The effect of shortening potential unemployment
benefit duration on the duration of unemployment
Arthur Croes
Studentnumber: 10095624
University of Amsterdam
09-‐07-‐2014
Abstract
In this paper the effect of shortening the potential unemployment benefit duration on
unemployment duration is researched. To research this effect a policy change in Germany was used that provided 6 age groups with a reduction in potential benefit duration and 1 age group without a reduction that acted as a control group. No significant differences in both exit rates and coefficients of the reform dummy between the control group and the treated group were found. Also there was no spike found in exit rates just prior to benefit exhaustion, contrary to prior research.
Introduction
Unemployment in the Euro area was 11.9% in April 2014, compared to 7.5% in April 2008 (Eurostat). Unemployment is thus on average a lot higher in the Euro area than it was in 2008 and countries within the Euro area are looking for ways to reduce the unemployment in their country. The Dutch government is planning to reduce the maximum duration of unemployment benefits to reduce unemployment (Rijksoverheid), and several countries like for example Germany and Poland took similar measures in the past to reduce unemployment. The link between unemployment benefits and the duration of unemployment has been researched several times in the existing literature. Katz & Meyer (1990), Lalive (2007), Caliendo, Tatsiramos and Uhlendorff (2009) and quite some other researchers have tried to find out whether there is a link between the duration of unemployment benefits and the duration of unemployment, and whether the unemployment benefits carry disincentive effects. Most of these studies find that longer benefit duration leads to longer
unemployment duration, with a spike in exit rates just before unemployment benefits are exhausted implying that there is a strong relationship.
In this study this relation between the potential duration of unemployment benefits and the duration of unemployment will be researched. Duration of unemployment benefits is chosen instead of the level of unemployment benefits because of two reasons. Firstly because the level of unemployment benefits is mainly dependent on wage earned before entering unemployment. This makes it difficult to assess the impact of the difference in unemployment benefit level because there are a lot of factors that influence the wage a worker earns and there might be intrinsic differences that are unobservable between people that earn high wages and people that earn low wages. Duration of unemployment benefits is only dependent on age and work experience in a certain time span before entering unemployment, which are observable characteristics. The second reason is that government unemployment insurance policy could have discontinuities with respect to age or a large change in unemployment insurance policy with a reduction in maximum unemployment benefits duration. A discontinuity or a policy change makes it ideal for research since the average unemployment duration just prior to the age threshold or the policy change date can be compared to the average unemployment duration just after. Differences in duration could be attributed to the change in benefit duration because there is no reason to assume people aged 54 and people just turned 55 are very different. It can also be assumed that unemployed people prior to a reform are not different from unemployed people after a reform.
For this research policy changes in Germany will be used called the Hartz reforms. The Hartz reforms took place from 2003-‐2008 and were aimed at reducing the number of unemployed workers. They
started with Hartz I-‐III in 2003 and 2004 before the change in maximum unemployment benefit duration that was called Hartz IV was implemented. The Hartz reforms contained a number of important changes in unemployment insurance policy. One of the most important parts of Hartz IV was that the potential duration of unemployment insurance was reduced significantly, with different changes for different age groups, starting on the 1st of February 2006 (table 1). Besides this change there were other important changes that could possibly influence the results of this research, these will be discussed in the section The Hartz reforms. The reduction in unemployment benefits will be used in this paper to research the relation between potential unemployment benefit duration and unemployment duration.
Panel data from the German Socio-‐Economic Panel (GSOEP) is used in this paper, a longitudinal survey that started in 1984 and has data up to 2012. However, for unemployment spells the data for 2011-‐2012 is not yet available so the range of years that will be considered will be 2001-‐2010. In this paper the difference in differences approach will be used. It is assumed that had there not been a reform, the changes in unemployment duration would have developed over time in the same way for both treated groups and control group. Since the potential unemployment benefit duration did not change for workers aged under 45 this age group will act as a control group to make sure that the analysis does not just capture a general labor market trend. Possible spikes in exit rates close to benefit exhaustion will also be researched in this paper.
In this paper first the standard job search theory will be shortly discussed along with the results of existing literature. After this the details of the reforms in Germany will be provided, followed by a description of the dataset and the specification of the econometrical model. Subsequently the results will be discussed and finally a conclusion will be given with respect to the influence that potential unemployment benefit duration has on unemployment duration.
