Unemployment Insurance Design and Its Effects: Evidence for Uruguay
| Autor | Verónica Amarante, Rodrigo Arim, Andrés Dean |
| Páginas | 7-42 |
7
D E S A R R O . S O C . 71, P R I M E R S E M E S T R E D E 2013, P P . X-X X , I S S N 0120-3584
Revista
Desarrollo y Sociedad
71
Primer semestre 2013
PP. 7-42, ISSN 0120-3584
Unemployment Insurance Design and Its Effects:
Evidence for Uruguay1
El diseño del seguro de desempleo y sus efectos:
evidencia para Uruguay
Verónica Amarante
Rodrigo Arim
Andrés Dean2
DOI: 10.13043/DYS.71.1
Abstract
Using unemployment insurance records and social security labor histories,
we provide evidence of the impacts of recent changes in the unemployment
insurance system of Uruguay. We estimate the effects on unemployment dura-
tion and post unemployment wages. Two main changes are considered: the
modification in the scheme of payments —from a lump sum during six months
to a decreasing system of payments during the same period— and extension
in the duration of the benefit up to one year for workers 50 or older. We con-
sider different impact evaluation techniques (propensity score, difference in
1 This article was prepared for the Latin-American Research Network sponsored by the Inter-American
Development Bank, as part of the project “Protecting Workers against Unemployment in Latin America
and the Caribbean”. We acknowledge useful comments and suggestions from Robert LaLonde, Carmen
Pagés-Serra, Verónica Alaimo, Jacqueline Mazza and Marisa Bucheli, as well as helpful comments received
from other researchers participating in the project. We thank two anonymous referees for helpful com-
ments. Finally, we are grateful to Banco de Previsión Social for providing the data for this study, and to
Gabriel Lagomarsino for his help in the administrative process to get the data. Any errors are our own.
2 The autors are Universidad de la República, Uruguay. Corresponding author: Verónica Amarante, e-mail:
vero@iecon.ccee.edu.uy. Other authors: Andrés Dean, e-mail: adean@iecon.ccee.edu.uy; Rodrigo Arim,
e-mail: rodrigo@iecon.ccee.edu.uy.
Este artículo fue recibido el 25 de abril de 2012; revisado el 6 de diciembre de 2012 y, finalmente,
aceptado el 15 de marzo de 2013.
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
8
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
differences and regression discontinuity), and find that the change in the pay-
ment scheme has implied a small reduction in unemployment duration, with
no effects in terms of subsequent earnings. The possibility of extension of the
unemployment insurance for older workers has led to an extension in unem-
ployment duration and it has not helped to subsidize better job matches in
the form of better paying jobs.
Key words: Unemployment insurance, impact evaluation.
JEL classification: J64, J65, J68.
Resumen
En este artículo se analizan los impactos de los cambios recientes en el seguro
de desempleo de Uruguay. Combinando información de las historias labora-
les de la seguridad social y de los registros administrativos del programa de
seguro de desempleo, se estiman los efectos de las recientes modificaciones
sobre la duración del desempleo y los salarios de reingreso al mercado laboral.
Los cambios analizados consisten en la variación del esquema de pagos, que
pasó de una transferencia de sumas fijas a un régimen de pagos decrecientes
en el tiempo, y la extensión de la duración potencial del seguro de desempleo
para los trabajadores de cincuenta años y más. Se utilizan distintas metodo-
logías (propensity score, diferencias en diferencias y regresión discontinua) y
se encuentra que el cambio en el esquema de pagos implicó una reducción
en la duración del desempleo, aunque de pequeña magnitud, sin modificacio-
nes en los ingresos al reingresar al mercado de trabajo. Para los trabajadores
de mayor edad, la duración del seguro se ha extendido como consecuencia
del cambio introducido y no ha estado acompañado de mejores salarios para
quienes reingresaron al mercado laboral.
Palabras clave: seguro de desempleo, evaluación de impacto.
Clasificación JEL: J64, J65, J68.
Introduction
The design of the unemployment insurance program may have important
consequences on labor market outcomes. In particular, it can affect both
Verónica Amarante, Rodrigo Arim y Andrés Dean
9
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
unemployment duration and the quality of subsequent job matches. On the
unemployment duration side, job search models show that higher benefits and
longer benefit duration may lead to longer unemployment spells (Hopenhayn
and Nicolini, 1997; Meyer, 1990; Moffit, 1985; Mortensen, 1977), as benefi-
ciaries of the UI have higher reservation wages and make less effort in the
search process. Theory then clearly predicts a positive relation between UI and
unemployment duration, and the empirical evidence confirms this relationship.
The well known empirical result about the spike in the exit rate from unem-
ployment next to the expiration date has been interpreted as an illustration
of the disincentive effects of the UI system. A different vision is provided by
Chetty (2008), who argues that most of the increase in unemployment duration
caused by UI is due to a liquidity effect -which is welfare enhancing- instead
of distortions in the marginal incentives to search. Among the more important
empirical contributions related to measuring the effects of potential benefit
duration on unemployment duration are Katz and Meyer (1990), Hunt (1995),
Card and Levine (2000), Van Ours and Vodopivec (2006).
On the effects of UI on subsequent employment outcomes, two channels can
be identified. If UI benefits increase reservation wages, one would expect UI
beneficiaries to earn higher wages after they are reemployed. Also, unemploy-
ment may operate as a subsidy, allowing the unemployed people to wait until
they receive an offer more suitable for their skills. This outcome favors post-
unemployment job stability, improving the efficiency of the matching process
(Marimon and Zilibotti, 1999). In those ways, UI can contribute to better job
matches (higher salaries, more stability). The effects of UI on post unemploy-
ment wage outcomes have been addressed by Addison and Blackburn (2000),
who report modest evidence in support of UI increasing post unemployment
wages of recipients when compared to non recipients in the US.3 Belzil (2001)
also finds a weak positive effect of UI on subsequent job duration for Canada.
More recent empirical evidence is provided by Tatsiramos (2009) for European
countries, suggesting that even if receiving UI benefits has a direct negative
effect in terms of reducing the duration of unemployment spell, it also has a
positive effect on subsequent employment stability.
3 The pioneering research in this topic is Ehrenberg and Oaxaca (1976), who find a positive relation
between the UI replacement ratio and post unemployment wages, and Classen (1977), who reports no
evidence of an increase in benefits leading to the acceptance of more lucrative job offers.
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
10
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Van Ours and Vodopidec (2008) find, for the Slovenian case, that reducing the
potential duration of unemployment benefits has no detectable effect on post
employment wages or job stability.
Empirical evidence is scant for developing countries, especially in the case of
subsequent employment outcomes, and it is almost missing for Latin Ameri-
can countries. On the one hand, the experience of the region with these pro-
grams is very limited (see Mazza, 2000), and on the other hand, for those
countries who do have UI programs, the restriction is the available microdata
sets. Recent evidence is provided for Argentina by Rozada et al. (2011), who
find that unemployment duration increases when unemployment insurance
transfers are higher or provided for a longer period.
In this paper, we provide new evidence on the effects of the UI on unemploy-
ment duration and subsequent employment wage in a developing country. Recent
changes in the legislation of the UI system in Uruguay allow us to undertake
this impact evaluation study. We assess the impacts of two main changes: the
modification of the scheme of payments, from a lump sum during six months
to a scheme of decreasing payments during the same period, and the possibility
of extension of the UI benefit up to one year for workers 50 or older.
Using unemployment insurance records and social security labor histories and
based on different evaluation strategies, we try to disentangle the effect of
each of these changes. For the first modification, the impact evaluation strategy
is based on propensity score and difference-in-differences estimators, for the
second change, effects are estimated using regression discontinuity design.
