 and imposing 0 for severance pay. I average these probabilities
across individuals for each 30 day time period and take the sum of the probability of the
period t and earlier periods to arrive at the cumulative probability.
These results highlight that the job finding probability is very different between those older
and those younger than 55. For those younger than 35, almost 75% find a job within six
months and this increases to about 85% after 3 years. For those between 35 and 55 this
is 60% and 80% respectively. In contrast, for those older than 55, the probability without
treatment is around 5% and barely increases over time. Again, apart from the pure age
effects, this could be related to older workers receiving longer unemployment benefits or
(early) retirement. Second, outplacement seems to be most effective for the elderly, where it
increases the job-finding probability by around 5 percentage points. Severance pay is even
more effective, leading me to suspect that selection plays an important role here.
28 Outplacement starts being significant and positive when I restrict the sample to age ≥ 46 and onwards. The
maximum is obtained with a sample restricted to age ≥ 55, after which the estimates turn insignificant again.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
Table 7—: Parameter estimates for hazard rate models.
Outplacement
Severance pay
Proportional Hazard
Mixed Proportional Hazard
0.0320
(0.0447)
0.1475∗
(0.0577)
0.0553
(0.0647)
0.1523†
(0.0866)
-0.2925∗∗
(0.0453)
-0.6835∗∗
(0.0504)
-2.0733∗∗
(0.0770)
-0.0440†
(0.0226)
-0.0093∗∗
(0.0018)
-0.1673∗∗
(0.0344)
-0.0085
(0.0386)
0.1019∗∗
(0.0351)
0.2719∗∗
(0.0393)
-0.5601∗∗
(0.0731)
-1.1417∗∗
(0.0833)
-3.0182∗∗
(0.1316)
0.0080
(0.0368)
-0.0145∗∗
(0.0027)
-0.2164∗∗
(0.0520)
0.0531
(0.0577)
0.1757∗∗
(0.0520)
0.3942∗∗
(0.0593)
-0.5277∗∗
(0.0545)
-0.7489∗∗
(0.0610)
-0.7952∗∗
(0.0651)
-1.0715∗∗
(0.0751)
-1.0298∗∗
(0.0768)
-1.2084∗∗
(0.0573)
-1.4696∗∗
(0.0665)
-1.4778∗∗
(0.0569)
-1.8783∗∗
(0.0709)
-2.2822∗∗
(0.0551)
-4.8509∗∗
(0.2130)
-0.3719∗∗
(0.0712)
-0.4609∗∗
(0.0894)
-0.3463∗∗
(0.1101)
-0.6193∗∗
(0.1268)
-0.4825∗∗
(0.1317)
-0.5561∗∗
(0.1075)
-0.7712∗∗
(0.1257)
-0.4126∗∗
(0.1362)
-0.8925∗∗
(0.1527)
-1.1376∗∗
(0.1160)
-5.3414∗∗
(0.3396)
1.0572
(0.0675)
6,502
Individual characteristics
Age: 35 – 45 years
Age: 45 – 55 years
Age: ≥ 55 years
ln(wage) in lost job
Tenure (months) / 10
Female
Partner
Children
Dutch
Duration dependence
Month 2
Month 3
Month 4
Month 5
Month 6
Month 7 – 9
Month 10 – 12
Month 13 – 18
Month 19 – 24
Month 25 +
Constant
σ2
N
6,502
Notes: Standard errors in parentheses. Controls for year and quarter of inflow, sector and firm size are included.
Omitted categories: Duration: Month 1 Age: < 35 years. Significance levels: † : 10% ∗ : 5% ∗∗ : 1%.
Source: Own calculations using registration data from Statistics Netherlands on displaced workers.
37
38
Table 8—: Parameter estimates for hazard rates out of unemployment for specific subgroups.
Baseline
Outplacement
Severance pay
N
0.0553
(0.0647)
0.1523†
(0.0866)
6,502
< 35
35 – 55
≥ 55
Low wage
High wage
Male
Female
0.2413
(0.2610)
-0.0846
(0.2892)
1,314
-0.2228∗
0.7587∗
(0.1055)
-0.0123
(0.1534)
3,199
(0.3347)
1.4155∗∗
(0.4434)
1,375
-0.0069
(0.0934)
0.0511
(0.1132)
3,242
0.1003
(0.1027)
0.2395
(0.1722)
3,260
0.0189
(0.0890)
0.2086
(0.1329)
4,008
0.0831
(0.1064)
0.1331
(0.1326)
2,494
Notes: Standard errors in parentheses. All controls are included. Significance levels: † : 10% ∗ : 5% ∗∗ : 1%. Low (high) wage is less (more) than the
in-sample median wage of 2,360 euros.
Source: Own calculations using registration data from Statistics Netherlands on displaced workers.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
39
Unemployment duration: sensitivity analysis
To check the sensitivity of my results for outplacement, I performed several different estimates. First, I checked whether the effect of outplacement is duration dependent by including
interaction effects of outplacement and duration. It could for example be that outplacement
only has an effect at the beginning of the unemployment spell, but if it doesn’t result in a job,
its effect quickly diminishes. On the other hand, it could be that people want to finish their
outplacement program and hence the effect only starts after a couple of months. However, I
found no evidence of a duration dependent effect.