Literature
Job search theory
Mortensen (1977) wrote an overview of standard or original job search theory, which is relevant for this review as it could provide theoretical explanations for the results found in this review. A
standard job search model consists of a person that gets a number of wage offers, n, during a certain period and each of these wage offers is randomly drawn from distribution F. An unemployed worker will only stop searching if the offered wage is at least equal to a certain reservation wage. For this wage the value of continuing the search and the value of accepting the wage and receiving this wage forever are the same. It is optimal for the worker to stop searching if the highest wage offer received in a period is higher or equal to the reservation wage. With the inclusion of unemployment benefits,
the value of continuing the search is also influenced by the benefit rate and duration of benefits. According to Mortensen (1977), the reservation wage unambiguously increases as a consequence of an increase in benefit duration. This prolongs the unemployment duration since it would take more time to get offered a wage as high as this reservation wage.
Standard job search theory thus predicts a longer duration of unemployment as a consequence of receiving unemployment benefits. This effect of unemployment benefits has been studied
extensively in existing literature.
The existing research has used either a change or a discontinuity in unemployment insurance policy to research the relation between potential benefit duration and unemployment duration.
The results of the existing literature are in line with the theory. Lalive (2007) uses a policy change in Austria in his research where the maximum duration of 30 weeks was extended to 209 weeks for unemployed workers aged 50 years or over in 1988. He finds that average unemployment duration is 14.6 weeks longer for men aged between 50 and 53 than for men aged between 46 and 49, but controlling for pre-‐existing differences between these two age groups the effect is 11.3 weeks. For women the difference in unemployment duration between the same age groups as the men is 68.2 weeks, according to Lalive this is because women could go into early retirement at a lower age than men. Using the same policy change, Lalive, Van Ours and Zweimüller (2006) find a relatively larger effect of UI extension for older workers, which they explain is because older workers face different labor conditions as well as a lower value from finding a job because they are close to retirement. Katz & Meyer (1990) also find a negative effect of UI on unemployment duration. They use data from the United States containing UI recipients and nonrecipients and also find a spike in hazard rate close to benefit exhaustion. Meyer (1990) uses different data that contains unemployed workers entitled to two different lengths of benefit. He finds that the workers with shorter entitlement are
unemployed for a shorter of time than workers with longer entitlement, and he also finds the spike in hazard rate close to benefit exhaustion. Caliendo, Tatsiramos and Uhlendorff (2009) use a
discontinuity also used in this research, the increase in potential unemployment benefit at the age of 45 in Germany. They find a spike in the hazard rate just prior to benefit exhaustion and find that extending benefit duration reduces the job-‐finding rate.
Also using a policy change, but in Slovenia, Van Ours and Vodopivec (2006) find that reducing UI from six to three months has no significant effect on job-‐finding rates but just on transition rates to other destinations, for example to active labor market programs. Other reductions, from 9 to 6 months, from 12 to 6 months and from 18 to 9 months have significant effects on both job-‐finding
and other transition rates. Their estimates also show spikes in the hazard rate out of unemployment in the month the UI is exhausted. These spikes are also found in several other studies, and they are an important factor through which the link between UI and the duration of unemployment can be researched.
Card & Levine (2000) use a unique temporary policy change in New Jersey that is unrelated to changes in macroeconomic factors like the state of the economy and specifically the state of the labor market. This makes it suitable for this particular research since often changes in UI policy are correlated to macroeconomic conditions. Since the policy change was temporary, they simulate the long run effects of a benefit extension of thirteen weeks. The result is that a benefit program
extension of thirteen weeks would prolong the regular UI claim by 1 week. They also find a spike like the one mentioned earlier, but they find no difference in this spike between the year with the policy change and other years, which indicates that the temporary UI extension has little or no influence on the spike.