The article is organized as follows: first we present the Uruguayan unemploy-
ment insurance program (section I). We then discuss our empirical strategy and
describe our data (section II). Later, we discuss the effects of the change on
the scheme of benefits on unemployment duration and reemployment wages
(section III). We then analyse the impact of the extension in the duration of
UI for workers aged 50 or more (section IV) and finally present some conclud-
ing remarks (section V).
Verónica Amarante, Rodrigo Arim y Andrés Dean
11
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
I. The Uruguayan Unemployment Insurance
A. Overview of the System and Recent Changes
The origins of the present Uruguayan unemployment insurance date back to
1958, when a very similar program was created. It was modified later on in
1962 and in 1982. This last version of the UI system operated until 2009, when
the program went through important modifications (law 18399). The program
depends on the Ministry of Labor, and is under the administration of the social
security main institution, Banco de Previsión Social (BPS).
There are three possible reasons or causes for entering the program: job loss
(being fired or permanently laid off), job suspension (total suspension of activi-
ties for a period, temporary lay-off) and job reduction (when days of work or
hours of work suffer from a reduction of at least 25%). The modality of job
suspension allows firms to lay off workers when facing demand fluctuations,
and recall them back when UI benefits are exhausted.4
Originally, the program was mandatory for private and rural workers, exclud-
ing domestic workers and workers from the financial system. Rural workers
were included as beneficiaries since 2001. To have this subsidy, workers should
have worked at least six months in the previous year, and they should have
been involuntarily unemployed. Unemployment insurance lasts for six months
or the equivalent to 72 days of labor for day laborers. Until February 2009, the
subsidy was 50% of the average wage of the last six months, or a monthly
subsidy equivalent to 12 working days. That amount could never be less than
half the minimum wage. In the case of job reduction, the amount of the benefit
is the difference between 50% of their average wage during the previous six
months, and the salary they continue to get from their employees.
Married workers or workers responsible for other people receive an additional
20%. The worker cannot re-enter the insurance program until a year has passed
since the last time he received the benefit. Although the worker may receive
the benefit for a maximum of six months, the Executive Power can extend this
period, in a rather discretional way. This extension is supposed not to surpass
4 This modality has led to the use of the program as a subsidy for firms whose activity presented important
seasonal features (see Amarante and Bucheli, 2008).
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
12
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
18 months, although this has been violated in some occasions. The general
rule is that if the worker does not return to his job after six months, he has
been the facto fired and has the right to get severance payment.
UI beneficiaries loose the benefit if they get another job, reject a job offer or
get a pension. The first requirement implies that workers receiving the unem-
ployment insurance could not have a job that implies a contribution to the
social security system, although if they are working in the informal sector this
may not be detected. The system does not include the monitoring of unem-
ployed workers or the application of punitive sanctions. UI beneficiaries may
voluntary apply to receive training, financed by the Fondo de Reconversión
Laboral (FRL).
Important modifications to the unemployment insurance program were intro-
duced with the approval of law 15.180, implemented in February 2009. The
coverage did not change: the unemployment insurance is mandatory for all
private (formal) workers. The most relevant has to do with the amount of the
benefits for those permanently laid off: instead of receiving an equal sum
every month during at most six months, the new system establishes a decreasing
scheme for benefits. This implies an average benefit of 66% of his previous salary
in the first month (instead of 50% as before). This modification is aimed at
fostering job search among beneficiaries. The maximum benefit is kept equal
on average, but adapted to the new decreasing scheme.
Another important change refers to workers aged 50 or more, who can now
keep the subsidy for six additional months. During this last additional six
months, they receive the same amount of benefit than during the sixth month
(40%). This change tries to address the difficulties that this group of workers
finds when trying to re-enter the labor market. They represent approximately
15% of total beneficiaries.
These two main changes, detailed in Table 1, are the ones evaluated in this
article.
Other modifications, not addressed in this article, include the reduction in
unemployment duration from six to four months for those beneficiaries under
the regime of suspension (temporary job loss). Workers under this regime were
Verónica Amarante, Rodrigo Arim y Andrés Dean
13
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
not considered in this evaluation. The new regulations also attempt to coordi-
nate UI with active labor market policies, and theoretically beneficiaries may
lose their UI benefits if they do not participate in training courses offered by
the Ministry of Labor. Another change is the introduction of compatibility
between the unemployment insurance and holding other economic activity.
Also, in the new regime, a worker can interrupt the benefits for a short time, in
case he gets a temporary job, and then return to the insurance system. Finally,
before February 2009 the benefit could only be claimed within 30 days after
the last day of work, in the actual regime there are no restrictions.
Table 1. Main Changes of Unemployment Insurance System in Uruguay
Old Regime New Regime (February 2009)
Benefit amount Lump sum:
50% of the average wage of •
the last six months or subsidy
equivalent to 12 days of labor
for day laborers (job loss or
suspension)
diffe rence betwe en 50% of •
the ir avera ge wage du ring
the p revious six months, a nd
the salar y th ey co ntinue to
get from their employees (job
reduction)
Minimum: half BPC / Maximum: •
8 BPC
Job loss: decreasing scheme (as % of average
wage of last 6 months): 1st month: 66%, 2nd
month: 57%, 3rd month: 50%, 4th month: 45%,
5th month: 42%, 6th month: 40%. For day laborers:
equivalent to 16 days of labor in the 1st month,
14 in the 2nd, 12 in the 3rd, 11 in the 4th, 10 in the
5th and 9 in the 6th.
Job suspension or job reduction: similar to the
old system
Minimum: 1 BPC/ Maximum: similar to the •
old system (adjusted to the new decreasing
scheme in the case of job loss)
Benefit duration -6 months
72 days of labor (day laborers)•
6 months in the modality of job loss or job •
reduction (or 72 days of labor)
4 months in the modality of suspension (or •
48 labor days)
can be extended to one year for workers older •
than 50
can be extended to 8 months for job loss in •
cases of economic recession
Note: BPC means Base de Prestaciones Contributivas. In December 2010, a BPC was equivalent to 2061 $
(103 USD), and represented 46% of the National Minumum Wage.
Source: Authors’ elaboration based on decree-law 15180 and law 18399.
B. Basic Statistics
Before considering some basic statistics about the UI program, it is worth
presenting some information about the Uruguayan labor market, and in par-
ticular the informal sector.
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
14
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
The role of unemployment insurance programs has special features in develop-
ing countries, as the size of the informal sector may exert a relevant influence.
Whereas in the traditional moral hazard problem employment status is observ-
able and workers cannot lie to the authorities, in economies with large informal
sectors workers can join the informal sector and receive unemployment bene-
fits. This means that the presence of a large informal sector may undermine the
utility of unemployment insurance programs by providing undesired incentives
to increase informal sector employment while receiving the insurance. Studies
about the consequences of UI benefits in dual labor markets are not abundant,
exceptions including Alvarez-Parra and Sánchez (2009), Vodopivec (2009) and
Bardey and Jaramillo (2011).5
In Uruguay, the informal sector, defined as workers who do not contribute to
the social security sector, comprises 32% of total workers in 2009. The first
comparable figure corresponds to 2001, and in that year the informal sec-
tor represented 36% of total employment. If only private workers are con-
sidered (they represent 57% of total employment in 2009), non contributors
were 30% in 2001 and 24% in 2009. During the last decade, contributions to
social security decreased between 2001 and 2004, but have been increasing
during the last years, mainly driven by private workers (see Table A1). This
configures a situation where private formal workers —those who may apply
for the insurance— represent around 44% of total workers in 2009. More-
over, according to household survey information, almost 25% of unemployed
in 2005 had lost their previous job within the prior six months, but that job
was informal (Amarante and Bucheli, 2008).
These structural characteristics of Uruguayan labor market explain the low
coverage of the UI program. Around 48% of those unemployed in 2009 were
not supposed to be covered by the insurance, because they were looking for
their first job or re-entering the labor market after a long absence.
