ARE SPANISH OPEN-ENDED CONTRACTS PERMANENT? DURATION AND TRAJECTORY ANALYSES *

Transcription

1 ARE SPANISH OPEN-ENDED CONTRACTS PERMANENT? SUMMARY: DURATION AND TRAJECTORY ANALYSES * Inmaculada Cebrián, Gloria Moreno and Luis Toharia (Universidad de Alcalá) VI JORNADAS ECONOMÍA LABORAL Alicante, de Julio de 2005 One of the assumptions generally accepted not only by social partners but also by labour market analysts in Spain is that open ended contracts can be considered as infinite or permanent. One of the foundations upon which the reform introduced in Spain in 1997, after an agreement between the social partners, was precisely the idea that fostering the use of open-ended contracts (by creating a new contract with lower firing costs and reductions in social charges) would reduce the proportion of workers under temporary contracts, by far the highest in the European Union. The evolution of this proportion, however, has shown a significant resistence to decrease. This paper deals with one of the reasons for this: the very fact that open-ended contracts are not as permanent as usually assumed and the lower stability of the new contracts introduced by the 1997 reform as compared both to the ordinary (more costly) ones and also the conversions from temporary into open-ended contracts (also eligible for the new contract). Using administrative data on open-ended contracts, a duration analysis, both through non-parametric survival Kaplan-Meier estimates and through a semi-parametric Cox model, is carried out. We complete our analysis with a trajectory study, which goes beyond the end of the first constract. The main conclusion is that, contrary to what is generally assumed, firing costs do not appear to be the main element in the determination of the proportion of employees with a temporary contract. Draft, april 2005 Corresponding address: Inmaculada Cebrián Departamento de Fundamentos de Economía e Historia Económica Universidad de Alcalá Plaza de la Victoria Alcalá de Henares (Madrid) Spain Telephone: (secr: 4202) Fax: inmaculada.cebrian@uah.es * This paper is based on a wider study on temporary contracts in Spain, funded by the Ministry of Labour and the Employment Public Services (INEM), in which other researchers have participated. Our views are clearly influenced by our discussions within that project. We thus gratefully acknowledge the inputs of Cecilia Albert, Javier Calvo, Jesús Cruz, Carlos García Serrano, Miguel A. Malo and Virginia Hernanz. They are not responsible, however, for the precise wording of the arguments in this paper and the possible errors therein. Financial help from Spanish Ministry of Education under CICYT project Obmetra, ref. SEC is gratefully acknowledged. Thanks are also due to INEM for providing the data which this paper uses.

2 1 Introduction The 1997 labour market reform, passed in the wake of the Acuerdo Interconfederal para la Estabilidad del Empleo (AIEE) signed by the social partners, was an effort to promote the use of open-ended contracts, with the aim of reducing the persistently high proportion of employees with a fixed-term (or temporary) contract. Its basic elements were the creation of a new so-called employment promotion openended contract, carrying lower firing costs for unfair cases than those applying to ordinary open-ended contracts, which could be used to hire certain groups of workers (basically people under 30 and over 45, but also long-term unemployed of any age) and also to transform former temporary workers into open-ended ones. The reform was fueled with substantial reductions in social charges to be paid by employers for a series of groups closely related to those that could benefit from the new employmentpromotion open-ended contract, though not exactly coincident. This reform, passed in principle for a 4-year period, was turned into permanent in Evaluations of this reform have tended to emphasise their positive effects (see e.g. Dolado et al, 2002) mostly linked to the reduction in firing costs which it implied for the newly created permanent contracts. The general impression, however, is that the reform did not succeed in reducing the proportion of temporary workers: between 1997 and 2003, this rate has dropped from 33.6 to 31.1, following a slow downward trend which started in 1995 and without any break in the series. While noticeable, more so when considering the private sector only (from 41 to 32.5 percent in the same 6-year period), this decrease cannot be considered high in terms of the expectations of the signatories of the 1997 agreement and in terms of the increase in the late 1980s and early 1990s. This paper aims at providing further empirical evidence on the effects of the reform and, especially, on the role played by the reduction of firing costs as a factor underlying the extensive use of temporary contracts in the Spanish economy. We use information on workers who were given an open-ended contract in 1998 and follow their labour market trajectories in two aspects: number and type of contracts received (up until mid 2001) and unemployment benefits received (if at all). We compare these trajectories for those whose initial contract in 1998 was ordinary open-ended, 1

