Directory UMM :Data Elmu:jurnal:L:Labour Economics:Vol7.Issue2.Mar2000:

28  Download (0)

Teks penuh

(1)

www.elsevier.nlrlocatereconbase

Spell durations with long unemployment

insurance periods

Espen Bratberg

)

, Kjell Vaage

1

Department of Economics, UniÕersity of Bergen, Fosswinckelsg. 6, N-5007 Bergen, Norway

Received 18 May 1998; accepted 22 June 1999

Abstract

This paper uses data from a natural experiment to investigate the potential incentive effect of a fixed unemployment insurance period. We compare two large groups of Norwegian unemployed persons who registered as unemployed in 1990 and 1991. The last group was affected by a rule change that in practice extended the length of unemployment benefits to more than 3 years. Our data are taken from official records, and we construct unemployment durations by combining information from the unemployment registers with employers’ records. We use a proportional hazard model with a flexible baseline. The results suggest that the main effect of benefits running out is to make people drop out of the unemployment register. We find neither clear evidence that the hazard into employment increased when the end of benefits approached in the pre-liberalisation group, nor that behaviour in this part of the spells changed after the reform. On the other hand, our results suggest that the reform had an all over negative effect on the employment hazard.q2000 Elsevier Science B.V. All rights reserved.

JEL classification: J64; C41

Keywords: Unemployment duration; Semiparametric hazard model; Natural experiment

)Corresponding author. E-mail: espen.bratberg@econ.uib.no 1

E-mail: kjell.vaage@econ.uib.no.

0927-5371r00r$ - see front matterq2000 Elsevier Science B.V. All rights reserved.

Ž .

(2)

( ) E. Bratberg, K. VaagerLabour Economics 7 2000 153–180

154

1. Introduction2

Standard search theory predicts that the duration of an unemployment spell is increasing in the level of unemployment benefits, because the cost of rejecting a job offer decreases. Furthermore, if benefits are cut after a fixed period, the reservation wage decreases, and the exit rate out of unemployment increases as the

Ž Ž ..

time of running out approaches Mortensen 1977 . Of these two hypotheses, the first one is fairly well established empirically. A number of contributions deal with the connection between benefit size and unemployment duration,3 even though

there are different results as to the magnitude of the effect. The second prediction is less researched, and our purpose with the present paper is to add to the knowledge of the potential incentive effects of a fixed unemployment insurance period. Doing so, we use two large samples of extensive Norwegian register data covering a period including a natural experiment — in 1992 a former rule of 13 weeks benefits withdrawal after 80 weeks was abolished. Thus, we can compare a group of unemployed who were affected by this policy change with another consisting of individuals who were not.

Compared to most other European countries, the Norwegian unemployment rate has been remarkably low through most of the 1980s, mostly staying between 1.5 and 3% until 1988. Then, in 1989, unemployment increased sharply and stayed at approximately 5–6% of the labour force during the early 1990s. Unemployment insurance is universal for all employees with earnings above a minimum level. Until 1991, the maximum entitlement period was 80 weeks, followed by 26 weeks

Ž .

without benefits. One could then receive unemployment benefits UB for a second 80-week period. The alternative to UB for those who did not manage to get a job, would be means tested social benefits. In May 1991, as a response to the increasing number of long-term unemployed, the length of the period without UB was reduced to 13 weeks. One year later, it was then decided that if the unemployment agencies had not offered an individual a new job or a labour market programme after 80 weeks, benefits should no longer be withdrawn in the 13 weeks period. Thus, from May 1992, it became possible to receive UB for a continuous period of 186 weeks.4

As we have already suggested, the empirical literature concerning the effect of

Ž .

fixed benefits periods is relatively scarce. Meyer 1990 and Katz and Meyer

Ž1990 find that spikes in the hazard out of unemployment may be explained by.

2 Ž .

The data used in this paper are provided by The Norwegian Social Science Data Services NSD . NSD is not responsible for the authors’ analyses.

3 Ž . Ž .

Surveys can be found in, e.g., Atkinson 1987 and Layard et al. 1991 . 4

(3)

5 Ž .

the end of benefits approaching. Fallick 1991 and Narendranathan and Stewart

Ž1993a , however, find that the effect of benefits decreases over time. These.

results are harder to reconcile with the prediction that the tendency to leave unemployment increases at the end of the benefits period. Micklewright and Nagy

Ž1996 , using Hungarian post-transition data, find no rise in the hazard near the.

Ž .

time of benefit exhaustion. Winter-Ebmer 1998 , using Austrian data, finds that males react to extended benefits duration but females do not.

Apart from the academic issue of whether the theory yields correct hypotheses, there are of course important policy implications. For instance, Layard et al.

Ž1991 advocate using labour market policies similar to those of Sweden to keep.

long term unemployment down: a fixed UB period combined with active man-power policy. However, it is not clear that a policy that consists of labour market training programmes and relief jobs does not distort the potential incentives from a

Ž .

fixed benefit period. Carling et al. 1996 , using Swedish data, investigate that question and conclude that such distortions probably do not take place. Their evidence of the effects of benefit exhaustion on the hazard into employment is,

Ž .

however, only marginally statistically significant. Korpi 1995 , in a study of youth unemployment, does not find benefits to have any significant effect on the rate of transition.

The Norwegian labour market policy includes a variety of training programmes and relief jobs, and thus resembles that of Sweden. Until 1992, these countries also were similar in having a fixed benefits period, even though it was longer in Norway. Researching Norwegian data can therefore have bearing both on the incentive and the labour market programme issues. In the present paper, we focus on the former. Recently extensive Norwegian register data have become available. Using data from these registers, we analyse two random samples of the inflow to the unemployed population, before and after the 1992 changes in unemployment

Ž .

insurance rules. Hernæs and Strøm 1996 have used data from the same registers, but focus on duration dependence in general. To our knowledge, the present analysis is the first that addresses the fixed UB period question with Norwegian data. Our data sets are large and utilise information from the unemployment registers, the employers’ register, and other sources. Using a semiparametric proportional hazard model, we estimate exit rates into employment. Search theory predicts, ceteris paribus, individuals with UB in the first sample to have a jump in the transition rate into employment when the first 80 weeks entitlement period has expired. Since persons in the second sample would not be faced with a benefit withdrawal after 80 weeks, a corresponding jump therefore should not be observed there. In addition, the prospect of benefit cuts should lead to increased search efforts in the entire unemployment period of the first sample compared to the second one.