Second, I constructed treatment and control groups using nearest neighbor propensity score
matching and excluded those not on the common support. As Table 3 shows, the group
who receives outplacement is in some respects quite different from the group who doesn’t
receive outplacement, so this method could control for possible selection effects. However,
the estimates turn out to be very similar.29
I also check for the sensitivity of severance pay. One could worry that my results for
severance pay are related to my choice of variables. After all, it could be that it is not so
much whether you receive severance pay or not, but rather the level that you receive. To
check whether these different measures matter, I ran the MPH model with two other sets of
measures. First, I simply included all five categories of severance pay: no severance, severance
unrelated to the kantonrechtersformule (KRF), severance with C < 1, C = 1 and C > 1
(model A). In model B I’ve calculated the actual level of severance pay people receive and
determined categories on that basis. This can only be done for those who receive severance
related to the KRF, so the sample size is smaller. Also, I only know whether the correction
factor is equal to, larger or smaller than 1. These are the two reasons why I chose to not use
this method in the baseline estimates. I calculate severance on the basis of the KRF, which
is
A (weighted years of tenure) · B (gross wage) · C (correction factor)
The weights are determined by age, with every year below 40 counting as 1, every year between
40 and 50 as 1.5 and every year after 50 as 2. For the correction factor I take C = 0.5 if
C < 1 and C = 1.5 if C > 1. I then determine four groups based on the calculated levels:
(1) no severance, (2) severance < 16, 514, (3) 16, 514 ≤ severance < 60, 000 and (4) severance
≥ 60, 000.30
Table 9 reports the results. I’ve also included the coefficient for outplacement, and it shows
that the estimates for outplacement remain insignificant with these different measures of
severance pay. The results from both models, especially model B, suggest that the actual
level of severance pay doesn’t matter as much as just receiving it. This supports the selection
effect discussed above, that those who don’t receive severance pay are likely those with the
worst prospects.
Finally, since age seems to matter a lot for the effects of both outplacement and severance
pay, it could be that the results depend on the exact definition of age categories. To test
this, I ran the same MPH model as in Table 7 with different age categories: < 30, 30 – 40,
40 – 50, 50 – 58 and ≥ 58. I also ran the model with seven age intervals, starting at age 30
until age 60, with five year intervals in between (both are not reported). For both models the
estimates for outplacement and severance pay are very similar to the baseline estimates and
29 This is in line with the argument in Angrist and Pischke (2008, 70), who claim that “[...] regression can be
motivated as a particular sort of weighted matching estimator, and therefore the differences between regression and
matching estimates are unlikely to be of major empirical importance.”
30 16,514 euros is the median level after excluding those without severance pay and 60,000 euros is a more or less
arbitrary point, roughly where it becomes fiscally interesting to set your severance aside, although that depends on
individual considerations.
40
1
1
0.8
0.8
Cumulative probability
Cumulative probability
Figure 6. : Cumulative probability of finding a job for those who experience an unemployment
spell.
0.6
0.4
0.2
0
0.6
0.4
0.2
0
0
6
12
18
24
30
Unemployment duration (months)
36
0
6
36
(b) < 35 years
1
1
0.8
0.8
Cumulative probability
Cumulative probability
(a) Full sample
12
18
24
30
Unemployment duration (months)
0.6
0.4
0.2
0
0.6
0.4
0.2
0
0
6
12
18
24
30
Unemployment duration (months)
36
0
12
18
24
30
Unemployment duration (months)
36
(d) ≥ 55 years
(c) 35 – 55 years
No treatment
6
Outplacement
Severance pay
Source: Own calculations based on registration data from Statistics Netherlands. Simulations based on the MPH
estimates in Table 8.
Notes: The labels on the x-axis refer to the beginning of the spell. I censored the graph at 40 months, since it remains
constant after that period.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
41
Table 9—: Parameter estimates for hazard rates out of unemployment estimated with different
measures of severance pay.
Outplacement
Severance pay
Baseline
Model A
Model B
0.0553
(0.0647)
0.1523†
(0.0866)
0.0798
(0.0668)
0.0638
(0.0698)
0.3875∗∗
(0.1333)
0.4222∗∗
(0.0891)
-0.0128
(0.1033)
0.3817∗∗
(0.1472)
C<1
Unrelated to KRF
C=1
C>1
< 16, 514 euros
[16, 514, 60, 000)
≥ 60, 000
N
6,502
6,502
0.3402∗∗
(0.0898)
0.4804∗∗
(0.1025)
0.4592∗∗
(0.1384)
5,106
Notes: Standard errors in parentheses. Significance levels: † : 10% ∗ :
5% ∗∗ : 1%. Model A includes all five categories of severance pay. Model
B includes four levels based on calculations of severance pay with the
KRF. The in-sample median severance pay (when those with no severance
pay are excluded) is 16,514 euros.
Source: Own calculations using registration data from Statistics Netherlands on displaced workers.
the effects for older workers remain. This means that my results for both outplacement and
severance pay are quite robust.
D.
Job quality: wages
While outplacement and severance pay might not significantly affect unemployment duration or the probability of moving from job to job, it could be that they still lead to an
improved match between employer and employee. I will start with estimating the effects on
wage in the new job – an important dimension of job quality – and continue to job duration
below.
Given that outplacement will typically make search more efficient so that workers can be
more selective with their new job, we would expect that it would lead to a higher reservation
wage and a higher wage (see the evidence in Arellano (2009). Severance pay could have a
similar positive effect if it relieves the liquidity constraints unemployed workers face. Table 10
shows the results from a simple OLS log-wage regression and a Heckman selection model. I
take the natural logarithm of gross monthly wage in the new job as the dependent variable.
