Labour market transitions and unemployment ... - Wiley Online Library

1 downloads 0 Views 1MB Size Report
unemployed and remaining without a job for a long period of time. We also ... Keywords: unemployment, labour market flows, micro-data, duration analysis.

Economics of Transition, Volume 6 (I),113-144, 1998

Labour market transitions and unernployment duration: Evidence from Bulgarian and Polish micro-data' Sandrine Cazes" and Stefan0 Searpetta"" *International Labour Office (ILO) 4, route des Morillons CH- 121 1 GENEVE 22 Tel: + (22) 799 61 11 Fax: + (22) 798 8685 E-mail: [email protected]

**OECD 2, Rue Andre Pascal 75775 PARIS Cedex 16 Tel: + (33) 1 4524 9023 Fax: + (33) 1 4524 9050 E-mail: [email protected]

Abstract The segmentation of the labour market is one of the most striking characteristics of the transition process in Central and Eastern European countries. Not only do the young, unskilled workers and women face a high risk of unemployment, but joblessness also varies significantly geographically. This paper sheds some light on labour market segmentation in transition countries by analysing individual records of individuals registered at the labour offices of two Polish regions (Warsaw and Ciechanov and two Bulgarian regions (Sofia and Botevgrad) over the initial three to four years of the transition to a market economy. The empirical results confirm the existence of highly selective firing and hiring processes in the Polish and Bulgarian labour markets. Overall, unskilled or poorly educated workers have the highest probability of becoming unemployed and remaining without a job for a long period of time. We also analysed the determinants of unemployment duration across regions and over time using a piece-wise constant hazard model with multiple destinations, i.e. employment and exit from the labour force. The results suggest that the unemployed with a high education and previous experience in the private sector have a higher probability of getting a new job, especially in the more dynamic labour markets, while those without previous work experience tend to stay unemployed for a longer period of time and often leave the labour market. The econometric results also suggest that the reforms of the unemployment benefit systems have produced important effects on unemployment flows.

JEL classification: J63, J64, J78. Keywords: unemployment, labour market flows, micro-data, duration analysis.


Labour market transitions and unemployment duration

1. Introduction The transition to a market economy in Central and Eastern Europe has been accompanied by the build-up of high and persistent unemployment. From being virtually zero before the transition, unemployment rates often rose above double-digit levels in a few years. The incidence of long-term joblessness has also increased rapidly and many long-term job-seekers have left the labour market altogether instead of being re-integrated into the work life. There is also clear evidence that certain vulnerable groups, including women, unskilled workers and those without work experience have been disproportionally affected by unemployment and long-term joblessness. Moreover, labour market imbalances have a clear geographical dimension in Central and Eastern Europe, with unemployment rates often ranging from less than 10 per cenr in the capitals and some large urban areas to more than 20 per cenr in backward areas (Boeri and Scarpetta, 1996). Resuming economic growth in the most recent years has brought about a stabilization or even some decline in the number of job-seekers in most Central and Eastern European countries. Yet unemployment rates are still high compared with the OECD average in most countries of Central and Eastern Europe and joblessness remains highly concentrated on certain regions and social groups. In this context, policy-makers have been facing the delicate challenge of providing an adequate social safety net while, at the same time, stimulating market-oriented behaviour among economic agents. Both income support schemes for the unemployed and active programmes to facilitate their (re)insertion into employment have been introduced in all countries since the beginning of the transition process. The unemployment insurance benefits have subsequently been reformed in most countries, tightening eligibility conditions and reducing the maximum duration of benefits (see OECD, 1996). At the same time, more generous conditions are generally offered to disadvantaged workers, such as those living in backward areas or elderly workers with limited chances of getting a new job. The rapid upsurge of unemployment together with the introduction-and subsequent revisions-of labour market programmes in transition countries have stimulated a great deal of research, but this research has until recently been largely based on aggregate data.* This study is based on detailed micro-data from the administrative records of Poland and Bulgaria. They make it possible to take into account the regional and time dimensions of unemployment and to link these factors to individual characteristics, such as the sex, the age, the level of education, the skills and the like. Moreover, the longitudinal data enable the experience of individual workers during their spell of joblessness to be followed, thereby offering useful insights on duration dependence. There is a number of issues which could be addressed with the help of micro-data. In particular, does the risk of becoming unemployed depend upon individual characteristics? Does this risk vary across regions for workers with similar characteristics? Has the risk of unemployment evolved over time? Have the reforms in the UB systems produced significant effects on unemployment duration and destination of the outflows? Do these effects change across different local labour markets? The paper is organized as follows. Section 2 presents the micro-data for two Bulgarian regions and two Polish regions and briefly discusses the labour market and overall economic conditions of these regions. A detailed analysis of inflows into unemployment, the lengths of the unemployment spells and the destinations of outflows by individual characteristics, time and location is the subject of Section 3. In particular, the analysis focuses on how the unemployment risk and probability of leaving unemployment vary across social groups, over time and local labour markets. Section 4

Cazes and Scarpetta


presents the results of an econometric model of unemployment duration. We specify a reduced form job search model where we allow for a flexible baseline hazard and multiple destination states out of unemployment, i.e. employment and exit from the labour force. Section 5 concludes.

2. The data 2.1. The regions selected for analysis The regions chosen for the analysis are characterized by significantly different economic and labour market conditions. On the one hand, we have the two capitals-Sofia and Warsaw-where structural economic factors have contributed to a rapid and successful transition with lower than average costs in terms of unemployment. In particular, the two capitals have a highly differentiated economic structure, with a significant role played by the service sector in the allocation of labour (even before the beginning of the transition). These regions have benefited more than other areas from price and trade liberalization and from the development of private activities. A booming demand for services as well as more developed infrastructures and financial conditions for the start-up of new ventures have fostered their economies. Moreover, the labour force in these regions is often better educated than the average and offers a wider distribution of skills than those in rural and highly industrialized areas. After a few years into the transition process, these regions already had most of the employment in private ventures, they absorbed significant fractions of total foreign direct investments and developed close links with foreign market^.^ Of the two other regions selected for this study, one is a rural area (Ciechanov in Poland) and the other is a highly industrialized area (Botevgrad in Bulgaria). In both cases, the transition process has been more difficult up to now. State subsidies to agriculture and industrial activities have been phased out and the orientation of demand severely hit existing firms. The development of a new private sector encountered greater difficulties in these regions due to the lack of demand and financial resources, while foreign direct investments have not played a comparable role of that in the capitals. The wide differences in economic performance of the two capitals compared to the other two regions resulted in significantly different labour market conditions (Table 1). In 1995, the two capitals had the lowest unemployment rates of their respective countries. On the contrary, Botevgrad and Ciechanov had higher than average unemployment rates, with the latter being at the bottom end of the regional unemployment rate distribution. Regional disparities are even stronger if one looks at both sides of the labour market: the ratios of unemployment to vacancies (the so called U N ratio) were the lowest in Warsaw and Sofia, while they were amongst the highest in Ciechanov and Botevgrad in 1993 and in Ciechanov in 1995.4 As shown in Table 1, flows into unemployment have been relatively limited in all four regions, even if lower flows were recorded in the capitals compared to the other regions. The employment adjustment has been largely obtained in the transition countries via attrition (e.g. normal and early retirements, invalidity pensions) and reduction of working hours rather than mass lay-offs, and thus flows into unemployment record only part of the overall adjustment. A significant contribution to the divergent pattern of aggregate unemployment comes from the very different rates at which unemployed workers left the pool and, in particular, the different number of those who came back to work as opposed to those who left the labour market.


Labour market transitions and unemployment duration

To sum up, the four regions selected for the empirical analysis can be classified into two broad labour market prototypes: the first-which includes the two capitals-is a dynamic labour market while the second-including Ciechanov and Botevgrad-is a rather stagnant labour market. Since the two countries are characterized by different unemployment benefit systems and, more generally, different labour market policy settings, the sample allows investigation of how the same policy strategy works in different labour market conditions (Warsaw and Ciechanov, on the one hand, Sofia and Botevgrad, on the other) and comparison of different policy options in broadly similar labour market conditions, across countries.