5 Ž .

(4)

( ) E. Bratberg, K. VaagerLabour Economics 7 2000 153–180

156

In both of our samples we expect to find different exit patterns between persons with and without entitlement to benefits. Furthermore, the 1992 rule changes are expected to have no effect on unemployed individuals not receiving UB. The non-receivers therefore function as a control group.

We now proceed by presenting our model specification in Section 2. In Section 3 we describe our data sets and discuss the definition of unemployment spells. Our results are presented and discussed in Section 4. It turns out that the analysis does not give empirical support to the notion that the 1992 reform produced significant changes in the behaviour around 80 weeks of unemployment duration. However, the reform appears to have had effects on the earlier stages of the search process, in that the first year’s exit rates for the second sample UB-receivers are lower than in the first sample. Section 5 contains a critical assessment of the findings.

2. Model specification

The hazard rate out of unemployment and into employment is defined as the limit of the conditional probability of a transition taking place in a small interval d t after time t if no transition occurred until t, when that interval approaches zero. Formally,

where it is understood that t is a realisation of a random variable, T, that measures unemployment duration. The conditional probability in the numerator is also the product of the chance of receiving a job offer and the probability of accepting, a fact that might be used as a starting point for a structural model. We shall, however, use the more common procedure of specifying the hazard directly. More specifically, we use a semiparametric approach to duration analysis that was

Ž .

proposed by Prentice and Gloeckler 1978 . The virtues of this method are that it is unnecessary to make parametric assumptions concerning the hazard’s time dependence, and account can be taken of the potential discrete nature of the data. Recently, this semiparametric approach has found several applications in the study of unemployment duration.6

Assume that conditional on staying in the pool of unemployed persons until

Ž .

time t, individuals leave unemployment with rate r t,x , specified as

r t ,x

Ž

.

su0

Ž .

t exp x

Ž

Xb

.

.

Ž

2.2

.

Ž . Ž .

Eq. 2.2 is known as the Cox proportional hazards specification. Here,u0 t is the

baseline hazard, which is some arbitrary nonnegative function of time, x is a

6

Ž . Ž . Ž .

(5)

vector of covariates, and b is a coefficient vector.7 Now define the discrete or grouped hazard, lt, as the probability of a transition taking place in the interval

w .

ats ty1,t , conditional on survival until ty1. Using the relation between the hazard rate and the survivor function, it follows that

t X t

Time-varying covariates may easily be introduced if they are assumed to change only at the endpoints of time intervals: simply subscript x with t.

The probability of surviving through any interval at after having survived the

Ž .

preceding interval is 1ylt . Therefore the likelihood contribution of someone who leaves unemployment in thetith interval is

tiy1

lti

Ł

Ž

1ylt

.

.

ts1

We assume that censoring takes place in the beginning of intervals.8 Then, defining dis1 if individual i’s spell ends in a transition, 0 otherwise, i’s likelihood contribution is

Collecting terms, using Eq. 2.4 , taking logs, and summing over i, the log-likeli-hood of a sample of N observations is then

ty1

where we have subscripted x to indicate that some variables may be time-varying. The hazard model as outlined above has no error term, and is correct only if all differences in the individual durations are due to differences in the vector of

7

Ž .

The exponential and Weibull models are also proportional hazard models withu0 t appropriately

parameterised. 8

(6)

( ) E. Bratberg, K. VaagerLabour Economics 7 2000 153–180

158

observed variables, x . However, there may also be unobserved sources oft

heterogeneity. The standard way of trying to deal with this problem is to assume that an unobserved random variable ´, which is time constant and independent of

Ž .

the observed covariates, enters the hazard multiplicatively. Eq. 2.2 is thus changed to

r t ,x

Ž

.

su0

Ž .

t exp x

Ž

Xb ´

.

,

Ž

2.6

.

With an additional assumption regarding the distribution of this unobserved variable,9such a model can be estimated. Usually a gamma distribution is chosen.

Ž .

Meyer 1990 implemented this approach in the semiparametric model. It may be shown10 that when ´ is gamma distributed with unit mean and variance s2, the

log-likelihood function becomes

Results from both specifications will be reported in this paper.11

One final modification of the model is made: we ask whether the hazard increases as benefit exhaustion approaches. To include this aspect in the analysis,

UB Ž

we define Is1 if benefits are received, 0 otherwise, and replacegt bygt q 1

. nonUB

yI gt . The interval length is four weeks, corresponding to the accuracy with which the spells are measured in the unemployment register.

3. Data

Unemployment insurance is compulsory for all employees in Norway. The premium is included in the contribution to the social insurance system. Employees with earnings above a certain minimum level during the last three years who lose their jobs involuntarily, are entitled to unemployment benefits amounting to 62% of the previous year’s wage income. To obtain benefits, one has to register at the local employment agency and be available for new placement. The agencies offer guidance of various sorts, and access to labour market programmes. These services

9 Ž

Alternatively, the distribution can be approximated nonparametrically Heckman and Singer

Ž1984 ...

10 Ž . Ž .

See Meyer 1990 or Dolton and van der Klaauw 1995 . 11

The choice of a gamma distribution for the error term is made for computational reasons, and may

Ž .

(7)

are available also to those who are not entitled to benefits, e.g., first entrants to the labour market. Therefore, even individuals without benefit entitlement have some incentives to register. Beneficiaries who stop reporting lose their benefits.