3 employment-promotion open-ended (subject to a lower firing cost) and also open-ended from conversion of a previous temporary contract (also subject to lower firing costs). The standard idea about firing costs and temporary work is that the main reason for a high use of temporary work is that firing costs impose too high a barrier to employers willing to hire workers. Lowering such barrier would thus provide a higher incentive to offer open-ended contracts. One should thus expect more open-ended contracts being signed after the 1997 Reform. In the limit, of course, if no firing costs existed, the distinction between open-ended and temporary would be meaningless: all workers would be temporary. However, this standard argument implicitly assumes that all open-ended contracts are similar; as a matter of fact, it is usually implied that all of these contracts are permanent. In this paper, we present evidence that this is not so. Moreover, we find that firing costs are not clearly related to labour market stability. The 1997 created three categories of open-ended contracts the older, ordinary ones, carrying the usual firing costs, and two others with lower firing costs: those offered after having being hired as temporary workers, known as conversion contracts; and those offered on a general basis to certain groups of workers (basically younger and older workers, but also long-term unemployed of any age, so that age in itself does not define this thrid category), known as employment promotion contracts. Using standard duration analysis, we find that conversion contracts are those showing longer durations, while employment promotion contracts are those with the shortest ones, ordinary contracts being in-between. These differences appear in descriptive non-parametric Kaplan-Meier survival analyses and also after controlling for the characteristics of the workers and the jobs in which they are hired. As a complement to this duration analysis, we also undertake a study of the full trajectories of workers in terms of the contracts they have signed until the present and the reception of unemployment benefits. All of these results suggest that firing costs are not the only element at play in the decision to offer workers an open-ended or temporary contract. The structure of the paper is as follows. Section 2 presents background information on the evolution of the proportion of temporary workers in Spain since 1987, as well as a discussion of the relationship between the stock of open-ended and temporary contracts, which jointly determine the proportion of workers under temporary contracts, and the flows of contracts on which we base our analysis. Section 3 discusses 2

4 the data used in the analysis. Section 4 presents the duration study. Section 5 deals presents the trajectories and the factors associated to each of them. Section 6 concludes. 2 The recent evolution of temporary and permanent contracts in Spain 2.1 The evolution of the proportion of temporary employees Figure 1 presents the evolution of the proportion of temporary employees in Spain between 1987 and 2003, by gender. Three main periods may be distinguished: first, a rapid increase which took place until 1992; secondly, the period between 1992 and 1995 when, after a receding period coinciding with the employment crisis, the rate of temporary work recovered its upward trend, peaking, especially for males, in 1995; finally, the slow but continuous decline after In the case of females, from its 1992 peak of 39%, the proportion of temporary employees fell to 37 and then recovered to 38% by 1995, it then fell to 34% by 1998 and has since remained more or less at that level, with yet again a small drop in As for males, from its peak of 33-34% in 1995, it very slowly fell to reach 32% by 1999, accelerating thereafter its trend to reach under 29% by Similar results are found when considering only the private sector of the economy. In the case of females, the proportion of temporary employees peaked at 46% in 1992, remained at that level until 1995 and then dropped steadily until the most recent figure (2003) of 35%. For males, the maximum was reached in 1995, at 38% and the decrease, which also started at that time, accelerated a bit after 1997, to reach 31% by

5 Figure 1. The evolution of the proportion of employees with temporary contracts, by gender, 1987-II IV (Source: INE, Labour Force Survey) Percentage of employees Females All Males (points in the graph correspond to the second quarter of each year) These trends appear to be quite independent of any of the various reforms introduced by the different governments since 1992, aimed at restoring the so-called causality principle by which temporary contracts can only be given to fill temporary jobs. This principle, common in various legal systems (most notably in Italy and France), was suspended in Spain in 1984 and was progressively recovered throughout the 1990s. Among these reforms, as already mentioned, the one passed in 1997 was especially significant, as it affected for the first time the firing costs of permanent workers. One can say that by 2001 the full circle had been rounded, and that legally, at least, this principle now holds in Spain. Yet the proportion of employees who at any given moment in time have a temporary contract remains much higher than elsewhere in Europe, including the closest (legally speaking) countries of Italy and France. 4

