UNEMPLOYMENT DURATION, LAYOFFS AND COMPETING RISKS

Transcription

1 UNEMPLOYMENT DURATION, LAYOFFS AND COMPETING RISKS J.M. Arranz, C. García-Serrano and L. Toharia (Univeridad de Alcalá, Spain) Abtract The main objective of thi paper i to invetigate the determinant of unemployment duration in a competing rik framework with two detination tate: employment and inactivity. In particular, we analyze labour market trajectorie of job loer and worker without previou labour experience uing data from the Spanih Labour Force Survey (EPA) for the period We find that male worker, people aged 25-34, higher educated job loer in region with low unemployment rate and worker with previou employment experience exhibit higher exit rate from unemployment into a job. On the contrary, the ame variable preent a negative effect on the probability of exiting from unemployment to inactivity. We alo find that non-claimant of unemployment benefit exit from unemployment fater than claimant do. Finally, the exitence of a dead-end job effect i detected in the Spanih labour market during the JEL Claification: J64, J65 Keyword: unemployment duration, job lo, competing rik, unoberved heterogeneity, dead-end job The author thank the Comunidad Autónoma de Madrid for the financial upport (project no. 06/HSE/0346/2004). Correponding author: Univeridad de Alcalá, Departamento de Etadítica, Etructura Económica y OEI, Plaza de la Victoria 3, Alcalá de Henare (Madrid), Spain. Univeridad de Alcalá, Departamento de Fundamento de Economía e Hitoria Económica, Plaza de la Victoria 3, Alcalá de Henare (Madrid), Spain.

2 1. Introduction Since the 1980, the peritence of high unemployment ha been one of the mot relevant economic problem in mot OECD countrie, and epecially worrying in Spain 1. The invetigation of the phenomenon of unemployment peritence and of what circumtance may influence it i an important tep toward developing economic policie. In that context, the analyi of the future labour market poition of job loer and if they may remain in unemployment for a long period of time may provide additional inight on the knowledge of the cot of unemployment. The notion that unemployed individual may be permanently trapped in joblene i explored theoretically in Blanchard and Diamond (1994), Ljungqvit and Sargent (1998) and Ridder and van den Berg (2001). Mot tudie in the literature regarding the effect of a job lo on diplaced worker focu their attention on the iue of earning loe aociated with pell of job interruption: Farber (1997), Jacobon et al. (1993) and Steven (1997) for the US and Arulampalam (2001) and Gregory and Juke (2001) for the UK. Evidence for other European economie i much more pare (Ackum, 1991, for Sweden; Burda and Merten, 2001, for Germany; García-Pérez and Rebollo, 2006, and Arranz and García-Serrano, 2004, for Spain). Other tudie analyze the effect of unemployment incidence and/or duration on future unemployment pell: Arulampalam et al. (2000) and Gregg (2001), for the UK; Muhleien and Zimmermann (1994) for Germany; Roed et al. (1999) for Norway; Arranz and Muro (2004) for Spain; and Heckman and Borja (1980) and Omori (1997) for the US. Finally, reearch on the labour market poition of job loer and whether or not they face problem in leaving unemployment i already extenive in North America (Ruhm, 1991, and Jacobon et al., 1993) but limited in the European literature (Bender et al., 2002, for France and Germany; Addion and Portugal, 2003, for Portugal; Lubyova and Van Our, 1999, for Slovakia; and Cebrián et al., 1996, Bover et al., 2002, and Alba-Ramírez et al., 2006, for Spain) 2. The analyi of the future labour market poition of the unemployed in a competing rik framework with two detination tate (employment and inactivity), while not yet commonplace in the duration literature, i becoming more familiar (Meyer, 1990; Narendranathan and Stewart, 1993; Addion and Portugal, 2003). Thi will be the main objective of thi paper. For that purpoe, we will conider an unemployment duration model in which hazard rate from unemployment may reult into the rik of entering a job or becoming inactive 3. In our analyi, we will ditinguih between worker with previou labour experience (job loer) and without it (firt-job eeker), allowing for oberved and unoberved characteritic to affect the unemployment exit proce. Furthermore, the invetigation will alo focu on the 1 In 1991, the Spanih unemployment rate wa 16.3 percent, increaing during the harp crii of the early-1990 up to 24.2 percent in Since then, it ha hown a continuing decreaing trend. Nowaday, the unemployment rate i below 9 percent, till high when compared with the US (4.8 percent), Japan (4.1 percent) and the average OECD countrie (6.3 percent), although cloe to the average Euro zone (8.2 percent) (ee OECD, 2006). Thi rate i not trictly comparable to that of 1991, due to the change in the definition of unemployment introduced in 2001, in application of a new EC Regulation. The comparable figure would probably be 2-3 point higher (for more detail, ee Garrido and Toharia, 2004). 2 There are alo tudie that invetigate the influence of individual' previou unemployment experience on future job duration; ee, for intance, Booth et al. (1999) and Böheim and Taylor (2002) in the UK and García-Pérez (1997) and Arranz and García-Serrano (2004) in Spain. 3 Flinn and Heckman (1983) addreed the quetion of whether unemployment and out of labour force are ditinct tate. 1

