Labour Market Flows: Facts from the United Kingdom

Transcription

1 DISCUSSION PAPER SERIES IZA DP No Labour Marke Flows: Facs from he Unied Kingdom Pedro Gomes November 2010 Forschungsinsiu zur Zukunf der Arbei Insiue for he Sudy of Labor

2 Labour Marke Flows: Facs from he Unied Kingdom Pedro Gomes Universidad Carlos III de Madrid and IZA Discussion Paper No November 2010 IZA P.O. Box Bonn Germany Phone: Fax: Any opinions expressed here are hose of he auhor(s) and no hose of IZA. Research published in his series may include views on policy, bu he insiue iself akes no insiuional policy posiions. The Insiue for he Sudy of Labor (IZA) in Bonn is a local and virual inernaional research cener and a place of communicaion beween science, poliics and business. IZA is an independen nonprofi organizaion suppored by Deusche Pos Foundaion. The cener is associaed wih he Universiy of Bonn and offers a simulaing research environmen hrough is inernaional nework, workshops and conferences, daa service, projec suppor, research visis and docoral program. IZA engages in (i) original and inernaionally compeiive research in all fields of labor economics, (ii) developmen of policy conceps, and (iii) disseminaion of research resuls and conceps o he ineresed public. IZA Discussion Papers ofen represen preliminary work and are circulaed o encourage discussion. Ciaion of such a paper should accoun for is provisional characer. A revised version may be available direcly from he auhor.

3 IZA Discussion Paper No November 2010 ABSTRACT Labour Marke Flows: Facs from he Unied Kingdom * This paper documens a number of facs abou worker gross flows in he Unied Kingdom for he period beween 1993 and Using Labour Force Survey daa, I examine he size and cyclicaliy of he flows and ransiion probabiliies beween employmen, unemploymen and inaciviy, from several angles. I examine aggregae condiional ransiion probabiliies, job-ojob flows, employmen separaions by reason, flows beween inaciviy and he labour force and flows by educaion. I decompose conribuions of job-finding and job-separaion raes o flucuaions in he unemploymen rae. Over he pas cycle, he job-separaion rae has been as relevan as he job-finding rae. JEL Classificaion: E24, J60 Keywords: worker gross flows, job-finding rae, job-separaion rae, ransiion probabiliies Corresponding auhor: Pedro Gomes Universidad Carlos III de Madrid Economics Deparmen C/ Madrid Geafe Spain pgomes@eco.uc3m.es * I would like o express my sincere hanks o Philip Bunn, Sephen Millard, Kaharine Neiss and Kenny Turnbull for heir conribuions o his work and o all a he Srucural Economic Analysis Division a he Bank of England for heir hospialiy. I am also graeful o Frank Cowell, Jennifer Greenslade, Chrisopher Pissarides, Simon Price and paricipans a he Bank of England seminars for heir useful commens. Financial suppor was provided by FCT.