2.2. The micro-data Our analysis is based on micro-data drawn from the administrative records of local labour offices. Administrative records of the unemployed include information on the age, sex, level of education, sector of origin (if the worker has a previous work experience), skills, date of beginning of the spell, level (if eligible) of unemployment benefits, etc. Each record is up-dated regularly so as to report any change in the labour market status of each individual. Thus, these data make it possible to study unemployment from a multi-dimensional perspective, comparing individual experiences over time and across regional and individual characteristics. In Poland and Bulgaria, the number of job-seekers who register at the local labour office is high since registration is a pre-condition for receiving-if eligibleunemployment benefits as well as most social assistance benefits (social benefits, health insurance e t ~ . )Inflows .~ into registered unemployment include workers coming from employment-after having been laid-off or having quit voluntarily-or from inactivityschool leavers, other new entrants into the labour force, etc. In Bulgaria and Poland, workers are allowed, under certain conditions, to combine earnings from work with unemployment benefitsa6Hence, registration data in transition countries do not fully conform to the ILO-OECD internationally agreed definition of unemployment. Yet, the comparison between these data and those from the Labour Force Surveys-which, in contrast, follow the standard definitions-suggests that registration data cover relatively well the ‘true’ unemployment, with roughly 70 per cent of those considered as unemployed according to the LFS standards being registered at the labour oEces.’ Data were extracted from the administrative registers in June 1994. They are based on a random extraction from the whole sample of those who entered into the unemployment pool in the last quarter of each year from 1990 (Poland) or 1991 (Bulgaria) to 1993.*The use of inflow data allows the analysis of unemployment duration for a sample of individuals who entered unemployment at the same time and, thus, it helps to avoid the risk of a sample selection bias, which is often present in the case of stock data.’

3. Unemployment flows 3.1. Inflows into unemployment Table 2 presents the main characteristics of the inflows into unemployment in the four regions. One of the most striking elements is the declining number of those eligible for UBs upon registration to the labour office in Poland. While in 1990 almost all new entrants received income support, only two-thirds of the newly registered unemployed in 1993 fulfilled the conditions for receiving such benefits. The tightening of the criteria for

C a w s and Scarpetta


UB provision at the beginning of 1992 is certainly responsible for this phenomenon. The observation that the decline in the proportion of insured unemployed upon registration occurred later in Ciechanov than in Warsaw can be explained by the higher incidence of laid-off workers in total inflows in Ciechanov, the majority of whom were eligible for UBs even within the new rules. A different picture emerges by reading the Bulgarian data over the 1991-93 period: in Sofia, a larger fraction of workers entering unemployment received unemployment benefits compared to Botevgrad. Since the UB system had not changed significantly over the period covered by the data, this different pattern can be explained by a greater number of school leavers (who are eligible for income support only for a short period of time) in Sofia than in Botevgrad (see Table 2). Table 2 also reveals significant differences in the demographic characteristics of workers entering into the unemployment pool. Both in Bulgaria and in Poland, there seems to be an inverse relationship between overall labour market pressures and the share of women in total inflows: the lower the unemployment rate, the higher the share of women in total inflows. Even if one looks at inflow rates (as opposed to inflow levels)", the picture does not change significantly. Several reasons could justify the different incidence of women in total inflows. Differences in regional economic structures are among them. In all regions, the employment adjustment in large state-owned enterprises (SOEs) has been highly selective, especially at the beginning of the transition process. Liquidation of certain activities of SOEs such as recreational services, health care, housing and transport (which were dominated by female employment) produced a disproportionate share of women in total inflows into unemployment. In backward regions, however, this type of labour adjustment in SOEs has often been insufficient to cope with the (generally more difficult) economic conditions, and downsizing touched both sexes rather than just women. The age structure of inflows into unemployment also indicates marked differences across types of regions. Backward regions (Botevgrad and Ciechanov) were characterized in the initial years of transition by a larger share of young workers in total inflows than the two capitals. Nevertheless, the proportion of young workers declined over time in backward regions, relatively more than in the capitals. The structure by education indicates that workers with low levels of education and/or unskilled are those most affected by unemployment, regardless of the region of residence. In the four regions of this study, the proportion of unemployed workers with low education (elementary, vocational or lower) increased between 1990191 and 1993: in Warsaw, for example, it rose from 27.4 p e r cent in 1990 to 59 p e r cent in 1993. Finally, there are significant regional differences in the sector of origin of the unemployed job seekers. Data for Poland reveal that the private sector is not only a destination of workers expelled from the state sector but also a source of unemployment. In Warsaw, which saw the most rapid development of private initiatives among the Polish regions, the proportion of inflows from the private sector in the total increased at a similar pace as the share of the private sector in total employment. In 1993, the private sector was the source of almost 45 per cent of total inflows into unemployment, from a modest 13 per cent of total inflows in 1990 when private initiatives were still limited. The situation in Ciechanov is partially different insofar as most agricultural firms were already in private hands at the beginning of the transition and thus a relatively larger number of unemployed came from the private sector even in 1990/91. The different role played by the private sectors in the two regions is also confirmed by the data on mode of separation. In Warsaw almost 44 per cent of workers left their job voluntarily or by means of a mutual agreement with their employers, while in Ciechanov 62 p e r cent of total inflows involved workers laid-off individually or by means of a group lay-off. Thus,


Labour market transitions and unemployment duration

unemployed workers in the capital include many who left their previous job attracted by the options opened up by a fast-growing private sector. To sum up, the micro-data confirm the marked differences among regions in a period of major transformation and reinforce our presumption that these differences deserve particular attention in the analysis of the labour market developments and in the design and assessment of policies.

3.2. Duration of the spells and outjlowsfiom unemployment Tables 3 and 4 present a breakdown of unemployment by duration for the whole population (i.e. also including uncompleted spells) and for those with completed spells." The first observation is the marked differences in the incidence of long-term unemployment (LTU), across types of regions, both in Bulgaria and Poland (based on data referring to the whole population). In the first case, the incidence of long-term unemployment in Botevgrad was three times that of Sofia in 1992. Differences in Poland are also marked: 53.7 per cent of those who become unemployed in Ciechanov in 1992 experienced more than 12 months of continued joblessness, while the proportion was 22 per cent in Warsaw. Table 3 suggests a different evolution of long-term unemployment in the two Polish regions. Focusing on the whole unemployed population, the incidence of long-term unemployment has not changed significantly in the first three years of transition (199092) in Ciechanov, while the proportion of long-term unemployment among those who left the pool (completed spells) fell significantly. The opposite picture emerges for Warsaw where long-term unemployment as a share of the total unemployed population fell over time, while the proportion of LTU among the completed spells remained broadly constant. In Bulgaria (Table 4) the evidence is less clear-cut, with a constant share of LTU in the total population of unemployed workers and a falling share of LTU among the outflows, in both regions. How can we explain these regional differences in the incidence of LTU? Did the reforms of the UB systems contribute to explain these differences? We should recall that the Polish UB system was reformed substantially over the 1990-92 period with a tightening of the eligibility criteria, a reduction in the maximum duration of benefits and the substitution of the earnings-related benefit system with a flat-rate system. In Bulgaria the reform of the UB system in 1992 was less marked, although it reinforced incentives to job search via a reduction of benefits, especially for those with high previous wages. These reforms are likely to have affected differently individuals with different demographic characteristics and living in regions with different overall economic and labour market conditions. Figures 1 to 4 present non-parametric (graphic) estimates of the hazard function for individuals who entered the unemployment pool in 1991 and 1992 in the four regions of this study. The hazard function corresponds to the conditional probability of leaving unemployment after a certain length of the unemployment spell. A natural estimator of the hazard is the ratio between the number of exits from unemployment at duration tj divided by the total number of individuals who could have left then. Essentially, this estimator is obtained by setting the estimated probability of completion of a spell at tj equal to the observed relative frequency of completion at $.I2 Formally:

i ( t i )='