3 poible diincentive effect of benefit on exit out of unemployment and the tranition from unemployment to either temporary or permanent job contract. The data come from the quarterly Spanih Labour Force Survey (EPA, Encueta de Población Activa) linked file. We elect thoe individual who have jut tarted an unemployment pell ( new entrant ) in each quarter of every year over the period 1992 (firt quarter)-2004 (fourth quarter). Thee individual are followed during the period they remain in the urvey (up to ix quarter), o that the duration of their unemployment pell range from one month to a maximum of 18 month. Obviouly, ome worker remain unemployed for the whole obervation window but other experience tranition into either employment or out of labour force 4. The paper i organized a follow. In ection 2, we preent the data, how the ample ha been extracted and the way the information on the unemployment pell ha been contructed. Section 3 outline the methodology employed in the empirical analyi: the dicrete time competing rik hazard rate model. Section 4 provide the etimation reult. Finally, ection 5 ummarize the main finding and offer ome concluion. 2. The data The data we ue in thi paper i taken from the Spanih Labour Force Survey (EPA). Thi a nationally repreentative urvey developed quarterly by the National Statitical Office (INE, Intituto Nacional de Etadítica). The period under analyi i 1992(1)-2004(4), the tarting and ending date being dictated by change in urvey deign after the firt quarter of 1992 and 2005, repectively. In each quarter, the EPA ample conit of approximately 60,000 houehold, or 200,000 individual. One ixth of them leave the ample each quarter. Thi allow following tranition out of unemployment for up to ix ubequent quarter. Hence, the individual i oberved over a time interval of up to ix quarter of length. Since our attention focue on the unemployed, we elect thoe individual who enter the unemployment tate in each quarter. The longitudinal nature of the urvey make it poible to check their labour market tatu in ubequent quarter, allowing for three excluive tate: unemployment (U), employment (E) and out of the labour force (OLF). Therefore, we are able to know whether an unemployed individual ha experienced no tranition from unemployment or he/he ha moved from U to E or from U to OLF between two ubequent quarter. The quetionnaire in each quarter provide u with individual demographic information (uch a gender, marital tatu, age, level of chooling), information on previou job match (the reaon for job lo, job occupation, intitutional ector, firm indutry affiliation) and other variable (tenure on current unemployment pell and whether the individual receive unemployment benefit). 4 In the Spanih literature, there i a previou work by Alba (1999) which analyze individual tranition from unemployment to either employment or inactivity. Our paper preent ome relevant difference with it. Firt, we ue the period in our analyi while he ued the period Second, we analyze tranition out of unemployment obtained by worker' labour market tatu in all ix conecutive quarter, while he ued only two conecutive quarter. Third, we ue time varying covariate, and oberved and unoberved heterogeneity in a competing rik duration model, while he ued a multinomial logit without time varying covariate and without unoberved heterogeneity term in the etimation. Finally, we conider all men and women who become unemployed at the time of the quarterly urvey, aged between year old, with and without previou labour experience, while he analyzed only men, aged 20-59, with previou work experience. 2

4 Data file have been linked thank to the exitence of individual identification number. The data on unemployment hitory i brought into peron-month format and then the variable neceary for the implementation of the etimation are contructed, in particular the pell identifier, the cenoring indicator and the hazard rate. The election conit of individual aged who become unemployed at the time of the quarterly urvey (in fact, thoe who have been unemployed for le than three month at the moment of the interview). We aume that unemployment pell that are not completed (i.e. the unemployed that have not moved from U either to E or to OLF) from the firt quarter up to the ixth quarter are right-cenored duration at that length percent of total pell are cenored unemployment pell, 58.6 percent end in a tranition from U to E and 27.7 percent in a tranition from U to OLF. After the ample election, the total inflow conit of 14,861 individual. Table A.1 in the Appendix provide decriptive tatitic for the entire ample and for individual with and without previou labour experience. Job loer repreent 81.3 percent of the total inflow into unemployment (65.6 percent lot their lat job due to the end of a temporary contract, 5.9 percent due to layoff, and 2.3 and 7.5 percent becaue of involuntary and voluntary reaon, repectively). Thi information i diplayed in Figure 1 for the whole time period under analyi. (FIGURE 1) The mean unemployment duration i around 7 month for job loer and 8 month for thoe without previou labour experience. We alo oberve a higher percentage (17.9 percent) of cenored unemployment pell for worker without previou job experience than for job loer (12.2 percent). The proportion of individual who exit from U to OLF i higher in the cae of thoe without previou job experience (48.9 percent) than in the cae of job loer (22.8 percent). On the contrary, the proportion of job loer who exit from U to E i higher than that for thoe without previou job experience: 64.5 percent againt 33.2 percent. The evolution of thee proportion may be viewed in Figure 2 and 3. (FIGURES 2 AND 3) Figure 4-6 how the proportion of unemployed worker who find a job, who exit to out of the labour force and who remain in unemployment by caue of exit from the lat job. In general, data from thee figure ugget that worker without previou employment experience are le prone to make tranition from U to E while thoe entering in unemployment due to the ending of a temporary contract are more likely to move to the employment tate. (FIGURES 4, 5 AND 6) Finally, figure 7-9 contain information on the urvival function of exiting from U to E and to OLF for the entire ample, for worker with and without previou job experience, and by caue of exit from the lat job, repectively. The duration of a pell of unemployment i defined a the period of time elaped between the firt quarter of inflow into unemployment and the lat quarter of outflow from unemployment. From Figure 7, we ee that there are more right cenored obervation in the exit from U to OLF (51.5 percent) than into E (26.7 percent). Therefore, the unemployed remain horter time in unemployment when they exit to E than to OLF. Ditinguihing job loer from the unemployed without previou job experience make it poible to detect a different behaviour in the individual: unemployment pell for job loer before 3

