Long-term unemployment as a screening device and its consequences for active labour market policies

Transcription

1 Working Paper No Long-term unemployment as a screening device and its consequences for active labour market policies Ricardo Welters and Joan Muysken CofFEE-Europe University of Maastrict April 2002 Centre of Full Employment and Equity Te University of Newcastle, Callagan NSW 2308, Australia Home Page: ttp://e1.newcastle.edu.au/coffee coffee@newcastle.edu.au

2 1 Introduction During te last ten years, active labour market policies (ALMPs) constituted a substantial part of te efforts governments put into figting long-term unemployment. Employment subsidies i.e. firms get a subsidy wenever tey take on long-term unemployed comprise a considerable part of ALMP. Figure 1 sows tat teir sare increased from 32 per cent in 1990 of expenditures on active labour market policies in te Neterlands to 56 per cent in Bearing in mind tat te expenditures on ALMP increased from two to six billion euros in te same period, sows te magnitude of tese programs (SZW 2001). But wy would a government want to subsidize employment for jobseekers tat firms do not want to ire? From an economic point of view, tis can only be justified wenever tere are market failures. Figure 1 Sare of employment subsidies in total expenditures on ALMP, te Neterlands General LTU Protected Source: SZW (2001). General employment subsidies comprise tax exemptions employers receive for all employees tey employ at a wage at or just above te minimum wage (for example SPAK). LTU employment subsidy scemes are targeted at long term unemployed only. Part of te money is spent on job creation in te public sector. Te rest is available for employers in te private sector wo ire long term unemployed (WIW, I/D, VLW). Protected unemployment aims at creating jobs for disabled unemployed wo will not find a non subsidized job (WSW). In our view market failures can arise since firms cannot observe ability levels of jobseekers witout a cost, due to asymmetric information. In order to minimize iring costs firms use screening devices. However, te use of suc devices leads employers to disregard some jobseekers wit sufficient ability levels. Tis constitutes a market failure, and government intervention by means of employment subsidies can be justified. Te Dutc economy experienced some considerable canges during te last decade. In te early nineties, te unemployment rate fluctuated around 6 to 7%, but at te end of te decade it declined to around 3% (CBS, 2001). Te tigtening of te labour market as consequences for te screening standards used by firms. Firms can be less 2

3 coosy, wic means tat long-term unemployed, wo were fallaciously disregarded in an easy labour market, will be taken into consideration wen te labour market tigtens. Moreover te tigtening of te labour market as encouraged a more tailormade approac to employment subsidies. We analyze ow bot te tigtening of te labour market and te policy cange affect te success of employment subsidies. 2 Modeling iring beavior using screening devices Weiss (1980) was one of te first to model firm iring beaviour under information asymmetry. According to Weiss, productivity differences between jobseekers can be attributed to costless observable and costly observable caracteristics. Costless observable caracteristics are for example te attained education level of workers, wereas costly observable caracteristics are for instance innate abilities. Bot caracteristics togeter determine total abilities and ence productivity. Employers are looking for te most productive jobseekers and terefore need to get grip on tese costly observable caracteristics. One way to be able to select igly productive workers is to offer ig wages. Weiss points at te fact tat workers aving ig productivity levels will also ave ig reservation wages. Hence offering ig wages will increase te sare of igly productive workers amongst te applicants. Guasc and Weiss (1980) propose a second approac to solve information asymmetries. Tey introduce ability tests, wic enable firms to locate te most productive workers (tests on a pass/fail basis). Guasc and Weiss argue tat te most productive workers will be most confident of passing te test and ence are more likely to do te test. Tey sow tat tere migt be a testing equilibrium in wic at least part of te iger productivity workers are screened. 1 Greenwald (1986) proposed a tird route to solve te problem of asymmetrical information. Greenwald argues tat employers introduce probationary contracts. During suc a probationary period, te employer observes te oterwise idden abilities of workers. At te end of tis period te employer decides weter to continue te employment relation or not, based on te abilities of te worker. By doing so, employers categorize te unemployment pool. Workers tat lose teir job at te end of te probationary period and return to te unemployment pool inadvertently 1 See also Guasc and Weiss (1981). 3

4 signal to oter employers tat tey do not ave ig idden abilities. 2 Consequently teir cances on reemployment diminis. Te categorization of workers also reflects in te wage. During te probationary period, te employer pays wages equal to te average marginal product of all workers concerned. At te end of te period, te employer picks out workers aving ig abilities and pays tem accordingly, wereas te fired workers find temselves in a second and labour market were below probationary period wages prevail. Tese tree models ave in common tat tey rely on self-selection among jobseekers. Entry barriers like a probationary period, ability tests and to a lesser extent offering ig wages deter unqualified. Tat is, it is not te employer wo selects potential employees, it is te employees wo self-select. Te existence of selfselection leads te autors to ignore to a certain degree te actual iring process starting from posting a vacancy to actually filling it firms undertake. Altoug we do not want to underestimate te importance of self-selection, te substantial expenditures firms devote to uman resource management point at te difficulties firms face wen trying to fill a vacancy (see for example Hale (1998)). Terefore we focus on te iring decision process firms undergo to fill a vacancy. Information asymmetry induces screening devices We assume a competitive labour market in wic tere are I identical firms. Next to tese firms tere are J jobseekers, wo differ wit respect to teir ability level α j. Firms want to secure a certain minimum productivity level p. However, information asymmetry prevents firms from observing te ability of an applicant and ence er productivity. Only during a toroug and costly assessment procedure firms can determine te true ability level of a jobseeker. Figure 2 represents te iring procedure, wic starts wen te firm posts a vacancy. Tis announcement will yield a rate of contacts of applicants. Since te firm is not able to assess te abilities of applicants rigt away, it will screen applicants using some easy obtainable device t. 2 Waldman (1984) notes tat firms not only use te post-period 1 job matc results at oter firms to infer te abilities of workers, but also te pre-period 1 job assignments at oter firms. Here we discard tis possibility. 4