h. "j


Cazes and Scarpetta

where h, is the number of unemployed workers who left the unemployment pool and n, is the number of unemployed workers in the risk set. The subscript j varies from 1 to 20 (and more) months of unemployment duration at which exits are observed. The standard deviations of these estimations were relatively small thereby confirming the appropriateness of the estimations. These aggregate estimates of the hazard function are remarkably different from those observed in many OECD countries. In particular, in most western European economies the hazard function tends to decline with duration, which implies that the chances of getting a job after a long spell of unemployment are much lower than the chances of a newly unemployed person.I3 Among the four regions of this study, Warsaw and-to a lesser extent-ciechanov seem to follow the experience of western European countries, with a declining hazard function up until the expiration of UBs. In the two Bulgarian regions, the hazard function increases with duration up to 7-8 months, then declines up to 13-15 months and rises again. The likely explanation for this unusual behaviour is the differentiated maximum lengths of UBs, whereby unemployed beneficiaries exhaust their benefits after 6 , 9 or 12 months. Some other elements emerge from Figures 1 to 4. Firstly, the effects of the UB reforms on the hazard function have been more marked in Poland than in Bulgaria. We considered three methods to test for the null hypothesis that the hazard functions were the same across populations of unemployed workers who entered into the register in the three years 1990, 1991 and 1992: the log rank test (Mantel-Cox test), the Breslow test (also known as the generalized Wilcoxon test) and the Tarone-Ware test.I4 In the case of Warsaw, the null hypothesis is rejected at the 1 per cent level by the three tests considered. These results are robust to stratification for sex and age. Also in the case of Ciechanov, the statistical tests tend to reject the null hypothesis in all but one case, that is, when we control for the gender of unemployed workers. As discussed below, these results are suggestive of a higher responsiveness of the male unemployed to changes in the unemployment benefit system. In the case of Bulgaria, on the contrary, the different tests cannot reject the hypothesis of constancy of the hazard function over changes in the UB regulations. In Poland, the exit probability towards the end of the entitlement period increased dramatically since the reform of the UB system in 1992 (Figures 1 and 2), while in Bulgaria-where only the level of unemployment benefits changed-the incentive of leaving the unemployment pool increased moderately (Figures 3 and 4). There has also been a stronger effect of reforms in backward areas than in the more developed areas. The increase in exit probabilities towards the end of the eligibility period-12 months in Warsaw and 18 months in Ciechanov (crisis regionfihas been higher in Ciechanov than Warsaw. In Ciechanov, the average exit probability between the tenth and twelfth month was four times lower than the average probability between the sixteenth and the eighteenth month, that is, just before the expiration of the right to UBs. Moreover, the exit probability, despite the length of the spell, was far lower in Ciechanov than in Warsaw. The economic and labour market conditions of the two Polish regions may help to explain the different effects of UB reforms on exit probabilities. In a stagnant region such as Ciechanov, workers remained on the unemployment register up until the expiration of the UBs because there were no alternatives. After the expiration of UBs, many of them left the register because of the limited support they could receive from the overloaded labour ofices and the lack of job opportunities in the formal labour market. These workers either entered the grey or black economy or left the labour market altogether. It should also be kept in mind that the reservation wage varies greatly across regions. If one



Labour market transitions and unemployment duration

assumes that during the period of fiuition of the UBs unemployed workers compare the UB level-which became uniform for all workers after the 1992 reform, regardless of the previous earnings-and the ‘prospective’ wage, that is, the wage implicit in the available (and acceptable) job offers, it is likely that disincentive effects are lower the higher the gap between the average prospective wage and the UB level. Since dynamic markets, such as Warsaw, are characterized by more vacancies which generally offer higher wages, the gap between UB levels and the prospective wage is large, and disincentive effects were contained. The effects of reforms may have affected the behaviour of certain individuals such as low-skilled workers and new entrants into the labour market, but not other workers with higher qualifications and previous work experience. On the contrary, in Ciechanov, where the few vacancies are likely to offer wages close to the minimum administrative wage and thus to the flat-rate unemployment benefit, the gap is lower and disincentive effects more marked, at least during the UB period. Demographic characteristics also affect unemployment duration. Tables 2, 3 and 4 reveal that during the period 1991-93, women not only had a higher probability of becoming unemployed (as shown on the basis of inflow rates in Table 2) but also a higher risk of remaining without a job for a longer period (whole population). Moreover, exit probabilities were generally lower in backward regions than in dynamic areas during the UB eligibility period but became higher after the expiration of benefits. However, since vacancies were scant, many outflows of women were towards inactivity rather than to jobs (see below). The incidence of LTU among young workers did not differ significantly from the aggregate values.16 However, the average duration of the spell for the successful cases, that is, those that left the unemployment pool, is much shorter than the average. These findings are suggestive of a duality of the unemployment experience for young workers: a first group is composed of those who successfully complete the spell with a job and the second group includes those who stay unemployed for long periods without receiving suitable job offers. As already stressed while analysing the inflow data, workersincluding young workers-with low levels of education andor low skills are the most disadvantaged in the labour market, especially in backward areas where competition for the few vacancies is tougher and selective procedures in the hiring process more severe. From Tables 2, 3 and 4, it emerges clearly that the UB reforms produced different effects on different categories of unemployed job-seekers. As stressed above, in backward regions, the reforms mainly affected those with some chances of getting a new job and, for this reason, the effects of reforms should be read from the second columns of Tables 3 and 4 (completed spells). For example, the incidence of LTU among women who left the register fell from 64.5 per cent for those who entered unemployment in 1991 to 38.9 per cent for those who entered in 1992 in Ciechanov, while it remained stable in Warsaw. In this latter region, on the contrary, the effects of reforms were mainly felt by those with a prospective wage close to the minimum, e.g. disadvantaged workers with long unemployment spells.

3.3. Destination of outflowsfiom unemployment Table 5 presents a breakdown of outflows by length of the spell and destination in Poland.” Available data were re-grouped into four main destinations: regular employment, subsidized employment (start-up projects, intervention works, public works and temporary subsidized works), inactivity and an additional category which includes those who left because of bad co-operation with the labour office. Table 5 suggests that the short-term unemployed (less than 6 months) often leave the register for a regular job or a subsidized job in the formal sector. At the same time, those

Cazes and Scarpetta


leaving towards the end of the UB entitlement are also likely to move to employment. Time spent as unemployed affects the probability of leaving the labour market, especially after UB exhaustion: in Ciechanov and Warsaw 47 per cenr and 60 per cent, respectively, of those who left the register after one year of a continuous spell became inactive. However, the comparison between the two regions should be made with care as the proportion of exits to the total number of unemployed workers is much smaller in Ciechanov than in Warsaw. Likewise, the ‘real’ destination of those who officially became inactive is unknown. It is likely that in Warsaw, outflows to inactivity include a significant migration towards the prosperous informal sector. On the contrary, a greater proportion of outflows to inactivity in Ciechanov is likely to be made by discouraged individuals who left the active life, or workers who performed occasional activities in the agricultural sector.

4. The determinants of unemployment duration This section aims at identifying the main determinants of unemployment duration by estimating hazard functions for the four regions under analysis. In particular, one of the key issues which has been discussed throughout this paper is whether changes in exit rates over time and across regions are due to changes in policy and overall economic conditions, or merely reflect heterogeneity among entrants. For example, cross-regional differences may be simply due to a larger than average proportion of low-skilled workers becoming unemployed in backward areas. We also study duration dependence whichalbeit to a lesser degree than in most western European countries-seems to be a problem, in Poland, in particular.