The Hartz Reforms
The German unemployment rate rose from 7.8 in 2001 to 11.1 in 2005 (Worldbank) and because of the high unemployment the German government wanted to reform the unemployment system. Eichhorst, Grienberger-‐Zingerle and Konle-‐Seidl (2007) provide a summary of activation policies in Germany. An important part of the reforms, and essential for this research, was the change in unemployment insurance policy. Before the Hartz reforms were implemented the unemployment insurance system in Germany consisted of three parts:
-‐ Unemployment benefits, eligible for up to 32 months if a certain work experience
requirement was met, 67% of net weekly remuneration if insured and at least one child and 60% if no child.
-‐ Unemployment assistance, assistance after exhaustion of benefits for an unlimited amount of time for the long-‐term unemployed
-‐ Social assistance, for people with or without work experience that do not have enough income from other sources and need extra assistance. It was to be paid by municipalities. For the last part, social assistance, any possible job was considered good enough for the unemployed worker, so the testing was strict. For the other two parts it was less strict but very ambiguous. There were no specific conditions a job had to fulfill to be acceptable. Contrary to the national
responsibility for unemployment benefits, the reemployment efforts for social assistance recipients were local responsibility, with local authorities effectively providing jobs that were covered by social
insurance for social assistance recipients in order to switch the workers to the unemployment benefit pillar.
Because with this system the number of long-‐term unemployed rose continuously and therefore the number of benefit recipients, reforms were introduced called the Hartz reforms. These reforms were aimed at reducing unemployment, mainly through stimulating labor market participation and
changing the generosity of the unemployment benefits. One important part of the reforms was the merge of the social assistance and unemployment assistance and with this a change to one single authority that is responsible for the payment and reemployment efforts.
Another change in the new unemployment insurance system was with respect to job search efforts. If an unemployed worker refuses suitable work or does not make enough effort to look for a job disqualification periods could be imposed and ultimately the benefits could cease to be paid after repeated cases. This could influence the results of this research because it could influence the duration of unemployment, which is the goal of the change. A disqualification period could prevent a person from refusing a certain job or possibly make more effort after the period would be imposed. This would add an extra factor that influences unemployment duration besides the change in benefit entitlement, and since it occurred simultaneous to the change in benefit entitlement it could
influence the results.
The generosity of unemployment benefits was reduced significantly, the maximum eligibility for unemployment benefits changed from 32 months to 18 months for people of 58 or over, from 24-‐30 months to 15 months for people aged 55-‐57 and from 20-‐26 months to 12 months for people aged 45-‐55. The precise eligibilities with respect to work experience and age can be seen in table 1. In table 1 it can also be seen that in 2008 the reductions were eased, especially for older workers. The maximum potential duration was extended from 18 to 24 months for workers aged 58 or over.
Table 1
Maximum duration of unemployment benefit and required work experience – before and after the Hartz reforms
Length of benefit entitlement (in months) Age (in years) Months worked in last 7 years
Length of benefit entitlement (in months) Age (in years) Months worked in last 5/7 years
Prior to the Hartz Reforms February 1, 2006 -‐ February 28, 2008
6 -‐ 12 6 -‐ 12 8 -‐ 16 8 -‐ 16 10 -‐ 20 10 -‐ 20 12 -‐ 24 12 -‐ 24 14 45 28 15 55 30 16 45 32 18 58 36 18 45 36 Since March 1, 2008 20 47 40 6 -‐ 12 22 47 44 8 -‐ 16 24 52 48 10 -‐ 20 26 52 52 12 -‐ 24 28 57 56 15 50 30 30 57 60 18 55 36 32 57 64 24 58 48
Source: Caliendo & Hogenacker (2012)
Data and Model
Data used in this paper comes from the German Socio-‐Economic Panel (GSOEP). In the GSOEP 26,866 unemployment spells were recorded, of which 6,357 were either left or right censored. For this analysis only the years 2001-‐2009 will be considered, so there were 17,508 unemployment spells removed because the starting date of the spell was outside this range. In order to see if a person
would be eligible for unemployment benefits, and for what length of benefit receipt the person would be eligible, the age of a worker at the start of the unemployment spell was included. Another factor that influences the duration of unemployment benefits is work experience. In Germany around the Hartz reforms the minimum work experience needed to receive a certain length of unemployment benefits was different for different ages. The minimum work experience had to be gained in the last 7 years, so there was a variable created with respect to work experience to make sure the workers from the sample qualified for the maximum duration of benefit entitlement. Workers with a work experience of less than 12 months in all of the years 2001-‐2009 were deleted from the sample, workers that did not receive benefits at any point in the chosen time period were also deleted from the sample since they would not be relevant in this research. These workers could have acted as another control group, they do not receive benefits both before and after the reforms and therefore experience no change, but one control group was deemed enough. The existing control group is fairly large and the results for this group are all statistically significant. To look at differences between sexes and married and single workers variables with respect to these characteristics were also included. Finally what is left are 7,150 unemployment spells.