5 Te relevance of screening in iring decisions as lead to an ongoing debate in te empirical literature: te screening versus uman capital debate. Te general consensus now is tat screening plays a role in iring decisions see for example Taubman and Wales (1973), Wolpin (1977), Albrect (1981), Miller and Volker (1984) and Lang and Kropp (1986). However, tese autors refer to te level of education as a screening device. Lync (1985), Lync (1989), Van den Berg and Ours (1996) and especially Omori (1997) sow tat firms also use unemployment duration as a screening device. In our view tis is a relevant practice, wic we will furter analyse. Figure 2 Overview of te iring procedure job applicant t>t screening device t<t p<p assessment procedure p>p iring procedure closed An applicant wo does not meet te screening device standard t (for instance no more tan one year unemployed ) will be rejected immediately; oterwise te applicant enters te assessment procedure. In te latter case te applicant will be assessed torougly and er true ability level will be revealed. Wen er ability level is sufficient and ence productivity exceeds p, te applicant will be accepted for te job and te iring procedure closes. Oterwise te applicant will not be accepted and ence te firm as to start te iring procedure all over again wit anoter applicant. 3 To function as a useful screening device, tere needs to be a ig correlation between te screening device and te unobservable caracteristic. From a teoretical point of 3 Te possibility tat a worker meets t but not p sows tat te screening device is no perfect indicator of ability/productivity. If it would be, tere would be no asymmetric information anymore. 5

6 view, tere are at least two reasons wy unemployment duration and productivity levels are negatively correlated. Firstly, te duration dependence argument explains wy tere is a negative relation between productivity and unemployment duration. Non-use of skills during unemployment spells leads to atropy of skills and ence to productivity loss. Secondly, eterogeneity explains tis negative relationsip. Te most productive workers will find a job quickly, leaving te less productive workers in te unemployment pool, wose sare in te total jobseekers pool terefore increases as unemployment duration increases. In our analysis we will not include eterogeneity effects as including tem would lead to complicating te model witout enricing it. Te negative relationsip between unemployment duration and productivity terefore solely stems from atropy. Productivity declines wit unemployment duration Firms want to secure a certain minimum productivity level p. Altoug, tey cannot observe te productivity of a worker, p j, nor er innate ability, α j, te employer knows te distribution of abilities over all workers. Te ability of a worker only corresponds to er productivity wen se as not been unemployed. During unemployment te potential productivity depreciates at a rate ω. Hence we assume te following link between ability and productivity for worker j: (1) p j = 1 ω( t j, α j) α j and ω(0, α) = 0, ωt > 0, ωα > 0, ωtt < 0, ωt α < 0 We briefly elaborate on te properties of te depreciation of abilities, since tey condition te results of te rest of te paper. Te first property, ω t > 0, indicates te positive relation between productivity depreciation and unemployment duration. Skills get outdated following non-use and ence productivity declines. Te empirical literature on post intermittence wage declines confirms tis see for example Mincer and Ofek (1982), Mincer and Polacek (1978), Kim and Polacek (1984) and Albrect et al. (1998). But not only economists find tis negative relationsip, also psycologists do (Artur et al., 1998). Te second property, ω α > 0, assures a positive relation between productivity depreciation and ability level. Hig ability workers face a iger productivity depreciation rate tan low ability workers following a given spell of unemployment. 6

7 Albrect et al. (1998) and Neumann and Weiss (1995) find empirical support for tis ypotesis. 4 Te tird property, ω tt < 0, implies a diminising increase in productivity loss as unemployment duration continues. Te rationale beind tis feature follows from te second property, wic intuitively says: te more productive you are, te more tere is to lose. Te fourt and final property, ω αt < 0, opens te possibility to model te productivity loss process in suc a way tat individuals aving ig abilities remain more productive tan individuals aving low abilities for a given t. Every unemployed as a personal combination of α j and t j, wic togeter determine er productivity p j. We assume tat employers can observe te duration of unemployment of applicants. Since tey also know te distribution of abilities of workers, tey can infer from tat te distribution over workers wit respect to productivity and duration, x(p, t). 5 Figure 3 presents te evolution of productivity loss following non-use of skills. At unemployment duration t=0, we observe te productivity level of individuals wo were fired in te current period. As firing is assumed to be a random process, 6 tese individuals are a representation of te productivity distribution of employees. Teir productivity ranges from p min (0) to p max (0). 4 We abstract in tis paper from unemployment spells originating from earlier completed spells of unemployment. 5 Note tat not only individuals aving te same ability level and unemployment duration face te same productivity level. If person A as an ability surplus over person B, tis surplus (in productivity terms) could be wiped out by a longer spell of unemployment. 6 Job separation follows job-specific idiosyncratic socks, wic make te job not productive anymore and ence te job breaks down, regardless te (innate) ability level of te employee carrying out te job. Tese socks occur at probability λ. 7