6 2.2 Open-ended workers: stocks and flows The evolution of the proportion of temporary workers depends on the evolution of the number of temporary contracts signed as well as of the number of open-ended contracts signed. As a matter of fact, the rate of variation of the percentage is approximately equal to the difference between the rate of change of the number of temporary workers and the rate of change of open-ended workers (times the proportion of open-ended workers). In this paper, we concentrate on the rate of change of the number of open-ended workers. The absolute change in this stock is equal to the difference between inflows into the open-ended status and outflows from such a status. Inflows in any given year are equal to the number of new open-ended contracts signed in that period. As for outflows, there are various elements: one is the natural wastage of the stock of openended contracts, mostly due to retirements; a second one is the number of dismissals of workers with an open-ended status at the beginning of the year, for whatever reason; a final one is the attrition of new open-ended workers (due to dismissals or any other separations). While inflows can be estimated from administrative data, as all contracts have to be registered with the Public Employment Services, outflows are more difficult to estimate. Our approach is not, therefore, to undertake a precise stock-flow accounting, as this is an almost impossible task. Rather, what we want is to estimate the number of outflows which would make consistent the evolution of the stock of openended workers and the flow of new open-ended hires. Table 1 presents such calculations. As can be seen, the total stock of open-ended workers started to grow in Surprisingly, the inflows did not follow the same trend, which means that estimated outflows decreased 1. The most interesting figures in Table 1, however, are thos referring to the post-reform period. Starting in 1997, the number of contracts sharply increased, but the change in the stock did not follow, thus implying an ever increasing estimated outflow. This is the fact we want to concentrate on and provide an explanation for. While the initial increase of outflows in 1997 might be attributed to earlier underreporting, the increase in later years cannot. 1 Of course, this does not make sense. One possible explanation is that the number of open-ended contracts is underestimated. Indeed, registration is compulsory but the penalty for not registering the contract is either not benefitting from any subsidy or advantage, or considering the contract open-ended. For regular open-ended contracts, thus, the incentive to register the contracts may be small. 5

7 Table 1. Open-ended workers: the necessary consistency between stocks and flows, (Source: Labour Force Survey and INEM and Ministry of Labour, Boletín de estadísticas laborales, various issues) Time(t) Stock (first quarter of t) Change of stock between t-1 and t (1 st qtr) Inflows (contracts signed in t) Estimated outflows in t Notes: - all figures are in thousands. - estimated outflows are calculated as the numbers which would make consistent the evolution of stocks and inflows Our argument thus centers on the rate of exit from open-ended status of the newly hired workers. What we want to see is whether the open-ended contracts signed after 1997 first are less stable than those previously signed, and also whether different open-ended workers behave differently in terms of leaving their permanent status. To complete the information provided in Table 1, Table 2 presents the evolution of the number of contracts in Spain since Here information is provided also on temporary contracts, and also on the distinction between the three types of open-ended contracts mentioned before. Although employment-promotion contracts existed since 1989, it is only after May 1997 when they carry lower firing costs and smaller social charges. 6

8 Table 2. Evolution of contracts registered with the PES, by type, (yearly totals) (Source: INEM and Ministry of Labour, Boletín de estadísticas laborales, various issues) TOTAL OPEN-ENDED CONTRACTS PART- EMPL.PROMOTION TEMPORARY (open-ended + TOTAL ORDINARY TIME temporary) Initial Conversion Note: starting in 2002, part-time contracts are no longer a separate category; rather, a part-time contract will thereafter be assigned to the specific contract type it belongs, thus introducing a long due correction in the way the contract statistics were collected. This explains why the three categories of open-ended contract increase their numbers in 2002 while the total decreases. It can be observed in Table 2 that the number of open-ended contracts registered with the PES doubled between 1996 and 1997, a trend which followed in the next two years. Since 1999, the number of open-ended contracts has remained stable at a level four times as high as it was in The number of temporary contracts also increased over the same period, although more slowly and also following a trend which had started in 1994; it has also remained more stable since As a consequence, the share in total contracts of open-ended contracts increased from around 4-5 percent in the early 1990s to around 9 percent after

9 3 Data We use microdata provided by the Public Employment Services on contracts registered between 1996 and More specifically, we were given all of the contracts signed by all persons who received at least one open-ended contract in 1996 through For each contract, information is provided on the date of hiring and several personal and job characteristics (such as age, gender, level of education, province, industry and occupation). From a separate file, we also have information on unemployment benefits which can be linked to the contract information, so that, among other things, we can identify who has received unemployment benefits and when such benefits started. We consider only the first contract received by a person in a given year. From the information provided, we estimate the duration of the contract (in number of days) as follows: - if the contract is followed by another contract, the duration is the difference between the two starting dates; the assumption is that this corresponds to a change without an unemployment break or with a such a short break that unemployment benefits are not received; - if the contract is followed by a spell of unemployment benefit, the duration is the distance between the starting date of benefit and the starting date of the contract; - if the person has no other contract, the contract is assumed to last until observation time (end of may 2003); obviously, these are right-censored observations. One problem of the way in which duration is computed is that we may interpret that a person is still at work while s/he may have dropped from employment altogether. The most likely situation for this is retirement. To try to avoid this problem, we have excluded people older than 55 from our analysis. Voluntary quits leading to non-labour market situations are assumed to be negligible. It is possible that we observe more than one spell per person, although this is unlikely. The way in which the database is constructed, we have been working by years, so in principle one could observe one person hired with an open-ended contract say in 1996 ending in a few months, and that same person could be hired again in 1997 (or 8