5 exiting to E (to OLF) are the hortet (the longet); on the contrary, the unemployment duration for individual without previou job experience i horter before exiting to OLF than to E. (FIGURE 7) Figure 8 how that individual without labour experience remain longer in unemployment before exiting to a job than the ret of worker, with a 48.2 percent of cenored obervation. Then, we have worker entering unemployment from layoff (with 35 percent of cenored obervation), due to voluntary reaon (32.4 percent) and due to involuntary reaon (29 percent). Finally, thoe who ended a contract exhibit the hortet unemployment duration (20.5 percent of cenored duration). (FIGURE 8) Finally, in analyzing the tranition from U to OLF, Figure 9 ugget that the unemployed without previou job experience remain the hortet in unemployment before exiting to OLF (with 37.2 percent of cenored obervation), followed by thoe who entered unemployment due to voluntary reaon (42.1 percent), due to the end of a contract (57 percent), and due to layoff and involuntary reaon (around 60 percent of cenored pell). 3. Econometric pecification 3.1. The econometric model (FIGURE 9) The model choen for the empirical analyi i a dicrete time competing rik hazard rate model. A dicrete time model i choen becaue the data i available in dicrete time interval (monthly data). A competing rik framework i elected ince we are able to ditinguih between two exit mode out of unemployment for each individual: employment and out of labour force. In the formulation of the model, we follow the terminology propoed by Allion (1982) and Jenkin (1995), extended by Steiner (2001), Lauer (2003), D' Addio and Roholm (2005) and Alba et al. (2006). In our cae, the baic idea of the hazard rate model i that one divide the unemployment duration into a dicrete and finite number of time interval and look whether the individual ha left or not the unemployment tate in each time interval. The hazard rate i aumed to be contant within time interval but i allowed to differ between them. In thi context, the hazard rate ( h ) that individual i leave her th pell of tate j (unemployment) to detination k (employment or out of labour force) in the interval I t, given that her pell lated in tate j until the beginning of interval I t, conditional on oberved characteritic x ijk (t) and unoberved characteritic v ij, i a follow: h ijk ( t xijk ( t), vijk ) = Pr( Tijk = t, dijk ) Tijk t, xijk ( t), vijk ) (1) ij i=1,,n; j,k=1,,k; t=1,, T ij ; S=1,,S i 4

6 Where dicrete number of interval and end in tate k, 0 otherwie (pell cenored or end in other tate than k). T ij i the time pent by individual i in the th pell of tate j, I t i a d ij equal 1 if the th pell of individual i in tate j Auming that all pell obervation -conditional on the explanatory variable and the unoberved factor- are independent and that cenoring i random, the ample likelihood function for the original tate j may be written a follow (ee Jenkin, 1995; Steiner, 2001; and Lauer, 2003) 5 : L j 1 yijk y k j ijk ( t) 1 hijk ( t). (2) k j N S T i ij = hijk i= 1 = 1 k j t = 1 Where y ijk i a binary variable that take value equal 1 if the th pell of individual i in tate j i cenored and 0 otherwie. In order to etimate empirically the likelihood function, it i neceary to aume further pecification choice. For the hazard rate we chooe the logitic pecification that, with multiple event, generate the multinomial logit model (Maddala, 1983). It allow for the three poible tate conidered: employment, out of labour force, and remaining unemployed (which i the reference tate category). For individual i, the tranition rate from tate j to k pecified a a multinomial logit can be written a (Steiner, 2001; Lauer, 2001): h ijk ( t x ijk exp( α ( t) + β x ( t) + v ) jk jk ijk ij ( t), vij ) = ' 1+ exp( α ( t) + β x ( t) + v ) l j jl ' jl ijl ij (3) In (3) x ijk i a vector of explanatory variable that may vary with time, β jk i a vector of parameter to be etimated; the term α jk tand for the o-called baeline hazard which repreent the pattern of duration dependence. The pecification of the baeline hazard i very important in duration model. A common but retrictive approach conit in pecifying a parametric form (for intance, gamma, weibull, lognormal, etc.). Neverthele, thi approach i very trong becaue the aumption over the form are difficult to jutify from an economic point of view and provoke a mipecification problem. To avoid that problem, we pecify a emi-parametric approach: a piecewie contant hazard, by pecifying monthly dummy variable for which coefficient for tranition from unemployment to employment can differ from thoe for tranition to inactivity. Thi method preent the advantage of being a flexible pattern of duration dependence auming that thi pattern may vary among the tate where the duration effect i found to be contant or the number of obervation i very mall. Finally, v ij account for unoberved heterogeneity characteritic in the model uch a motivation, ability, effort, etc. We aume that the unoberved heterogeneity effect i a pecific detination tate, time contant, and independent of the oberved characteritic 6. Unoberved heterogeneity i dicretely ditributed with unknown 5 We omit temporarily x ijk (t) and v ij to implify notation. 6 Thi i a tandard aumption in duration model (Jenkin, 1995; Steiner, 2001 D Addio and Roholm, 2005).If we relax the aumption and v i correlated with X, then the probability of exiting from 5