4.1. Description of the model In this study the individual probability of leaving unemployment is modelled using a reduced form job search model. The (conditional) probability of leaving unemployment is defined via the hazard function. Given our lack of knowledge as to the precise functional form of the instantaneous probability of leaving unemployment, following Lancaster (1990) we use a flexible functional form, a piecewise constant hazard:

The survivor function is:

F( t )=exp c, < t

{ IS -

(s) ds} = e x p { - c ; ,

b j B j -(t-cm )&+I


and m = 0,1,2, ..., M-1, and bj = c j - cj-1

The main advantage of using a flexible hazard function specification is that no


Labour market transitions and unemployment duration

unjustified restriction is imposed on the data. As stressed by Lancaster (1990), the piecewise-constant hazard is essentially a way to let the data tell us how the hazard behaves as a function of time. The non-parametric estimates of the hazard discussed above showed indeed that the hazard evolves over time in a non-monotonic way with a sudden increase towards the end of the benefit eligibility. Consequently, alternative specifications, such as the well-known Weibull, are not suitable. The regressors enter the model as follows:

where x is a vector of time-invariant regressors. The model has a proportional hazard characteristic insofar as the impact of covariates is to shift the baseline hazard by a constant factor. The pure dependence on time of the instantaneous probability of leaving unemployment is captured by the piecewise-constant baseline hazard 6,. As stressed in the previous section, the Polish data suggested that the destination of those who exit from unemployment varies with the duration of the unemployment spell and with personal characteristics. Hence, an analysis of the hazard function, which did not distinguish the different destinations out of unemployment, would lead to inappropriate conclusions. Given the data availability, it is possible to consider two different destinations out of unemployment: 1) employment, whether ‘regular’ or subsidized; and 2) inactivity. Moreover, right censoring is treated as an additional destination as suggested by Lancaster (1990). By definition, these destinations are mutually exclusive and exhaust the set of possible exit options. We assume that v k (k = I , . . . , K ) is a set of dummy variables that take value one if the individual exits to state k and zero otherwise. Therefore, the K transition intensities are: e k (t

)= lim dt -to

P( t IT < ( t+dt ), q k = I I T 1 t ) dt


The total of the survivors at time t who leave in the following period is the sum over k of those who leave for destination k, which also provides the relationship between the hazard function B ( t ) and the transition intensities 9( (f). Formally:


i.e. the hazard function is the sum of the transition intensities over the different destination states. The probability of exiting to state k in a small time interval (t, t + dt) can be written as : P ( l e f t f o r k a t t h e t i m e t ) = P( ) = & ( t ) e x p

From the above, the log-likelihood contribution for a single person is given by:

Cazes and Scarpetta


where 7 are the dummy variables for the k possible destinations. From the previous discussion, we assume that the f&are specified as functions of the observable explanatory variables x. As suggested by Narendranathan and Stewart (199 1)-under the assumption of absence of omitted heterogeneity terms correlated across the cause-specific hazards-the log-likelihood for the competing risks model is additively separable into terms each of which is a function of the parameters of a single cause-specific hazard." Therefore, maximizing the likelihood for the entire competing risks model is equivalent to separately maximizing the likelihood for the k" retirement state, while treating unemployment spells ending in other labour market states as right censored. The full log-likelihood is the sum of terms like ( 7 ) over i =I, 2, ..., N and is therefore:

4.2. Thefinal specification Duration information for each unemployed person is expressed in months of continuous unemployment spell. All specifications include age dummies (-25, 25-49, 50+),gender dummies, industry dummies-1) agriculture, 2) manufacturing, construction and energy; and 3) services for Poland and, given the sectoral breakdown of employment, 1) manufacturing construction and energy; 2) public services; and 3) other services for Bulgaria-and education dummies. On the basis of available data, four education levels have been considered: 1) higher education; 2) secondary general education; 3) elementary vocational education (not available for Bulgaria); and 4) primary education or lower. Given data availability, we also included in the Polish model a dummy identifying persons with a partial invalidity. The Polish data also allowed us to take into account the sector of origin of unemployed workers: 1) public sector; 2) co-operative; and 3) private sector. Three possible destinations out of unemployment are considered: 1) employment; 2) out-of-the-labour-force; and 3) censored data.'' Each model also includes dummies for the different years at which the job-seeker registered at the labour oftice. This enables us to take into account the effects of changes in the aggregate regional labour market conditions on the individual probability of leaving unemployment. The analysis focuses on individuals who registered in the last quarter of 1990 (only for Poland), 1991, 1992 and 1993. Given the structure of the dummies used in the empirical analysis, the 'reference' person is a prime-age male, with a primary or lower


Labour market transitions and unemployment duration

level of education coming from the state industry and who was entitled to unemployment benefits. Thus, when the coefficients of the explanatory variables are positive, this implies that the hazard is enhanced with respect to this ‘reference’ person, while the contrary holds if the sign of the coefficients is negative. The econometric analysis of this section enables us to test whether the changing patterns of the hazard in Poland can in fact be directly linked to the reform of the UB system, or rather to differences in the characteristics of the unemployed, before and after the reform of the UB systems.

4.3. Empirical results We present our empirical results in two ways.” Table 6 shows the estimated parameters of the hazard function referring to all registered unemployed in the four regions of the sample. The effects of the unemployment benefit system on the duration of unemployment are captured by a dummy indicating whether the individual was entitled to unemployment benefits. The relatively larger size of the Polish dataset also allows us to focus on the sub-sample of the recipients of unemployment benefits (Table 7 ) . Here we tested the effects of the unemployment benefit replacement rate on unemployment duration and destination of the outflows. Moreover, in the case of the two Polish regions, we estimate the hazard model for the period prior to the reform of the unemployment benefit system (1990-91) and for the period after the introduction of the reforms (199293). 4.3.1. Results for all unemployed workers At first glance, the results for the two Polish regions suggest that individual characteristics have assumed a greater importance after the reform of the UB system than before. This is consistent with the idea that the restrictions introduced in the Polish UB systems-either explicitly as the reduction in the length of the maximum eligibility, or implicitly as the tightening of the screening achieved through reforms of the eligibility criteria and by a learning-by-doing process by the staff of the labour offices-have increased the incentive towards job search. The dummy also captures the incentive effects for UB recipients. In the two capitals (1992-93, in the case of Warsaw) where more vacancies were available, those without UBs had a higher overall hazard (results from the single destination model, Table 6 ) , after controlling for individual characteristics.*’ However, Table 6 suggests that people without unemployment benefits exit the unemployment register to inactivity rather than to employment. Indeed, the overall effect of the UBs on the hazard out of unemployment masks two opposite effects for the two destinations: the absence of UBs ceteris paribus lowers the probability of moving from unemployment to employment and raises the probability of moving to inactivity. The effects are not statistically significant in Ciechanov prior to the reform of the UB system-when almost all unemployed were in receipt of benefits and therefore there is not enough variability to assess the importance of unemployment benefits on the hazard. It is important to notice, however, that even in Warsaw, where there were more job opportunities for the unemployed, those without UBs had a higher probability of moving to inactivity than those with benefits in the 1992-93 period. We should recall here that those who are not entitled to unemployment benefits arc workers without previous work experience and those with repeated unemployment spells. Their higher risk of moving from unemployment to inactivity can be interpreted as a sign of the preference of employers for experienced workers as well as the higher ability of the latter in the job search process. The results suggest that, ceteris paribus, women had a lower hazard to employment in