The Heckman selection model takes into account the selection bias resulting from some people
not having a job. It assumes a normal selection process, and I use whether someone has a
partner and children as selection variables, since these likely have little or no effects on wages,
but could influence the job-finding probability in multiple ways.
The coefficient on the inverse Mill’s ratio is significantly different from zero, which provides
evidence of selection on wages.31
31 The coefficient on the inverse Mill’s ratio is the covariance between the error terms of the selection and the wage
equation and hence tests for independence of the two equations. In other words, it is equal to σ12 in equation 10.
42
The OLS results suggest that both outplacement and severance pay don’t affect wages. The
Heckman results confirm this for outplacement, but indicate that the effect for severance pay
seems to be a positive effect on the job finding probability and a negative effect on wages. The
results for wage in the previous job in the Heckman model are in line with the OLS results,
with a large positive effect on wage and a small negative effect on the job finding probability.
This variable captures a large part of the unobserved characteristics, such as education or
ability. The elasticity is high, around 0.65, indicating that, conditional upon other observed
characteristics, if the previous wage was 10% higher, the current wage will be 6.5% higher.32
The coefficient on “job to job” indicates the impact of having moved to another job without
an intervening unemployment spell. This is another indicator of worker quality, because it is
likely that those who directly moved to another job have different unobserved characteristics
that will also affect their wages. And indeed, moving from job to job has a large significant
impact on wage in the next job.33 Finally, the coefficients on age are also interesting. The OLS
results suggest negative effects for older workers on wage, but the Heckman estimates suggest
that this is mostly due to age having a large negative effect on the job finding probability.
This is a clear case of selection: older workers find it more difficult to find a job, but once
they find it, they are not paid any less. Tenure could be taken as an indicator of firm-specific
human capital. As such, it should have a negative impact on wages and this is confirmed in
both the OLS and Heckman results. As we already saw above, tenure negatively impacts the
job-finding probability, possibly due to out-dated experience on the job market for those with
longer tenure. But tenure also negatively affects wages, although the effect is small: a year
more tenure translates into a 1% lower wage in the Heckman model.
I have also examined the effects, using the same Heckman model as before, for different
age groups. I found only positive effects of outplacement on wage for those younger than 35,
where outplacement is associated with a 17% higher average wage in the new job. For those
older than 55 both outplacement and severance pay have a significant positive effect on the
job finding probability, in line with the results we saw before.
E.
Job quality: employment stability
Another indication of job quality is employment duration. Since outplacement makes search
more efficient, it could lead to a better match, which should lead to longer employment duration of the first job. Severance pay will have similar effects if it relieves liquidity constraints.
I first estimate the probability of job loss within one year using a probit model with the same
control variables as above. I found no significant effects of either outplacement or severance
pay (not reported). However, a probit model ignores right-censored data and also doesn’t
include jobs shorter than 1 year. To get a better sense of the effects of outplacement and
severance on job duration, we should use hazard rate models.
Table 11 shows the estimates on the exit rate out of the first job. For this model, job
duration was measured in months, rather than days. And, note that while the model is in
other respects the same as the hazard model from Table 7, the interpretation of the coefficients
is reversed, because we are now looking at the exit rate out of employment. The table shows
that outplacement has a barely significant (at the 10% level) and small positive impact on the
exit rate out of the first job. These estimates, together with the estimates on wage, suggest
that outplacement doesn’t seem to affect the quality of the job or the match. Severance pay
32 I calculate the effect for the Heckman model and take into account the small negative effect wage has on the
probability of finding a job. See e.g. Cameron and Trivedi (2005, 552) for the formulas for calculating the marginal
effects while taking into account selection.
33 “Job to job” is not included in the selection equation of the Heckman model, because it almost perfectly predicts
selection.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
43
Table 10—: Parameter estimates for wage in first new job.
Log-wage regression
(OLS)
Outplacement
Severance pay
Heckman: wage
(conditional upon
finding a job)
Heckman: selection
(probability of finding
a job)
0.0415
(0.0294)
-0.0411
(0.0428)
0.0294
(0.0336)
-0.1104∗
(0.0472)
0.0586
(0.0483)
0.2253∗∗
(0.0648)
0.0134
(0.0284)
-0.1120∗∗
(0.0361)
-0.7362∗∗
(0.0698)
0.4978∗∗
(0.0203)
0.6551∗∗
(0.0248)
-0.0094∗∗
(0.0016)
-0.2194∗∗
(0.0277)
0.0111
(0.0286)
-0.0068
(0.0253)
0.1351∗∗
(0.0313)
2.1137∗∗
(0.2073)
0.0676†
(0.0374)
0.0835
(0.0598)
0.2445
(0.2248)
0.4958∗∗
(0.0246)
0.6803∗∗
(0.0214)
-0.0056∗∗
(0.0017)
-0.1494∗∗
(0.0327)
-0.3035∗∗
(0.0615)
-0.7403∗∗
(0.0630)
-1.9767∗∗
(0.0725)
7,043
8,751
Individual characteristics
Age: 35 – 45 years
Age: 45 – 55 years
Age: ≥ 55 years
Job to job
ln(wage) in previous job
Tenure (months) / 10
Female
Partner
Children
Dutch
Constant
0.0657†
(0.0358)
2.2294∗∗
(0.1892)
Inverse Mill’s ratio (λ̂)
N
-0.0728∗
(0.0355)
-0.0086∗∗
(0.0018)
-0.2214∗∗
(0.0457)
-0.0419
(0.0478)
0.2062∗∗
(0.0402)
0.2341∗∗
(0.0468)
1.5061∗∗
(0.3165)
-1.0610∗∗
(0.2338)
8,751
Notes: Robust standard errors in parentheses. Controls for year and quarter of inflow into unemployment,
sector and firm size of old job are included. Omitted categories: Age: < 35 years. Significance levels: † :
10% ∗ : 5% ∗∗ : 1%.