7 upport point. Thee point can be interpreted a latent individual' variable. Heckman and Singer (1984) decribed the non-parametric approach baed on the exitence of ome latent clae of individual. In thi approach, one aume that v ij may be divided in a limited number of ma point v mj with a given probability π 7. Then, the likelihood function for an individual may be obtained integrating the following conditional likelihood ditribution: L j N S T 1 yijk k j π hijk ( t xijk, vzj ) (4) k j Z = i ij yijk z hijk ( t xijk, vzj ) 1 z = 1 i= 1 = 1 k j t = 1 In the next ection we etimate thi likelihood function by maximum likelihood to know how peronal, job and labour market characteritic influence unemployment duration of pell that end either in employment or in inactivity Specification tet The main limitation of the multinomial (or competing rik) pecification i the property of the independence of irrelevant alternative (IIA) that ha to be fulfilled. In a etting with three alternative, the IIA mean that the ratio of the probabilitie of any two alternative doe not depend on the characteritic of a third alternative. The validity of the IIA aumption will be teted by mean of two tet: the Hauman (HM) tet (Hauman and Mc Fadden, 1984) and the Small-Hiao (SH) tet (Small and Hiao, 1985). Thee popular tet conit in partitioning the choice et of alternative into ubet and therefore comparing the coefficient (HM) or the likelihood function (SH) from the complete model and from the retricted model obtained by leaving out one or more alternative. We alo have Wald and LR tet to tet whether ome of the outcome categorie hould be combined (for intance, whether the parameter etimate differ ignificantly acro outcome categorie). Table 1 provide the reult of thee tet. The HM and SH tet how that the null hypothei i accepted; then, the IIA aumption would be fulfilled. Therefore, thee tet indicate that the data upport the multinomial logit pecification for each departure tate. Wald and LR tet examine whether ome tate can be pooled into a ingle tate, in which cae the pecification hould binomial rather than multinomial. A the reult how, the null hypothei that the coefficient of the two candidate for pooling are not ignificantly different i rejected for any pair of potential alternative. The rejection mean that unemployment, employment and OLF are ditinct tate. Then, the multinomial pecification eem to be appropriate, ince none of the categorie could be combined. (TABLE 1) Finally, in order to ee whether the number of ma point found a optimal i robut toward the pecification with unobervable in the tandard multilogit model which i implicit in the text, three alternative information criteria are ued: Akaike unemployment through employment or inactivity will be affected, and a tet for endogeneity will be required. 7 The number of ma point will be determined by the approach of Baker and Melino (2000). A comparion of information criteria i computed from the etimation reult of model with a different number of ma point: the number of ma point i increaed until the addition of a further ma point top improving the model. 6

8 information criterion (AIC), Hannan-Quinn criterion (HQ) and Bayeian info criterion (BIC) 8. Table 2 report the value of thee information criteria. The preferred model i that yielding the lowet IC value. A can be een, all information criteria lead to the ame concluion. For all the unemployed, for job loer and for the unemployed without previou labour experience, accounting for individual unoberved heterogeneity by ditinguihing two ma point doe not improve the fit of the model, which mean that the bet model hould not include any ma point. (TABLE 2) Alternatively, a imple likelihood ratio tet of a model with unoberved heterogeneity againt another without unoberved heterogeneity confirm the concluion that unoberved heterogeneity i not ignificant. The value of the likelihood ratio tet tatitic for the entire ample of a model with unoberved heterogeneity againt that without it i Thi value exceed the critical chi quare value of 5.99 for 2 d.f. at ignificance level of 5 percent and, therefore, the unoberved heterogeneity component hould not be included in the pecification of the model. The value of the likelihood ratio tet tatitic are and for the etimation regarding job loer and thoe unemployed without previou labour experience ample, repectively. Both value exceed the critical chi quare value (previouly mentioned); therefore, unoberved heterogeneity i alo not ignificant Empirical analyi 4.1. Explanatory variable In the etimation, we conider the following et of variable: - Socio-demographic variable: gender, age dummie, marital tatu, attained educational level and relation with the peron of reference in the houehold. - Previou job variable: previou employment experience, type of ector, caue of exit from the lat job, occupation, indutry affiliation, and lat job duration. - Job earch variable: previou poition in the labour market before job earch (working, engaged in education, dometic work, etc.), whether worker receive unemployment benefit, type of job he/he i looking for (only fulltime job, preferably full-time, preferably part-time, only part-time, or whatever type), and whether the unemployed would accept a job that implied to move to a different city, a change in occupation, le income given hi/her qualification, or a job category below what wa expected. - Other variable: the quarterly regional unemployment rate to control for labour demand condition, the quarterly GDP rate to capture the effect of the buine cycle, and four dummie to control for the quarter of exit and entry to 8 A decription of the AIC, BIC and HQI criteria are preented in Baker and Melino (2000) and Lauer (2003). 9 We have alo etimated a third order polynomial pecification for the baeline hazard function with two upport point for the unoberved heterogeneity (that wa ignificant) rather a piecewie-contant baeline hazard function. Although the likelihood ratio tet cannot be ued to differentiate betweeen both pecification becaue the model are not neted, we choe a piecewie contant baeline hazard function in thi paper ince the data we ue i dicrete and it provide a more flexible repreentation of the baeline hazard function. Neverthele, the reult of the third order polynomial for the baeline hazard function are very imilar to thoe preented in the text. 7