5 Alvarez-Parra and Sánchez (2009) find that payment profile must be relatively flat to avoid participa-
tion and keep search effort high, and there must be no payments for long run unemployed. Bardey and
Jaramillo (2011) conclude that one shot UI programs would not necessarily have negative consequences
on labor market in developing countries, as they may not reduce the effort made by unemployed to secure
a new job in the formal sector during the same period. Vodopedic (2009) discusses how to adjust UI designs
for developing countries. Suggested adaptations include self-insurance financing, complemented by
solidarity funding, simpler eligibility conditions and even weaker monitoring.
Verónica Amarante, Rodrigo Arim y Andrés Dean
15
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
According to administrative records, the number of beneficiaries of the UI
program has shown some oscillations until 1999 and a sharp increase during
the economic crises. Average beneficiaries in 2002 more than doubled those
of 1998 (37302 versus 17652) (Graph 1).
Graph 1. Beneficiaries of the Unemployment Insurance. 1993-2009
0
1993 1995 1997 1999 2001 2003 2005 2007
5.000
10.000
15.000
20.000
25.000
30.000
35.000
40.000
Total Montevideo Rest of the country
Source: Based on Banco de Previsión Social (2010).
Data from BPS allows analyzing the profile of UI beneficiaries. Most of them
are men (70% in 2008). At the beginning of the period beneficiaries from Mon-
tevideo represented more than 55% of total beneficiaries, but by 2009 they
were just 44%. Beneficiaries are concentrated in central ages (around 50%
are between 30 and 49 years old). During the last years, efforts were made,
in terms of more requirements, to dissuade firms from using the suspension
modality, whose importance has decreased. Whereas in 2000 almost 60% of
beneficiaries corresponded to this modality, in 2008 the figure was around
33%. Finally, most of the beneficiaries have family dependents (Table 2).
The program is small in terms of the resources involved. It represents around
2% of total BPS expenditures, and less than 1% of GDP. Its financial impor-
tance increased in 2002, during the economic crises (Table 3).
The program’s coverage can be analyzed based on data from the household
survey. In this survey, unemployed are asked if they receive the unemployment
insurance. As shown in Table A2, data from the household survey very much
0
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
16
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Table 2. Characteristics of Unemployment Insurance Beneficiaries
1992 1995 2000 2005 2008 2009
Men 66.9 69.8 68.3 65.1 70.1 70.0
Women 33.1 30.2 31.7 34.9 29.9 30.0
Montevideo 55.3 63.1 59.6 51.2 43.5 43.8
Rest of the country 44.7 36.9 40.4 48.8 56.5 56.2
Younger than 20 3.0 3.4 2.1 2.1 2.4 2.1
20-29 33.0 31.7 33.6 26.6 29.5 32.6
30-39 26.1 27.4 22.1 29.9 25.0 29.6
40-49 20.5 19.9 17.4 21.1 19.6 19.6
50-59 12.2 12.7 12.7 12.0 12.4 13.0
60 and more 2.6 2.8 2.5 2.8 3.0 3.1
Job loss 43.4 41.6 43.0 60.0 65.5 62.1
Suspension 55.2 57.9 56.9 31.3 25.6 33.3
Job reduction 1.4 0.5 0.1 8.8 8.5 4.6
With family 67.7 62.9 64.1 65.7 63.1 63.4
Without family 32.3 37.1 35.9 34.3 36.9 36.6
Source: Authors’ calculations based on BPS statistical yearbook.
Table 3. Amount of UI Benefits. 1993-2009
Total Benefit Payments (Constant
Terms, Index Base Year = 1993)
Benefit Payments/
BPS Expenditure Benefit Payments/GDP
1993 100.0 2.2% 0.2%
1995 128.9 2.6% 0.2%
2000 169.6 3.0% 0.2%
2005 67.3 1.5% 0.1%
2008 105.8 2.4% 0.3%
Source: Authors’ calculations based on BPS statistical yearbook.
resembles that from administrative records, which in turn include all unem-
ployed in Uruguay. The percentage of unemployed receiving the benefit has
been between 2.4 and 6.2% during the last two decades. The higher coverage
of 6.2% of unemployed corresponds to the worst moment of the economic
crisis in Uruguay (2002) (Graph 2).6
6 Those workers who receive the unemployment insurance under the modality of suspension are considered
as employed by the household survey, and so are not included in these figures.
Verónica Amarante, Rodrigo Arim y Andrés Dean
17
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Graph 2. UI Coverage and Economic Growth
0.0
1991
1992
1993
1994
1995
1996
1997
1998
1999
2000
2001
2002
2003
2004
2005
2006
2007
2008
2009
1.0
% of unemployed covered by UI
2.0
3.0
4.0
5.0
6.0
7.0
-10
-8
GDP growth rate
-6
-4
-2
2
6
10
8
4
0
% of unemployed covered by UI GDP growth rate
Source: Authors’ calculations based on household surveys and data from Banco Central del Uruguay.
II. Empirical Strategy and Data Description
Our impact evaluation of the unemployment insurance program is based on two
data sets: administrative records from the unemployment insurance program
and a sample of longitudinal data on social security records. The main out-
comes analyzed in this paper are mean duration of unemployment and wage at
reemployment, the latter being an indicator of the quality of job matching.
Three different evaluation strategies are used: propensity score, difference in
difference and regression discontinuity design (see table 4). In the propensity
score and difference in difference strategy, the wage outcome reflects the
change between wage before the unemployment event and wage at reem-
ployment. In the regression discontinuity design, wages at reemployment of
control and treatment groups are compared.
The unemployment data cover the universe of all unemployed workers who
entered the program 15 months before and 15 months after the modifica-
tion of the program (February 2009). These data come from the administra-
tive records of BPS, and include information on sex, date of birth and sector of
activity, as well as the exact amount of money they received and the months
they were in the program. For these workers that entered the unemployment
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
18
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
insurance, we also have all their labor history until April 2010, so we can
know if they returned to work once the UI expired, and in case they returned
to employment, their wage at reemployment. One of our strategies to analyze
the effects of the change in the scheme benefits is to compare similar work-
ers before and after the modifications were implemented, comparing them by
means of propensity score matching. A sub-sample of this data set, including
workers aged 46 to 53 at the moment of unemployment, allows evaluating
the impact of the extension of duration for older workers, using regression
discontinuity design.
As a second strategy to evaluate the impact of the change in the scheme of
benefits, data on social security records were used to construct a control group
of workers who lost their formal job but were not covered by the UI. This control
group was compared to treated workers (those who entered the UI program),
before and after the change in the design, using difference in difference esti-
mates. The following table describes the evaluation strategy used to analyze
each change, detailing the treatment and control groups in each case.
Table 4. Impact Evaluation Strategy
Reform of UI
Evaluated
Evaluation
Strategy
Definition of Treatment
and Control Groups
Data Bases Used in the
Analysis
1. Change in
scheme of
payments
1. 1 Propensity
Score Matching
(PSM)
T: unemployment
beneficiaries after the
change
C: unemployment
beneficiaries before the
change
Both treatment and control
groups come from the
administrative records of the
UI program
1.2 Differences in
differences (DD)
T: unemployment
beneficiaries before and
after the change
C. Out of the labor force,
without insurance
Treatment group comes from
the administrative records of
the UI program.
Control group comes from
labor histories (social
security data)
2. Increase in
maximum duration
for 50 & older UI
recipients
2.1 Regression
Discontinuity
(RD)
T: 50-53 after the change
C: 46-49 after the change
Both treatment and control
groups come from the
administrative records of the
UI program
Source: Authors’ elaboration.