8 Figure 3 Link between unemployment duration and productivity productivity p max (0) p max (t ) p min (0) D p B A C p max e -at p min (t ) p min e -ct t= t unemployment duration As te unemployment spell increases, skill atropy causes productivity to decline, wic is represented by te two downward sloping dotted curves. Te upper dotted curve represents te productivity depreciation of te igest ability worker: (2) max () = α max p t e at Hence a is te per period depreciation rate of te igest ability worker and for a given unemployment duration t it produces te igest productivity level available. Te lower dotted curve represents productivity depreciation of te lowest ability worker: ct (3) pmin () t = α mine and a > c For a given t, tis function delivers te lowest possible productivity level. Since ig ability workers experience more productivity deterioration following non-use of skills tan low ability workers, ω α > 0 in equation (1), a must be larger tan c. Te productivity distribution Te maximum and minimum productivity level given unemployment duration define te boundaries of x(p,t), te distribution of workers in te unemployment pool over 8

9 productivity. Te igest productivity level tat can be found in te unemployment pool is p max (0), wic is an unemployed aving te maximum ability level, α max, and no unemployment record. Te lowest possible productivity level in te unemployment pool must be p min (T), were T is te maximum duration of unemployment. We assume te intelligence distribution over te population to be a Bell curve, wic empirically appears to be close to a normal distribution cf Herrnstein and Murray (1994). Since we assume entry into unemployment to be random, eac cross section of x(p,t) in Figure 3 at a particular unemployment duration will be Bell-saped. 7 Figure 4 presents te productivity distribution of unemployed experiencing t periods of unemployment, x(p;t ), as a simplified version of te Bell curve. It is obvious tat te mode productivity level of x(p;t ) is decreasing in t, as productivity depreciation follows non-use of skills. Figure 4 Productivity distribution given t pr x(p;t ) p min (t ) p max (t ) productivity given t=t Unemployment duration as a screening device Te longer unemployment spells last, te lower te productivity level on average becomes. Terefore using unemployment duration as a screening device increases te probability tat a firm meets a qualified job applicant. In te absence of using a screening device te probability to immediately find a qualified worker is 1/[1 + (B + C)/(A + D)] cf Figure 3, were A, B, C and D represent te surfaces of four different areas. Setting a screening device standard t increases tis probability to 1/[1 + B/D]. Crucial in tis respect is tat tis probability will only be iger wen B/D < (B + 7 Here we make te assumption tat tere is a close link between intelligence and innate ability, wic enables us to present x(p;t) as a (simplified) normal distribution, too. 9

10 C)/(A + D). Tis is te case wenever te reduction of type 1 error costs (retaining unqualified jobseekers in te selection process) outweigs te increasing costs of generating type 2 errors (excluding qualified jobseekers from te selection process). By imposing a screening device standard t, only unemployed experiencing t or less tan t periods of unemployment remain relevant for a job (tose to te left of t in Figure 3). Hence decreasing t, increases te sare of qualified unemployed in te unscreened pool. Let te function s of t be te distribution of unemployment duration of te jobseekers pool, i.e.: (4) pmax ( t) s() t = x( p,) t dp pmin ( t) Ten te sare of unemployed eligible to be screened, i.e. te screening survival rate, equals: (5) t= t t= t max ( ) p t St ( ) = stdt ( ) = xptdpdt (, ) t= 0 t= 0 pmin ( t) Figure 5 sows tis process in a different way, using a cumulative distribution of productivity of all job seekers: (6) t f ( p;0, t ) = x( p, t) dt 0 Te productivity distribution ranges from p min (T) to p max (0). And te probability of finding a qualified applicant is 1/[1 + (B + C)/(A + D)]. Wen te iring standard t is introduced, te cumulative distribution is reduced. Te maximum productivity level remains p max (0). However, te minimum productivity level increases due to te use of te screening device to p min (t ) tis is elaborated in te annex to te paper. As a consequence, te probability to find a qualified applicant increases to 1/[1 + B/D]. In tis case te condition B/D < (B + C)/(A + D) follows from te sift of te distribution to te rigt. 10

11 Figure 5 Introducing a screening device pr f(p;0,t) A f(p;0,t ) C D B p min (T) p min (t ) p p max (0) productivity Hiring costs and optimal screening Te firm will encounter two types of iring costs; te costs of carrying out an assessment k and per period foregone productivity z. 8 Te latter costs refer to te productivity loss of a vacancy. 9 Total iring costs equal te average cost per assessment times te average number of assessments needed to find a qualified jobseeker. Since labour market tigtness migt cause vacancies difficult to fill wen a strict screening device is used, an applicant migt not be available in every period te firm as a vacancy. Terefore total z costs do not ave to be in line wit te number of assessments carried out. To model te arrival rate of applicants we use a stocastic job model (Pissarides, 2000). After te firm as posted a vacancy, jobseekers start contacting te firm. Te contact rate q is by definition te number of contacts m.l over te number of vacancies 8 We abstract from oter costs like te costs related to posting te vacancy, as teir contribution to total iring costs is small. Implicit to our reasoning is tat we assume tat posting a vacancy will eventually lead to filling it. However, te time span between opening and filling te vacancy remains unknown. 9 Wic we define to be a fraction, κ, of p. Since p is no endogenous variable in our model, z is neiter. 11