Cazes and Scarpetta


the two Polish regions. However, despite their lower probability of finding a job, women do not quit the labour market as quickly as men, as indicated by the negatively signed coefficients for the transition from unemployment to inactivity. The age coefficients suggest a somewhat surprising higher hazard to inactivity for young workers compared with prime age workers (the reference group). Moreover, young workers have a significantly lower probability of moving into employment, in Warsaw and in the two Bulgarian regions, even if in the latter the coefficients have a high standard error. Several factors are likely to affect the observed differences in the sign and magnitude of the coefficients for young workers across the four regions. As stressed above, the shorter duration of unemployment benefits and the fact that most of them were not eligible for means-tested social benefits upon completion of the UB period can certainly explain the higher hazard to inactivity, even if it should be stressed that leaving the register does not necessarily mean becoming inactive but may also mean moving to the informal economy. The results for older workers offer a rather mixed picture: in Warsaw they nave a lower hazard to employment, while in Sofia and Botevgrad they have a higher risk of moving to inactivity. The education coefficients offer a fairly consistent picture across countries and regions. In all regions, workers’ education is crucial for the transition from unemployment to a job, with workers with higher education experiencing the highest hazard to job and those with primary or less (the reference group) having by far the lowest hazard to job and the highest hazard to inactivity.22In Poland, workers with higher or secondary levels of education had a consistently higher hazard to employment and, in Warsaw, a significantly lower risk of moving to inactivity. These results are broadly confirmed in Botevgrad, while Sofia is the only region where education did not play a significant role. While cross-country differences are likely to be affected by differences in the classification of workers in these broad education categories, the results seem to confirm the relatively better position of workers with high education levels. 4.3.2. Results for recipients of Unemployment benefits Table 7 focuses on the unemployment benefit recipients in Poland. The results do not lend strong support to the hypothesis of a disincentive effect of UBs on job search. The coefficients for the unemployment benefit replacement rates are wrongly signed in the period prior to the reform on the UB system, probably because the unlimited duration of the benefits makes it difficult to properly estimate their impact on the unemployment transition. However, in Ciechanov the replacement rate seems to play a significant role in lowering the hazard to employment in the most recent period. This result can be justified by considering that in Poland since the 1992 reform the level of benefits is equal for all beneficiaries, while the wage levels of potential vacancies are well below the national average. Consequently the effective replacement rate is higher than the national average and higher the lower is the previous wage and the potential wage the unemployed can obtain in the market (i.e. the unemployment trap). The results in Table 7 suggest some mild variation in the hazard across workers depending on their sector of origin. Holding everything else constant, workers coming from the service sector experienced a somewhat higher risk of moving from unemployment to inactivity than those coming from industry (the reference category) in Warsaw. The transition from unemployment to employment is significantly quicker for workers with previous experience in the private sector. This is consistent with the idea that experience in the private sector does help the job-seeker in future job searches, through the contacts which helshe may have developed, and a better attitude in the search process.


Labour market transitions and unemployment duration

5. Concluding remarks This paper focuses on the functioning of local labour markets in two Central and Eastern European countries-Bulgaria and Poland-over the first four years of transition to a market economy. The analysis is based on micro-data from the unemployment register of local labour offices in Botevgrad and Sofia in Bulgaria and Ciechanov and Warsaw in Poland. These represent considerably different labour market cases, with the two capitals accounting for dynamic markets and the other two areas accounting for more stagnant markets. Botevgrad is, indeed, a region dominated by heavy industry with severe restructuring problems, while Ciechanov is a rural area with different, though comparably arduous, problems of transformation. The micro-data suggest that certain workers faced a higher risk of becoming unemployed than others during the first phases of the transition. Regardless of the local labour market conditions, those with low levels of education (primary education or less) generally experienced a higher risk of becoming unemployed. Likewise, the econometric analysis of the hazard function suggests lower than average lengths of unemployment among workers with medium and high education levels and a higher probability of obtaining a new job. Workers with low education not only have higher probabilities of being laid off but also a lower probability of being hired while unemployed. The unemployment risk and the length of joblessness for other social groups are considerably affected by local labour market conditions. Our findings confirm the results obtained at the more aggregate levels which pointed to a higher risk of unemployment for women relative to men in the relatively more developed areas, such as the two capitals considered in this paper. Women, in these contexts, also face a higher risk of becoming long-term unemployed. In contrast, in less developed areas such as Botevgrad and Ciechanov, young workers are those most affected by joblessness even if their hazard rate is generally higher than that of prime-age workers. The shorter duration of unemployment benefits and the fact that, in general, young workers are not entitled to means-tested social benefits can explain these higher hazard rates which, at least in the case of Poland, are mainly due to inactivity or perhaps unrecorded activities. Several results can be drawn from the empirical analysis of unemployment duration. We first presented Kaplan-Meier estimates of the total hazard and concluded that the probability of leaving unemployment declines with duration in Poland until the typical exhaustion dates for unemployment benefits, where it shows a dramatic spike. In Bulgaria, the hazard follows a more complicated pattern, though in Botevgrad a spike at the date of exhaustion of unemployment benefits is also observed. The econometric analysis of the hazard function also reveals that the reforms of the unemployment benefit systems have produced considerable effects on the labour market. Overall, the tightening of the eligibility criteria has produced a significant reduction in the number of unemployed who receive unemployment benefits in all labour markets. Notwithstanding this widespread effect on the coverage of UBs, the reforms have produced very different effects on the behaviour of those eligible for benefits in the different labour markets. For example, the move from an almost universal and unlimited UB system in Poland to the current system which offers a flat level of benefits for a limited period produced very different effects on workers’ behaviour in the two Polish regions. Since the tightening of the UB system, workers with no right to benefits exit the unemployment pool much more quickly than UB recipients, ceteris paribus. Those who are entitled to benefits tend to stay in the pool until the exhaustion of the benefits and leave it afterwards. However, a number of those who leave the unemployment pool upon exhaustion of the benefits move to inactivity rather than to employment, especially in backward areas.

Cazes and Scarpetta


Endnotes 1.

2. 3.

4. 5.




9. 10. 11.




We thank Robert Plasman for helpful discussions. A preliminary draft of this paper was presented at the 1997 EALE Conference in Aarhus (Denmark, September 25-28) and we gratefully acknowledge comments from several participants. The views expressed herein are those of the authors and do not necessarily reflect those of the organisations of affiliation. Notable exceptions include Ham et al. (1995); Kotzeva et al. (1996); and Terrell el al. (1996). See amongst others Bobeva and Hristoskov (1995); Fazekas (1994); Gora and Lehmann (1995) and Scarpetta (1995). For comparison, the U N ratios were on average 22 in the United Kingdom and 53 in Spain in 1991. For this reason, outflows from unemployment due to the expiration of the rights to unemployment benefits are relatively low, at least when compared with the experience of many OECD countries. However, only a limited number of unemployed workers do in practice combine earnings with unemployment benefits. In Poland it occurred especially in 1990 when all job-seekers could register as unemployed. In Bulgaria this possibility is open to workers with special work contracts (civil contracts) and with a wage below 150% of the minimum wage. See OECD (1993) and the OECD-CCET Labour Market Database, various issues. In Bulgaria, the records selected refer to individuals born on the 7th, 17th and 27th of each month. In Poland, data were selected by random extraction using the SPSS statistical programme. Further details on the micro-data are in Cazes and Scarpetta (1995). The long-term unemployment spells are generally over-represented if one uses stock data, which may give rise to sample bias. See Kiefer (1988). The inflow rate is the ratio between the number of women entering unemployment and the population of origin, that is, the female working age population. The first variable is useful for assessing the importance of changes in the unemployment benefit systems on the behaviour of the job-seekers; the second is more appropriate for assessing the incidence of long-term unemployment (more than 12 months of continuous spell). The corresponding estimator for the survivor function is:

See Layard et al. (1991) for a survey and Katz (1986). The three tests are all based on the weighted difference between the observed (Oj) and expected (Ej) number of exits at each of the time points. A component of the statistic can be written as follows: ’

where wi is the weight for time point j , and K is the maximum number of months. The log rank test uses equal weights for all observations, while in the Breslow test



16. 17. 18.



21. 22.

Lrrbour market transitions and unemployment duration

the weights are the number at risk at each point in time, and for the Tarone-Ware test the weights are the square root of the number at risk. All three tests are distributed as a ~ ~ ( 2 ) . These results seem in line with previous findings for OECD countries. According to several studies, the exit probability declines during the first months of unemployment and then rises again towards the end of the entitlement period. See, in particular, Katz and Meyer (1990); and Joutard and Ruggiero (1994). Data are not reported in the table but can be found in Cazes and Scarpetta (1995). Data reported in the table are the averages over the 1990-93 period, which allowed for more degrees of freedom than annual data. We also tested for unobserved heterogeneity under the simplifying assumption that it is not correlated across cause-specific hazards. Following Meyer (1990), we assumed that unobserved heterogeneity can be summarised by a Gamma distributed random variable with unit mean and variance cf? which is distributed independently of both x and t. The general hazard model in this case is:

In the case of the Bulgarian data, we considered those who withdrew from the register as censored cases. There is indeed no reason to assume that these persons were no longer looking for a job even if they were no longer registered. The log-likelihood ratio tests for the presence of unobserved heterogeneity were not statistically significant at the one per cent level in the four regional models. Thus we report in Tables 6 and 7 the results without control for unobserved heterogeneity. Kotzeva et al. (1996) obtained similar results for Bulgaria using a logit model. Ham et al. (1995) obtained similar results in the case of male unemployed in the Czech Republic.