Our two data sets are drawn from the KIRUT 12 database, which collects extensive information from various administrative registers for a 10% random sample of the Norwegian working aged population. The data, linked with personal identifiers, are organised in an event-oriented fashion, and presently covers the period from 1989 until 1994, inclusive. Data providers are the Directorate of Labour, the National Insurance Administration, and Statistics Norway. Using register data has obvious advantages. At relatively low costs, the researcher gets access to large amounts of data on the individual level. The problems with sample dropouts so often encountered in surveys, are to a large extent avoided. Also, the researcher is able to construct case histories based on information collected for bureaucratic reasons, rather than relying on individual retrospection. Some exactly recorded information, e.g. related to earnings histories, may be less precisely recalled by the individuals themselves. Admittedly, using register data also has its problems, which we shall go into in the next paragraphs. To a large extent they are related to the fact that the records are generated for other purposes than research. Our sampling strategy is to use two samples of individuals who face different rules with respect to UB duration, but similar labour market conditions. The first sample was constructed by picking everybody in the database who started report-ing as unemployed at the public employment agencies from June 1 through December 1, 1990, and the second sample consists of people that started reporting in the same period the next year. The samples include several dates pertaining to change of labour force status, where the two most important are: the day a person left the unemployment register, and the day he was recorded in the employers’ register.13 The data from the unemployment register cover the length of the

Ž .

current spell, and succeeding spells up to December 31, 1992 1993 for data set 1

Ž .2 . From the employers’ register, we have records of all spells that started after

Ž .

the unemployment spell and before December 31, 1992 1993 . We remind the reader that the UB withdrawal period after 80 weeks was abolished in May 1992. Consequently, the reform affected the search incentives around benefits expiration for the second group, but not the first. As for the beginning of the unemployment period, the incentive effects for the second group are less clear. Until May 1992, people from both groups were faced with identical UB rules. However, to the extent that the unemployed in the second group knew the content of the reform before it was passed by the Parliament, we must assume that their expectations to

12

This Norwegian acronym roughly translates to ‘‘Clients into and through the Social Insurance system’’.

13

(8)

( ) E. Bratberg, K. VaagerLabour Economics 7 2000 153–180

160

the coming reform might have had disincentive effects on their search behaviour right from the beginning of their unemployment period. A sample based on people reporting as unemployed after the reform of May 1992 and followed through 1994 would cast light on this assumption, but was not available at the time when the research was conducted.

Sampling from the unemployment registers may underestimate the number of persons who would like to have a job if they could get one, i.e. who are in the labour force but out of work. Presumably, that would mainly be individuals without benefit entitlement. If one chooses to use register data, this problem remains anyway. Furthermore, given the available registers, the duration of an unemployment spell could be defined in several ways. First, the duration could simply be defined as the time spent in the unemployment register. Then we would not know if the spell ended with a transition to employment or out of the labour force. The individual could also stop registering, e.g., because he was not entitled to benefits, but still be a job seeker. Alternatively, the length of a spell could be defined as the time from the unemployment register record started until the individual was recorded in the employer’s register. A new problem introduced by

Ž

this approach is that individuals who spent time out of the labour force e.g.,

.

education, young males in the military services, females having children etc. may be perceived as having a continuous spell of unemployment. It seems fairly obvious that the first approach may underestimate the true unemployment dura-tion, and the second may overestimate it. The problem may be alleviated by combining the registers, only counting transitions directly from the unemployment register into the employers’ register as transitions, and treating unemployment records ending without an according employment record as censored. This defini-tion may still be a downward biased measure of unemployment duradefini-tion. We have chosen to use the last definition, but with a somewhat less strict censoring criterion: individuals are only censored when spending more than two months out of the unemployment register without having an employment record.14 Some

Ž .

inaccuracies in the employers’ register which may be likely for small enterprises and some potential spurious unemployment registration behaviour is thus allowed

14

Ž .

(9)

for. This reduces the problem of potentially biasing durations downwards, but obviously without resolving it. It seems that problems of this kind, inherent in the use of register data, remains whatever definition of unemployment is used, and a reader must bear in mind the consequences of the different definitions when comparing results from different studies.

The analysis is also complicated by a tendency for individuals to disappear from the unemployment register for a couple of months, and then to turn up again. These features may reflect that individuals actually leave the labour force, or may be due to errors in the registers. It is not obvious that everyone who leaves the unemployment register for a couple of months really has been out of the labour

Ž

force meanwhile. This is not to say that such behaviour can be completely ruled

.

out. To cope with these possible inconsistencies, we assume that periods two months or less apart belong to the same spell.15A transition into employment is

then defined as having a record in the employers’ register that begins before

Ž . Ž .

December 31, 1992 1993 for those who registered on December 1, 1990 1991 . Individuals who registered earlier in the sampling periods are observed for a period of the same length. Persons with a gap of more than two months between the unemployment and the employment record are, however, treated as censored.16 In behavioural terms, we regard them as having left the labour force.

Obviously, combining the two registers as described affects the number of transitions and also the length of spells compared to using the ‘‘raw’’ registers. Table 1 reports spell characteristics for first transitions out of the unemployment register,17 first transitions into the employers’ register, and the combination used

in this paper. Within the observation period, almost all the observed individuals

Ž .

left the unemployment register first panel , and about two-thirds showed up in the

Ž . Ž .

employers’ register second panel . When combining third panel , the number of transitions is reduced compared to when looking only at records into the employ-ers’ register, but the mean duration of completed spells is shorter.18 It is

important to note that within the framework of a single spell model, the rather few transitions do not imply that the others did not get a job within the observation period. Those who are censored for the reasons discussed above, may nevertheless have found employment later, but that is outside the scope of the present analysis.

15

Results from running the model on spells with one month-gaps — not reported here but available on request — increased the number of censorings but otherwise did not affect the results significantly.

16

Some jobs may be relief jobs that are part of labour market programmes. Controlling for this is tricky because there are no clear administrative rules as to whether such jobs should be recorded in the employers’ register. We have tried to identify ‘‘ordinary’’ jobs by searching for labour market programme records in the unemployment register with a starting date that fall within"two months of the starting date in the employers’ register. In such cases, we use the next employment record, if any.

17

Without any checks for succeeding unemployment records. 18

(10)

()

E.

Bratberg,

K.

Vaage

r

Labour

Economics

7

2000

153

180

162

Table 1

Ž .