12 v.l. 10 Using te notion of a matcing function, te rate of contacts q depends negatively on labour market tigtness θ, wic is defined as v/u: 11 ml ul vl u (7) = m(, ) = m(,1) = q( θ ) vl vl vl v However, since te firm uses unemployment duration as a screening device cf equation (5) te product of te screening survival rate S(t ) and te arrival rate q(θ), gives te final arrival rate: (8) St ( ) qθ ( ) Te average total costs incurred during an assessment ten are: (9) z k + St ( ) qθ ( ) A more stringent screening device standard leads to a smaller final arrival rate wic leads to more forgone productivity and ence to iger per assessment costs: a typical type 2 error. Te firm is interested in keeping te number of assessments it as to complete to find a qualified worker (p>p ) low. Terefore, te success rate of assessments needs to be ig i.e. te probability tat an applicant wo enters te assessment procedure indeed meets te productivity standard, wic is a reduction in type 1 errors. Te success rate of a firm setting t= t follows from equation (6): (10) pmax p 0, t ( ;0, ) 1 = ( ) f p t dp F p As te average number of assessments needed to find a qualified worker is te inverse of te success rate of te assessment procedure, total iring costs H(t ) are defined by: H ( t 1 z ) = 0, k + t 1 F ( p ) S ( t ) q( θ ) (11) 10 Were m is te number of jobseekers tat apply for te job as a fraction of te labour force l and v is te number of vacant jobs as a fraction of l. 11 We exclude on te job searc, wic means tat all jobseekers experience positive unemployment spells. 12

13 Te rigt-and side of te equation indicates te trade-off, wic te firm faces. A iger iring standard increases te success rate (reduces type 1 errors) but at te same time increases te per assessment costs (iger type 2 errors). Te firm will coose its optimal screening standard, t *, suc tat iring costs are minimised. In te Annex we sow tat t * will increase wen te labour market becomes tigter. In tat case it is optimal for firms to relax teir screening standards. Wage determination and market failures Given te screening standard, t *, te wage level is determined in a Nas bargain between employers and employees. Depending on te bargaining strengt of bot parties, te wage will be a weigted average of on te one side te outside option and on te oter side te productivity level of te employee plus te iring costs te firm incurred. If te employers ave full bargaining strengt, te wage equals te outside option. If te employees ave full bargaining strengt, te wage equals productivity plus iring costs since te firm as to incur iring costs again if wage negotiations break down. Actually, te iring standard t migt influence te wage outcome, since it defines te relevant sare of te unemployment pool and it also influences te outside option. However, tese are aggregate effects and it seems reasonable to assume tat individual firms will not take tese into account. Figure 6 sows ow te optimal screening device standard, t *, defines te fraction of all unemployed wic are taken into consideration. Tis means tat te labour supply curve is cut off at E 0, and a wage w 0 results as te outcome of te bargaining process. 12 Altoug t * is optimal from te firm s point of view, it is socially not. Unemployed wo do not meet t *, but wo do meet p, are excluded from te iring procedure. Tese unemployed pay a price for te inaccuracy of te screening device. But since te firm does not face te externalities (i.e. unemployment benefits) of tis type 1 error, it does not take tese costs into consideration wen setting t. Hence, we are dealing wit a market failure. 12 For expositional simplicity we ave drawn vertical lines in te figure, of course tese sould be increasing, depending on te bargaining strengt of employees. 13

14 Figure 6 Te effects of te screening device standard on te employment outcome wage l s (t * ) l s (t ) w 0 w 1 l d E 0 E 1 employment To internalise te market failure, te government as to induce te firm to increase t to say t. As iring costs are iger after t *, te only option to induce firms to increase t is to subsidize te increase in te iring costs. Following subsidization, te firm will increase t and subsequently take (part of) te unemployed in area A (see Figure 3) into consideration for te job. In Figure 6 tis means tat labour supply increases to l s (t ), wic leads to a wage reduction and ence an increase in employment. Tis increase in employment and consequently decrease in government expenditures on unemployment benefits could be used to finance te subsidy Employment subsidies to correct market failures Employment subsidies can encourage firms to recruit from long-term unemployed for different reasons. We discuss two forms of subsidies: compensation for increased iring costs, and scooling or training subsidies. In bot cases a temporary or once and for all subsidy can lead to permanent job creation. Compensation for increased iring costs Te objective of te government is to persuade firms to recruit from unemployed in area A of Figure 3. Recruiting from jobseekers experiencing more tan t periods of unemployment leads to increasing iring costs for te firm. Terefore te government 13 We do not elaborate te optimal size of t ere. 14