10 1998) with another open-ended contracts. However, since all our estimates will be broken by years, it is very unlikely (albeit not impossible) that we observe two spells of the same person within the same year. Since the 1997 reform establishes age conditions for some of the contracts, we have subdivided the sample in three age brackets: people under 30, people aged 30 to 44, and people aged Table 3 presents the breakdown of cases analyzed by age brackets and contract type for the different years studied. Data for 1996 actually correspond to inflows up until may 1997, when the reform came into force. Data for 1997 only correspond to contrracts signed after the reform was implemented. For 1996, only ordinary contracts are considered, the idea being that the employment promotion contracts entered at that time corresponded to a different policy (in any case, they were very few relative to the numbers in the following years, as seen in Table 2). Table 3. Distribution of the sample of contracts included in the analysis by contract type (Source: contract database, INEM) Horizontal percentages Number of Vertical Employment Ordinary Conversions Total cases percentages promotion (until 16th May) Less than Total (17th May-31st December) Less than Total Less than Total Less than Total Note: For , only ordinary contracts are considered; see text. 9

11 Conversions represented more than half of all open-ended contracts in 1997, the share being even higher for middle-aged people, and clearly lower for older workers. In 1998 and 1999, its share decreased for all ages. For young people, employment promotion contracts are the most widely used in 1998 and A non-negligible 12 percent of workers under 30 did receive ordinary contracts, although they were eligible for the employment promotion one. As for middle-aged workers, the share of employment promotion contracts is significantly smaller; this is normal as only longterm unemployed of this age bracket are eligible for such contracts. In the case of older workers, employment promotion contracts are the most widely used, but still 14-16% of all contracts are ordinary. On the whole, age clearly influences the choice of contract. Still, across all age brackets sufficient contract variation appears to exist, so that estimations for each age bracket are meaningful. 4 Duration analysis 4.1 Non-parametric profiles We now turn to the duration analysis of open-ended contracts. We first present the non-parametric Kaplan-Meier survival profiles, for each type of open-ended contract and for the different years under analysis. Here all age brackets are considered together, as the idea is to simply provide some initial empirical evidence. Figure 2 presents the results. Figure 2 indicates that the three types of open-ended contracts show clearly differentiated profiles. First, it is interesting to note the difference between the prereform profile and the post-reform ones. Even for contracts with similar legal characteristics (namely ordinary contracts, for which regulations was left untouched), the survival profiles are lower after the reform. One explanation is that the characteristics of those being offered ordinary contracts after 1997 were more prone nonetheless to shorter durations even before the reform. In any case, what is more relevant is that conversion contracts are the ones with more stable patterns, especially in 10

12 1998 and 1999: their survival rate after three years (1080 days 2 ) is around 70% (65% for 1999). Ordinary contracts, at that same point, show survival rates around 60% (a bit more in 1997 and 1998 and somewhat less in 1999). Finally, employment promotion contracts are the less stable contracts: after three years, half of them are no longer in force (the proportion is a bit more in 1997 and a bit less in 1998 and 1999). On the whole, these survival profiles raise interesting questions. First, it is clear that open-ended contracts are not as permanent as might be though. Secondly, after the 1997 reform, open-ended contracts of all types became more volatile. However, those carrying lower firing costs were not necessarily the least stable. This was the case indeed for employment promotion (initial) contracts. However, conversion from temporary workers happened to be the most stable. Survival Figure 2. Kaplan-Meier estimates of the duration of open-ended contracts, by starting year and type of contract (Source: estimated from INEM contract and unemployment insurance databases) Ord Ordinary 1997-Conversions 1997-Empl.Prom Ordinary 1998-Conversions 1998-Empl.Prom Ordinary 1999-Conversions 1999-Empl.Prom Days 2 In INEM data, months are normalizad to last 30 days. 11