Spell characteristics days . Standard deviations in parentheses

Ž . Ž .

1990 Ns9936 1991 Ns12054

All With UB Without UB All With UB Without UB

( )

Out of unemployment register first record

a Ž . Ž . Ž . Ž . Ž . Ž .

Mean duration, completed spells 160 156 202 170 95 103 183 173 225 185 101 109

acensored spells 330, 3.3% 254, 4.2% 76, 2.0% 518, 4.3% 444, 5.5% 74, 1.9%

( )

Into employers’ register first record

a Ž . Ž . Ž . Ž . Ž . Ž .

Mean duration, completed spells 258 210 270 205 236 217 267 206 275 204 248 210

acensored spells 3572, 36.0% 1996, 32.9% 1576, 40.7% 4459, 37.0% 2756, 34.1% 1703, 42.8%

Into employment — combined registers

Ž . Ž . Ž . Ž . Ž . Ž .

Mean duration, completed spells 179 167 207 170 116 145 216 190 254 197 142 157

Ž . Ž . Ž . Ž . Ž . Ž .

Mean duration, censored spells 322 246 391 249 230 208 346 249 403 247 228 208

acensored spells 6313, 63.5% 3598, 59.3% 2715, 70.2% 7943, 65.9% 5340, 66.1% 2603, 65.4%

a

(11)

Table 2

Sample characteristics

Standard deviations in parentheses.

1990 1991

a Ž . Ž .

Income prev. year 76 726 69 743 80 502 70 832

aŽ . Ž . Ž .

Spouse income if married 133 302 100 695 134 672 106 391

a Ž .

UBrweek – – 1462 600

Ž . Ž .

Years of education 10.7 1.8 10.9 1.9

Ž . Ž .

Age 29.8 10.8 30.5 10.9

Ž . Ž .

Years of experience 7.8 7.2 8.5 7.6

Ž . Ž .

achildren-11 0.30 0.67 0.31 0.68

Ž . Ž .

achildren)11,-18 0.16 0.47 0.16 0.47

b Ž . Ž .

% local unemployment 6.11 2.02 6.71 1.94

% female 43.8 40.6

% married 28.4 28.5

% divorcedrwidow 9.9 10.2

% non-Scandinavian 1.95 1.65

After excluding 1152r11 088 and 1550r13 604 observations with missing background information, the 1990 and 1991 samples consist of 9936 and 12054 persons, respectively. A full description of the elements of the covariate vector xt is found in Appendix A Table 2 provides descriptive statistics for some important variables, revealing few differences in the sample means. The average person in the 1991 sample is slightly older, with somewhat longer labour force experience and higher previous earnings, possibly reflecting the increasing national unem-ployment rate. There are also more females in the 1990 sample. However, the relatively low average age and previous income are conspicuous in both samples. The low percentage of married persons is complemented by the fact that spouse income for those married averages higher than the sample incomes. Considering the well-known fact that female labour market behaviour differs from that of males, in the analysis we have added interaction effects between gender and some variables relevant to family background.

To catch the effect from unemployment benefits, we allow the baseline hazard to be different for UB-receivers and non-receivers,19 and between the samples.

19

(12)

( ) E. Bratberg, K. VaagerLabour Economics 7 2000 153–180

164

Ž

The hypothesis is that being a UB-receiver and belonging to the 1991 post rule

.

change sample adds negatively to the exit hazard.

As an indicator of labour demand, we use the monthly local unemployment rate. Where the 4-week periods that the durations are split into do not fall within a single month, we use a weighted mean of two succeeding months. Although the national unemployment rate showed an increasing trend in the observation period, there is considerable variation in this variable.

4. Results and discussion

We estimated the model separately for each sample, allowing the baseline hazard to vary between UB receivers and non-receivers. Both specifications, Eqs.

Ž . Ž . 20

2.2 and 2.6 , were estimated. For the specification without gamma hetero-geneity, we also estimated the model on the pooled sample to test ifgt,1990sgt,1991

for UB receivers and non-receivers, respectively.21

Ž .

Before turning to the results from estimating Eq. 2.6 , a look at some nonparametric estimates also is instructive. Figs. 1–3 plot Kaplan–Meier hazard rates for the same three spell definitions as in Table 1. In the first sample, there is

Ž .

a marked jump upwards in the hazard out of the unemployment register Fig. 1 at about 20 four-week periods, when benefits cease — suggesting that for some ‘‘survivors’’, benefit entitlement may be an important motivation for reporting.22

Ž .

For the hazards into the employment register Fig. 2 , no such effect is detectable from the nonparametric estimates alone. Fig. 3 suggests an increasing hazard after 72 weeks for the 1990 sample. The right-hand side panel of Fig. 1 shows that also in the 1991 sample, there is an increase in the hazard around 80 weeks, even if it is not as marked as in the 1990 group. One possible explanation is that lack of information about the change in benefit rules kept some unemployed individuals from reporting after having finished the 80 weeks. Finally, it is important to note that the hazard out of the unemployment register is about four times the hazard into recorded employment, suggesting that the unemployment register alone is a poor indicator of unemployment duration.

The estimation results favour the gamma heterogeneity specification, neverthe-less we report results from both specifications.23 To improve readability, the

20 Ž .

The program for estimating the model with gamma heterogeneity is written by Jenkins 1997 . 21

The specification with heterogeneity did not converge, probably due to the number of parameters necessary to estimate the unrestricted model — four baselines and a full set of sample interaction dummies on the covariates.

22

Stratification by the UB receiverrnon-receiver dichotomy — not shown here — affirms that the jump in the hazard is caused by UB receivers.

23

Likelihood ratio tests of H : s2s0 yield x2 test statistics of 21.92 and 15.38, respectively 0

(13)

Fig.

1.

Out

of

unemployment

(14)

( ) E. Bratberg, K. VaagerLabour Economics 7 2000 153–180

166

Fig.

2.

Into

employment

(15)

Fig.

3.