15 needs to allocate a subsidy to firms wo take on unemployed experiencing unemployment spells ranging from t to say t r > t te subsidy range ten covers te area A. Introducing a subsidy for employers wo take on unemployed out of te range t to t r, can be seen as a sift of te screening device standard cf te sift from t to t r in Figure 7. Terefore te additional iring costs of firms wo ave to recruit from te complete subsidy range t to t r equal H(t r ) H(t ), cf equation (11). As iring costs keep on increasing to te rigt of t, te subsidy as to depend on unemployment duration positively, to ensure tat firms take into consideration all unemployed experiencing unemployment spells ranging from t to t r. 14 Finally, since te subsidy leads to te selection of persons wose productivity meets te standard p, te worker occupies a permanent job. Subsidy to scool unemployed workers Besides extending te searc process for a qualified worker, te firm can also decrease te productivity standard and use te subsidy to upgrade te productivity of te ired jobseeker to te originally required productivity level. Figure 7 illustrates tat decreasing required productivity from p to p r, means tat te firm takes more unemployed (t r -t ) into consideration. Hence allocating subsidies to firms wo use te subsidy to scool workers also leads to job opportunities for unemployed wo are excluded oterwise. Lowering p to p r, induces te firm to adapt t to t r. Tat is, te optimal screening device standard sifts to te rigt. Now te firm finds it optimal to recruit up to t r, ence te subsidy can be uniform. 15 Altoug te design of te subsidy differs between bot options, te outcome is te same: by means of te subsidy, employers recruit from long-term unemployed and find qualified personnel witin tis group. Moreover, in bot cases tey will continue to employ te workers after te termination of te subsidy as tey meet te firm s ability standard. 14 Moving to te rigt of t means tat iring costs increase as te gains in terms of more arrivals do not outweig te losses in terms of rejections during assessments. 15 Te level of te subsidy depends on te costs of scooling, wic we take for granted ere. 15

16 Figure 7 Te impact of employment subsidies productivity p max (0) p min (0) p B A p max e -at p r p min e -ct t= t t= t r unemployment duration 4 Employment subsidies: te case of te Neterlands Te two oil socks in te seventies tat it te Dutc economy, caused unemployment rates to increase to levels unprecedented in te post war period. Altoug te unemployment rate decreased gradually, it did not fall back to pre oil sock levels. Consequently, long-term unemployment increased rapidly and remained ig trougout te nineties. Te use of employment status as a screening device is probably one of te driving forces beind tis long-term inactivity. Te Dutc autorities observed te problem unemployed faced and increased te use of ALMP (bot in financial terms as in terms of volumes). Prime objective of ALMP was mobilizing long-term unemployed. A typical example of ALMP was te so-called Melkert1 project, wic aimed at 40,000 subsidised jobs for long-term unemployed in te public sector. Te objective of tis program was to meet te needs of bot unemployed and society (te subsidized jobs were created in socially valuable sectors sort of personnel). Te government justified te use of ALMP to elp long-term unemployed escape social isolation and grant teir rigt to work. We will entitle tis by equity justification As opposed to efficiency justification, wic refers to using ALMP to repair market failures. 16

17 Te empasis on equity arguments also appeared from te design of te programs. Te subsidy period was often unlimited, wic assured tat long-term unemployed did not fall back in isolation, but it did not elp as an outflow stimulus. Te subsidy projects were mainly created in te public sector, wic indeed led to socially valuable work, but te work to be done, contributed little to increasing te productivity level of participants wic would be useful in finding a non-subsidized job. Hence it was no surprise tat te results in terms of outflow to unsubsidised employment were disappointing cf Welters (1998). However, as tis was no primary objective of ALMP, low outflow rates were no major concern to policymakers. Tis justification sould be compared to te efficiency justification, wic primarily focuses on te unemployed in area A of Figure 3, i.e. aims at reducing errors of type 2. An example of te latter is te Melkert2-project. Tis program provides temporary subsidies for taking on long-term unemployed to employers in te private sector only. Tis programme was a reaction to te tigtening of te labour market in te nineties. We sow tat deadweigt losses are a serious problem ere. Finally te Dutc government as canged it policy towards profiling, i.e. targeting te unemployed. Tis policy primarily aims at reducing te type 1 errors. We sow tat tis policy is igly successful in principle, but tat deadweigt loss migt become a serious problem again. Te equity justification: low deadweigt loss Te success rate of employment subsidies depends, amongst oter factors, on te type of jobseekers tat enter te program. If te government requires firms to select from long-term unemployed close to te rigt of t, te success rate will be iger tan firms aving to recruit from unemployed experiencing unemployment spells considerably larger tan t (for example te subsidy range from t r to t r in Figure 8). Te more te subsidy range sifts to te rigt, te iger te subsidy as to be to induce firms to participate, as iring costs increase. Someting, wic is typically not found in reality. In te present example of Figure 8, tere would not be a subsidy level ig enoug to induce firms to recruit from te range t r to t r, since tere are no 17

18 unemployed in tis range tat meet te productivity standard. Tis is probably one of te reasons wy Melkert1 jobs were created in te public sector. 17 Wen te unemployed close to t are targeted, efficiency arguments come to te fore and one sould beware of te deadweigt loss. Deadweigt loss equals te sare of participants in employment subsidies tat would ave found a job witout te subsidy. However, te risk of provoking deadweigt loss by subsidizing te Melkert1 unemployed is low. In particular wen te government really succeeds in targeting te range t r to t r in Figure 8, and tus also can correctly identify t r. It is obvious tat tis is not always te case. In particular te improved situation on te labour market in te 1990s and te increased dynamics made it more difficult for te government to determine te subsidy range and increased te deadweigt loss risk. Moreover, te increased labour market tigtness caused a cange from an equity to an efficiency justification of active labour market policies. Figure 8 Targeting unemployed and deadweigt loss productivity p max (0) p min (0) p F G p max e -at p min e -ct t t t r t r unemployment duration Te efficiency justification: deadweigt loss During te late nineties te labour market became muc tigter. As a result of falling unemployment rates te sare of long-term unemployed in total unemployment 17 An alternative would of course be to provide scooling subsidies. 18