Source: Own calculations using registration data from Statistics Netherlands on displaced workers.
44
has no effect on the hazard rate.
There seems to be some evidence of duration dependence, because the hazard rate significantly declines after the first year. This indicates that job separations are most likely in the
first year, especially the first few months. There could be at least two reasons for this. First,
employers generally give new employees a trial period of one or a few months, during which
they can easily fire them. Second, and probably more important, many workers take temporary jobs. Personal characteristics have the expected signs. The two indicators of worker
quality, wage in the previous job and whether someone moved directly from job to job, have
significant negative effect on the exit rate, particularly large for the “job to job” variable. On
the other hand, with each successive age group the exit rate increases significantly relative to
younger age groups. There could be at least two explanations for this. First, older workers
find it much harder to find a good match on the labor market. Second, older workers only
need to work a short time period until (early) retirement, although this really only holds for
workers older than 55 years. Finally, having dependents, in the form of children and possibly
a partner, has a significantly negative effect on the exit rate. I have repeated the analysis
with the MPH model for different age groups, but found no effects.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
Table 11—: Parameter estimates for hazard rate models on exit out of first job.
Proportional Hazard
Mixed Proportional Hazard
Outplacement
0.1014†
Severance pay
(0.0557)
0.0293
(0.0696)
0.1199†
(0.0660)
0.0743
(0.0864)
0.1795∗∗
(0.0572)
0.5562∗∗
(0.0648)
1.1107∗∗
(0.0909)
-0.6139∗∗
(0.0437)
-0.3819∗∗
(0.0303)
0.0000
(0.0023)
-0.0046
(0.0441)
-0.1356∗∗
(0.0432)
-0.2158∗∗
(0.0479)
-0.3004∗∗
(0.0478)
0.2431∗∗
(0.0694)
0.6883∗∗
(0.0827)
1.4416∗∗
(0.1390)
-0.7167∗∗
(0.0539)
-0.5319∗∗
(0.0506)
-0.0007
(0.0029)
-0.0166
(0.0546)
-0.1857∗∗
(0.0540)
-0.2805∗∗
(0.0615)
-0.3843∗∗
(0.0628)
-0.0177
(0.1457)
0.0652
(0.1395)
-0.3447∗
(0.1522)
-0.3232∗
(0.1496)
-0.2144†
(0.1206)
-0.3768∗∗
(0.1227)
-0.6382∗∗
(0.1129)
-1.2303∗∗
(0.1113)
-0.5231†
(0.3089)
-0.0102
(0.1521)
0.1092
(0.1483)
-0.3190∗
(0.1614)
-0.2733†
(0.1604)
-0.1140
(0.1306)
-0.2433†
(0.1355)
-0.4385∗∗
(0.1290)
-0.9328∗∗
(0.1400)
0.6939
(0.4553)
0.6143
(0.1253)
7,043
Individual characteristics
Age: 35 – 45 years
Age: 45 – 55 years
Age: ≥ 55 years
Job to job
ln(wage) in previous job
Tenure (months) / 10
Female
Children
Partner
Dutch
Duration dependence
Month 3
Month 4
Month 5
Month 6
Month 7 – 9
Month 10 – 12
Month 13 – 18
Month 25 +
Constant
σ2
N
7,043
Notes: Standard errors in parentheses. Controls for year and quarter of inflow, sector and firm size are included.
Omitted categories: Duration: Month 1 Age: < 35 years. Significance levels: † : 10% ∗ : 5% ∗∗ : 1%.
Source: Own calculations using registration data from Statistics Netherlands on displaced workers.
45
46
VII.
Conclusions
Both outplacement and severance pay are important instruments in the social plans that
workers are offered when they are involved in mass layoffs with large firms. However, there
has been little research on the effects of these instruments on unemployment duration and
subsequent job quality. I find that both outplacement and severance pay are not associated
with a higher probability of moving from job to job, except for older workers. For those who
do experience an unemployment spell, I also find that outplacement is not associated with a
higher exit rate for the average worker. When I restrict my sample to specific subgroups, I
find that outplacement seems to be most effective for elderly workers, with the estimates being
largest when the sample is restricted to 55 years and older. For younger workers I find no,
or even negative, effects of outplacement. For severance pay I find that it is associated with
a slightly higher exit rate out of unemployment. Other measures of severance pay, both in
levels and in relative indicators, confirm that indeed just receiving severance pay is associated
with a shorter unemployment duration, which likely indicates a selection effect.
Aside from job search, outplacement and severance pay could also influence job quality. I
found little evidence for this however. Outplacement has no average effect on wages, although
for those younger than 35 it seems to be effective. Severance pay positively affects job finding
probabilities and has a small negative affect on wages. There is no effect of severance pay on
the duration of the first job, whereas outplacement seems to have a small negative effect on
the duration of the first job.
To recap, I find very little evidence of an average effect of outplacement on the labor market
prospects of those who receive an offer when they are laid off. The only group that seems
to benefit from outplacement are elderly workers, although it is unclear whether this is due
to a selection effect, since this group also benefits significantly from receiving severance pay.