9 examine the exitence of a poible eaonal effect and the calendar time effect, repectively. Furthermore, a duration dependence pattern through a number of monthly dummie ued for the baeline hazard pecification i included for each exit tate Expected effect Turning to the effect of particular individual and job related characteritic on the exit rate of unemployment, we expect the following effect. If inactivity were a true detination tate for women when they are in active age, then men would exhibit a poitive effect on the rik of exiting from unemployment to employment. Elderly unemployed worker have kill that are more likely to become obolete, o they would experience more difficultie in adapting to a new job and will how le probability of exiting to employment and more probability of exiting to inactivity (Narendranathan and Nickell, 1985) 10. Single worker are more able to accept a job becaue they are more mobile. On the contrary, married unemployed may earch more intenively ince they have familiar reponibilitie and need more income. The effect of the educational level i ambiguou. On the one hand, the higher the worker educational level, the higher hi/her reervation wage and, hence, the lower the probability of accepting a job offer. On the other hand, employer may prefer thoe unemployed with higher educational level expecting a higher productivity, o the probability of exiting to a job may be higher. The prediction on the influence of benefit on the exit from unemployment i alo ambiguou, ince there are an incentive and a diincentive effect on the intenity of job earch by the unemployed (Atkinon and Micklewright, 1991). On the one hand, the tandard prediction i that the unemployed receiving benefit will exhibit a lower probability of tranition from unemployment to employment ince the incentive of unemployed worker to earch for a job i reduced. On the other hand, there i an incentive effect ariing from the fact that the level of benefit increae the reource devoted to earch and hence increae the probability of finding a job. Regarding the influence of previou labour market experience, it may be anticipated that thoe individual with previou experience may be conidered a wage earner and, therefore, they would how a higher probability of experiencing tranition from unemployment to employment than thoe without previou labour market experience. An additional inight come from the fact that individual become unemployed due to different reaon. In thi ene, we may expect that worker who have been laid off face more difficultie to find a job, if job-match pecific human capital ha been lot (Podgurky and Swaim, 1987; Ruhm, 1991). Moreover, tenure on the lat job may have a carring effect on the duration of the current unemployment pell (Kuhn, 2002). We expect GDP growth rate to have a poitive effect on the hazard rate of finding a job. Concerning the tate of local labour market demand, the unemployed who live in region with low regional unemployment rate will enjoy a higher probability of finding a job ince there would be le competition for exiting vacancie. Finally, regarding the influence of indutry affiliation, we expect that worker in ervice and building ector will face expanding employment opportunitie and higher chance to find a job. 10 Elderly unemployed worker normally accumulate more labour experience which generate a higher reervation wage. Thi provoke that they have le probability of getting a new job (Folmer and van Dijk, 1988). 8

10 4.3. Reult Table 3 report parallel reult from fitting a competing rik hazard model to analyze the determinant of exiting from unemployment through two detination tate: employment (E) and out of the labour force (OLF). The firt two column provide the determinant of exit from U through E or through OLF for the entire ample. The following two column offer the reult regarding the ame pecification of the model but for the unemployed with previou labour experience, and the lat two column for thoe worker without previou labour experience. (TABLE 3) The reference individual for the etimation without previou experience i a man, aged year old, non-married, without tudie, main peron of the houehold, who declare that he/he doe not know if would accept a job in another city, in another occupation, of le income and with a lower job category, and looking for a job of whatever type regarding working time. Supplement regreor are incorporated in the etimation for job loer. The reference here i that lat job wa in a white-collar highkilled occupation, in the private ector, and in indutry; in addition, he/he exited from the lat job by voluntary reaon and doe not receive unemployment benefit. We firt preent the reult for ocio-demographic variable; later, we hall report reult for previou job characteritic; finally, we will comment reult on job earch variable and will conduct ome extenion Socio-demographic variable We oberve difference in the regreion coefficient etimated acro detination tate. With repect to the exit from U to E, gender, marital tatu and age eem to matter for the two group of unemployed. Female job loer exhibit a ignificantly 11 percent lower probability of exiting from U to E than imilar men. Thi negative effect i larger (around 20 percent) for thoe women without previou labour experience. The familiar pro-upply effect of marital tatu i quite confirmed in the cae of job loer: being married increae the probability of tranition from U to E. Neverthele, thi effect i larger for married men. On the contrary, while women without previou labour experience how a lower rik of exiting from U to OLF than men (a 22 percent), women who are job loer and married are more likely to be le attached to the labour force than men. Thi can be related to family reaon. Regarding the effect of age, difference are very marked. While individual aged have more chance of getting an employment (for job loer and for worker without previou job experience a well), the oldet worker tend to be the mot diadvantaged ince they exhibit the lowet tranition rate from U to E (thee are even lower for thoe without previou labour experience). On the other hand, the age variable doe not affect ignificantly the rik of exiting from U to OLF for the unemployed without previou labour experience. But for the group of job loer the oldet and the younget are thoe who how the larget likelihood of exiting from U to OLF. Thi mean that younger job loer are more inclined to come back to tudy or to dometic or family care, while older job loer are more inclined to retire. Finally, attainted educational level affect tranition from U to OLF: all worker with any educational level different from having no tudie (ave for pot-compulory econdary) are le likely to drop out the labour force. However, the educational level i 9