19 started to decline, toug at a slow rate. Besides equity arguments to elp long-term unemployed, a second argument arose: employers are in need of personnel. Tigtness on te labour market meant tat firms could not easily fill teir vacancies, toug at te same time tere was a core group of unemployed. Government policy canged from te Melkert1-project to te Melkert2-project, wic was focused on te private sector and intended to matc long-term unemployed to vacancies. Te merits of tis type of ALMP come muc closer to te ALMP as we discussed in section 3. However, deadweigt loss ten becomes a serious problem. Minimizing deadweigt loss is important, as it constitutes a waste of subsidy transfers. A government tat uses employment subsidies sould not only care about maximizing te success rate but it sould at te same time try to minimize deadweigt loss. Figure 8 sows tat wit respect to te latter te ideal situation would be, a government knowing t and setting te subsidy range from t to for instance t r. If it does so, no unemployed would find a subsidized job se would ave found in te absence of te subsidy, ence no deadweigt loss. However, if te government does not know t and accidentally sets te subsidy range from t to t r, deadweigt loss equals F/(F+G). A factor, wic is often overlooked in te policy discussion, is tat te increased labour market tigtness induced employers to relax teir selection standards for unemployed. In te Neterlands policymakers ardly adapted te subsidy range to increasing tigtness. Te entrance criterion remained one year out of unemployment trougout te nineties. Tis as increased te apparent success rates of te subsidies, but at te expense of deadweigt loss. In te Annex we sow ow indeed increased labour market tigtness will lead to an increase in te optimal screening device standard t * for te employers. Tis rigtward sift of t * following an increase in tigtness affects te success of employment subsidies, as firms are more willing to participate. However, tis increased interest of firms to participate stems from te fact tat tey receive subsidies for unemployed tey would ave taken on witout te subsidy. Figure 8 also illustrates te deadweigt loss as labour market tigtness increases. Government starts te subsidy range at t, wic coincides wit te initial value of t *. Hence initially tere is no deadweigt loss. However, as labour market tigtness 19

20 increases, t * sifts to te rigt to for example t. A government tat now does not sift te subsidy range accordingly, will induce te development of deadweigt loss, wic amounts F/(F+G). Profiling unemployed: more successful Altoug, te Dutc autorities did not cange te subsidy range following te canging stance of te labour market, tey implemented oter measures to adapt to te canging environment. Te allocation of long-term unemployed towards te various programs became more structured. Employment Offices set up profiles of unemployed ( fasering ), wic provide te distance unemployed face to te labour market. Tese profiles determine te reemployment probabilities of long-term unemployed, based on individual caracteristics (education level, occupation, age etc.). Profiling enables policymakers to allocate unemployed to te program tat suits tem best see for example Kooreman (1999). Tat is, if te outcome is tat unemployed face severe difficulties to find employment, programs like te Melkert1- project (te name canged into In-/Doorstroom banen ) are still available. But if tese difficulties appear to be less severe oter alternatives are available, ranging from temporary employment subsidies at private sector firms to re-scooling possibilities. Profiling offers unemployed a tailor-made solution, but not only unemployed will benefit from successful profiling. Also employers will benefit from profiling as profiles offer employers a clear indication of wat to expect from unemployed. In our model, profiling plays te role of decreasing te uncertainty margin of te screening device if Employment Offices succeed in filtering out te most promising long-term unemployed. In our analysis tis means tat te lower bound of te uncertainty margin sifts up, following te provision of additional information about te qualities of long-term unemployed. Less uncertainty, means less failures during te assessment procedures and ence total assessment costs go down, for a given screening device standard. Figure 9 illustrates tat te additional information reduces area B to area B. However, as a result of reduced risk, firms sift te screening device standard to te rigt, from t to t, tereby taking more unemployed into consideration. Tis sift causes deadweigt loss to arise, if te government does not adapt te subsidy range. Hence not only targeting needs to be done cautiously, te government also needs to monitor external labour market developments. 20

21 Figure 9 Te impact of profiling: more success productivity p max (0) p max (0) p A p max e -at B B p min e -ct t t unemployment duration 5 Conclusion Using a Pissarides stocastic job model, we sowed tat employers use employment status as a screening device to minimize iring costs during iring procedures. Te use of tis device leads to te exclusion of long-term unemployed in iring decisions and leads to market failures. A well-tried policy measure to elp long-term unemployed is employment subsidies. Tere are two independent reasons wy firms participate in employment subsidy scemes. Firms use te subsidy as compensation for extended searc among longterm unemployed or firms use te subsidy to scool long-term unemployed up to te required level. Bot applications of te subsidy open job opportunities for long-term unemployed. However, te design of te employment subsidy sould be different. Wen firms extend teir searc process, te level of te subsidy sould depend on unemployment duration wereas wen te firm uses it to scool workers te subsidy sould be uniform. We discussed tree canges tat influence te success rate of employment subsidies, using te Dutc experience as an example. One is related to te economy: increased tigtness. Increased tigtness does lead to more success, but at te expense of deadweigt loss if te government does not adapt te subsidy categories. Te second and tird cange, are related to te employment subsidy policy: targeting of 21