Table 2
Descriptive Statistics
Male Female Total
Gender 3,926 3,224 7,150 No Yes Family 5,030 2,120 7,150
Min Max Mean Std Dev Median
Unemployment Duration 0 121 8.94 12.77 5 Age 16 65 33.89 12.99 31 Work-‐exp 12 84 45.10 30.66 48
In table 2 descriptive characteristics are provided for the important variables. The 7,150 unemployed workers consist of 3,926 males and 3,224 females. There are 5,030 unmarried workers and 2,120 married workers. The maximum unemployment duration is 121 months and the maximum work experience is 84, as could be expected since 7 years work experience with 12 months a year makes 84. The average age of the sample is about 34, which means that the sample contains significantly more unemployed workers that do not experience a change in benefit entitlement. Hence, there is a large control group.
Model
There is a quite a lot of right-‐censoring in the data, and because of this a hazard rate model will be used to estimate the probability of leaving unemployment. In existing literature there are different kind of hazard rate models that are used. Most used model in survival analysis is the Cox semi-‐ parametric proportional hazard model. This model is a proportional hazard model, which means that it is assumed that for each amount of time passed the effect of a covariate on the hazard rate is the same. In this analysis a non-‐proportional hazard model will be used since it is likely that the possible ‘treatment’ effect of the Hartz reform dummy variable has a different effect for different durations of unemployment. The difference in hazard rate before and after the reforms might be bigger if the duration of unemployment is longer since then it might be the difference between receiving unemployment benefits and receiving no unemployment benefits. Testing the proportionality assumption using Schoenfeld residuals confirms this. Instead of a proportional hazard model an Accelerated Failure Time (AFT) model will be used. To see which distribution fits the data best a generalized gamma model was estimated. The models nested within this model were compared using the likelihood ratio test. It seems the lognormal distribution fits the data better than the exponential or loglogistic distribution. With the lognormal distribution it is assumed that the hazard function first increases to a maximum and then decreases over time to approach 0.
In order to separate the general labor market trend from the effects of the reduction in potential unemployment benefits, the difference-‐in-‐differences approach will be used. As mentioned earlier, in table 1 it can be seen that for workers aged under 45 there is no change in potential
unemployment benefit duration. For six other age groups, 45-‐46, 47-‐51, 52-‐54, 55-‐56, 57 and >57, there are varying changes in potential unemployment benefit duration. The <45 age group will act as the control group and the six other age groups will act as treatment groups. The presence of this many treatment groups creates the opportunity to see if different changes in benefit entitlement also lead to different changes in unemployment duration, since the change in entitlement is different for the different age groups. In table 3 the average unemployment duration is shown before and after the Hartz reforms along with the change in entitlement.