References Bobeva, D. and Hristoskov (1999, ‘Unemployment in Agricultural Regions: An Overview of Central and Eastern Europe and a Case-Study from the Bulgarian Experience’, In: Scarpetta S. and Worgotter A., eds., The Regional Dimension of Unemployment in Transition Countries, Paris: OECD-CCET. Boeri, T. and Scarpetta S. (1996), ‘Regional Mismatch and the Transition to a Market Economy’, Labour Economics, 3(3), pp.233-54. Cazes, S. and Scarpetta S. (1993, ‘Caractkristiques individuelles, marchis du travail locaux et ch6mage en Pologne et en Bulgarie: l’apport des micro-donnies’, Revue de 1 ’OFCE,54, July. Fazekas, K. (1994), ‘Types of Micro-Regions, Dispersion of Unemployment and Local Employment Development in Hungary’, Institute of Economics of the Hungarian Academy of Science, Discussion Paper, No. 19. Gora, M. and Lehmann H. (1995), ‘How Divergent is Regional Labour Market Adjustment in Poland?’ In: Scarpetta s. and Worgotter A., eds., The Regional Dimension of Unemployment in Transition Countries, Paris: OECD-CCET. Ham, J. C., J. Svejnar and K. Terrell (1995), ‘Unemployment, the Social Safety Net and

Cazes and Scarpetta


Efficiency During the Transition to a Market Economy: Evidence from Micro Data on Czech and Slovak Men’, mimeo, May. Joutard, X. and Ruggiero M. (1994), ‘Taux de sortie du chbmage A l’approche de la fin des droits A l’indemnisation. Une ttude de pBriodes atypiques durant 1’Cpisode de chbmage’, Economie et Prhision, No. 113-4. Katz, L.F. (1986),‘Layoffs, Recall, and the Duration of Unemployment’, National Bureau of Economic Research Working Paper, No. 1,825. Katz, L.F. and Meyer B. D. (1990), ‘The Impact of the Potential Duration of Unemployment Benefits on the Duration of Unemployment’, Journal of Public Economics, No.72. Kiefer, N. M. (1988), ‘Economic Duration Data and Hazard Functions’, Journal of Economic Literature, 26(2), pp.649-79. Kotzeva, M.D., D. Mirheva and A. Worgotter (1996), Evidence from Evaluation Research in Selected Countries: Bulgaria, in OECD (1996). Lancaster, T. (1990), ‘The Econometric Analysis of Transition Data’, Econometric Society Monographs, No. 17, Cambridge: Cambridge University Press. Layard, R., S. Nickel1 and R. Jackman (1991), Unemployment: Macroeconomic Performance and the Labour Market, Oxford: Oxford University Press. Meyer, B. D. (1990), ‘Unemployment Insurance and Unemployment Spells’, Econometrica, 58(4), pp.757-82. Narendranathan, W. and M. B. Stewart (1991), ‘Simple Methods for Testing for the Proportionality of Cause-Specific Hazards in Competing Risk Models’, Oxford Bulletin of Economics and Statistics, 53(3), pp.33 1-40, OECD (1993), ‘The Labour Market in Poland’, Paris: OECD. OECD (1996), Labour Market Policies in the Transition Countries: Lessons from their Experiences, Paris: OECD. Scarpetta, S. (1995), ‘Spatial Variations in Unemployment in Central and Eastern Europe: Underlying Reasons and Labour Market Policy Options’, In: Scarpetta S. and Worgotter A., eds., The Regional Dimension of Unemployment in Transition Countries, Paris: OECD-CCET. Terrell, K., M. Lubyova and M. Strapec (1996), ‘Evidence from Evaluation of Research in Selected Countries: The Slovak Republic’, in OECD (1996).

Labour market transitions and unemployment duration


Data appendix Figure 1. Hazard function, Warsaw. Inflows:1991-92







0.25 -


1 -

0.1 0.05

0 ’ 0














Figure 2. Hazard function, Ciechanov. Inflows:1991-92



0.12 0.1 -

w 2

0.08 0.06 0.04 0.02






-+- 1992 inflows -4-


Cazes and Scarpetta


Figure 3. Hazard function, Sofia. Inflows: 1991-92


0.3 0.2s


O.l 0 -
















-+-1992 inflows &1991 inflows Figure 4. Hazard function, Botevgrad. Inflows: 1991-92














14.9 7.7 22.2


13.6 5.9 19.8



15.5 5.7 21.3

75.9 12.2 155.2


85.2 14.5 171.3





U N ratio

733 23.0 234.2





1.2 0.5 0.8






0.9 0.6 1.4


0.8 1.5 0.6







Idlow rates

Source: OECD, Regional labour market database. ”Data for 1995 refer to the whole macro-region (Sofia vicinity) including Botevgrad. bats not available.

Warsaw Ciechanov
















Unemployment rate (a)

43 5.9 2.8





4.8 5.2 3.6





Oufflow rates

7.8 11.7 6.8





Table 1. Regional labour market disparities in Bulgaria and Poland, 1992-95 (annual averages of monthly data)

23 3.6 1.7





23 2.2 1.7

3.9 3.7 3.1


2.0 0.9

2.4 1.o

1995 0.9


Oufflow to job rates







1 a i% r



g z.=.3







Cazes and Scarpetta

Table 2a. Structure of quarterly inflows*: Warsaw (in per cent of total inflows) 1990'

Gender Female Male




57.8 42.2

54.2 45.8

53.8 46.2

52.6 47.4


28.3 61.9 9.8

21.3 69.5 9.2

23.9 62.4 13.6

27.1 60.6 12.3

Educational attainment Higher Secondary Vocational Primary

20.1 52.6 11.2 16.2

13.3 46.1 21.5 19.2

9.2 40.2 23.3 27.3

7.3 33.7 22.2 36.8

Status at registration Benefit recipient Non-benefit recipient

85.4 14.6

86.2 13.8

70.1 29.9

68.0 32.0

SectorP Agriculture Industry-construction Services

0.9 36.5 62.6

0.7 31.4 61.9

1.3 33.9 64.8

1.3 39.4 59.3

Mode of separationb Mutual agreement Individual layoff Group layoff Quit Others

26.5 19.3 43.0 9.3 1.9

21.2 29.4 32.4 11.6 5.4

24.7 35.1 23.6 7.9 8.7

29.9 33.2 11.5 13.9 11.6

Status of last employe# State sector Co-operatives Private sector

76.6 10.3 13.1

63.9 9.7 26.4

56.6 8.6 34.8

46.5 8.8 44.7

Age -25 years 25-49

Source: Administrative records of unemployment. *Last quarter of each year. aDatasince July 1990. bForthose who left a job.


Labour market transitions and unemployment duration

Table 2b. Structure of quarterly inflows*: Ciechanov (in per cent of total inflows) 1990




38.6 61.4

40.0 60.0

34.5 65.5

31.5 68.5


34.9 62.7 2.4

34.6 63.1 2.3

37.0 61.3 I .7

30.8 65.0 4.2

Educational aftainment Higher Secondary Vocational Primary

2.4 33.7 39.8 24.1

2.3 28.5 36.9 32.3

1.7 31.9 39.5 26.9

0 32.2 34.3 33.6

Status at registration Benefit recipient Non benefit recipient

100.0 0

97.7 2.3

84.0 16.0

61.5 38.5

Sectof Agriculture Industry-construction Services

2.4 50.0 47.6

6.5 57.9 35.5

9.2 52.0 38.8

14.5 44.4 41.0

Mode of separation" Mutual agreement Individual layoff Group layoff Quit Others

35.7 12.9 30.0 14.3 7.1

26.9 22.2 32.4 7.4 11.1

15.1 44.1 25.8 4.3 10.8

19.0 53.4 8.6 6.9 12.1

Status of last employef' State sector Co-operatives Private sector

66.7 16.7 16.7

64.5 11.2 24.3

51.0 11.2 37.8

53.4 11.0 35.6

Gender Female Male Age -25 years 25-49

Source: Administrative records of unemployment. *last quarter of each year. aFor those who left a job.