22 unemployed to te rigt program and profiling job seekers. Targeting leads to an improvement of te ability of te subsidy to elp long-term unemployed back to employment. However, targeting leads to te emergence of deadweigt loss if targeting is not executed carefully. Te effects of profiling are comparable to te effects of increased tigtening; tey lead to deadweigt loss. Hence not only targeting and profiling need to be implemented cautiously, government also sould adapt te design of employment subsidies to canges in labour market tigtness. 22

23 References Albrect, J. (1981), A Procedure for Testing te Signalling Hypotesis, Journal of Public Economics, 15(1), Albrect, J., P. Edin, M. Sundström and S. Vroman (1998), Career Interruptions and Subsequent Earnings: A Reexamination Using Swedis Data, Journal of Human Resources, 34(2), Artur, W., W. Bennett, P. Stanus and T. McNelly (1998), Factors Tat Influence Skill Decay and Retention: A Quantitative Review and Analysis, Human Performance, 11(1), CBS (2001), Werklooseid gedaald, Persberict, PB01-086, 19 April Devine, T. and N. Kiefer (1993), Te empirical status of job searc teory, Labour Economics, 1(1), Greenwald, B. (1986), Adverse Selection in te Labour Market, Review of Economic Studies, 53(3), Guasc J. and A. Weiss (1980), Wages as Sorting Mecanisms in Competitive Markets wit Asymmetric Information: A Teory of Testing, Review of Economic Studies, 47(4), Guasc, J. and A. Weiss (1981), Self-Selection in te Labour Market, American Economic Review, 71(3), Hale, J. (1998), Strategic Rewards: Keeping your best talent from walking out te door, Compensation & Benefits Management, 14(3), Herrnstein and Murray (1994), Te Bell Curve: Intelligence and class structure in American life. New York: Te Free Press. Kim, M. and S. Polacek (1994), Panel Estimates of Male-Female Earnings Functions, Journal of Human Resources, 29(2), Kooreman, A. (1999), De kansmeter: fasering werkzoekenden, Tijdscrift voor loopbaanadviseurs en begeleiders, 4(6), Lang, K. and D. Kropp (1986), Human Capital Versus Sorting: Te Effects of Compulsory Attendance Laws, Quarterly Journal of Economics, 101(3),

24 Layard, R. and G. Psacaropoulos (1974), Te Screening Hypotesis and te Returns to Education, Journal of Political Economy, 82(5), Lync, L. (1985), State Dependency in Yout Unemployment, Journal of Econometrics, 28(1), Lync, L. (1989), Te Yout Labour Market in te Eigties: Determinants of Re- Employment Probabilities for Young Men and Women, Review of Economics and Statistics, 71(1), Miller, P. and P. Volker (1984), Te Screening Hypotesis: An Application of te Wiles Test, Economic Inquiry, 22(1), Mincer, J. and H. Opek (1982), Interrupted Work Careers: Depreciation and Restoration of Human Capital, Journal of Human Resources, 17(1), Mincer, J. and S. Polacek (1978), Women s Earnings Reexamined, Journal of Human Resources, 13(1), Ministerie van Sociale Zaken en Werkgelegeneid (2001), Aan de slag, eindrapport van de werkgroep Toekomst van et arbeidsmarktbeleid, Den Haag. Neumann, S. and A. Weiss (1995), On te effects of scooling vintage on experience-earnings profiles: Teory and evidence, European Economic Review, 39(5), OECD (1998), Wat works among active labour market policies: evidence from OECD countries experiences, Labour Market and Social Policy, Occasional Papers No. 35. Omori, Y. (1997), Stigma Effects of Nonemployment, Economic Inquiry, 35(2), Pissarides, C. (2000), Equilibrium Unemployment Teory, 2nd edition, MIT Press. Taubman, P. and T. Wales (1973), Higer Education, Mental Ability, and Screening, Journal of Political Economy, 81(1), Van den Berg, G. and J. van Ours (1994), Unemployment Dynamics and Duration Dependence in France, Te Neterlands and te United Kingdom, Economic Journal, 104(423),

25 Waldman, M. (1984), Job assignments, signalling, and efficiency, Rand Journal of Economics, 15(2), Weiss, A. (1980), Job Queues and Layoffs in Labour Markets wit Flexible Wages, Journal of Political Economy, 88(3), Welters, R. (1998), De effectiviteit van et Nederlandse actieve arbeidsmarktbeleid, Maandscrift Economie, 62(8), Wiles, P. (1974), Te correlation between education and earnings: Te external-testnot-content (ETNC) ypotesis, Higer Education, 31, Wolpin, K. (1977), Education and Screening, American Economic Review, 67(6),