4 1 Inroducion The behaviour of flows beween employmen, unemploymen and inaciviy drive movemens in aggregae indicaors, such as he employmen and unemploymen rae. They are criical o our undersanding of labour marke dynamics and business cycle flucuaions. Furhermore, worker gross flows and ransiion raes lie a he hear of sae-of-he-ar models of unemploymen, anchored in he Morensen and Pissarides (1994) search and maching framework. The objecive of his paper is o esablish a number of key facs abou he properies of he UK labour marke flows, by examining daa from he Labour Force Survey over he pas seveneen years. In so doing, i exends he work by Bell and Smih (2002) and provides a sysemaic sudy of worker gross flows based on UK daa, along he lines of he pioneer work of Blanchard and Diamond (1990) for he Unied Saes. 1 One main conribuion is o add o he debae recenly revived by Shimer (2007) regarding he relaive imporance of job-finding and separaion raes for flucuaions in unemploymen. I provides evidence for he Unied Kingdom using differen decomposiion mehods proposed in he lieraure. The addiional ineres, relaive o Perongolo and Pissarides (2008) is ha he sample covers a complee business cycle: he expansion beween 1993 o 2001, he slowdown and he curren recession. I find ha he job-separaion rae is as imporan as he job-finding rae. The increase of he job-separaion rae a he onse of he curren recession gives srengh o he poin made by Davis, Faberman, and Haliwanger (2006) ha changes in he job separaion rae explain mos of he variaion in unemploymen during sharp recessions. I hen go on o analyse paricular elemens of he labour marke ha can be useful for economiss in oher areas of research. Given he size of he flows from and ino inaciviy I have explored in more deail heir role over he business cycle. In paricular, I have disaggregaed he inacive ino wo subgroups: hose ha wan a job (and herefore can be considered marginally close o he labour marke) and hose ha do no wan a job and evaluaed he differences beween hem. In he las few years he Unied Kingdom experienced srucural changes in he level of educaion of he labour force. Therefore, i seems imporan o examine he size and he behaviour of labour marke flows by educaion. 1 There are some sudies on he UK labour marke flows, noably classical sudies by Nickell (1982) and Pissarides (1986) or more recenly Burgess and Turon (2005), bu hey only consider inflows and ouflows of unemploymen using claiman coun daa. Bell and Smih (2002), on he oher hand, use Labour Force Survey daa bu heir sample only runs unil 2000 and hey resric heir analysis o he size and cyclical properies of gross flows, job-o-job flows and job separaions by reason. 2

5 I have also provided evidence on job-o-job flows and on-he-job search, on he causes of employmen separaions and on aggregae condiional ransiion probabiliies. These sylised facs, summarised in he conclusion, are of ineres o policymakers and macroeconomiss alike. For policymakers hey can help improve he monioring of business cycles, he deecion of urning poins and he assessmen of labour marke ighness. For macroeconomiss, his paper can be seen as a reference for he calibraion of a number of parameers, and also provide a guideline of he empirical feaures ha heoreical models should ideally have. 2 Preliminary conceps 2.1 Labour marke dynamics In order o analyse labour marke dynamics I make use of some fundamenal equaions ha describe he evoluion of he sock of employed E and he sock of unemployed U. The pool of inacive is denoed as I. Adding he hree pools gives us he working-age populaion W, while summing employmen and unemploymen corresponds o he labour force L. The unemploymen rae is defined as u = U L and he paricipaion rae as p = L W. Toal employmen evolves according o he following equaion: E +1 = E + N UE + N IE N EU N EI, (1) where N is he gross flows beween he pools indicaed by he superscrip. If we normalise his equaion by he oal working-age populaion, we ge he following equaion ha focuses on he oal gross flows as he deerminan of changes in he employmen rae. E +1 E W = N UE W + N IE W N EU W N EI. (2) W Alernaively, (2) may be wrien in erms of ransiion probabiliies raes (λ ij ) E +1 E E = λ UE u + λ IE (1 p ) 1 u p (1 u ) λeu λ EI. (3) We can perform a similar decomposiion of he changes in unemploymen U +1 = U N UE + N EU N UI + N IU, (4) 3