13 4.2 Duration models What explains the results found in the preceding subsection? Is there any specific characteristic of the employment promotion, ordinary and conversion contract which makes each of them behave differently, or is it a composition effect due to the differing use of each of these contracts for people with varying characteristics and in different jobs? To explore these issues, we now turn to a more complete multivariate analysis by estimating full duration models. In this section, we present the results of estimates of semi-parametric proportional hazard Cox models. We emphasize the role played by the type of contract in affecting the hazard rate of ending the contract. Other regressors include gender, level of education, region, industry and occupation. We estimate separate models for each of the groups identified in Table 3, for each of the years of the post-reform period. The results are summarised in Table 4 in terms of the coefficients of the contrat type variables. All the coefficients estimated are significant at the 99.9% level. The results presented in Table 4 are very interesting. They confirm the ideas found in the empirical survival profiles, namely, that conversion contracts reduce the hazard rate, i.e. they are linked to longer durations, and that employment promotion contracts increase the hazard rate, i.e. they are linked to shorter durations. The results are similar for all age brackets and for the three periods observed. These coefficients, in terms of relative hazard ratios, indicate that receiving a conversion rate reduces the hazard rate relative to that of ordinary contracts by 20% for youngsters, 30% for middle-aged workers and 40% for older workers, while having an employment promotion initial contract increases by 25-30% for youngsters and by 15-25% for middle-aged and older workers. In the case of this latter contract type, the ratios appear to be decreasing over time. We have also tested for the proportional hazards assumption implicit in the Cox model, by comparing predicted and empirical values of the survival profiles. The results indicate that such an assumption cannot be nullified. We have also estimated parametric Weibull models, which have provided essentially the same results. On the whole, a robust conclusion follows from our analyses, in the sense that employment promotion contracts increase the hazard rate and conversion contracts reduce it relative to ordinary contracts. The interesting point to be remembered is that 12

14 these two contracts carry the same firing costs, lower than those applying to ordinary open-ended contracts, suggesting that firing costs are not the only element in the decision to use open-ended contracts rather than temporary ones. Table 4. Proportional hazards semi-parametric Cox regression of the duration of open-ended contracts, 1997, 1998 and 1999, three age brackets. Coefficients for contract type variables and hazard ratios Age brackets and contract type variables COEFFICIENTS Less than 30 Conversions Employment promotion years Conversions Employment promotion years Conversions Employment promotion HAZARD RATIOS Less than 30 Conversions Employment promotion years Conversions Employment promotion years Conversions Employment promotion Notes: - reference contract is ordinary contract - other regressors: gender, level of education (9 leves), region (17 regions), industry (10 industries) and occupation (9 occupations). 13

15 5 Trajectory analysis In the previous section, a duration analysis of the first open-ended contract was undertaken. One problem of such a study is that only one contract is considered. However, given the substantial amount of outflows observed, it may be worth considering a more complete study of the contract trajectories followed by workers. Does the end of an open-ended contract lead to another open-ended contract, to unemployment benefits, to temporary contracts? Do workers leaving open-ended employment face long periods of unemployment? Do they enter an unstable pattern of multiple contracts? If so, are there specific characteristics which lead to such unstable patterns? These are interesting questions which nicely complement the duration analysis presented so far. 5.1 The trajectories followed by workers with an open-ended contract In this section, we deal with the patterns of evolution of workers whose first contract in 1998 was open-ended, distinguishing between the three types we have been considering so far 3. Given the information available, already discussed, it is possible to establish a tipology of workers based on their labour market experience since they received the open-ended contract. At least, the following eight groups may be distinguished: 1. workers who receive only one contract since 1998 and have had no unemployment benefit spell; these may be considered the group with long-lasting record. Although we may not be sure that they remained working during all the four years analysed, it is quite safe to assume so. They represent the most stable group. They represent 56.0% of all the open-ended workers under consideration. 2. workers who receive two open-ended contracts since 1998 without any unemployment benefit spell either in between the two contracts or after the second contract. This group may be assumed to be also quite stable, because they have two contracts without any significant period in between (because otherwise they 3 The analysis presented in the section is based on a slightly different database, particularly because the sample was extracted in 2001 instead of In the final version of the paper, the updated analysis shall be included. 14