Into

employment

combined

(16)

( ) E. Bratberg, K. VaagerLabour Economics 7 2000 153–180

168

Ž .

estimates ofgt are reported separately in Table 3a estimated by sample and 3b

Žpooled estimation . The estimates of. b can be found in Table 4. We can use Table 3a and 3b to evaluate two hypotheses. First, comparing receivers and non-receivers of UB, we predict higher hazard rates for the latter group. Second, if a fixed benefits period increases the hazard to employment, we predict higher

Ž

hazard rates for the UB receivers in the 1990 sample when there was a fixed

.

period than in the 1991 sample. For the non-receivers there should be no such decrease. This is our natural experiment.

Ž

Inspection of the gt-estimates without heterogeneity for the 1990 sample in

.

Table 3a suggests a tendency towards negative duration dependence for both UB

Ž

receivers and non-receivers. As one would expect, in the first three periods i.e.,

.

three months the latter group’s risk of leaving unemployment is higher than for those receiving UB, but from then on there are basically no differences in the estimated baselines. As for the effect from receiving benefits around the time of

Ž .

exhaustion period 20 , there are, surprisingly, no signs of increase in exit rates. There is, however, a slight increase in the periods after period 20. The estimates

with heterogeneity show flatter rates, consistent with the general result that

unobserved heterogeneity implies downward biased duration dependence. Here,

Ž .

too, the rate does not increase in period 20 in fact, it decreases , but it then increases slightly in the next 4 periods.24

Turning to the 1991 sample, there seem to be only small changes in the risk of leaving unemployment for the group of non-receivers. For UB-receivers, on the other hand, there is a distinct decrease in exit rates. Again, the differences between the two groups appear to be significant in the early periods. When approaching benefit exhaustion, the differences by and large become insignificant. This goes for both specifications. Note, however, that this time the lack of significance is not counter-intuitive. With the new rules there is no reason to expect any peak in the baseline hazard for benefit receivers.25

Table 3b shows estimates of the baseline parameters obtained from pooled

Ž .

estimation without heterogeneity of the samples, making statistical tests of sample differences possible. It confirms the impressions from the findings reported in Table 3a. First, it indicates no statistically significant change in the behaviour of unemployed without benefit entitlement. For those who were affected by the 1992 reform, namely the benefit receivers, the improved insurance for long term

24

We do not present formal tests of whether the increases after period 20 are statistically significant, but inspection of confidence intervals indicate that they are not. In the specification without

heterogene-w x

ity, the 95% confidence interval forg20 is y5.885,y4.877 for UB receivers, in the heterogeneity

w x

specification it is y5.674,y4.479 . Thegt’s for the following periods fall within these intervals. 25

(17)

unemployment appears to have lowered the probability of exiting the unemploy-ment state.26 It is surprising, however, that the decrease is statistically significant only at the early stages of the unemployment spell. When it comes to the time when benefits expire, the conclusion from this part of the analysis is that it cannot be verified that the behaviour has changed because of the 1992 policy change.27

For visualisation, we refer the reader to the figures below. For the specification with gamma heterogeneity, Fig. 4 and 4b plot the estimated baseline hazards for UB receivers and non-receivers for the 1990 and 1991 sample, respectively. Fig. 5 compares UB receivers in 1990 and 1991, while Fig. 5b does the same for non-receivers.

Turning to the rest of the covariates, the effects are very much as one would expect. Previous income, which can be interpreted as an indicator of the opportu-nity cost of rejecting a job offer, has a significant and positive effect on the

28 Ž .

hazard. Having had a job the previous year Experience2 increases the hazard,

Ž .

as does the number of years with income above a minimum level Experience1 . Education has a positive and significant effect, almost identical in both samples.

Ž .

Those younger than the reference age group 36–45 years have a larger probabil-ity of getting a job, whereas the hazard is lower for people belonging to the older age groups. Having a non-Scandinavian citizenship also decreases the hazard. The local unemployment rate is a rough indicator of local labour demand, and should have a negative effect, which it has. Relatively to the base category, eastern Norway including Oslo, residing in other regions has a positive and significant effect in the 1991 sample that cannot be found in the other group.

26 Ž

The expected duration of the mean individual calculated as the integrated survivor function in the

.

specification with heterogeneity with UB increased from 605 to 776 days. For non-receivers, it decreased from 512 to 306 days. Because of censoring, these estimates by far exceed the observed

Ž .

mean durations. This, however, also seems to be the case in similar work, cf. e.g., Carling et al. 1996

Ž .

and Hernæs and Strøm 1996 . 27

We have also estimated UB receivers and non-receivers separately, but pooled across samples

Žwith separate baselines for the 1990 and 1991 samples . The heterogeneity-corrected estimates in.

general show a slight increase in the baseline hazards over the observation period. The reforms all over negative effect on the employment hazard remains unaltered, with a significant drop in employment risk for receivers from 1990 to 1991. The number of periods where the difference is significant also increases from 12 to 18. Among the non-receivers, there are only two periods where the difference between 1990 and 1991 are significant at the 5% level. The results are available from the authors upon request.

28

(18)

()

Ž .a Maximum likelihood estimates ofgafor UB-receivers and non-receivers

t

Ž .

With P-values for equal parameters receivers and non-receivers Wald tests .

Period 1990 1991

Without gamma With gamma Without gamma With gamma

heterogeneity heterogeneity heterogeneity heterogeneity

Pooled test all periods 0.000 0.000 0.000 0.000

Žx2 . Ž 2 . Ž 2 . Ž 2 .

s266.92 x s252.3 x s502.52 x s368.9

(19)

()