Table 3 d Avg unemp. Dur. Age Before After Diff. Entitl. change <45 8.18 6.05 2.13 0 45-‐46 14.27 8.37 5.9 -‐6 47-‐51 12.2 6.92 5.28 -‐10 52-‐54 17.47 9.03 8.44 -‐14 55-‐56 18.31 11 7.31 -‐11 57 24.62 12.34 12.28 -‐17 >57 18.37 12.38 5.99 -‐14
Hazard rate models will be estimated for each age group containing the dummy variable Hartz that indicates if the spell is after the reforms, a variable with respect to whether the worker is married or not, an age control variable and a gender dummy. Variables with respect to children and education have also been considered. However, the datasets containing information about these characteristics only contained this information for about half the sample that is used in this research. Since the sample already was pretty small for some age groups, for example only 27 females aged 57, these variables were left out. Also not included in the model is a variable for work experience. This is because the hazard rate models will be estimated only for workers that have a certain work
experience. The reason for this is that only for these workers the change in benefit entitlement is as large as shown in table 3, for workers with work experience below this level the change is different. In order to get results that are easily interpreted considering the change in benefit entitlement, workers of a certain age that do not satisfy the work experience required for the maximum length of unemployment benefits are left out. Because the workers contained in the models already have to satisfy a certain work experience requirement it was deemed unnecessary to include a variable with respect to work experience in the model. The specification of the model is:
log( 𝑇) = 𝛼!+ 𝛼!∗ 𝐻𝑎𝑟𝑡𝑧 + 𝛼!∗ 𝑀𝑎𝑟𝑟𝑖𝑒𝑑 + 𝛼!∗ 𝐺𝑒𝑛𝑑𝑒𝑟 + 𝛼!∗ 𝐴𝑔𝑒!− 𝐴𝑔𝑒! +
1 𝑝𝜖 Where T is the survival time, Hartz is the dummy coefficient with respect to the reforms and Married is the marriage dummy. 𝐴𝑔𝑒!− 𝐴𝑔𝑒! controls for age within the age groups. 𝐴𝑔𝑒! is the age of the
unemployed worker and 𝐴𝑔𝑒! is the age at which the age group starts. For unemployed workers
parameter p is the shape parameter. For each age group a separate model will be estimated. If the age and the work experience is known, and the worker has received unemployment benefits, then the maximum duration of eligibility for unemployment benefits is known for each age group. The largest change is for workers aged 57, for whom the maximum unemployment benefit duration is reduced from 32 months before the reforms to 15 months after the reforms. In table 1 it can be seen that on the 1st of March the reduction were eased a little bit, mainly for the older workers. The changes in 2008 will also be tested for the workers aged over 57, but these results may be less important because the changes in 2008 are significantly smaller than the changes in 2006, only for workers aged 58 or older the change is larger than 3 months.
An indication of the link between unemployment benefit duration and unemployment duration is a possible peak in the hazard from unemployment just before unemployment benefits are exhausted, which has been found quite often in existing literature. A significantly higher hazard rate in the month that benefits are exhausted could mean that job search effort is higher just before benefit exhaustion or the reservation wage is lowered. In this paper the month of benefit exhaustion will also be considered to see if the hazard rate is higher in this month.
Results
In this section the results of the econometrical analysis will be discussed. A problem with the data could be that workers could choose or agree with their boss that they would enter unemployment just before the reforms, since the policy change was known to be implemented in February 2006 and for some people it could have a major influence. The number of unemployment spells in the sample after the reforms are less than the number of unemployment spells before the reforms, so for each month the number of unemployment spells declines after the reforms. To see if there were
disproportionately large numbers of unemployment spells that started in the months prior to the reform a comparison was made with the same months in other years. For the months November, December, February and March this is not the case. For the months November and December the changes between 2004 and 2005 were 2 spells more and 2 spells less respectively, so these changes are not significant. For February and March the differences between 2006 and 2007 are 1 spell less and 4 spells less respectively. For February and March these years are used because it could be that right after the reforms the number of spells started are lower and since the number of spells before and after the reforms already differ it has to be compared with a year later. For January the number of unemployment spells started in the year of the reforms is significantly higher than in the other years and this month is thus dropped from the sample.
In table 1 it can be seen that on the 1st of March the reduction were eased a little bit, mainly for the older workers. Now we will test the difference in means to see if this small change in potential benefit duration had an influence on the unemployment duration. When testing for the age group >57 the results indicated that average unemployment duration was lower after the increase in unemployment benefit duration, which is contradictory to findings in practically all existing literature. The result is significant at a 1% level. What is more likely is that the effects of the first policy change caused the unemployment duration to decrease for several years after the first policy change, and thus disabling any possible effects the subsequent small increase might have. It is not unlikely that it took more than two years to implement the policy change and for the labor market to adjust to these changes. Because of this and the fact that the change is relatively small compared to the changes two years before, the easing of the reforms in 2008 will not be used in the rest of the analysis.