It must be kept in mind that for outplacement I only estimate the effect of the offer of
outplacement – the intention to treat – which means that I likely underestimate the actual
effect, although it is unclear by how much. Nevertheless, my findings for outplacement are
in line with many of the findings on ALMP, where most studies find small positive or zero
effects of policies in many ways similar to outplacement. In addition, Van den Berg and Van
der Klaauw (2006) also find that the effect of counseling is most effective for older workers,
whereas they find no effects for younger workers. My results for severance pay stand in
contrast to the literature, but this is likely due to selection effects.
My findings also provide evidence of some overall trends in the labor market. Older workers
typically find it much harder to find a job than younger workers, although part of this could
be due to early retirement by older workers or longer unemployment benefits. Attempts
to include this failed due to lack of good data on early retirement schemes. Females and
non-Dutch workers also have worse labor market prospects. Wage in the previous job – a
proxy for unobservable characteristics, such as education, ability and motivation – shows
mostly positive significant effects on average labor market prospects. Similar effects hold
for indicators of job quality, such as wages and job duration. At the firm level I found no
significant relation between firm characteristics, such as whether a firm pays relatively high
wages or the median age of the workers in a firm, and the content of its social plans.
A major limitation of my results is selection bias. First, my sample of firms is non-random,
but since especially outplacement is typically only offered by larger firms, this might be a relatively small problem. Second, since the best workers typically move to another job without an
intervening unemployment spell, the effects of outplacement and severance on unemployment
duration only hold for those people who actually experienced an unemployment spell. Third,
workers who don’t receive severance pay are very different from those who do receive sever-
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
47
ance pay: they are younger, earn much less and more likely to work in sectors that require less
education, such as retail. For outplacement the heterogeneity isn’t quite as substantial, but
still exists, where those who don’t receive outplacement earn less and work in different sectors than those who do receive outplacement. While I control for most of these factors, such
as sector and wage, they could be related to unobserved characteristics which will bias my
results. If those who don’t receive severance pay or outplacement have worse prospects than
those who do, my results are an overestimation of the actual effect. The evidence suggests
that this is particularly the case for severance pay, where I find significant positive effects for
those who do receive it on most dependent variables, whereas theory and previous studies
suggest it should either have no or negative effects.
Policy implications
Partly in response to the crisis and growing unemployment rates, we have seen recent
policy discussions on what to do with the relatively large sums of severance pay people
receive and the low job finding chances of some groups of workers, such as elderly workers.
One seriously considered proposal is to limit the sums of severance pay workers receive and
to set aside some of this for a so-called “transfer budget”. This budget would be used to
help people transit to another job, through services such as outplacement, or, if they require
it, retraining. My results are relevant for this discussion, although we can’t draw any strong
policy conclusions. They suggest at least that older workers find it much harder to find
a job, and that outplacement can be effective for them, both in reducing unemployment
durations and increasing direct job to job transitions. For younger workers I find no effects of
outplacement, which is not surprising given that they typically find a job quite easily (almost
75% of those younger than 35 are estimated to have a job within 6 months and for those
between 35 and 55 this was still around 65%).
This suggests that a policy similar to Belgium might be an improvement. In Belgium outplacement is mandatory for every laid off worker 45 years or older and studies on gross effects
suggest that the policy is particularly helpful for elderly workers. It is however important to
carefully study the experiences in Belgium. While the policy has led to some positive effects,
it has also, due to its design with a fine for non-complying employers, led to a reduction in
price and quality of outplacement services. Such effects should be taken into account when
considering a similar policy.
Future research
This is one of the few studies dealing with outplacement and severance pay, and certainly
the first doing so with data from social plans in the Netherlands. My results suggest many
avenues for further research.
First and foremost, it is important to deal with selection bias. It is clear that those who
receive severance pay or outplacement are selected non-random, so a different setup might
be required. An obvious improvement would be to have a dataset where it is clear who
actually made use of the offer of outplacement. It would be even better if we could use a
proper (quasi-)experimental setup, but this would require some sort of discontinuity. Perhaps
the recent policy discussions on cutting down severance pay and focusing more on job-to-job
support will provide an opportunity to get a better sense of the causal effects of outplacement
and severance pay. Another option to handle selection bias would be to jointly estimate the
probability of moving from job to job and the effects on unemployment duration, so that at
least part of the selection based on unobserved characteristics will be captured.
48
Second, I find no effects from outplacement, except for older workers. This result suggests
that younger workers found it relatively easy to find a job, so that outplacement support
hardly matters, whereas for older workers finding a job was much harder, and support can
be helpful. It would be interesting to see if these differences hold up with more recent data
taking into account the recession, when the labor market for younger workers is much worse
than it was in my study period from 2003 to 2007. It would also be interesting to examine
whether this also holds for other workers who find it more difficult to find a job, such as
workers in declining sectors.
Third, not every outplacement program is created equally and it could matter whether
someone receives a very minimal program or a more extensive program. I haven’t examined
such differences, but they could be important drivers of effectiveness.
Finally, I don’t observe some important variables, which could lead to an omitted variable
bias. These include education, but also for how long people receive unemployment benefits.
The effects are currently partly captured by such variables as age (for unemployment benefits)
and wage (for education), biasing the results for these variables.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
49
References
Angrist, Joshua D., and Jörn-Steffen Pischke. 2008. Mostly Harmless Econometrics.
Princeton University Press.
Arellano, Alfonso. 2007. “The Effect of Outplacement on Unemployment Duration in
Spain.” FEDEA Working Paper.
Arellano, Alfonso. 2009. “The effect of outplacement services on earning prospects of unemployed.” EC Working Paper Series.