One drawback of our data for both the PSM and DD strategies is that we are
not considering the same length of time after being out of the labor force for
all workers. In fact, for those workers who entered the UI program 15 months
Verónica Amarante, Rodrigo Arim y Andrés Dean
19
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
before the change, we have information for the 30 subsequent months, whereas
for those workers who entered the UI program 10 months after the reform,
we have information only on the 5 subsequent months. In other words, the
probability that a worker gets a formal job is higher for those workers who
entered the UI before the change, because we have a longer spell of time. Fur-
thermore, the potential duration of a spell of unemployment is related to an
individual’s treatment status.
To avoid this problem and make both groups as comparable as possible, we
recoded unemployment duration for the first group of workers, allowing the
same window of time for them as that for the post reform group. For example,
if a worker became unemployed one month before the reform, and he gets
a formal job after 15 months, we consider he didn’t get a formal job in the
period (this universe is considered as sample 1).
As a second strategy to limit problems derived from the observation of incom-
plete spells, we constructed another subsample, extracted from this one, which
only considers workers with complete unemployment duration observed (sam-
ple 2). All estimations were undertaken for both samples.
III. The Effects of the Change in the Scheme of Benefits
To analyze the effects of the change in the scheme of benefits for permanently
laid off workers, we used a cohort design and propensity score matching using
individuals who entered the unemployment in the modality of job loss before
and after the change in the scheme of UI payments.
As a second strategy we used difference in difference estimator, comparing
UI beneficiaries before and after the change, with a control group of workers,
who lost their formal jobs, but did not enter the UI program.7 The following
equation was estimated:
YTtT tX
it ii iit
=+ ++++ab
11
(1)
7 This group is comprised by those unemployed who do not fulfill the tenure requirement (having worked
at least six months in the previous year), were voluntary unemployed, or have already received the
benefit during the previous twelve months.
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
20
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Where T1 = 1reflects the presence of the new UI program at t = 1, whereas
T1 = 0 denotes lack of treatment at time t = 1, and t is a time variable, being
one after the moment of the modification of the unemployment program. The
b coefficient corresponding to the interaction between the treatment and
the time variables gives the average DD effect of the program. The vector Xi
includes controls for age and sex, and controls for the month of the year were
also included in the specification.
Density functions of unemployment duration for treated individuals (laid
off workers under UI) before and after the change in the scheme of benefits
(groups B and A respectively) show some changes, as the mode detected in
the six months before the change vanishes after the change (Graph 3). The
control sample of workers who did not enter the UI program, which were used
for DD estimation (groups C and D, after and before the change respectively),
present very similar density functions.
Graph 3. Density Function of Unemployment Duration
0
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
0.02
0.04
0.06
0.08
0.10
0.12
0.14
0.16
0.18
0.20
Unemployed duration
Treated after change (A)
Treated before change (B)
0
0
1
2
3
4
5
6
7
8
9
10
11
12
13
14
15
16
17
18
19
20
0.05
0.10
0.15
0.20
0.25
0.30
0.35
Unemployed duration
Untreated before change (D)
Untreated after change (C)
Source: Authors’ calculation using a sample from administrative records from BPS.
Density functions of changes in earnings differ between treated individuals before
and after the change in the UI regime (Graph 4). Treated individuals after the
modification of the UI present a clearer mode around zero, but considerably
less mass for higher order changes. Density functions for untreated individuals
before and after the change are similar.
Verónica Amarante, Rodrigo Arim y Andrés Dean
21
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Graph 4. Density Function of Earnings Change
Treated after change Treated before change
Untreated before change Untreated after change
0
-100% -80% -60% -40% -20% -100% 20% 40% 60% 80% 100%
0.2
0.4
0.6
0.8
1.0
1.2
Wage loss
Source: Authors’ calculation using a sample from administrative records from BPS.
The simple comparison of unemployment duration means for treated and con-
trol groups indicates that unemployment duration was shorter for the former,
for the two samples considered (Table 5).
Propensity score matching between UI beneficiaries before and after the
change in the scheme of benefits indicates that the average treatment effect
on unemployment duration is negative, indicating that this change caused a
reduction in unemployment duration. The matching was done considering age,
age squared, sex and the interaction between sex and age.8
These results could indicate that the reform produced a significant but very
small reduction in the unemployment duration. To the extent that the depen-
dent variable is measured in months, a coefficient of 0.06 represents a reduc-
tion of two days, a very small magnitude.9
8 Note that the density functions of de propensity score are almost perfectly overlapped (Graph A.1).
9 This reduction in unemployment duration has taken place simultaneously with the decrease in the size
of the informal sector reported in section I. Unfortunately, our empirical strategy does not allow us to
rigorously connect both facts.
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
22
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Table 5. Mean Unemployment Duration and Average Treatment Effect On the Treated
(ATT) of Reduction in UI On Unemployment Duration (PSM Estimates)
Sample 1 (all) Sample 2 (restricted)
Average Duration
Control group 4.45 4.48
Treatment group 4.40 4.40
Unadjusted difference -0.05 -0.08
Average Treatment Effect On Treated (ATT)
Nearest neighbor matching -0.06 -0.078
(0.02) *** (0.029) ***
Stratification matching -0.073 -0.078
(0.029) *** (0.028) ***
Nº of treated observations 49,961 23,567
Nº of control observations 35,683 16,356
Note: Dependent variable: unemployment duration, in months. Standard errors in parenthesis. *** significant
at 1%.
Source: Authors’ calculations using administrative records from BPS.
Results on average earnings’ change depend on the sample considered. On aver-
age, job loss is associated with a reduction of 20 percentage points of wages
for workers that return to labor activity for both samples. But although average
change in earnings is negative for both treatment and control groups, the loss
is higher for treatment groups for the restricted sample (workers with complete
unemployment duration), whereas the contrary happens for sample 1.
Propensity scores estimations also give different results for the different sam-
ples. Under the restricted sample, which we prefer for being more demanding,
the change in the scheme of unemployment duration has implied a reduction
of average earnings loss (Table 6). The propensity score estimates then indi-
cate that the performance would be slightly better after reform, since the loss
would be approximately three points lower. This indicates that the decrease
in duration is not associated with a worse job matching in terms of earnings.
The reform did not cause the unemployed to take poorer paying jobs because
their UI benefits ran out.
Difference-in-differences estimates confirm our previous results in relation
with unemployment duration. In this case, treatment are permanent laid off
workers covered by UI and the control group are unemployed workers not cov-
ered by UI, in both cases before and after the change in the regime (Table 7). Our
Verónica Amarante, Rodrigo Arim y Andrés Dean
23
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
variable of interest, the interaction between the treatment and time variable,
indicates that the change in UI benefits caused a decrease in unemployment
duration of less than one week. The reduction is higher for men (gender = 1)
and presents a non linear effect in age. Results also indicate a reduction of
wage loss of around 9%. In this case, the reduction is higher for women. Simi-
lar results are obtained with the unrestricted sample (see Table A3).
Table 6. Mean Earnings’ Change and Average Treatment Effect On the Treated (ATT)
of Reduction in UI On Earnings Change. (PSM Estimates)
Sample 1 (all) Sample 2 (restricted)
Average wage change
Treatment group -0.21 -0.21
Control group -0.23 -0.17
Unadjusted difference -0.02 0.04
Nearest neighbor matching 0.028 -0.033
(0,004) *** (0.005) ***
Stratification matching 0.028 -0.033
(0.004) *** (0.005) ***
Nº of treated observations 25,921 20,934
Nº of control observations 21,557 14,348
Note: Dependent variable: earnings’ change, in percentage points. Standard errors in parenthesis.*** signi-
ficant at 1%.
Source: Authors’ calculations using administrative records from BPS.
The key assumption in difference in difference strategy is that the outcome
variables would have followed a similar trend in the absence of the treat-
ment (similar trend assumption). This assumption is difficult to verify. A simple
graph of the outcome variables before and after the intervention suggests that
this assumption is reasonable for both outcomes (see Graphs A1.1 and A1.2).