Table 3 already showed that the average duration of unemployment was lower after the reforms than before the reforms. Although there is a significant difference in unemployment duration for the control group consisting of workers aged under 45, the difference is much larger for the other age groups. The group with the largest change in maximum benefit entitlement, workers aged 57, is also the group with the largest change in unemployment duration. Also for the age group 52-‐54 the relatively large change in unemployment duration is in line with the large change in potential benefit duration. From this table it can be calculated that on average a 1 month reduction in potential unemployment benefit reduces unemployment duration by 0,45 month. The change of the control group is subtracted from the other changes to control for changes in labor market conditions. As mentioned earlier, the econometrical analysis will first be done for the whole sample after which the differences between men and women will be discussed. In table 4 the empirical hazard rates and the standard errors of the hazard rates both before and after the reforms are shown. The monthly hazard rate of the first twelve months are shown, with intervals of three months.
Table 4 Hazard rates
Before Reforms
≤3 months Std Err 3-‐6 months Std Err 6-‐9 months Std Err 9-‐12 months Std Err <45 0.1516 0.0042 0.1407 0.0051 0.1229 0.0059 0.1354 0.0074 45-‐46 0.096 0.0156 0.0855 0.017 0.0645 0.0166 0.0984 0.0229 47-‐51 0.135 0.0136 0.0958 0.0138 0.0711 0.0136 0.0981 0.018 52-‐54 0.0667 0.013 0.0589 0.0135 0.0633 0.0153 0.0613 0.0168 55-‐56 0.0617 0.0137 0.0474 0.0131 0.0417 0.0132 0.0686 0.0182 57 0.0408 0.0166 0.0549 0.0207 0.0175 0.0124 0.0185 0.0131 >57 0.0329 0.0066 0.0427 0.0079 0.04 0.0082 0.0557 0.0103
After Reforms
≤3 months Std Err 3-‐6 months Std Err 6-‐9 months Std Err 9-‐12 months Std Err
<45 0.182 0.0058 0.1609 0.0072 0.129 0.0082 0.1769 0.0118 45-‐46 0.1122 0.0268 0.0901 0.0282 0.1538 0.0432 0.1414 0.0522 47-‐51 0.1631 0.0209 0.1315 0.0239 0.0897 0.0238 0.1667 0.038 52-‐54 0.1333 0.0239 0.0637 0.0202 0.102 0.0279 0.1754 0.0437 55-‐56 0.0533 0.0188 0.0905 0.027 0.0984 0.0324 0.1212 0.0421 57 0.0364 0.021 0.1037 0.0387 0.0571 0.0329 0.1235 0.0543 >57 0.0436 0.0121 0.0585 0.015 0.087 0.0203 0.11 0.0263
The empirical hazard rates shown in table 4 can be interpreted as the monthly exit rate in a given time interval. It is worth noting that a large part of the empirical hazard rates have relatively high standard errors and should be interpreted with caution. For the age group 57 the standard error is often more than 50% of the hazard rate. When looking at this table it can be seen that for the control group the hazard rate also increases for each number of months, which is in line with the decrease in unemployment duration of the control group discussed earlier. This confirms that there is an upward trend in the labor market, probably connected to the other parts of the Hartz reforms. This means the results should be interpreted with a little caution and that the results should be compared to the results for the control group in order to see if the changes in hazard rates for the treated groups are significantly larger than the changes for the control group.
Based on the changes in potential benefit entitlement shown in table 1, one might expect that for workers of the age groups 52-‐54, 57 and >57 the change will be largest. The difference in hazard rate is consistently large for the age group 52-‐54, but for the other age groups the difference grows larger as the time period grows larger. This seems logical because even if the shorter unemployment benefits would have influence it is difficult for older workers to find a job on short notice. However, since the standard error is large for most of the hazard rates there is not much that can be concluded from this table. The only thing that can be concluded is that for workers aged under 45 the hazard rate after the reforms is on average 1.175 times the hazard rate before the reforms.