16 would claim unemployment benefits). In most cases, the type of the second contract is the same as that of the first one. They represent 5.8% of the workers under study. 3. workers who receive more than two open-ended contracts since 1998 and have had no unemployment benefit spell; these are similar to the preceding group, except that they seem to be moving to a larger extent. They cannot be considered stable because their contracts do not seem to last long, even if they are formally open-ended. They represent 7.5% of the workers under study. 4. workers whose last contract in the four-year period studied is not open-ended 4, no matter how many contracts they might have received (as a matter of fact most of them have received more than 3 contracts), and have had no unemployment benefit spell. They look similar to the preceding groups, except that the last observed contract happens to be temporary, and the number of contracts tends te be larger. They represent 12.9% of the total number of workers under study. 5. workers whose last contract is open-ended and have experienced at least an unemployment benefit spell in between (they may have more than one contract and also more than one spell). The main difference from the preceding groups is that these workers clearly interrupt their working activity and spend some time as unemployment benefit claimers. They represent 2.9% of total workers. 6. workers whose last contract is not open-ended and have experienced at least an unemployment benefit spell in between (they may have more than one contract and also more than one spell). They are similar to the preceding group, except that the last contract observed happens to be temporary instead of open-ended. They represent 6.8% of total workers. 7. workers who, at the end of the observation period, are receiving unemployment benefits, and have received more than one contract during the period of observation. They represent 3.7% of total workers. 8. workers who, at the end of the observation period, are receiving unemployment benefits, without having received any other contract after the initial one. They represent 4.5% of total workers. 4 Not open-ended is to be interpreted as either temporary or other, where this included various contract types such as seasonal workers or part-time workers. These look closer to temporary contracts than to open-ended contracts. 15

17 These trajectories identify workers whose labour market situation has experienced various degrees of stability. Thus, trajectory 1 may in principle be considered the most stable one; trajectories 2 and 3 may also be considered relatively stable, especially if all the contracts signed were open-ended and not too numerous. Actually, it might be possible that trajectories 2 and 3 are in some sense better than trajectory 1, inasmuch as they reflect a voluntary, and hence improvement-related, move by workers. Achieving the best match at once is not necessarily the most likely outcome. Trajectories 7 and 8 may be interpreted as leaving the labour market, despite the obvious right censorship problems. The remaining trajectories may be considered unstable. How do the different type of open-ended contracts fare in terms of the trajectories just described? Table 5 presents the results. Conversions appear to be the most stable type of open-ended contracts in terms of the first match. There are significant differences with the two other categories. Employment-promotion contracts are the worst matching contracts. When considering two contracts, the differences narrow a bit between ordinary open-ended contracts and conversions, but they get wider with respect to employment-promotion contracts ( ). This worse initial match of employment-promotion contracts translates into higher percentages of trajectories with a non-open-ended contract or unemployment benefit (after multiple spells) as the ending point: adding trajectories 4, 6 and 7, we find 34% of employment-promotion workers, as opposed to 24% of ordinary workers and 15% of conversions. Table 5. Trajectories of workers who received an open-ended contract in 1998, (Source: INEM, Contract and UB databases) Initial contract in 1998 (open-ended) TRAJECTORIES Employment- Ordinary promotion Conversions Total 1.One open-ended contract; no UB (1OE;noUB) Two open-ended contracts; no UB(2OE;noUB) Three or more contracts, last open-ended; no UB (3+C;noUB) Last contract temporary or other; no UB (LastTO;noUB) Last contract open-ended, UB (LastOE;UB) Last contract temporary or other, UB (LastTO;UB) Last observation UB (multiple contracts) (EndUB;>1c) Last observation UB (single contract)(endub;1c) Total Number of workers

18 To complete the information in Table 5, Table 6 presents the distribution by number of contracts of the different trajectories, and for the three main categories of open-ended contracts. Some of the trajectories are, by definition, linked to a certain number of contracts (this happens in trajectories 1, 2 and 8) but it is interesting to know the number distribution of the other trajectories. Table 6. Distribution of workers receiving an open-ended contract in 1998, by total number of contracts, trajectories and initial type of contract (Source: INEM, contract and unemployment benefits database) Inicial contract Number of contracts Average Trajectories number of type (1998) 2(*) Total contracts 1. 1OE;noUB OE;noUB c;lastOE;noUB Ordinary 4. LastTO;noUB LastOE;UB LastTO;UB EndUB;>1c Employment promotion (initial) Conversions 8. EndUB;1c OE;noUB OE;noUB c;lastOE;noUB LastTO;nov LastOE;UB LastTO;UB EndUB;>1c EndUB;1c OE;noUB OE;noUB c;lastOE;noUB LastTO;noUB LastOE;UB LastTO;UB EndUB;>1c EndUB;1c Note: (*) by definition, 1 in the case of trajectories 1 and 8. Trajectory 6 turns out to be the one with the largest number of contracts. Trajectories 3, 4 and 5 also show high number of contracts although less so. Trajectory 7 is clearly under the others. The information contained in Tables 5 and 6 suggests that the eight trajectories described may be aggregated into four main types: stable pattern: trajectories 1 and 2 17