Cazes and Scarpetta


Table 2c. Structure of quarterly inflows*: Sofia (in per cent of total inj7ows)




60.2 39.8

65.9 34.1

65.3 34.7


31.4 61.9 6.8

31.1 62.9 6.1

28.8 66.1 5.1

Educational attainment Higher Secondary Primary

47.5 40.7 11.9

52.3 31.1 16.7

44.9 37.3 17.8

Situation at registration Benefit recipient Non benefit recipient

62.7 37.3

56.8 43.2

64.4 35.6

Status at registration Layoff School-leaver Other' Student

54.2 5.1 37.3 3.4

43.9 9.1 43.2 3.8

47.5 11.9 35.6 5.1

2.0 58.8 11.8 3.9

6.4 53.2 8.5 4.3

2.4 65.9 9.8 4.9

Gender Female Male Age -25 years 25-49

Sect02 Agriculture Industry-construction Trade Transport-communication Financial sector Health and social sector Public administration Qualification Blue collar White collar Source: Administrative records of unemployment, *last quarter of each year. aNon benefit recipient. $or those who left a job.




19.6 3.9

21.3 6.4

9.8 4.9

63.9 36.1

71.3 28.7

73.5 26.5


Labour market transitions and unemployment duration

Table 2d. Structure of quarterly inflows*: Botevgrad (in per cent of total inj7ow.s) 1991



47.6 52.4

53.9 46.1

62.8 37.2


42.7 50.0 7.3

24.7 69.7 5.6

34.5 58.4 7.1

Educational attainment Higher Secondary Primary and less

17.7 37.1 45.2

25.8 23.6 50.6

8.8 38.9 52.2

Situation at registration Benefit recipient Non benefit recipient

50.0 50.0

64.0 36.0

45.1 54.9

Layoff School-leaver Other' Student

45.2 3.2 50.0 1.6

59.6 3.4 36.0 1.1

38.9 0 54.9 6.2

Sect04 Agriculture Industry-construction Trade Transport-communications Health and social sector Public administration

23.2 53.7 1.2 1.2 19.5 1.2

47.1 26.5 2.9 23.5

19.3 55.4 1.2 1.2 22.9

Qualifiation Blue collar White collar

86.4 13.6

84.7 15.3

87.7 12.3

Gender Female Male Age -25 years 25-49

Status at registration

Source: Administrative records of unemployment. *last quarter of each year. aNon benefit recipient. %or those who left a job.


Cazes and Scarpetta

Table 3. Distribution of unemployed by duration, Poland 1990-92 (%) Duration

Inflows I990 5 3 months 4-6 months 7-9 months 10-12 months > 12 months

Inflows 1991 5 3 months 4-6 months 7-9 months 10-1 2 months k 12months

Ciechanov Whole Completed Whole Completec Iopulation spells female spells for popula- women tion

Warsaw Whole Completed Whole Completed iopulaspells female spells for tion populawomen tion

9.6 3.6 6.0 3.6

13.3 5.0 8.3 5.0

9.4 0.0 0.0 6.3

14.3 0.0 0.0 9.5

18.2 11.2 7.5 6.4

32.8 20.1 13.5 11.6

15.0 9.2 9.0 6.6

28.0 17.1 16.7 12.3









3.8 11.5 6.2 6.9

5.5 16.5 8.8 9.9

3.8 7.1 1.9 7.7

6.5 12.9 3.2 12.9

18.8 22.5 9.2 8.3

25.1 30.1 12.3 11.1

14.7 19.8 7.4 9.8

20.5 27.6 10.3 13.7









7.6 7.6 4.2 6.7

13.4 13.4 7.5 11.9

9.8 7.3 2.4 7.3

22.2 16.7 5.6 16.7

15.2 20.8 9.8 12.0

20.6 28.2 13.3 16.2

13.0 17.0 9.3 14.6

17.4 22.8 12.5 19.6









k Inflows I992 5 3 months 4-6 months 7-9 months 10-12 months > 12 months

Source: Administrative records of unemployment and authors’ calculations. Note: More details on the distribution of unemployment by duration for the youth and the unskilled are in Cazes and Scarpetta (1995).

Table 4. Distribution of unemployed by duration, Bulgaria, 1991-92 (%) Duration

Sofia Botevmad Whole Completed Whole Completed Whole Completed Whole Completed )opulation spells female spells for iopulaspells female spells for popula- women tion populawomen tion

Inflows 1991 5 3 months 4-6 months 7-9 months 10-12 months > 12 months

17.8 33.1 28.8 8.5

18.4 34.2 29.8 8.8

11.3 31.0 39.4 9.9



Infro ws I992 5 3 months 4-6 months 7-9 months 10-12 months > 12 months

26.5 22.7 26.5 12.9

28.7 24.6 28.7 13.9



9.7 17.7 29.0 12.9

10.9 20.0 32.7 14.5

5.1 18.6 25.4 16.9

5.7 20.8 28.3 18.9


11.4 31.4 40.0 10.0 7.1





19.5 21.8 29.9 14.9

21.5 24.1 32.9 16.5

11.2 14.6 30.3 12.4

13.9 18.1 37.5 15.3

12.5 10.4 33.3 12.5

15.4 12.8 41.O 15.4




13.8 5.1 31.5 -

Source: Administrative records of unemployment, authors’ calculations. Note: More details on the distribution of unemployment by duration for the youth and the unskilled are in Cazes and Scarpetta (1995).


Labour market transitions and unemployment duration

Table 5. Destination of oufflows by ‘completed’ duration, Poland 1990-93 Regular job

Subsidized emploment’





< 3 months





4-6 months 7-9 months







32.3 20

10-12 months


> 12 months





5 3 months





4-6 months





7-9 months





10-12 months





> 12 months






Source: Administrative records of unemployment, authors’ calculations.

‘Includes loans for enterprise creation, public works etc.

1991 1993 Log-likelihood N.obs.


Secondary educ. Vocational educ. No UB

-7458.438 1,976

-0.0510 0.0765 -0.0225 -1.5216 -0.1079

5.2325 5.5269 2.7523 1.6510 0.7184 0.9638 0.3273 6.8964 1.9961

-0.2810 0.3459 -0.2560 0.1583


Age 15-24 Age >50 Higher educ.


A: Single destination



4.4021 -552 1.420 2,049

-0.3303 0.4050 -03555 03136 0.0425 -0.0960 0.4515 -15233


Table 6a. All unemployed, Poland: piece-wise hazard estimations


5.3286 5.8499 3.3095 2.761I 0.5461 I . 0940 6.4689 8.3487


-886.292 213

-03410 -0.0862 0.3462 0.6174 0.2135 0.3745 -0.2256 -1.2854 0.3456

1-91 It-statl I.9028 0.4700 0.6440 0.9929 0.8895 I.7326 0.3084 I.2440 2.0042


0.6154 -588.478 262

-0.5394 0.0015 -0.621 1 2.8476 0.1887 0.0781 0.0132 0.7043



It-statl 2.1390 0.0073 0.8147 2.9803 0.7006 0.3235 0.0493 1.7437

n Q


fi Q




N h cb

-7554.126 1.945



0.2752 -0.4864


-03598 -0.3590 -0.2433 0.3326




0.0566 0.2657



-0.2167 0.5745

5.0780 3.7419 1.9618 2.7897 1.1387 2.1505 4.0776 0.9517 3.7366

2.1597 5.1678 0.9888 1.6577 0.4073 1.6366 3.5218 1.6094 10.9899

-6074.396 2.049




-03524 1.2748 0.8435 0.4322


3.0819 1.1720 2.2029 7.6566 6.1345 2.7333 5.9977 1.6646



4.8992 8.1890 2.4490 2.6663 4.2323 2.7766 12.5333 5.8957 6.3718



-0.4050 0.7203 -03557 -0.5051 -0.4354 -03018 1.0686 4.1919


-962.589 213

-13.7534 -13.6403 0.3287


-0.2320 -0.0356 0.2937 0.3194 0.4487

-0.1623 0.4226 0.8975 -0.0923 0.1806 0.4705 4.7300 0.3534



0.9786 0.1459 0.3865 0.3088 1.3743 1.7999 0.0159 0.0185 1.4225

1.7817 0.5857 0.5549 1.1239 0.2554 0.5779 0.6221 0.6818 1.3646



-636.003 262


-0.6332 1.1948


1.1228 0.8837 0.9016 3.1478 1.9418 0.4453 1.6564 2.4720

-0.3255 -0.2301 -0.8864 3.1899 0.6397


2.0022 1.2653 0.2183 0.0094 1.5997 0.3768 2.4107 0.1729 1.4931



0.4455 -0.2345 -9.6995 -0.9718 0.1415



Note: Statistically significant coefficients (at the 10 per cent level or lower) are in bold. T-statistics (in absolute terms) are in italics. The unemployment benefit replacementrate is entered in logarithm.