26 Appendix Te distribution function x(p,t) and f(p) ave a clear link, wic we will explain in tis appendix. Figure A1 sows two cross sections of x(p,t) namely x(p;t+1) and x(p;t+2). Obviously, x(p;t+2) is to te left of x(p;t+1) as skills deteriorate following an additional period of non-use. Te figure also sows tat ig skilled unemployed suffer more from productivity depreciation tan low skilled unemployed. We also included te iring standard, p, in te figure to illustrate tat te part of te unemployed in eac segment tat meets p decreases as te unemployment spell prolongs. Figure A1 Decreasing t leads to a sift in x(p;t) pr x(p;t+2) p x(p;t+1) p min (t+2) p min (t+1) p mod (t+2) p mod (t+1) p max (t+2) p max (t+1) Te effects (in terms of increasing te probability to find a qualified jobseeker) of decreasing te iring standard (from t+2 to t+1) are visualized in Figure A2. Excluding jobseekers experiencing t+2 periods of unemployment, means tat te left tale of f(p) sifts to te rigt as te lowest available productivity level increases from p min (t+2) to p min (t+1). As is apparent from x(p;t+2) in Figure A1, te vast majority of jobseekers in tis segment does not meet te iring standard, p, only a small fraction does. Hence if a firm excludes tis segment from te iring procedure, unqualified jobseekers are excluded disproportionably, wic is te very nature of te screening device. In Figure A2, tis means tat te peak of f(p) sifts to te rigt. 26

27 Figure A2 Excluding segments of x(p,t) leads to a sift in f(p) pr f(p;0,t+1) f(p;0,t+2) A mod(e;t+2) mod(e;t+1) C B D p min (t+2) p min (t+1) p p max (0) To determine te probability to find a qualified jobseeker after imposing a screening device standard, t =t+1, we need to define f(p;0,t+1). 18 Te distribution function f(p;0,t+1) consists of two parts, an upward sloping part and a downward sloping part. We define te upward sloping part as: (A1) p min ( 1) ( ;0, 1) p t + f u p t+ = mod( et ; + 1) p ( t+ 1) mod( et ; + 1) p ( t+ 1) min min Were mod(e) is a function of te modus ability level of te jobseekers wo survived te screening device standard. Tis function depends on te effectiveness (in terms of discriminatory power to distinguis qualified jobseekers from unqualified) of te used screening device. Clearly te stricter t is set, te iger te discriminatory power is (e t >0). However, for small values of t, e tt <0 as firms start making a considerable amount of type 1 errors : excluding qualified jobseekers. Te downward sloping part is defined as: (A2) max (0) f d ( p;0, t+ 1) = p p p (0) mod( e; t+ 1) p (0) mod( e; t+ 1) max max Determining te probability to find a qualified worker after imposing t =t+1 comes in te specific case of Figure A2 down to integrating equation A2 from p to p max (0), wic is equation (10) in te paper. 18 If t =t+1, all jobseekers wo experience spells of unemployment of t>t+1 are excluded from te iring procedure. 27

28 Our model sows tat tere are two types of iring costs, te costs of te assessment procedure k and foregone productivity z. Increasing tigtness does not directly influence costs k. However, it as a direct effect on total z costs as an increase in tigtness leads to a drop in te contact rate, wic increases te number of periods needed to fill te vacancy. Hence, te optimal screening device standard is influenced by labour market tigtness. To sow tis matematically, we first need to define te functions 1-F 0,t (p ) and S(t ) out of section 2. Te former represents te probability to ave a successful assessment, wic depends negatively on t. Terefore we replace it by a function ν(t ). Te probability to meet te screening device standard depends positively on t (or negatively on sarpening t ). Terefore we replace te integral S(t ) by a function σ(t ). For notational convenience we coose some simple representations: (A3) γ t ν = ae γ < 0 ηt σ = be η > 0 1 Important to see is tat γ<0, wic assures tat decreasing (sarpening) te screening device standard leads to an increase of te probability tat someone wo enters te assessment procedures meets te iring standard. Moreover, η>0, wic assures tat decreasing te screening device standard leads to a reduction of te number of candidates tat enter te assessment procedure. Substituting bot equations into equation (11), and maximizing to t yields te optimal t. Moreover, after some resuffling, tis optimality condition can be represented as a second representation of σ called σ 2. (A4) ( γνσ ησν ) 2 ( νσ q) H kγν zq + η z = 0: + = 0 σ = 1 + t kq 2 2 ν γ ( θ) Now we ave two expressions for σ (σ 1 and σ 2 ). Tese two equations can ten be used to derive dt /dθ: (A5) σ1 σ2 σ1 t σ2 t η zq'( θ) = = ησ = θ θ t θ θ θ γ kq( θ) t η+ γ z q'( θ) = > 0 η+ γ > 0 θ ηγ σq( θ) k q( θ) Tis derivative is only positive wenever decreasing te iring standard leads to a bigger decrease in accuracy of te screening device tan te increase it provokes in 28

29 te inflow rate into te iring procedure, ence η+γ>0. Henceforward in our model, firms will accept a somewat iger failure rate during te assessment procedure as long as te contact rate increases, wic reduces total foregone productivity costs. Tis derivative also sows te impact costs z and k ave on te relation between tigtness and te optimal screening device standard. Te per period costs of not filling a vacancy z and te per period costs of assessment procedures σq(θ)k influence te magnitude of te derivative. Te iger costs z in comparison to per period assessment procedure costs, te more a firm will increase its screening device standard as te labour market tigtens. 29