Black, Dan A., Jeffrey A. Smith, Mark C. Berger, and Brett J. Noel. 2003. “Is the
Threat of Reemployment Services More Effective than the Services Themselves? Evidence
from Random Assignment in the UI System.” The American Economic Review, 93(4): pp.
1313–1327.
Boeri, Tito, and Jan van Ours. 2008. The Economics of Imperfect Labor Markets. Princeton University Press.
Boos, C., A. Nagelkerke, and T. Serail. 2001. “Sociale zekerheidsafspraken tussen
werkgevers en werknemers.”
Borghouts-van de Pas, Irmgardt W.C.M. 2012. Securing job-to-job transitions in the
labour market. Wolf Legal Publishers.
Bosch, Nicole, and Bas ter Weel. 2013. “Labour-market outcomes of older workers in the
Netherlands: Measuring job prospects using the occupational age structure.” De Economist,
161(2): 199 – 218.
Cameron, A. Colin, and Pravin K. Trivedi. 2005. Microeconometrics: Methods and
Applications. Cambridge University Press.
Card, David, Jochen Kluve, and Andrea Weber. 2010. “Active Labour Market Policy
Evaluations: A Meta-Analysis*.” The Economic Journal, 120(548): F452–F477.
Card, David, Raj Chetty, and Andrea Weber. 2007. “Cash-on-Hand and Competing
Models of Intertemporal Behavior: New Evidence from the Labor Market.” The Quarterly
Journal of Economics, 122(4): 1511–1560.
Chetty, Raj. 2008. “Moral Hazard versus Liquidity and Optimal Unemployment Insurance.”
Journal of Political Economy, 116(2): pp. 173–234.
De Cuyper, Peter, Anneleen Peeters, Debbie Sanders, Ludo Struyven, and Miet
Lamberts. 2008. “Van werk naar werk: de markt van outplacement.”
EIM. 2008. “Werk op maat: curatieve van werk naar werk activiteiten in de praktijk.”
EIM. 2010. “Van Werk naar Werk in het MKB.”
Eurostat. 2012. “Total unemployment, total age and sex.”
Frenk, Myrthe, and Gerard Pfann. 2009. “Is het Nederlandse ontslagstelsel nu echt aan
verandering toe?” TPEdigitaal, 3(2): 104–25.
Gibbons, Robert, and Lawrence Katz. 1991. “Layoffs and Lemons.” Journal of Labor
Economics, 9(4): 351–380.
50
Graversen, Brian Krogh, and Jan C. van Ours. 2008. “How to help unemployed find
jobs quickly: Experimental evidence from a mandatory activation program.” Journal of
Public Economics, 92(1011): 2020 – 2035.
Heckman, James J. 1979. “Sample Selection Bias as a Specification Error.” Econometrica,
47(1): pp. 153–161.
Heckman, James J, Robert J LaLonde, and Jeffrey A Smith. 1999. “The economics
and econometrics of active labor market programs.” In . Vol. 3, 1865–2097. Elsevier.
Hujer, Reinhard, Stephan Lothar Thomsen, and Christopher Zeiss. 2006. “The
Effects of Short-Term Training Measures on the Individual Unemployment Duration in
West Germany.” ZEW - Centre for European Economic Research Discussion Paper No.
06-065.
Jacobsen, Louis S., Robert J. LaLonde, and Daniel G. Sullivan. 1993. “Earnings
Losses of Displaced Workers.” The American Economic Review, 83(4): 685–709.
Jacobs, Laura, and Peter De Cuyper. 2013. “Wat na collectief ontslag? Processen en
effecten van outplacement in kaart gebracht.” KU Leuven / HIVA.
Lancaster, Tony. 1979. “Econometric Methods for the Duration of Unemployment.” Econometrica, 47(4): pp. 939–956.
Mortensen, D. 1986. “Job Search and Labor Market Analysis.” In Handbook of Labor Economics Vol. II. , ed. O. Ashenfelter and D. Card, 849–919. Elsevier, Amsterdam.
OECD. 2008. “Employment Protection in 2008 in OECD and selected non-OECD countries.”
OVAL. 2012. “BrancheMonitor 2012 Samenvatting.”
Rogerson, Richard, Robert Shimer, and Randall Wright. 2005. “Search-Theoretic
Models of the Labor Market: A Survey.” Journal of Economic Literature, 43(4): 959–988.
The Wall Street Journal. 2009. “Outplacement Firms Struggle to Do Job.”
Thomsen, Stephan. 2009. “Job Search Assistance Programs in Europe: Evaluation Methods and Recent Empirical Findings.” FEMM Working Paper No. 18.
Tros, Frank H., C.W.G. Rayer, and E. Verhulp. 2005. “Flexibiliteit en zekerheid in
sociale plannen. Naar meer activering?” SMA, 60(11/12): 528–538.
Uusitalo, Roope, and Jouko Verho. 2010. “The effect of unemployment benefits on reemployment rates: Evidence from the Finnish unemployment insurance reform.” Labour
Economics, 17(4): 643 – 654.
Van den Berg, Gerard J. 1994. “The Effects of Changes of the Job Offer Arrival Rate on
the Duration of Unemployment.” Journal of Labor Economics, 12(3): pp. 478–498.
Van den Berg, Gerard J. 2001. “Chapter 55 Duration models: specification, identification
and multiple durations.” In Handbook of econometrics. Vol. 5 of Handbook of Econometrics,
, ed. J.J. Heckman and E. Leamer, 3381 – 3460. Elsevier.