Another way to test for the similar trend assumption is to restrict the data
to the pre treatment period, and assume that the unemployment insurance
reform was implemented in any time, for example in the beginning of 2008
(see Duflo, 2001). Computing the difference in difference estimator for this
change estimator can help to disentangle if the outcome variables differed
significantly between treatment and control groups before the change in the
system was introduced. Results from such a placebo regression, reported in
Table A4, are not significant, suggesting that our difference and difference
results are not driven by mistaken identification assumptions.
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
24
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Table 7. Differences in Differences Estimation. Effects of the Change in UI Benefits
On Unemployment Duration and Wage Change. Sample 2 (Restricted)
Coeffi-
cient Std. Err. T P > t Confidence
Interval
Unemployment Duration
Treatment 0.782 0.034 23.25 0.000*** 0.716 0.848
Time 0.133 0.039 0.34 0.733 -0.063 0.090
Treatment*t -1.036 0.227 -4.56 0.000*** -1.481 -0.591
Treatment*t*gender 0.228 0.118 1.94 0.052 -0.002 0.459
Treatment*t*age 0.056 0.011 4.91 0.000*** 0.034 0.079
Treatment *t*age cuad -0.001 0.000 -5.24 0.000*** -0.001 0.000
Treatment* t* age*gender -0.002 0.003 -0.52 0.605 -0.008 0.004
Gender -0.387 0.030 -12.9 0.000*** -0.445 -0.328
Age -0.070 0.007 -9.86 0.000*** -0.083 -0.056
Age cuadratic 0.001 0.000 11.84 0.000*** 0.001 0.001
Nº of treated obs. Before 16,355
Nº of treated obs. After 23,568
Nº of control obs. Before 8,862
Nº of control obs. After 8,126
Wage Change
Treatment -0.031 0.007 -4.39 0.000*** -0.045 -0.017
Time -0.027 0.009 -3.06 0.002*** -0.044 -0.010
Treatment*t 0.091 0.047 1.93 0.053 -0.001 0.184
Treatment*t*gender -0.082 0.023 -3.48 0.000*** -0.128 -0.036
Treatment*t*age -0.006 0.002 -2.46 0.014 -0.011 -0.001
Treatment *t*age cuad 0.000 0.000 2.35 0.019 0.000 0.000
Treatment* t* age*gender 0.002 0.001 3.98 0.000*** 0.001 0.004
Gender 0.022 0.006 3.62 0.000*** 0.010 0.034
Age -0.002 0.002 -1.38 0.166 -0.005 0.001
Age cuadratic 0.000 0.000 1.03 0.305 0.000 0.000
Nº of treated obs. Before 14,348
Nº of treated obs. After 20,934
Nº of control obs. Before 5,622
Nº of control obs. After 5,118
Note: *** significant at 1%. Estimation included months’ fixed effects controls.
Source: Authors’ calculations using administrative records from BPS.
Verónica Amarante, Rodrigo Arim y Andrés Dean
25
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
IV. The Effects of the Extension of Benefits for Older Workers
In order to identify the causal effect of extending UI benefits for workers aged
50 or more, we compare these workers with those who fall just short of this
age of requirement. These two groups are basically similar, and their main
difference after the modification in the legislation is that these older work-
ers may stay in the UI program for a year (instead of six months). If there is a
discontinuity in the outcome variable after the intervention, it is interpreted
as a consequence of the change. A similar strategy was proposed in Lalive
(2008), although the increase in duration they analyzed was much more dra-
matic (3.5 years). As stated in that paper, this strategy could be invalidated
if firms manipulate the UI system, offering workers not to lay them off until
they are 50 years old. In our case, this may be mitigated by the fact that we
are taking the first immediate year after the modification and that this change
has not been in the public discussion of unemployment reforms, reducing the
probabilities of manipulation.
We use information on individuals entering unemployment 15 months before and
15 months after the change in the UI system, so our data covers from November
2007 to April 2010 (the change was on the 1st February 2009). Regression dis-
continuity estimations consider as treated group those who entered UI system
in February 2009 and after, and where aged 50-53 when becoming unemployed,
and control group those aged 46-49 in the same period.
Mean unemployment duration is higher for individuals aged 50 or more when com-
pared to younger ones both before and after the change in the duration of benefits.
Nevertheless, after the change the difference in means is bigger (Table 8).
Table 8. Mean Unemployment Duration (in Months)
Before After Total
46-49 5.75 4.01 4.81
50-53 5.86 5.05 5.41
46-53 5.80 4.51 5.09
Source: Authors’ calculations using administrative records from BPS.
Average unemployment duration by age at entry into unemployment before
and after the change in the UI system is reported in Graph 5. Results are pre-
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
26
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Graph 5. The Effects of the Extension in UI On Duration: Age Threshold
46 48 50
Edad
52 54
4
5
6
7
8
a) Before (men and women) b) After (men and women)
46 48 50
Edad
52 54
3.5
4.0
4.5
5.5
5.0
6.0
46 48 50
Edad
52 54
2
4
6
8
a) Before (women) b) After (women)
46 48 50
Edad
52 54
3
4
6
5
7
46 48 50
Edad
52 54
4
5
7
6
8
a) Before (men) b) After (men)
46 48 50
Edad
52 54
3
4
5
6
Source: Authors’ elaboration using administrative records from BPS.
Verónica Amarante, Rodrigo Arim y Andrés Dean
27
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
sented for all workers and for men and women separately. There seems to be a dis-
continuity in at age 50, both for men and women, before the change in the policy.
When the previous period is considered, differences in unemployment duration at
the 50 years threshold do not seem to exist, especially in the case of men.
Following the RD estimation strategy, we run the following linear regression:
YTAA TA A
iiiiii
=+ +−+−+aa aa
01 2030
()()
(2)
Where Yi is the outcome variable (duration of unemployment and wage at
employment), T is the treatment variable and A is the assignment (or the forc-
ing) variable, in our case reflecting age, with A0 = 50. We also include quadratic
and cubic expressions of Ai – A0. The parameter a1 measures the average causal
effect of the extension on UI benefits on outcome variables. As shown in Table
9, our estimates indicate that average unemployment duration is almost 4 weeks
Table 9. Effect of UI Extension On Unemployment Duration (in Months). 46-53
Linear Quadratic Cubic Linear + Sex
Control
Quadratic +
Sex Control
Cubic + Sex
Control
After the Change in UI duration
All
0.881 0.881 0.859 0.883 0.883 0.862
(0.1347)*** (0.1352)*** (0.1814)*** (0.1348)*** (0.1352)*** (0.1815)***
Nº obs. 8502 8502 8502 8502 8502 8502
Women 0.821 0.829 0.528
(0.2444)*** (0.2447)*** (0.3219)
Nº obs. 2,789 2,789 2,789
Men 0.91 0.895 1.015
(0.1612)*** (0.1617)*** (0.2190)***
Nº obs. 5,713 5,713 5,713
Before the Change in UI Duration
All 0.231 0.234 0.412 0.23 0.233 0.415
(0.2092) (0.2097) (0.2731) (0.2092) (0.2097) (0.2731)
Nº obs. 6,994 6,994 6,994 6,994 6,994 6,994
Women -0.344 -0.331 0.108
(0.3588) (0.3596) (0.4547)
Nº obs. 2,294 2,294 2,294
Men 0.527 0.522 0.571
(0.2573)** (0.2577)** (0.3398)*
Nº obs. 4,700 4,700 4,700
Note: *** significant at 1%.
Source: Authors’ calculations using administrative records from BPS.
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
28
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
longer for those aged 50-53 when compared to those aged 46-49. If the same
regression is run with data from the period before the change was intro-
duced, the treatment variable is only weakly significant in some of the speci-
fications for men, indicating that for all workers, the effect can be explained
by the change in the policy. The difference is never significant for women.