11 not ignificant in explaining tranition from U to E ince it appear to have no influence on the earch proce. The reaon may be that the educational level egment the labour market and therefore worker earch for a job in their own labour egment. Then, the probability that a job i offered doe not depend on the educational level, and although the reervation wage i higher for the unemployed with more formal education, their probability of accepting a job offer i not lower ince their wage offer are alo higher. Gonzalo (2002) ha alo detected thi inignificant effect of educational level on tranition from U to E for Spain Previou job experience variable There are everal job experience variable that provide intereting inight into the way worker exit from U either into E or into OLF: whether or not the individual ha been employed previouly, the caue of exit from lat job, the type of ector, occupation and indutry affiliation. One of the mot intereting finding relate to the firt of thoe variable. In fact, having been employed previouly increae 37 percent the probability of returning to employment in the future after unemployment. Thi reult eem to indicate that worker who have been omehow attached to the labour market in the pat poe ome characteritic making them more prone to coming back to employment relative to thoe individual who have never been employed. On the contrary, previou job experience doe not play any role in the tranition from U to OLF. Comment on the ret of variable are only for job loer. Thoe worker who entered unemployed due to a layoff exhibit a 22 percent lower probability of exiting from U to E vi-à-vi the reference category (voluntary reaon a early retirement, quitting, etc.), while worker who entered unemployment following the termination of a fixed-term contract have noticeably a 11 percent higher probability of moving from one job to another with an intermediate unemployment period. Therefore, the latter are more employable but at the ame time they have more rik of experiencing unemployment. Böheim and Taylor (2002) in the UK and by Arranz and García-Serrano (2004) in Spain have alo found previouly thi effect. Not urpriingly, thoe who entered unemployed for voluntary reaon how a larger probability of exiting to OLF than the ret of individual. With regard to the indutry affiliation, job loer from the agriculture and building ector exhibit a higher hazard rate from U to E acro the period of tudy than worker from the indutry ector. Thi i an expected reult ince thoe are the ector with the highet proportion of temporary employment in the Spanih economy. Therefore, worker turnover i high, individual move from one job to another very frequently and job duration are hort. We alo find that job loer from the agriculture and the ervice ector how higher probability of dropping the labour force. Finally, hazard rate from U to E are larger for worker who previouly worked in blue-collar job and in the public ector Job earch variable The variable that capture the ituation of the unemployed previouly to begin the proce of earching for a job provide ome intereting reult. Compared to the unemployed who were working before they tarted looking for a job, job loer who were in dometic work or were in education are tigmatied becaue they have le probability of entering to a job and have more probability of dropping out of the labour 10

12 force. Again, thi reult tree the importance of being attached to the labour market to increae the employability of individual and to avoid definitive exit to inactivity. Second, we have contructed a et of variable a indicator of either preure to accept a job or labour force attachment. Thee are baed on four variable that capture the effect of willingne to accept a job in a different city, in a different occupation, commanding le wage for the given qualification or in a job category lower than expected (poible anwer to thee variable are ye, no, and don t know ). The firt dummy variable take value 1 if the worker indicate that he would be ready to accept a job in all the above-mentioned four dimenion. The econd dummy variable take value 1 if he i ready to accept a job in at leat one of thoe option. Finally, the third variable (reference category) take value 1 if he would not accept (or he doe not know) a job in neither of thee option. Reult indicate that job loer and worker without previou labour experience who report that they would accept a job under either all the referred condition or any of thoe condition have around percent le probability of finding a job. Moreover, they exhibit much le probability of leaving the labour force. Thee are the ame reult obtained by Alba-Ramírez (1999) uing two conecutive wave of the Spanih LFS. Another variable trying to capture worker chooine in the job earch proce i that relating to the type of job (working time) they are looking for. In thi cae, our finding ugget that job loer who declare that they are earching only for a full-time job exhibit more probability of exiting from U to E. At the ame time, job loer and thoe without previou labour experience who are willing to accept only a part-time job are more likely to drop out the labour force. The receipt of unemployment benefit 11 ha the tandard expected effect on the exit from U into E and OLF: non-claimant exit from unemployment fater than claimant do. The effect on the exit from U into E agree with that found previouly by Alba-Ramírez (1999), Bover et al. (2002) and Gonzalo (2002) with the Spanih LFS data. Reult alo ugget that the benefit diincentive effect tranlate into a lower probability of exiting to inactivity: tranition rate from U to OLF are two time more important than from U to E. Thi effect wa previouly documented by Wadworth (1991) for the UK Other variable There i a clear indication that the higher the quarterly regional unemployment rate the lower the probability of exiting from U to E and the higher the probability of leaving the labour force for the two group analyzed, being both effect lower in abolute value for job loer. Thi reult ugget that there i a dicouragement effect from participation when labour market i lack. Similarly, the buine cycle variable indicate that the higher the quarterly GDP rate, the higher the probability of finding a job for job loer. Thi mean that in expanion the hazard rate from unemployment increae becaue firm create new vacancie and offer better wage. However, thee opportunitie are only available to job loer: the buine cycle doe not alter the tranition from unemployment of worker without previou experience. Year dummie variable alo allow oberving the effect of buine cycle on tranition from U to E and OLF. Concerning the tranition from U to E, the coefficient for year dummie are poitive but only ignificant for job loer, increaing from Thi variable contain information on individual who receive either unemployment inurance or unemployment aitance benefit. 11