Coefficient SE Coefficient SE P Coefficient SE Coefficient SE P

1 y5.650 0.194 y6.154 0.195 0.067 y4.440 0.182 y4.922 0.185 0.064

2 y4.717 0.180 y5.313 0.182 0.018 y3.754 0.176 y4.012 0.175 0.297

3 y4.550 0.179 y5.216 0.181 0.009 y4.177 0.181 y4.325 0.179 0.562

4 y4.742 0.182 y5.296 0.183 0.032 y4.540 0.199 y4.352 0.186 0.492

5 y4.777 0.184 y5.398 0.186 0.018 y4.802 0.220 y4.814 0.207 0.969

6 y4.711 0.185 y5.182 0.185 0.072 y4.828 0.231 y4.762 0.213 0.833

7 y4.809 0.189 y5.210 0.187 0.132 y5.052 0.254 y4.810 0.220 0.473

8 y4.727 0.189 y5.243 0.189 0.054 y5.054 0.263 y4.907 0.232 0.676

9 y4.923 0.196 y5.383 0.194 0.095 y5.101 0.279 y4.597 0.222 0.157

10 y4.910 0.198 y5.264 0.193 0.200 y4.900 0.271 y4.635 0.230 0.455

11 y4.729 0.196 y5.341 0.196 0.027 y4.888 0.279 y5.040 0.266 0.693

12 y4.778 0.199 y5.529 0.203 0.008 y4.725 0.275 y5.070 0.274 0.374

13 y5.428 0.226 y5.563 0.206 0.658 y4.832 0.294 y4.975 0.273 0.721

14 y5.172 0.218 y5.528 0.208 0.238 y5.225 0.357 y5.153 0.306 0.879

15 y4.982 0.214 y5.330 0.204 0.240 y5.116 0.357 y4.933 0.293 0.692

16 y5.234 0.230 y5.587 0.216 0.263 y5.545 0.441 y4.845 0.292 0.185

17 y5.212 0.232 y5.682 0.223 0.144 y5.189 0.390 y5.309 0.356 0.821

18 y5.183 0.235 y5.436 0.215 0.427 y5.596 0.477 y5.336 0.371 0.668

19 y5.346 0.250 y5.179 0.209 0.610 y5.750 0.527 y5.081 0.343 0.287

20 y5.380 0.257 y5.478 0.225 0.774 y5.277 0.441 y4.864 0.322 0.449

21 y5.245 0.252 y4.911 0.205 0.304 y5.418 0.477 y4.940 0.343 0.416

22 y5.279 0.260 y5.018 0.213 0.436 y4.897 0.391 y5.316 0.412 0.460

23 y5.180 0.260 y5.241 0.227 0.860 y4.827 0.391 y4.998 0.371 0.751

24 y5.163 0.266 y5.062 0.222 0.770 y5.455 0.527 y4.814 0.356 0.314

25 y5.895 0.346 y5.550 0.254 0.421 y5.421 0.527 y5.458 0.476 0.958

26 y5.619 0.325 y5.450 0.253 0.682 y5.338 0.527 y5.186 0.440 0.824

27 y5.373 0.309 y5.387 0.251 0.972 y5.248 0.527 y5.569 0.526 0.666

Pooled test 0.000 0.030

2 2

Žx27s75.5. Žx27s42.4.

a

(20)

()

E.

Bratberg,

K.

Vaage

r

Labour

Economics

7

2000

153

180

172

Table 4

Maximum likelihood estimates of covariate effects With asymptotic standard errors and P-values.

1990 1991

Without heterogeneity With heterogeneity Without heterogeneity With heterogeneity

Coefficient SE P Coefficient SE P Coefficient SE P Coefficient SE P

Income 1.49E-06 3.22E-07 0.000 2.37E-06 4.68E-07 0.000 7.03E-07 3.04E-07 0.021 1.04E-06 3.91E-07 0.008

Female 0.227 0.044 0.000 0.310 0.059 0.000 y0.089 0.042 0.035 y0.106 0.052 0.043

Spouse inc. y1.08E-06 6.97E-07 0.120 y1.57E-06 9.52E-07 0.099 y5.85E-07 5.75E-07 0.309 y8.49E-07 7.51E-07 0.258 Spouse inc.=Female 6.42E-07 7.90E-07 0.416 1.16E-06 1.08E-06 0.286 1.94E-06 6.80E-07 0.004 2.37E-06 8.76E-07 0.007

Childr-11 0.062 0.048 0.199 0.047 0.064 0.462 y0.018 0.044 0.682 y0.034 0.057 0.553

Ch)11,-18 y0.050 0.064 0.441 y0.054 0.085 0.528 0.171 0.051 0.001 0.214 0.068 0.002

Childr-11=Female y0.234 0.060 0.000 y0.267 0.078 0.001 y0.176 0.058 0.002 y0.210 0.072 0.004 Ch)11,-18=Female 0.308 0.078 0.000 0.385 0.105 0.000 y0.036 0.073 0.621 y0.072 0.092 0.433

Education 0.092 0.010 0.000 0.117 0.014 0.000 0.093 0.009 0.000 0.116 0.013 0.000

Exper. 1 0.027 0.005 0.000 0.036 0.007 0.000 0.028 0.005 0.000 0.036 0.007 0.000

Exper. 2 0.381 0.037 0.000 0.474 0.050 0.000 0.247 0.034 0.000 0.302 0.044 0.000

Age 17–20 0.382 0.102 0.000 0.508 0.133 0.000 0.661 0.101 0.000 0.786 0.128 0.000

(21)

()

E.

Bratberg,

K.

Vaage

r

Labour

Economics

7

2000

153

180

173

Age 46–55 y0.138 0.079 0.080 y0.178 0.102 0.081 y0.077 0.073 0.289 y0.110 0.091 0.223

Age 56–62 y0.192 0.118 0.104 y0.207 0.153 0.177 y0.710 0.127 0.000 y0.858 0.156 0.000

Age 63–67 y1.664 0.268 0.000 y2.043 0.317 0.000 y1.951 0.323 0.000 y2.294 0.359 0.000

Married 0.178 0.099 0.072 0.252 0.134 0.059 0.149 0.090 0.096 0.196 0.116 0.092

Prev. marr. y0.207 0.070 0.003 y0.276 0.090 0.002 y0.136 0.064 0.033 y0.171 0.079 0.031

Married=Female y0.053 0.129 0.678 y0.138 0.173 0.424 y0.520 0.127 0.000 y0.605 0.159 0.000

Region 2 y0.011 0.049 0.829 y0.065 0.064 0.309 0.116 0.045 0.009 0.135 0.055 0.015

Region 3 0.046 0.044 0.298 0.030 0.057 0.596 0.152 0.042 0.000 0.179 0.052 0.001

Region 4 y0.017 0.054 0.753 y0.073 0.070 0.296 0.140 0.050 0.005 0.163 0.062 0.008

Unemp.rate y0.039 0.009 0.000 y0.044 0.011 0.000 y0.023 0.008 0.006 y0.030 0.010 0.003

Non-Scand y0.341 0.138 0.013 y0.427 0.173 0.013 y0.260 0.139 0.061 y0.331 0.168 0.048

2

s 1.049 0.254 0.000 0.910 0.267 0.001

Log likelihood y14 748.05 y14 737.09 y17 515.20 y17 507.51

(22)

( ) E. Bratberg, K. VaagerLabour Economics 7 2000 153–180

174

Fig. 4.