The effect detected for men before the policy change is consistent with the
hint of a discontinuity for men before the change (Graph 5). The increase in
unemployment duration due to the extension of benefits is mainly explained
by women’s behavior.
Estimations were also done considering narrower age bins, instead of the group
46-54. In particular, we considered 49-50, 48-51 and 47-52. As Tables 10 to 12
show, results are maintained for these groups. As the age bin is wider, the effects
become stronger. The effect is quite robust: the extension in the UI duration for
older workers leads to an increase in unemployment duration for older workers.
Table 10. Effect of UI Extension On Unemployment Duration (in Months). 49-50
Linear Quadratic Cubic Linear +
Sex Control
Quadratic +
Sex Control
Cubic + Sex
Control
After the Change in UI Duration
All
0.629 0.631 0.582 0.618 0.62 0.564
(0.2717)** (0.2719)** (0.3625) (0.2715)** (0.2717)** (0.3625)
Nº obs. 2,112 2,112 2,112 2,112 2,112 2,112
Women
-0.12 -0.121 -0.163
(0.4754) (0.4743) (0.6370)
Nº obs. 690 690 690
Men
0.976 0.984 0.994
(0.3302)*** (0.3297)*** (0.4387)**
Nº obs. 1,422 1,422 1,422
Before the Change in UI Duration
All -0.0794 -0.0485 -0.113 -0.0769 -0.0459 -0.0794
(0.3863) (0.3880) (0.5227) (0.3860) (0.3876) (0.3863)
Nº obs. 1,752 1,752 1,752 1,752 1,752 1,752
Women -0.398 -0.442 -0.109
(0.6510) (0.6566) (0.9058)
Nº obs. 591 591 591
Men 0.0627 0.141 -0.0682
(0.4762) (0.4771) (0.6340)
Nº obs. 1,161 1,161 1,161
Note: *** significant at 1%.
Source: Authors’ calculations using administrative records from BPS.
Verónica Amarante, Rodrigo Arim y Andrés Dean
29
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Table 11. Effect of UI Extension On Unemployment Duration (in Months). 48-51
Linear Quadratic Cubic Linear + Sex
Control
Quadratic +
Sex Control
Cubic + Sex
Control
After the Change in UI Duration
All 0.853 0.845 0.857 0.858 0.849 0.861
(0.1932)*** (0.1939)*** (0.2575)*** (0.1932)*** (0.1939)*** (0.2575)***
Nº obs. 4,201 4,201 4,201 4,201 4,201 4,201
Women 0.453 0.457 0.374
(0.3405) (0.3400) (0.4487)
Nº obs. 3,903 4,083 4,122
Men 1.042 1.029 1.056
(0.2336)*** (0.2347)*** (0.3127)***
Nº obs. 4,119 427 4,256
Before the Change in UI duration
All 0.28 0.284 0.143 0.292 0.296 0.163
(0.2874) (0.2882) (0.3720) (0.2874) (0.2883) (0.3719)
Nº obs. 3,516 3,516 3,516 3,516 3,516 3,516
Women -0.0264 -0.0197 -0.12
(0.4788) (0.4808) (0.6350)
Nº obs. 1,172 1,172 1,172
Men 0.432 0.432 0.275
(0.3574) (0.3582) (0.4562)
Nº obs. 2,344 2,344 2,344
Note: *** significant at 1%.
Source: Authors’ calculations using administrative records from BPS.
The same analysis was done considering earnings at reemployment as out-
come variable. The graphical analysis (Graph 6) is less clear than in the case of
duration. In any case, it indicates that older workers tend to find worse jobs,
in terms of payment, after the reform. The extension in the UI benefit does
not help workers to get better jobs by subsidizing job search.
Regression analysis shows that there are no differences in wages at reemploy-
ment when treated individuals are compared with untreated ones (Table 13).
The effect is positive for the linear and quadratic specification, and negative
for the cubic one, but never significant. In all cases, we are only considering
workers who reenter the labor market. The treatment coefficient is not sig-
nificant for men or woman, and when estimations are done considering
narrower age bins, results remain the same (Tables A.5 to A.7).
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
30
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Table 12. Effect of UI Extension On Unemployment Duration (in Months). 47-52
Linear Quadratic Cubic Linear +
Sex Control
Quadratic +
Sex Control
Cubic + Sex
Control
After the Change in UI Duration
All 0.783 0.788 0.92 0.786 0.791 0.924
(0.1559)*** (0.1564)*** (0.2097)*** (0.1560)*** (0.1565)*** (0.2098)***
Nº obs. 6,332 6,332 6,332 6,332 6,332 6,332
Women 0.598 0.608 0.352
(0.2795)** (0.2798)** (0.3665)
Nº obs. 2,078 2,078 2,078
Men 0.873 0.866 1.183
(0.1875)*** (0.1882)*** (0.2549)***
Nº obs. 4,254 4,254 4,254
Before the Change in UI Duration
All 0.35 0.352 0.156 0.351 0.353 0.168
(0.2386) (0.2388) (0.3096) (0.2386) (0.2388) (0.3096)
Nº obs. 5,216 5,216 5,216 5,216 5,216 5,216
Women -0.129 -0.102 -0.21
(0.4030) (0.4039) (0.5117)
Nº obs. 1,704 1,704 1,704
Men 0.602 0.591 0.322
(0.2953)** (0.2953)** (0.3864)
Nº obs. 3,512 3,512 3,512
Note: *** significant at 1%.
Source: Authors’ calculations using administrative records from BPS.
Graph 6. The Effects of the Extension in UI On Wages: Age Threshold
46 48 50
Edad
52 54
6.000
8.000
10.000
12.000
a) Before (men and women) b) After (men and women)
46 48 50
Edad
52 54
4.000
5.000
6.000
8.000
7.000
Verónica Amarante, Rodrigo Arim y Andrés Dean
31
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Graph 6. The Effects of the Extension in UI On Wages: Age Threshold (continued)
a) Before (women) b) After (women)
a) Before (men) b) After (men)
46 48 50
Edad
52 54
3.000
4.000
5.000
6.000
7.000
8.000
46 48 50
Edad
52 54
2.000
3.000
4.000
7.000
5.000
6.000
46 48 50
Edad
52 54
6.000
8.000
10.000
12.000
14.000
46 48 50
Edad
52 54
4.000
6.000
8.000
10.000
Source: Authors’ calculations using administrative records from BPS.
Table 13. Effect of UI Extension On Wages At Reemployment ($U dec 2009)
Linear Quadratic Cubic Linear +
Sex Control
Quadratic +
Sex Control
Cubic + Sex
Control
After the Change in UI Duration
All 564.8 556 -532.5 393.5 392.5 -555.4
(554) (560) (702) (531) (538) (673)
Nº obs. 4,439 4,439 4,439 4,439 4,439 4,439
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
32
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Table 13. Effect of UI Extension On Wages At Reemployment ($U dec 2009)
(continued)
Linear Quadratic Cubic Linear +
Sex Control
Quadratic +
Sex Control
Cubic + Sex
Control
After the Change in UI Duration
Women -36 -34.32 -908.8
(541) (541) (703)
Nº obs. 7,669 7,647 8,029
Men 594.5 589.3 -424.3
(736) (747) (932)
Nº obs 12,856 12,903 13,361
Before the Change in UI Duration
All -99.12 -92.07 -139.3 -27.7 -24.74 -205.9
(447) (448) (613) (432) (433) (592)
Nº obs. 5,822 5,822 5,822 5,822 5,822 5,822
Women 427.1 429.1 -192.2
(540) (542) (728)
Nº obs. 6,897 6,850 7,125
Men -237.3 -233.2 -218.4
(573) (575) (781)
Nº obs. 12,204 12,160 12,152
Note: *** significant at 1%.
Source: Authors’ calculations using administrative records from BPS.