Now the model specified in the previous part will be estimated. Below the time ratios for each of the predictors is shown, along with the accompanying time ratios. A time ratio of 0.656 means that the survival time, in this case time unemployed, of unemployed workers after the reforms is 0.656 times the survival time of unemployed workers before the reforms. A lower time ratio thus indicates a larger difference between the unemployment duration before the reforms and after the reforms. With an average unemployment duration of 17.47 this means unemployed workers aged 52-‐54 are 17.47-‐(17.47*0.656)= 6.01 months less unemployed after the reforms. The reported z-‐values indicate
that none of the time ratios of the Hartz dummy is statistically significant at the 5% level. The only time ratio for the Hartz dummy that is significant at the 10% level is the time ratio for the age group 52-‐54. Table 5
Hartz tr z-‐value Fam tr z-‐value Age1-‐
Age0 tr z-‐value Gender tr z-‐value
<45 0.951 -‐1.00 0.906 -‐1.56 1.02 5.87 0.87 -‐2.92 45-‐46 0.756 -‐1.01 0.61 -‐2.01 1.64 2.01 1.07 0.28 47-‐51 1.24 0.93 0.79 -‐1.19 1.02 0.25 1.01 0.06 52-‐54 0.656 -‐1.82 0.82 -‐0.91 1.10 0.76 0.98 -‐0.09 55-‐56 0.886 -‐0.46 0.908 -‐0.4 1.37 1.33 0.63 -‐1.91 57 0.714 -‐0.90 0.957 -‐0.13 1 -‐ 2.24 2.28 >57 0.864 -‐1.12 1.3 2.24 0.90 -‐3.40 0.88 -‐0.99
The relative time ratios of the age groups would have been in line with the expectations based on the change in potential benefit duration, but the results are insignificant. For other variables the time ratios are sometimes significant at the 5% or 1% level. For example being married and living together shortens the unemployment duration for workers aged 45-‐46 by 5.57 months. For workers aged over 57, however, the unemployment duration is prolonged by 5.51 months if you are married and living together. One surprising result is the time ratio of the age control variable for workers aged over 57. The time ratio is smaller than 1 which indicates that within the age group age has a negative effect on unemployment duration, so older people are unemployed for a shorter period of time on average. For the age groups under 45 and 45-‐46 the time ratio for the age control is also significant. For under 45 the effect is very small but for unemployed workers aged 45-‐46 the effect is large. It seems unemployed workers aged 46 are unemployed for a longer period on average than people aged 45. The gender dummy only has a significant time ratio for unemployed workers aged 57, and this ratio is extremely large. Unemployed men aged 57 are on average unemployed for 30.53 months longer than unemployed women aged 57.
In order to formally test treatment against control the equality of the Hartz dummies for the different age groups is tested. First the Hartz dummy of control group is tested against each of the treatment groups and then the Hartz dummy of the control group is tested against all of the treatment groups combined. Below are the results of the test are shown.
Table 6
Age group against
control 45-‐46 47-‐51 52-‐54 55-‐56 57 >57 All
X² 0.75 1.28 2.65 0.07 0.71 0.43 0.73
prob > X² 0.386 0.258 0.104 0.788 0.399 0.514 0.394
None of the time ratios of the Hartz dummy of the treatment groups was significantly different from the time ratio of the Hartz dummy of the control group. This means that even though there were differences in the time ratios the differences were too small or had too large standard errors to make them statistically significant. In this case it is most likely that the standard errors of the time ratios were a little too large to get significant differences with this test. In table 5 it was already shown that none of the age groups had a large z-‐value for the Hartz dummy and so no time ratios were
significant at the 5% level.
In Table 1 it can be seen that for every worker aged 45-‐54 and a work experience of at least 24 the maximum eligibility period changed to 12 months with the implementation of the Hartz reforms. According to the theory, the reservation wage would be lowered as benefits are almost exhausted. When looking at the hazard rate in the last month of receiving benefits for these workers, both men and women, there is a large peak that was not present before the reforms. However, a peak in hazard rates in the 12th month is also found for other age groups. There is no noticeable peak in hazard rates after the reforms in the 15th month for workers aged 55-‐57 and not in the 18th month for workers aged over 57. The hazard rate after the reforms in the 15th month for workers aged 55-‐57 is 0.0426, compared to 0.0800 in the 14th month and 0.1905 in the 16th month. This means that the peak in the 12th month can not be linked to the exhaustion of benefits. It can be seen below that there is a peak in the hazard rate in the 12th month for all ages.