13 to The probability i epecially large in the period A for the probability of leaving the labour force, it increaed from 1998 to Thi reult i worrying ince it ha happened in a period of economic expanion and net employment creation. Finally, Figure 10 diplay the etimated duration dependence for job loer and unemployed without previou work experience at mean of covariate. A can be oberved, etimated hazard rate from U to E are greater than from U to OLF for the entire ample. Furthermore, it can be oberved how in a competing rik framework the U-E and U-OLF etimated hazard rate teadily exhibit poitive duration dependence during the firt month up to the fourth month, negative duration dependence until the tenth month, and a poitive duration dependence thereafter. Notice that the mobility pattern of both collective for each outcome i imilar. The etimated hazard rate from U to E for job loer keep over the hazard rate from unemployment to OLF and from U to E and OLF for the unemployed without previou work experience. Then, expected unemployment duration for job loer who exit to a job i horter than for the ret of outcome. On the contrary, the unemployed without previou work experience exhibit horter unemployment duration when they exit from U to OLF than job loer for the ame outcome Further extenion Previou job tenure (FIGURE 10) Table 4 report an extenion to the reult offered in Table 4. In thi new etimation, we examine the effect of previou job tenure on tranition out of unemployment. The purpoe i to try and analyze whether duration in the lat job influence (poitively or negatively) the probability of exiting from unemployment. For that, we ue the whole et of variable included in the previou etimation except the caue of exit from lat job and the unemployment benefit variable ince there i ome correlation between previou job tenure and thee variable. A it i well known, previou job tenure and the reaon for leaving lat job provide indirect information on unemployment benefit: whether the individual ha the right to receive benefit and it duration. (TABLE 4) The reult ugget that the influence of previou job tenure on the probability of leaving the labour force i non-exitent, but the longer the duration in the previou job, the lower the hazard rate from U into E. Thi effect might be related to the lo of pecific human capital when worker are eparated from their previou job: if firm and worker are engaged in long-term relationhip, the rupture of a job match make the pecific component of human capital to vanih and individual may find more difficult to earch for a new job uitable for their kill. In thi ene, there i a rik of human capital deterioration and qualification get obolete with time, which may reduce productivity and, a a reult, re-employment probabilitie. Caue of exit from previou job: end of contract and layoff The ditinction acro individual according to the reaon of exit from the previou job i potentially relevant ince it may illuminate the effect of pecific human 12

14 capital on re-employment probabilitie. In order to invetigate thi, Table 5 diplay the reult for the etimation of tranition from U to E and OLF for worker who were previouly employed but were eparated from their lat job due to either the end of a contract or a layoff. (TABLE 5) Regarding thoe worker who became unemployed due to contract ending, their reult are imilar to thoe mentioned for the entire ample and for job loer (Table 3), ave for the benefit and job category variable that do not influence tranition from U to E. Hence, the unemployed with le probability of finding a job are women (pecially married), the elderly worker, worker in the private ector, in indutry ector, in region with the highet regional unemployment rate, who became unemployed in a receion, who were in dometic work previouly to tart looking for a job, ready to accept a job in another city and/or occupation and/or of le income and/or with lower job category, and thoe with large previou job duration. We alo find that the coefficient for year dummie are poitive, uggeting that the probability of tranition from U to E increaed from 1994 to at leat 2000 (not hown). On the contrary, married women, older and younger job loer, blue collar worker, worker in agriculture and ervice ector, in region with high unemployment rate, who were in education or in dometic work previouly to tart earching for a job, who do not receive benefit, do not accept a job in another city and/or occupation and/or of le income and/or with lower job category, are more likely to be le attached to the labour force. The coefficient for year dummie are poitive after 1997, but only ignificant from 2000 to 2003 (not hown). If we focu our attention on tranition of worker who were laid-off from their previou job, we find that married women, worker aged more than 35 year old, worker from the private ector, white collar high killed worker, thoe working in region with the highet regional unemployment rate and with longer previou job tenure are more prone to experience lower tranition from U to E. In contrat, married women, individual without tudie, who worked in the public ector, with longer previou job tenure, who are not ready to accept a job in another city and/or occupation and/or of le income and/or with lower job category are the group of worker with more probability of leaving the labour force. Thee tranition have increaed teadily from 2000 to 2003 (not hown). Stepping-tone veru dead-end job for the unemployed Temporary contract are often regarded a an important component of labour market flexibility in Spain in the lat twenty year. There i an extenive debate on the effect of temporary employment on the economy and on welfare, in general, and on the labour hitory of worker, in particular. On the one hand, ome author (Booth et al., 2002) argue that the exitence of temporary job help to currently unemployed worker ince they provide them with opportunitie to gain work experience and acquire human capital and act a a poitive ignal (tepping-tone) toward a permanent and more deirable job. On the other hand, other author conider temporary job a dead-end job ince they are undeirable job compared to permanent one: wore labour condition (lower wage, greater labour ecurity and bad time chedule) and le workrelated training, which implie that worker are more likely to have lower motivation at work. Hence, worker trapped in a chain of temporary job and unemployment have more rik of developing wore labour career and of excluion from the labour market. 13