19 mobile employment pattern: trajectories 3 and 4 (many contracts, but no unemployment benefits) entry and exit pattern: trajectories 5 and 6 (many contracts, with unemployment benefits spells in between) leaving the labour market: trajectories 7 and 8 (last observation is unemployment benefits) Needless to say, the notion of instability deriving from this aggregation may be questioned, because our information is right-censored, i.e. we have no information on the future after the last contract, and it might well happen that after a period of wandering around from one contract to another, people may find the correct match and stay there for good. The analysis we have just presented ought to be regarded with caution. Stability is not necessarily a single-sided variable. It is a dynamic variable, which develops over time. The trajectories we have defined make this point forcefully. We have taken trajectories 1 and 2 as the stable ones, but some of the workers in the other trajectories could be wandering around through several contracts and for some period of time before settling in a longer-lasting job. Stability at first sight is not necessarily the optimum outcome of job matching. This is something that should not be forgotten when interpreting later results. At any rate, what this analysis makes clear is that the usual implicit assumption that open-ended contracts are permanent is not warranted on a general basis. Even the contracts carrying the highest firing costs experience noticeable mobility; lowering those costs increases mobility probably because employers exert less care in achieving good matches at first. A significant exception occurs when workers and employers have had a previous match, allowing them to get to know each other; in this case, the proportion of conversions may be increased by the lowering of firing costs, although at the deadweight cost of the conversions which would have taken place in any case (a cost not only in terms of the distribution between employers and workers stemming from severance pay, but also in terms of the subsidies provided by the state to the new contracts). 18

20 5.2 Multivariate analysis Introduction In the last sub-section, we have defined the trajectories followed by workers whose first contract in 1998 was open-ended. The main result has been that employment-promotion contracts are less stable than ordinary contracts which in turn are less stable than conversions. The question that arises is whether this result may be attributed to other characteristics of the workers. For example, if employmentpromotion contracts were more oft-used in the construction sector, which is by its characteristics, less stable, this could be behind the lower stability of this type on contract. In order to analyze the determinants of stability, thus, we follow a double road. First, we run a logit regression of the probability of being stable in the sense of following trajectory 1, using as additional regressors the information on contracts available (age, gender, education, region, industry and occupation). Secondly, we extend our analysis by considering the full array of trajectories, although grouped in the four main categories identified before. A multinomial logit is run for such a study. 19

21 5.2.2 The probability of being stable Of all the trajectories identified, the first one is the most relevant. This is to be expected; in fact, the surprising fact is that there were no more cases in this trajectory. It it thus useful to identify the factors which make workers more likely to belong to this trajectory as opposed to the rest. Table 7 presents the results of a logistic regression of the probability of being stable, in the sense of belonging to trajectory 1. Both the coefficients of the regression and the odds-ratios are presented. As already seen in the previous basic analysis, the contract type is a very significant variable in determining stability. The probability of being stable increases 50% from ordinary to conversion contracts and decreases by almost 30% when going from ordinary to initial employment promotion contracts. When taking into account the various characteristics of workers and jobs, thus, this fundamental result remains. Among the other covariates, females show higher probability of being stable and age shows a consistent pattern on increasing stability. As for education, the pattern is not so clear, although university education (together with lower vocational education) shows a higher propensity to stability. Finally, occupations and industries show a less clear pattern, although agriculture is less stable and manufacturing and collective services have higher stability. 20

22 Table 7. Logistic regression of the probability of being stable (i.e. following trajectory 1), workers having received an open-ended contract in 1998 (Source: INEM, contracts and unemployment benefit databases); sample restricted to workers under 55 years of age INDEPENDENT VARIABLES Coefficient Odds-ratio Signif. CONTRACT TYPE Ordinary (&) Employment-promotion * Conversions * AGE * * * (&) * * * * GENDER Males (&) Females * LEVEL OF EDUCATION No education Primary education(&) Lower secondary * Upper secondary * Lower vocational * Upper vocational * Lower university * Higher university * OCCUPATIONS Managers * Professionals Technicians * Clerks * Service workers * Agricultural wkrs * Skilled manual mfrg.,constr.(&) Semi-skilled mfrg., constr * Unskilled serv/agric/constr/mfrg * INDUSTRIES Agriculture * Mining, quarrying, chemicals * Manufacturing * Engineering * Construction(&) Distribution, Hotels/restaurants Transport, communications Financial/producer services Collective services * Personal and domestic services * Constant * Number of observations Notes: controls also included for region of residence; (&) characteristics included in constant; (*): significant at the 99.9% level. Regression globally significant at the 99.9% level. 21