Van den Berg, Gerard J., and Bas Van der Klaauw. 2006. “Counseling and Monitoring
of Unemployed Workers: Theory and Evidence from a Controlled Social Experiment.”
International Economic Review, 47(3): 895–936.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
51
Winter-Ebmer, Rudolf. 2001. “Evaluating an Innovative Redundancy-Retraining Project:
The Austrian Steel Foundation.” IZA Discussion Paper No. 297.
52
Appendix
A1.
Sample construction
I draw my sample of social plans from a database of collective labor agreements from the
Ministry of Social Affairs and Employment. This database contains 672 unique social plans
for the period 2002 to 2010. I coded the contents of these plans by hand. I took the following
steps in connecting my dataset to registration data from Statistics Netherlands (CBS).
1) Firms are identified by a number they get from the Chamber of Commerce. These
numbers are different for each legal entity, so that a large corporation can have many
different numbers associated with it. Since this number was not available in the database
with social plans, I had to collect them by hand from a database of the Chamber of
Commerce. Since the sample goes back quite some time, some firms could not be found,
for example because they went bankrupt. Ultimately I couldn’t find the ID number for
35 firms.
2) I continue to connect my sample with social plans and the associated numbers of the
Chamber of Commerce to a database with company characteristics from Statistics
Netherlands. Each number of the Chamber of Commerce is associated with a unique
ID by the CBS. These correspond to the legal entities found in the world. I couldn’t
find a corresponding ID for 47 firms.
3) Finally, the firms in the CBS database upon which statistics are based and which are
connected to firm characteristics (e.g. sector) and jobs are special statistical constructs.
They are constructed in such a way that each firm has one defining economic activity.
So whereas a firm might consist of many different legal entities, if all of these carry out
the same main economic activity, they are constructed as one so-called “company unit”
(bedrijfseenheid ) in the database. I was unable to connect 27 ID’s to the statistical
company units.
4) Finally it turned out that some companies had multiple social plans in one year associated with them. The likely reason is that, since a company unit can consist of many
different legal entities, these different entities had their own social plans. Unfortunately,
since I only have data on the level of the company unit a choice had to be made between
the multiple social plans at the level of the company unit. Most of these plans actually
turned out to be the same, and where there were differences these were small. In case
of differences in the social plans, I consistently picked the most generous one. I lost 11
social plans here and ended up with a dataset containing 552 social plans.
To use the contents of these plans to estimate their effects on both the individual and the
firm level again required connecting some databases. For the analyses at the firm level in
section VI.A I ended up with 495 firms. For the analyses at the individual level I ended up
with 148 social plans out of 294 plans from 2003 to 2007. Table A1 compares the different
samples used throughout this thesis. It shows that, quite remarkably, there are almost no
significant differences between the samples of plans. I compare the final sample with the
sample that excludes missing firms using a probit model. These results, together with the
comparison of my sample with earlier studies in section IV.A, mean that at least at the level
of social plans my samples are quite representative.
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
53
Table A1—: Contents of the social plans in my sample after successive trimming.
Contents of
social plans
(%)
Full sample
Excluding
missings
2003-2007
Final sample
Severance pay
76.5
76.8
72.8
77.0
C=1
C<1
C>1
Unrelated to KRF
29.3
26.8
10.4
10.1
31.2
26.5
10.1
9.1
31.0
23.5
8.5
9.9
30.4
25.7
10.1
10.8
Incentive to leave
Supplement to UB
Incentive to replace
44.3
16.5
23.8
44.6
17.6
23.9
51.0
22.8
22.8
47.3
25.0∗∗
23.0
62.3
62.9
64.3
66.2
7.6
3 - 30
7.6
3 - 30
3742
8.2
3 - 30
3741
8.4
3 - 30
2953
66.9
9.5
23.3
21.4
66.7
9.4
23.7
20.5
65.0
7.5
22.1
24.5
64.2
5.4
14.2∗
19.6
38.9
10.1
37.5
10.9
40.1
10.9
41.9
13.5
672
2002 - 2010
552
2002 - 2010
294
2003 - 2007
148
2003 - 2007
Financial
Job-to-job
Outplacement
Duration (avg months)
Duration (min - max)
Budget (avg e)
Internal replacement
Mobility centre
Training
Income suppletion
Arrangements for
the elderly
Bridge to pension
More job-to-job support
N
Year
Notes: The final sample is compared to the sample without missing firms using a probit analysis.
Significance levels: † : 10% ∗ : 5% ∗∗ : 1%.
Source: Sample of social plans gathered from database of Ministry of Social Affairs and Employment.
54
A2.
Description of arrangements
• Financial arrangements
– Severance pay: severance pay is often related to the cantonal judge formula (kantonrechtersformule, KRF). The correction factors are as follows.
1) KRF with C = 1.
2) KRF with C > 1.
3) KRF with C < 1.
4) Unknown: not related to KRF. Usually less than the KRF.
– Incentive to leave: a financial incentive to voluntarily leave earlier than the determined dismissal date.
– Supplement to UB: supplement to unemployment benefits if dismissed employee
needs it. UB are usually at 70% of the last earned wage, a supplement often brings
them up to the last earned wage.
– Incentive to replace: financial incentive to leave for those who are not to be dismissed, if by doing so they replace someone that would be dismissed.
• Job-to-job support
– Internal replacement: arrangements that try to find a new place for the to be
dismissed employee within the organization. Usually a job that is one or two pay
ranges lower is deemed acceptable and the employee has to accept. Could be
accompanied with some training.
– Outplacement: usually a specified outplacement track with a given maximum
length and sometimes also a given budget. Employees are usually free to choose
whether they want this.