Gender and some family relevant factors have effects that vary across the samples somewhat surprisingly. The effect of being female is significantly positive

Ž .

in the 1990 sample, and negative but smaller in the 1991 sample. Spouse income

Ž .

(23)

Fig. 5.

Žnegative coefficient on female , and married women to a larger extent retreat to.

Ž .

(24)

( ) E. Bratberg, K. VaagerLabour Economics 7 2000 153–180

176

in the out-of-unemployment-register Kaplan–Meier hazard at 80 weeks may show that information about the policy change is less than perfect among the receivers. Furthermore, if there are some unobserved factors characterising people who have a hard time finding a job, these could dominate the potential incentive effects — after 80 weeks, there might be a majority of ‘‘hard’’ cases. The same goes for ‘‘true’’ duration dependence — if the employment probability deteriorates over time due to, e.g., loss of human capital, that effect may wash out the incentive effect under study here.

Besides the explanations offered above, there is a fundamental problem with the kind of data that are available. Ideally, one would wish to have data on persons who differ considerably with respect to UB receipts, but are comparable on other dimensions. As UB is related to individual earnings histories, typically this will not be the case. The uniformity of the Norwegian unemployment insurance system, which makes the potential benefit duration identical across individuals, and also prevents people from opting out of the insurance system, strengthens this general problem. Our data suffer from lack of the variation in benefit rules that can

Ž .

be found, for instance, in those of Katz and Meyer 1990 .

Reminding the reader of the introductory discussion in Section 1, our findings are not completely at odds with existing empirical research. Narendranathan and

Ž .

Stewart 1993a report that in the UK, the effect of unemployment income

Ž .

declines with the length of the spell. Fallick 1991 draws similar conclusions from

Ž .

US data. Carling et al. 1996 do find positive exhaustion effects on the hazards into employment and labour market programmes. However, only the effects on exits to labour market programmes are significant at the 5% level. The data and specification employed by the different authors may not be directly comparable with the present analysis,29 but they express a common tendency — the potential

exhaustion effects of a fixed benefit period may be hard to detect. This does not imply that such effects are non-existing. They have, however, turned out to be hard to identify in several investigations. It may be tempting to conclude that the lack of the incentive effects in the present study is due to disincentives caused by the extensive use of labour market programmes. We think that some caution is called for here. Maybe the effect of a decline in the reservation wage becomes

Ž .

dominated by a reduced job offer probability long before 80 weeks of unemploy-ment duration. If so, it would not be enough to reverse the liberalisation of 1991 and 1992 to observe the incentive effect. It would also be necessary to reduce the 80 weeks period of benefits entitlement. Comparing with the Swedish study in

29 Ž . Ž .

Fallick 1991 and Narendranathan and Stewart 1993a; b let the hazard vary with benefits over the course of the spell, the first author uses a flexible baseline, whereas the others use a time-varying

Ž .

(25)

Ž .

Carling et al. 1996 one cannot be quite certain that even reducing the period to 60 weeks would give clear effects.

In addition to the exhaustion effects, the prospect of benefit cuts should lead to increased search efforts in the entire unemployment period of the first sample compared to the second one. This form of disincentive effect from the abolition of fixed benefit durations appears to be present in our material. For nearly the whole first year of the search period the hazard to employment for UB receivers is significantly lower for the 1991 compared to the 1990 sample, while there is no such effect for non-receivers. In conclusion, therefore, the incentive effect of a fixed benefits period is not rejected by this analysis.

5. Concluding remarks

We have investigated the potential incentive effect of a fixed unemployment insurance period by comparing two large groups of Norwegian unemployed persons, where the last was affected by a rule change that in practice extended the benefits to more than 3 years. The main body of the analysis was performed within the framework of a proportional hazard model with a flexible baseline. Our results suggest that the main effect of benefits running out is to make people drop out of the unemployment register. There is no evidence that the hazard into employment increased when the end of benefits approached in the group that was not affected by the liberalisation of compensation rules. Consequently, we cannot reject the null hypothesis that behaviour around benefits expiration was unaffected by the change. Our results conform with some previous research, but not with US studies that find that benefits expiration significantly increases the exit rate into employ-ment. One explanation may be the attractiveness of labour market programmes. Alternatively, the differences may be explained by the incentive effects being outweighed by a reduced job offer probability because of the long benefit period. On the other hand, our results indicate that the reform appeared to have an all over negative effect on the employment hazard. This suggests that fixed duration benefit periods do have positive incentive effects, even in cases where the periods are as long as in Norway. Taken together with previous research, one may still argue in favour of the policy recommendation that the unemployment benefits period should be fixed.

Acknowledgements

We would like to thank Anders Bjorklund, Josef Zweimuller, and Alf Erling

¨

¨

(26)

( ) E. Bratberg, K. VaagerLabour Economics 7 2000 153–180

178

Appendix A. Table A1. Variable descriptions

Ž .

Income Gross income previous year 1989 NoK

Ž .

Spouse inc. Spouse’s gross income previous year 1989 NoK Female Dummy variables1 if female

Education Years of education

Ž

Experience1 Years with gross income)1‘‘G’’ regulated annually,

.