V. Concluding Remarks
Important modifications in the Uruguayan UI program were introduced in
2009. In this article, we presented an impact evaluation of two of them: the
change in benefit payments from a lump sum system to a decreasing scheme
and the extension of UI duration for workers 50 or older.
The first of these modifications has implied a reduction in unemployment
duration. This result holds both for propensity score and difference in differ-
ence estimations, but the magnitude of the effect is small. This decrease in
duration is not associated with a worse matching in terms of earnings. Peo-
ple tend to take a shorter time to find a new job under a decreasing scheme
Verónica Amarante, Rodrigo Arim y Andrés Dean
33
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
of benefits, and this shorter time does not affect the quality of job matching.
Although estimated effects are small in magnitude, they indicate that there
is a certain margin for improving the design of UI programs in countries with
lump sum schemes.
The possibility of extension of UI duration for workers aged 50 or more has
implied an extension in unemployment duration for older workers, and it has
not helped to subsidize better job matches in the form of better paying jobs.
The extension in unemployment benefits we analyzed was considerable (24
weeks more) and had a sizeable negative impact on unemployment duration
without implying an improvement in terms of wages in the next job. This result
casts doubts about the efficiency of this type of modification.
In all cases, the lack of effect on earnings at reemployment indicates that the
UI program in Uruguay acts mainly as a temporary income insurance, and not
as a subsidy for more productive job search.
References
ADDISON, J. y BLACKBURN, L. (2000). “The effects of unemploy-1.
ment insurance on postunemployment earnings”, Labour Economics,
7(1):21-53.
ÁLVAREZ-PARRA, F. y SÁNCHEZ, J. M. (2009). “Unemployment insur-2.
ance with a hidden labor market”, Journal of Monetary Econom ics,
56(7):954-96.
AMARANTE, V. y BUCHELI, M. (2008). “El seguro de desempleo en 3.
Uruguay”, Cuadernos del CLAEH 96-97, 2º serie, año 31, 96-97:
175-207.
BANCO DE PREVISIÓN SOCIAL (2010). Boletín Estadístico 2010. Avail-4.
able at www.bps.gub.uy.
BARDEY, D. y JARAMILLO, F. (2011). Unemployment insurance and infor-5.
mality in developing countries (Working Paper 11-257). TSE Toulouse
School of Economics.
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
34
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
BELZIL, C. (2001). Unemployment insurance and subsequent job duration: 6.
Job matching vs. unobserved heterogeneity (Working Paper 2001s-21).
Cirano.
CARD, D. y LEVINE, P. (2000). “Extended benefits and the duration of 7.
UI spells: Evidence from the New Jersey extended benefit program”,
Journal of Public Economics, 78(1-2):107-138.
CHETTY, R. (2008). “Moral hazard versus liquidity and optimal unemploy-8.
ment insurance”, Journal of Political Economy, 116(2):173-234.
CLASSEN, K. (1977). “The effect of unemployment insurance on the 9.
duration of unemployment and subsequent earnings”, Industrial and
Labor Relations Review, 30(8): 438-444.
DUFLO, E. (2001). “Schooling and labor market consequences of school 10.
construction in Indonesia: Evidence from an unusual policy experiment”,
American Economic Review, 91(4):795-813.
EHRENBERG, R. y OAXACA, R. (1976). “Unemployment insurance, dura-11.
tion of unemployment, and subsequent wage gain”, American Economic
Review, 66:754-766.
GONZÁLEZ ROZADA, M., RONCONI L. y RUFFO, H. (2011). Protecting 12.
workers against unemployment in Latin America and the Caribbean:
Evidence from Argentina (Research Department Publications 4759).
Inter-American Development Bank, Research Department.
HOPENHAYN, H. y NICOLINI. J. (1997). “Optimal unemployment insur-13.
ance”, The Journal of Political Economy, 105(2): 412-438.
HUNT, J. (1995). “The effect of unemployment compensation on 14.
unemployment duration in Germany”, Journal of Labor Economics, 13:
88-120.
KATZ, L. y MEYER, B. (1990). “The impact of the potential duration of 15.
unemployment benefits on the duration of unemployment”, Journal of
Public Economics, 41:45-72.
Verónica Amarante, Rodrigo Arim y Andrés Dean
35
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
LALIVE, R. (2008). “How do extended benefits affect unemployment 16.
duration? A regression discontinuity approach”, Journal of Econometrics,
142:785-806.
MARIMON, R. y ZILIBOTTI, F. (1999). “Unemployment vs. mismatch of 17.
talents: Reconsidering unemployment benefits”, Economic Journal,
109:266-291.
MAZZA, J. (2000). Unemployment insurance: Case studies and lessons 18.
for Latin America and the Caribbean (Working Paper 342). IDB.
MEYER, B. (1990). “Unemployment insurance and unemployment spells”, 19.
Econometrica, 58(4):757-782.
MOFFIT, R. (1985). “Unemployment insurance and the distribution of 20.
unemployment spells”, Journal of Econometrics, 28:85-101.
MORTENSEN, D. (1977). “Unemployment insurance and job search deci-21.
sions”, Industrial and Labor Relations Review, 30:505-517.
TATSIRAMOS, K. (2009). “Unemployment insurance in Europe: Unem-22.
ployment duration and subsequent employment stability”, Journal of
the European Economic Association, 7(6):1225-1260.
VAN OURS, J. y VODOPIVEC, M. (2008). “Does reducing unemployment 23.
insurance generosity reduce job match quality?”, Journal of Public
Economics, 92:684-695.
VAN OURS, J. y VODOPIVEC, M. (2006). “How shortening the potential 24.
duration of unemployment benefits entitlement affects the duration of
unemployment: Evidence from a natural experiment”, Journal of Labor
Economics, 24:351-378.
VODOPIVEC, M. (2009). Introducing unemployment insurance to 25.
developing countries (Paper 0907). Social Protection Discussion, World
Bank.
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
36
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Annex
Table A1. Workers without Social Security Contributions
Año Private Workers All Workers
2001 30% 36%
2002 31% 37%
2003 34% 40%
2004 36% 41%
2005 34% 39%
2006 28% 35%
2007 27% 34%
2008 25% 33%
2009 24% 32%
Source: Authors’ calculations based on household surveys.
Table A2. Comparison of Sex and Age Profile of Unemployed Receiving UI.
2007-2009
Household Survey Administrative Records (BPS)
Women 34.2% 33.6%
Men 65.8% 66.4%
Age
< 21 5.4% 4.3%
21 to 30 32.4% 33.4%
31 to 40 25.4% 29.0%
41 to 50 20.3% 19.5%
> 50 16.5% 13.9%
Source: Authors’ calculations based on household surveys and administrative records from BPS.
Verónica Amarante, Rodrigo Arim y Andrés Dean
37
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Table A3. Differences in Differences Estimation. Effects of the Change in UI Benefits
On Unemployment Duration and Wage Change. Sample 1 (Unrestricted)
Coefficient Std. Err. T P > t Confidence Interval
Unemployment Duration
Treatment 0.784 0.034 23.38 0.000 0.718 0.850
Time 0.069 0.038 1.79 0.074 -0.007 0.144
Treatment*t -1.125 0.224 -5.02 0.000 -1.565 -0.686
Treatment*t*gender 0.256 0.116 2.20 0.027 0.028 0.484
Treatment*t*age 0.059 0.011 5.22 0.000 0.037 0.082
Treatment *t*age cuad -0.001 0.000 -5.57 0.000 -0.001 -0.001
Treatment* t* age*gender -0.002 0.003 -0.63 0.528 -0.008 0.004
Gender -0.387 0.030 -13.05 0.000 -0.446 -0.329
Age -0.070 0.007 -9.95 0.000 -0.083 -0.056
Age cuadratic 0.001 0.000 11.90 0.000 0.001 0.001
Nº of treated obs. Before 16,422
Nº of treated obs. After 24,267
Nº of control obs. Before 8,907
Nº of control obs. After 8,575
Wage Change
Treatment -0.101 0.006 -17.59 0.000 -0.113 -0.090
Time -0.017 0.008 -2.16 0.031 -0.033 -0.002
Treatment*t 0.104 0.044 2.36 0.018 0.018 0.191
Treatment*t*gender -0.086 0.023 -3.72 0.000 -0.132 -0.041
Treatment*t*age -0.003 0.002 -1.31 0.190 -0.008 0.001
Treatment *t*age cuad 0.000 0.000 1.43 0.151 0.000 0.000
Treatment* t* age*gender 0.002 0.001 3.78 0.000 0.001 0.004
Gender 0.024 0.005 4.78 0.000 0.014 0.033
Age -0.005 0.001 -3.72 0.000 -0.007 -0.002
Age cuadratic 0.000 0.000 3.14 0.002 0.000 0.000
Nº of treated obs. Before 25,920
Nº of treated obs. After 21,558
Nº of control obs. Before 8,479
Nº of control obs. After 5,434
Source: Authors’ calculations using administrative records from BPS.