– Duration: the maximum duration of the outplacement track.
– Mobility center: usually a center setup by a large company with the help of external
support (outplacement bureaus). This will provide both internal and external
support to dismissed employees, aimed at finding a job within the organization
or outside. Can also provide training. For some sectors, such as metal-electro,
there is a sector-specific mobility center that also mediates between employers and
employees, but aims at keeping employees within the sector. These are all included
here and could be regarded as similar to outplacement.
– Education or training: training specifically provided for employees with the aim of
achieving an external job.
– Income supplement: if an employee finds an external job and the wage is lower
than the wage she had with the previous employer, the previous employer will
supplement this for a given period of time.
• Special arrangements for the elderly
– Bridging to pension: financial arrangements that help dismissed elderly workers
(usually starting from 50-55 years) bridge the period to pension. With the decline
of pre-pension and early retirement starting in 2006 this has become less attractive.
– More job-to-job support: elderly sometimes receive more job-to-job support in the
form of longer support or a higher budget.
A3.
Full estimation results individual effects
THE EFFECTS OF OUTPLACEMENT AND SEVERANCE PAY
55
Table A2—: Omitted parameter estimates for models in Tables 5, 7, 10, 11.
Probit model
(Table 5)
Unemployment
exit (MPH,
Table 7)
Heckman wage
(Table 10)
Exit rate out
of first job
(MPH,
Table 11)
-0.5265∗∗
(0.1091)
-0.0136
(0.0659)
0.1403
(0.1402)
0.0669
(0.2053)
0.2822∗∗
(0.0788)
0.2362
(0.2104)
-0.3565∗∗
(0.0728)
-0.0440
(0.0765)
-0.2541
(0.2834)
0.8463∗∗
(0.1716)
0.0214
(0.0974)
0.5294†
(0.2780)
-0.0112
(0.3706)
-0.3051∗
(0.1265)
0.3298
(0.3150)
0.3295
(0.1007)
-0.1807†
(0.1082)
-0.1768
(0.2665)
0.1336
(0.0870)
0.1016†
(0.0519)
0.0311
(0.1161)
0.1316
(0.1726)
0.2323∗∗
(0.0691)
0.0449
(0.1685)
0.1408∗
(0.0569)
0.1054†
(0.0606)
0.1168
(0.2314)
0.1524
(0.1511)
-0.2489∗
(0.0987)
-0.0630
(0.2288)
0.7982∗
(0.3322)
-0.2841∗
(0.1244)
-0.4356
(2862)
-0.1420
(0.1148)
-0.2576∗
(0.1179)
-0.1656
(0.4525)
0.0865
(0.1096)
0.1982†
(0.1141)
-0.0900
(0.1178)
-0.1692
(0.1495)
0.0766
(0.1008)
0.0090
(0.1047)
0.0359
(0.1009)
0.0313
(0.1014)
0.6498∗∗
(0.1460)
0.6342∗∗
(0.1594)
0.6670∗∗
(0.1489)
1.0151∗∗
(0.2153)
0.4307∗∗
(0.1272)
0.5143∗∗
(0.1346)
0.5942∗∗
(0.1238)
0.3433∗∗
(0.1246)
-0.1307
(0.0907)
-0.1167
(0.0960)
0.1182
(0.0932)
-0.1209
(0.1157)
-0.0589
(0.0795)
-0.0514
(0.0838)
0.1682∗
(0.0800)
-0.1201
(0.0833)
0.0345
(0.1581)
-0.0616
(0.1727)
-0.2055
(0.1691)
0.1391
(0.2075)
0.1361
(0.1447)
0.1129
(1527)
-0.4539∗∗
(0.1522)
0.0856
(0.1482)
-0.0445
(0.0985)
-0.1086
(0.0981)
0.1463
(0.1025)
0.3185∗∗
(0.1083)
-0.0455
(0.1555)
0.2284
(0.1546)
0.1485
(0.1644)
0.7058∗∗
(0.1771)
-0.1091
(0.0798)
-0.1743∗
(0.0798)
0.0941
(0.0835)
-0.0558
(0.0898)
0.3456∗
(0.1581)
0.3243∗
(0.1577)
0.0849
(0.1649)
-0.1593
(0.1740)
0.0345
(0.0525)
0.2036∗∗
(0.0526)
0.0891†
(0.0520)
8,751
-0.0401
(0.0781)
0.0844
(0.0810)
0.2352∗∗
(0.0771)
6,502
-0.0050
(0.0417)
-0.0010
(0.0428)
-0.0299
(0.0433)
8,751
0.0532
(0.0803)
0.0252
(0.0837)
0.0796
(0.0791)
7,043
Sector (ref: Manufacturing)
Construction
Wholesale and retail trade
Transport and storage
Accommodation and food serving
Information and communication
Financial institutions
Other specialised business services
Health and social activities
Other
Firm size (ref: < 50 employees)
50 – 99 employees
100 – 149 employees
150 – 199 employees
200 – 249 employees
250 – 499 employees
500 – 999 employees
1,000 – 1,999 employees
≥ 2000 employees
Year of job loss (ref: 2003)
2004
2005
2006
2007
Quarter of job loss (ref: Q1)
2nd quarter
3rd quarter
4th quarter
N
Notes: Standard errors in parentheses, robust standard errors for the probit and wage models. Significance
levels: † : 10% ∗ : 5% ∗∗ : 1%.
Source: Own calculations using registration data from Statistics Netherlands on displaced workers.