Table 7 Hazard rate 45-‐54 Hazard rate 55-‐65 Month Before
reforms After reforms Month
Before reforms After reforms 5 0.0988 0.0879 5 0.0502 0.0831 6 0.0608 0.1027 6 0.05 0.069 7 0.0647 0.1145 7 0.0289 0.0588 8 0.0624 0.0974 8 0.0389 0.0706 9 0.0702 0.1525 9 0.0436 0.1207 10 0.0677 0.0962 10 0.0291 0.0664 11 0.0937 0.1064 11 0.0436 0.04 12 0.094 0.2968 12 0.082 0.2273 13 0.068 0.0465 13 0.0188 0.137 14 0.0515 0.1 14 0.0347 0.0606
Conclusion
In this paper the relation between the potential duration of unemployment benefits and the duration of unemployment has been researched. An Accelerated Failure Time hazard model was estimated and the empirical hazard rates were analyzed in order to research this relation, with one control age group and 6 treatment age groups. The coefficients of the treatment groups were tested against the coefficient of the control group. The results of these tests indicate that there is no significant
difference between the impact of the reforms on the control group and the impact on the treatment groups.
The empirical hazard rates also do not indicate a significant difference between the control group and the treated groups. There is a difference in change in hazard rates between the control group and the treated groups but the standard errors of the hazard rates are too large to draw a conclusion from it.
When looking at possible spikes in hazard rates close to benefit exhaustion no significant spikes were found, so this does not support the findings in some previous studies. There were spikes in the hazard rates in the 12th month, but these spikes were present in all age groups and thus can not be linked to the exhaustion of unemployment benefits.
These results are not in line with the results in most of the existing literature. The most probable reason for the insignificance of the differences between the control group and the treatment groups is the large standard error that accompanied most coefficients and hazard rates.
Literature
Caliendo, M. & Hogenacker, J. (2012). “The German labor market after the Great Recession: successful reforms and future challenges” IZA Journal of European Labor Studies
Caliendo, M., Tatsiramos, K. & Uhlendorff, A. (2013). Benefit duration, unemployment duration and job match quality: A regression-‐discontinuity approach. Journal of Applied Econometrics 28 (4)604-‐ 627
Card, D. & Levine, P.B. (2000) . “Extended benefits and the duration of UI spells: evidence from the New Jersey extended benefit program.” Journal of Public Economics, 78, 107-‐138
Eichhorst, W., Grienberger-‐Zingerle, M. and Konle-‐Seidl, R. (2007). “Activation policies in Germany: from status protection to basic income support.” IAB discussion paper no. 6
Eurostat, Unemployment rate by sex, available at
http://appsso.eurostat.ec.europa.eu/nui/submitViewTableAction.do;jsessionid=9ea7d07d30dbf06b1 43ff9a447539b24ad9fb2b62d24.e34MbxeSaxaSc40LbNiMbxeNbhyKe0, visited on 6th of April 2014 Katz, L.F. & Meyer, B.D. (1990). “The impact of the potential duration of unemployment benefits on unemployment duration” Journal of Public Economics 41, 45-‐72
Lalive, R. (2007). “How do extended benefits affect unemployment duration? A regression discontinuity approach.” Journal of Econometrics, 142, 785-‐806
Lalive, R., Van Ours, J.C. and Zweimüller, J. (2006) “How changes in financial incentives affect the duration of unemployment.” The Review of Economic Studies, 73, 1009-‐1038
Meyer, B.D. (1990). “Unemployment and unemployment spells” Econometrica, 58 (4), 757-‐782 Mortensen, D.T. (1977). “Unemployment insurance and job search decisions” Industrial and Labor Relations Review, 30(4), 505-‐517
Rijksoverheid, Kabinetsplannen aanpassing WW, available at
http://www.rijksoverheid.nl/onderwerpen/werkloosheid-‐en-‐ww/kabinetsplannen-‐aanpassing-‐ww, visited on 6th of April 2014
Van Ours, J.C. and Vodopivec, M. (2006). “How shortening the potential duration of unemployment benefits affects the duration of unemployment: Evidence from a natural experiment.” Journal of labor Economics, 24(2), 351-‐378
Worldbank, Unemployment total(% of the labor force), available at
http://data.worldbank.org/indicator/SL.UEM.TOTL.ZS. Visited at 17th of April 2014