23 5.2.3 Multinomial analysis of trajectories To complete our analysis, we have run a multinomial regression of the probability of belonging to the various trajectories considered. In this case, we take the stable status (in the wider meaning of trajectories 1 and 2) as the reference and study the determinants, relative to that status, of belonging to one of the other three main groups identified, namely, mobile employment, entry and exit and going out of the labour market. The results are presented in Table 8. In this case, only coefficients of the regression are shown. As in previous cases, the contract type is highly significant as an explanatory variable. In all cases, conversion contracts are less conducive to non-stable trajectories and initial employment promotion contracts more. Females are less prone to enter mobile employment patterns but they are more likely to belong to entry and exit patterns and especially to leaving the labour market altogether. In terms of age, younger people show a higher probability of belonging to both unstable patterns with a somewhat more pronounced shape in the case of the mobile employment pattern. In all three cases, more education appears to be associated with a lower probability of being non-stable. As before, the results are less clear for occupations and industries. 22

24 Table 8. Multinomial logistic regression of the probability of belonging to various trajectories, workers having received an open-ended contract in 1998 (Source: INEM, contracts and unemployment benefit databases); sample restricted to workers under 55 years of age; reference category: stable pattern INDEPENDENT VARIABLES Mobile employment pattern Entry and exit pattern Leaving the labour market Coefficient Sign. Coefficient Sign. Coefficient Sign. CONTRACT TYPE Ordinary (&) Employment-promotion * * * Conversions * * * AGE * * * * * * * (&) * * * * * * * * * * * GENDER Males (&) Females * * * LEVEL OF EDUCATION No education Primary education(&) Lower secondary * * * Upper secondary * * Lower vocational * * * Upper vocational * * * Lower university * * * Higher university * * * OCCUPATIONS Managers * * * Professionals * * * Technicians * * Clerks * * Service workers * * * Agricultural wkrs * * Skilled manual mfrg.,constr.(&) Semi-skilled mfrg., constr * * Unskilled serv/agric/constr/mfrg * * INDUSTRIES Agriculture * * * Mining, quarrying, chemicals * * Manufacturing * * Engineering * * Construction(&) Distribution, Hotels/restaurants * * * Transport, communications * Financial/producer services * * Collective services * * * Personal and domestic services * Constant * * * Number of observations Notes: controls also included for region of residence; (&) characteristics included in constant; (*): significant at the 99.9% level. Regression globally significant at the 99.9% level. 23

25 6 Concluding remarks In this paper, we have analyzed administrative records on open-ended contracts in Spain since These are particularly relevant given that the labour market reform passed in 1997 created one new type of open-ended contract, applicable to both conversions from former temporary contracts and to some groups of workers newly hired as open-ended, the main characteristic of which was that it carried lower firing costs as compared to the ordinary open-ended contract, left unchanged. The main thrust of the empirical analysis has been that since 1997 open-ended contracts have shown lower stability (higher hazard rates of ending), but also that stability is not linked to firing costs (together with economic incentives in the form of lower social charges). This result appears in all of the empirical analyses undertaken: a duration study (empirical Kaplan-Meier profiles and semi-parametric Cox duration model) and a trajectory study (trying to go beyond the single-event analysis of duration models). The consequence of our results is that approaches based on the reduction of firing costs of open-ended workers appear not to be well-suited. First, if the reduction of firing costs leads to less stable trajectories and shorter durations, it means that a substitution of open-ended contracts for otherwise temporary ones is taking place. So reducing firing costs is equivalent to make open-ended contracts more equal to temporary workers, a process which would be fully achieved if firing costs were eliminated altogether. In this sense, achieving a reduction of temporary employees might be due to a mere redefinition of open-ended workers. However, the Spanish experience suggests that the new so-called open-ended contracts are closer to temporary ones on a more fundamental issue: their duration is shorter than that corresponding to other, ordinary, open-ended contracts. This might explain why, despite the high increase in the number of open-ended contracts signed after 1997 the stock of openended contracts did not react accordingly. Secondly, the fact that conversions are more stable than ordinary contracts means that temporary contracts may be playing the role of testing the worker. In this case, what is unclear is the need to foster such conversions by subsidising them, in the sense that they might occur anyway in the absence of such subsidy (thus implying a significant deadweight loss). Here, the proposal that the trial periods for open-ended contracts should be extended makes sense, although if conversions are possible, it is a 24