NoK 32275 in 1989

Experience2 Dummys1 if record in employment registerF1 year previous to unemployment period

Married Dummys1 if married

Div.rwidow Dummys1 if previously married Childr-11 achildren below 11 years

Childr)11-18 achildren between 11 and 18 years

Non-Scand Non-Scandinavian citizenship

Age groups Set of 6 dummy variables, reference is 36-45 years

Ž

Unemp Local unemployment rate time-varying, updated each

.

4-week period

Ž .

Region 1 Base category Eastern Norway, including Oslo Region 2 Dummys1 if residing in Central Norway

Region 3 Dummys1 if residing in WesternrSouthern Norway Region 4 Dummys1 if residing in Northern Norway

UB-receiver Dummys1 if receiving unemployment benefits

UB-period Set of five dummy variables indicating that UB-receivers

1 and periodsa, whereacorresponds to weeks 69–72,

73–76, 77–80, 81–84, and 85–88, respectively.

Month Set of six dummy variables controlling for the month the

Ž . Ž

spell started reference is June . Coefficients not

.

reported.

Appendix B. Table A2. Individuals at risk, exits and censored, by period

Period 1990 sample 1991 sample

At risk Exits Censo- At risk Exits

Censo-red red

0 9936 308 0 12 054 262 0

1 9628 629 0 11 792 595 0

(27)

( ) Appendix B. Table A2 continued .

Period 1990 sample 1991 sample

At risk Exits Censo- At risk Exits

Censo-red red

6 5030 174 296 6802 216 341 7 4560 167 268 6245 187 302 8 4125 126 247 5756 164 307 9 3752 119 211 5285 161 268 10 3422 126 170 4856 126 275 11 3126 112 171 4455 102 117

12 2843 62 78 4236 94 193

13 2703 63 175 3949 84 265

14 2465 69 127 3600 94 189

15 2269 48 127 3317 72 149

16 2094 48 88 3096 57 104

17 1958 43 90 2935 65 124

18 1825 34 75 2746 78 108

19 1716 33 76 2560 59 94

20 1607 34 50 2407 86 97

21 1523 34 59 2224 68 106

22 1430 34 88 2050 53 120

23 1308 28 74 1877 58 105

24 1206 15 27 1714 33 38

25 1164 17 104 1643 34 114

26 1043 19 75 1495 33 107

References

Atkinson, A.B., 1987. Income maintenance and social insurance. In: Auerbach, A.J., Feldstein, M.

ŽEds. , Handbook of Public Economics, Vol. 2, North-Holland, Amsterdam..

Atkinson, A.B., Micklewright, J., 1991. Unemployment compensation and labor market transitions: a critical review. Journal of Economic Literature 29, 1679–1727.

Carling, K., Edin, P.-A., Harkman, A., Holmlund, B., 1996. Unemployment duration, unemployment benefits, and labor market programs in Sweden. Journal of Public Economics 59, 313–334. Dolton, P., van der Klaauw, W., 1995. Leaving teaching in the UK: a duration analysis. Economic

Journal 105, 431–444.

Fallick, B.C., 1991. Unemployment insurance and the rate of re-employment of displaced workers. Review of Economics and Statistics 73, 228–235.

Heckman, J.J., Singer, B., 1984. A method for minimizing the impact of distributional assumptions in econometric models for duration data. Econometrica 52, 271–320.

Hernæs, E., Strøm, S., 1996. Heterogeneity and unemployment duration. Labour 10, 269–296. Jenkins, S.P., 1997. Discrete time proportional hazards regression. STATA Technical Bulletin 39,

(28)

( ) E. Bratberg, K. VaagerLabour Economics 7 2000 153–180

180

Katz, L.F., Meyer, B., 1990. The impact of the potential duration of unemployment benefits on the duration of unemployment. Journal of Public Economics 41, 45–72.

Korpi, T., 1995. Effects of manpower policies on duration dependence in re-employment rates: the example of Sweden. Economica 62, 353–371.

Layard, R., Nickell, S., Jackman, R., 1991. Unemployment, Oxford University Press, Oxford. Meyer, B.D., 1990. Unemployment insurance and unemployment spells. Econometrica 58, 757–782. Micklewright, J., Nagy, G., 1996. Labour market policy and the unemployed in Hungary. European

Economic Review 40, 819–828.

Mortensen, D.T., 1977. Unemployment insurance and job search decisions. Industrial and Labor Relations Review 30, 505–517.

Narendranathan, W., Stewart, M.B., 1993a. How does the benefit effect vary as unemployment spells lengthen? Journal of Applied Econometrics 8, 361–381.

Narendranathan, W., Stewart, M.B., 1993b. Modelling the probability of leaving unemployment: competing risks models with flexible base-line hazards. Journal of the Royal Statistical Society, Series C: Applied Statistics 42, 63–83.

Prentice, R.L., Gloeckler, L.A., 1978. Regression analysis of grouped survival data with application to breast cancer data. Biometrics 34, 57–67.

Figur

Table 1Spell characteristics days . Standard deviations in parentheses

Table 1Spell

characteristics days . Standard deviations in parentheses p.10
Table 2Sample characteristics

Table 2Sample

characteristics p.11
Fig. 1. Out of unemployment register.
Fig. 1. Out of unemployment register. p.13
Fig. 2. Into employment register.
Fig. 2. Into employment register. p.14
Fig. 3. Into employment — combined registers.
Fig. 3. Into employment — combined registers. p.15
Table 3a Maximum likelihood estimates of

Table 3a

Maximum likelihood estimates of p.18
Table 4Maximum likelihood estimates of covariate effects

Table 4Maximum

likelihood estimates of covariate effects p.20
Gender and some family relevant factors have effects that vary across theFig. 4.samples somewhat surprisingly

Gender and

some family relevant factors have effects that vary across theFig. 4.samples somewhat surprisingly p.22
ŽFig. 5.negative coefficient on female , and married women to a larger extent retreat to.non-market activities negative coefficient on the marriage interaction term .
ŽFig. 5.negative coefficient on female , and married women to a larger extent retreat to.non-market activities negative coefficient on the marriage interaction term . p.23

Referensi

Related subjects :

Pindai kode QR dengan aplikasi 1PDF
untuk diunduh sekarang

Instal aplikasi 1PDF di