woman Age 15-24 Age >SO Higher educ. Secondary educ Vocational educ No UB Disabled 1991 1993 Employment woman Age 15-24 Age >SO Higher educ. Secondary educ. Vocational educ. No UB Disabled 1991 1993 log-likelihood


B: Multiple destinations


Table 6a (cont). All unemployed, Poland: piece-wise hazard estimations

Cazes and Scarpetta


Table 6b. All unemployed, Bulgaria: piece-wise hazard estimations Botevgrad 1991-93

A: Single destination Woman Age 15-24 Age >50 Higher educ. Secondary educ. No UB 1992 1993 Log-likelihood N.obs. B: Multiple destinations Inactivity Woman Age 15-24 Age >50 Higher educ. Secondary educ. No UB 1992 1993 Employment


-0.1553 0.6545 0.5773 0.3553

0.4736 0.0406 0.0287 -1.2394

-0.0899 0.4466 0.6741

0.3620 0.4613 0.1150

0.1270 -1.1082

-0.4460 -0.1661 -1.0643

Higher educ. Secondary educ. No UB

2.1859 1.6552



1.105 3.355 3.395 2.359 1.639 0.272 0.180 6.295

-776.737 326

woman Age 15-24 Age >50

1992 1993

Sofia 1991-93

-0.3260 -0.4030 -2.0542

-842.491 326


-0.0980 0.3593

0.0280 0.2016 -0.0793 0.5289

-0.0853 -1.8392

0.754 2.381 0.105 1.056 0.392 4.017 0.636 9.102

-843.819 368 0.588 2.729 2.291 1.628 2.528 1.711 0.727 5.314

1.225 0.422 0.007 3.917 2.975 0.815 0.999 3.214

-0.0257 0.3535

0.12398 0.1939 -0.0023 0.8238

-0.0975 -1.8353

-0.4843 0.2354 -0.3158 0.1182 -0.5855 -1.9007

-0.0991 -1.7571

0.I79 2.160 0.420 0.951 0.011 5.796 0.667 8.371 1.496 0.576 0.508 0.209 0.902 3.067 0.289 3.341

-936.391 368

Statistically significant coefficients (at the 10 per cent level or lower) are in bold. T-statistics (in absolute terms) are in italics.

0.8090 0.9581 0.9850 0.3941 15.0996 4.7265

-0.4865 0.0982 0.1798 0.0456 7.7345 0.8139

-2,020.7 507

0.0184 1.2596 0.4671 0.5480 0.0487 2.0927 1.9758

0.0019 0.1957 -0.0747 -0.1069 -0.0070 0.3255 -0.9303



0.6201 0.7899 0.4119 1.3874 0.6818

3.6845 1.4930 2.9091 4.4001 2.5685 0.64#1 1.3223



-3.106.9 1,157


4.5686 1.3878 0.0664 0.1229 6.0262

-0.3223 0.1729 -0.4178 0.6781 0.2813 0.0806 -0.77 19


0.7118 -383.466




0.4242 0.9406 0.8867 2.7314 0.8022

1.1351 -5.1643 0.4352 0.7049 -22.0999 0.7876 0.4696 0.4214 0.8374 7.1127 2.8369


2.3289 1.3459 0.3874 3.5219 2.1614 0.2042 1.1993

-0.7770 0.3432 0.4026 4.0921 0.7141 0.0636 1.2991

1992-93 0.5894 0.0647 0.1394 1.6766 0.7673 1so94 0.0033




0.486 0.115 -0.184 -0.247 11.782 0.893


-0.171 0.022 0.100 1.736 0.275 0.521


Note: Statistically significant coefficients (at the 10 per cent level or lower) are in bold. T-statistics (in absolute terms) are in italics. a There is insufficient variation in the sample to estimate this coefficient. The unemployment benefit replacement rate is entered in logarithm.

log-likelihood N.obs.

1991 1993

Cooperatives Private sector UB repl. rate




Secondary educ. Vocational educ.

Age 15-24 Age >SO Higher educ.


A :Single destination


Table 7. Experienced workers, Poland: piece-wise hazard estimations




Secondary educ.

Vocational educ. Disabled Agriculture Services Co-operatives Private sector UJ3 repl. rate 1991 1993 0.0686 0.2342 0.5977 1.4877 7.3852 2.2475


1.0102 1.7557 1.4765 2.0875 0.5795 1.6654

2.0369 2.7664 1.3908 1.0468 3.1614 1.5475 0.0241 1.2458 1.7826 0.1519 0.0106 1.7212 2.7605



-0.2877 0.4870 -0.3007 -0.3002 -05299 -0.2785 -1 3.9 198 -2.0173 5.9520 0.0368 0.0015 2.8609

Warsaw 1990-91


-22.3 120 11.5954 1.3535 0.8090 6.2284

-14.3054 -1 5423 0.0312

1.0851 2.1572


Ciechanov 1992-93

Note: Statistically significant coefficients (at the 10per cent level or lower) are in bold. T-statistics (in absolute terms) are in italics. The unemployment benefit replacement rate is entered in logarithm.

63810 0.5348

0.0710 0.0386 0.1731 0.2714





B :Multiple destinations Inactivity woman Age 15-24 Age >50 Higher educ.


Table 7 (cont). Experienced workers, Poland: piece-wise hazard estimations


0.0125 1.1046 1.4247 1.4761 1.1436

1.1686 2.0140 1.8054 0.oood 1.3821 0.0583


-2262.454 507

-0.1052 0.0354 4.4115 0.1672 0.0678 0.2665 -0.4047 -0.7060 0.1325 0.1843 -0.0941 8.2812 1.0463

1990-91 0.8092 0.1669 1.8021 0.7186 0.3628 z.3094 0.8308 0.9507 1.0100 0.7769 0.6269 12.8769 4.0823

3.1394 0.2459 2.6365 6.6897 5.6852 2.4171 0.1578 0.1440 0.2442 0.1449 1.8316 0.3257 0.2479

-03495 -0.0380 -05084 1.3279 0.8902 0.4377 -0.0928 -0.1693 -0.5170 0.03 15 0.2053 -3.4902

0.0474 -3452.5% 1,157


I 1 S 9 1


4.9244 -12.4766 0.1431 0.6282 -55357 0.7794 -412.173 Z 70

0.0046 3.6480 2.9602 0.2808

-15.4922 4.4276 1.1255 0.1074 a


1.7790 0.2441 2.1342 I . 6769

1.9631 0.1791

199293 -0.7016 0.0537


Note: Statistically significant coefficients (at the 10per cent level or lower) are in bold. T-statistics (in absolute terms) are in italics. a There is insufficient variation in the sample to estimate this coefficient. The unemployment benefit replacement rate is entered in logarithm.

Age 15-24 Age >SO Higher educ. Secondary educ. Vocational educ. Disabled Agriculture Services Co-operatives Private sector UI3 repl. rate 1991 1993. log-likelihood N.obs.,




Table 7 (cont). Experiencedworkers, Poland: piece-wise hazard estimations

5 a





z3 x s

& E













Suggest Documents