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
38
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Table A4. Differences in Differences Estimation. Effects of the Change in UI Benefits
On Unemployment Duration and Wage Loss. P Placebo Regressions
Coef. Std. Err. T P>t Confidence Interval
Unemployment Duration
Treatment 0.88 0.091 9.73 0.000 0.706 1.062
Time 0.02 0.017 0.80 0.490 -0.090 0.100
Treatment*t -0.07 0.713 -0.09 0.927 -0.146 0.133
Treatment*t*gender -1.59 0.395 -3.44 0.001 -2.133 -0.058
Treatment*t*age 0.03 0.036 0.95 0.342 -0.036 0.104
Treatment *t*age cuad 0.00 0.000 -1.74 0.083 -0.002 0.000
Treatment* t* age*gender 0.03 0.010 3.05 0.002 0.011 0.051
Gender -0.25 0.080 -3.07 0.002 -0.403 -0.089
Age -0.10 0.019 -5.34 0.000 -0.014 -0.064
Age cuadratic 0.00 0.000 6.48 0.000 0.001 0.002
Wage Loss
Treatment -2.22 1.692 -1.31 0.190 -5.534 1.098
Time 0.01 0.015 0.82 0.540 -0.120 0.100
Treatment*t -9.85 1.370 -0.72 0.472 -3.670 1.700
Treatment*t*gender -1.37 7.452 -0.18 0.854 -1.598 1.324
Treatment*t*age 0.33 0.695 0.47 0.635 -1.033 1.693
Treatment *t*age cuad 0.00 0.009 -0.52 0.603 -0.021 0.012
Treatment* t* age*gender 0.08 0.191 0.44 0.657 -0.290 0.460
Gender -2.01 1.443 -1.39 0.164 -4.839 0.818
Age -0.47 0.364 -1.30 0.195 -0.118 0.242
Age cuadratic 0.01 0.005 1.09 0.277 -0.004 0.014
Source: Authors’ calculations using administrative records from BPS.
Verónica Amarante, Rodrigo Arim y Andrés Dean
39
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Table A5. Effect of UI Extension On Wages At Reemployment ($U dec 2009). 49-50
Linear Quadratic Cubic Linear + Sex
Control
Quadratic + Sex
Control
Cubic + Sex
Control
After the Change in UI Duration
All -1,812 -1,799 7,492 -1,743 -1,730 -253.1
(1,059)* (1,067)* (1,452) (1,016)* (1,024)* (1,400)
Nº obs. 1,123 1,123 1,123 1,123 1,123 1,123
Women -858.4 -860.7 -125.7
(1,060) (1,060) (1,445)
Nº obs. 8,048 8,102 7,813
Men -2,133 -2,107 -302.6
(1,412) (1,427) (1,961)
Nº obs 14,075 13,545 12,594
Before the Change in UI Duration
All 398.5 388.3 994.7 147.3 147.2 762.4
(930) (922) (1200) (895) (886) (1153)
Nº obs. 1,442 1,442 1,442 1,442 1,442 1,442
Women -148.3 -77.76 -651.1
(1,030) (1,019) (1,144)
Nº obs. 7,179 6,022 6,303
Men 272.5 213.1 1,239
(1,181) (1,169) (1,534)
Nº obs. 12,175 12,637 12,059
Source: Authors’ calculations using administrative records from BPS.
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
40
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Table A6. Effect of UI Extension On Wages At Reemployment ($U dec 2009). 48-51
Linear Quadratic Cubic
Linear
+ Sex
Control
Quadratic
+ Sex
Control
Cubic
+ Sex
Control
After the Change in UI Duration
All -744.6 -719.5 -1,735 -836.1 -819 -1,876
(742) (747) (1,036)* (711) (715) (996)*
Nº obs. 2,175 2,175 2,175 2,175 2,175 2,175
Women -845.6 -847.7 -815.7
(742.7619) (746.1784) (985.6998)
Nº obs. 701 701 701
Men -864.3 -835.1 -2,408
(987) (994) (1,402)*
Nº obs 1,474 1,474 1,474
Before the Change in UI Duration
All -289.2 -343.2 208.1 -418 -464.3 -8.289
(661) (658) (891) (638) (636) (860)
Nº obs. 2,919 2,919 2,919 2,919 2,919 2,919
Women 10.96 11.06 -180.3
(774) (777) (993)
Nº obs. 889 889 889
Men -597.2 -682.9 50.24
(845) (839) (1133)
Nº obs. 2,030 2,030 2,030
Source: Authors’ calculations using administrative records from BPS.
Verónica Amarante, Rodrigo Arim y Andrés Dean
41
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Table A7. Effect of UI Extension On Wages At Reemployment ($U dec 2009). 47-52
Linear Quadratic Cubic Linear + Sex
Control
Quadratic +
Sex Control
Cubic + Sex
Control
After the Change in UI Duration
All 109.7 120.8 -1,343 -5.175 0.762 -1,301
(623) (632) (806)* (597) (605) (771)*
Nº obs. 3,302 3,302 3,302 3,302 3,302 3,302
Women -426.7 -412.1 -1,209
(606) (608) (800)
Nº obs. 1,062 1,062 1,062
Men 177.9 172.4 -1,360
(829) (845) (1,074)
Nº obs 2,240 2,240 2,240
Before the Change in UI Duration
All -113.1 -108.8 -166.3 -97.37 -87.89 -402.5
(519) (520) (719) (502) (503) (695)
Nº obs. 4,336 4,336 4,336 4,336 4,336 4,336
Women 208.1 201.9 -209.3
(631) (634) (840)
Nº obs. 1,294 1,294 1,294
Men -243.8 -224.3 -468.2
(663) (663) (916)
Nº obs. 3,042 3,042 3,042
Source: Authors’ calculations using administrative records from BPS.
Unemployment Insurance Design and Its Effects: Evidence for Uruguay
42
D E S A R R O . S O C . NO. 71, B O G O T Á , P R I M E R S E M E S T R E D E 2013, P P . 7-42, I S S N 0120-3584
Graph A1.1. Unemployment Duration for Treated and Control Groups Before and After
the Change in UI
0
2007-12
2008-02
2008-04
2008-06
2008-08
2008-10
2008-12
2009-02
2009-04
2009-06
2009-08
2009-10
2009-12
2010-02
2
4
6
8
10
12
Treated Control
Before UI reform After UI reform
Source: Based on administrative records from BPS.
Graph A1.2. Wages for Treated and Control Groups Before and After the Change in UI
0
2007,12
2008,02
2008,04
2008,06
2008,08
2008,10
2008,12
2009,02
2009,04
2009,06
2009,08
2009,10
2009,12
2010,02
2
4
6
8
10
12
14
Treated Control
Before UI reform After UI reform
Source: Based on administrative records from BPS.