15 In the literature, we may find tudie analyzing empirically the effect of temporary job on worker labour career. There i not a clear concluion ince reult differ acro countrie. Thu, work by Hagen (2003) for Germany, Zijl et al. (2004) for the Netherland, Glagiarducci (2005) for Italy and Engelland and Riphahn (2005) for Switzerland find that temporary job have a poitive ignal (tepping-tone) toward a permanent job. On the contrary, Amuedo (2000) for Spain find that temporary job have a negative ignal on the probability of getting a permanent job, and Booth et al. (2002) for UK, Güell and Petrongolo (2003) for Spain, and D'Addio and Roholm (2005) for the UE find evidence for both theorie. In thi ection, we analyze the extent to which temporary employment make it eay for unemployed worker to move from unemployment either to a permanent work (tepping-tone) or to other temporary job (dead-end). Table 6 report determinant of tranition for the entire ample from unemployment directly to a permanent job, from unemployment to a temporary job, from unemployment to other type of job (elf-employment) or to inactivity. We firt focu our attention on the etimation for the exit from unemployment either to a temporary or to a permanent job. Our finding indicate that the unemployed who were eparated from a previou job match due to the ending of a temporary job and experience an intermediate unemployment period how a 22.6 percent higher probability of exiting to a temporary job. By contrat, they have an 18 percent le probability of finding a temporary job if the prior job match finihed due to a layoff. Thee reult ugget the preence of a dead-end effect ince worker who accepted a temporary job in the pat are trongly attached to a temporary job in the future. Thu, a we mentioned previouly, they are trapped in a chain of temporary job and unemployment with more rik of developing wore labour career and of excluion from the labour market. Furthermore, we alo find that the unemployed whoe previou job termination wa due to the end of a contract (temporary job) exhibit a 30 percent lower probability of acceing to a permanent job. (TABLE 6) Given the preence of a dead-end effect, the comparion of the parameter of the tranition from unemployment to permanent job and from unemployment to temporary job i informative of the relevance of thi effect for obtaining temporary job. Older unemployed have le probability of moving from unemployment to permanent or temporary job. Male have higher probability of finding temporary job. Being married (and being female) ha a trong negative effect on the tranition from unemployment to temporary job. Having worked in the public ector increae the probability of exiting to temporary job and reduce the probability of acceing to permanent job. Thi evidence ugget the exitence of a temporary jobunemployment-temporary job channel in the public ector that reflect it increaing ue of temporary job 12. The tranition rate into temporary job are alo higher in the agriculture and the building ector compared to worker from the indutry. And finally, worker in blue-collar (high and low killed) occupation and thoe with previou job experience are more likely to exit to temporary job. 12 Thi evidence i not a urprie ince Dolado et al. (2002) highlighted it previouly: the public ector exhibit a teadily increae of the hare of temporary employment ince the mid The reaon may be a change in the hiring behaviour after the Growth and Stability Pact and that a high proportion of the EC Structural Fund received by the Local Adminitration for promoting active labour maket policie have been ued to hire worker in targeted group under temporary contract. 14

16 Other informative reult come from the regional unemployment rate and the buine cycle variable. The unemployed in region with high unemployment rate (aociated with higher job detruction and lower job creation rate) preent le probability of exiting to either temporary or permanent job. A for the buine cycle variable, we obtain the expected reult: in eaon when the quarterly GDP rate are higher, the tranition from unemployment to temporary job increae. 5. Concluion In thi paper, we have invetigated the tranition from unemployment to employment and inactivity uing a competing rik framework. In particular, we have ditinguihed between job loer and worker without previou job experience, allowing for oberved and unoberved characteritic to affect the unemployment exit proce. The unoberved heterogeneity wa not ignificant in our etimation. The poible diincentive effect of benefit on exit out of unemployment and the tranition from unemployment to temporary or permanent contract are other iue that have been analyzed. Data ued have come from the quarterly Spanih Labour Force Survey (EPA) linked file over the period. In what follow, we ummarize the main finding. Firt of all, it i worth noting the notion that employment and inactivity are two very ditinct behavioural tate. For it part, the role of previou labour market experience and the type of exit from the lat job i critical. We have found that previou labour market experience enhance the probability of returning into employment after unemployment in the future but doe not eem to influence the leaving from the labour force. Among job loer, thoe who entered unemployed due to layoff exhibit a lower probability of tranition from U to E, while worker who entered unemployment following the termination of a fixed-term contract have noticeably higher probability of tranition between job with an intermediate unemployment period. Worker unemployed due to a voluntary exit from a job are more likely to exit to OLF than the ret. Secondly, ome peronal, job and firm characteritic appear to have an effect on the tranition of the unemployed. The influence of eniority on the lat job held by the individual eem to be a relevant factor: the higher the length of ervice in previou job, the lower the hazard rate from U into E. Moreover, non-claimant exit from unemployment (either to E or to OLF) fater than claimant do. Thi finding ugget that benefit not only exert a negative effect in lowing tranition from unemployment to employment but alo that the benefit diincentive effect tranlate into a lower probability of exiting to inactivity. And male worker, thoe aged 25-34, and individual living in region with low unemployment rate (either with and without previou labour experience) are thoe more prone to experience exit from U to E. On the contrary, women, the younget and the oldet job loer exhibit more probability of leaving the labour force. Finally, the ditinction of different detination tate within employment (either fixed-term or permanent job) make it poible to detect the preence of a dead-end job effect in the Spanih labour market during the 1990, ince worker who accepted a temporary job in the pat were trongly attached to temporary employment in the future. They are omehow trapped in a chain of temporary job and unemployment with more rik of developing wore labour career and of excluion from the labour market. Thi iue deerve more attention in order to deign adequate policie aimed at preventing unemployment and reducing the proportion of temporary employment. 15