6 eiher focusing on he gross flows or on he ransiion raes: U +1 U W = N UE W + N EU W N UI W + N IU, (5) W U +1 U U = λ EU 1 u u + λ IU (1 p ) p u λ UE λ UI. (6) Some auhors like Blanchard and Diamond (1990) or Davis (2006) focus on gross flows, while ohers, such as Shimer (2007) or Fujia and Ramey (2009) give more emphasis o ransiion raes. The wo perspecives are complemenary in he analysis of he labour marke and he ineres in one or he oher depends ulimaely on he heoreical model one has in mind. Thus I explore boh of hem. 2.2 Labour Force Survey The daa are consruced from he Labour Force Survey (LFS). The LFS is a quarerly survey of households living a privae addresses in he Unied Kingdom. Is random sample design, is based on he Poscode Address File, a lis prepared by he Pos Office wih all addresses which receive fewer han 25 aricles of mail a day. The LFS panel samples around 60,000 households for five successive quarers. The households are inerviewed face-o-face when firs included on he survey, and by elephone hereafer. The respondens are asked abou he household s characerisic, educaion, labour marke saus, economic aciviy, as well as oher elemens. The sample is spli ino five waves. Every quarer one wave of approximaely 12,000 leaves he survey and a new wave eners. In his way, we can observe he changes in he labour marke saus of 80% of he households ha ook par in he survey and, herefore, obain he gross labour marke flows. 2 Alhough he quarerly survey sars in 1992, he 5 waves only run since he firs quarer of 1993, so he sample is resriced o he period beween 1993:2 and 2010:1. There is a break in he survey in 1996 as before, i did no include Norhern Ireland. As Norhen Ireland represen less han 3% of he working-age populaion of he Unied Kingdom, he break does no affec he size of he gross flows as a percenage of he working-age populaion or in hazard raes. The consruced flows series are weighed using he Office for Naional Saisics (ONS) census populaion weighs. Esimaing gross flows on he basis of survey daa has wo shorcomings: hey suffer from non-response bias, and response-error bias. For he LFS, he non-conac rae is around 5% 2 A comprehensive discussion of survey s mehodology can be found in Clarke and Tae (2000). 4

7 while he refusal rae ranges beween 10% and 15%. The response error bias is a more serious problem because, in longiudinal daa he errors are cumulaive and lead o an overesimaion of flows. There is no pracical way o deal wih response-error bias. We should bear in mind ha he resuls migh be biased upwards, paricularly in he flows beween unemploymen and inaciviy. Neverheless here is no reason o believe ha he response-error bias affecs he cyclical properies of he gross flows. 3 Worker gross flows in he Unied Kingdom 3.1 Average gross flows Figure 1 summarises he average quarerly worker flows over he period. I repors he oal number of people ha changed saus in housands (), as a percenage of he working-age populaion (p) and as a ransiion probabiliy or hazard rae (h). Figure 1: Average quarerly worker flows, Labour Force Survey, Ino W.A.P. E 38 () U 24 () I 132 () 756 () 2.1 (p) 2.8 (h) Ou of W.A.P. E 71 () U 2 () I 86 () 451 () 1.3(p) 27.3 (h) Unemploymen 1680 () 4.6 (p) 364 () 1.0 (p) 1.3(h) Employmen () 74.3 (p) 303 () 0.8 (p) 18.1 (h) 378() 1.0 (p) 4.9 (h) 474 () 1.3 (p) 6.2 (h) 511 () 1.4 (p) 1.9 (h) Inaciviy 7651 () 21.1 (p) Noe: he worker flows are expressed as oal number of people in housands (), as a percenage of he working-age populaion (p) and as a hazard rae (h). The wo boxes show he movemens in and ou of he working-age populaion. The saisics are for he period saring in 1996:2 o include Norhern Ireland. 5

8 Over he sample period here was an average 50,000 ne increase in employmen per quarer. Subsanial quarerly gross flows hide, however, behind his value. An average of 875,000 people move ou of employmen every quarer, approximaely 60% of whom go ino inaciviy. An average of around 925,000 people move ino employmen, where he spli is broadly similar beween unemploymen and inaciviy. In addiion o he 2.6% of he oal working-age populaion ha join he pool of employed, here is an addiional 2.1% ha change employer every quarer. Demographic change represens a very small fracion of worker urnover, as shown in he wo boxes wihin he char. Only a minoriy of young people (less han 16 years of age) joining he working-age populaion ener he labour force direcly. Similarly, more han half of he people ha reach reiremen age (65 plus for men, 60 plus for women) are already inacive. For his reason, I exclude from he analysis new enries and exis from he working-age populaion. How do hese numbers compare o hose for he Unied Saes? Table 1 compares he quarerly figures for he Unied Kingdom wih he monhly values for he Unied Saes aken from Bleakley, Ferris, and Fuhrer (1999). If we inerpre he size of he gross flows beween unemploymen, employmen and inaciviy as a proxy for labour marke flexibiliy, one could be emped o say ha he labour marke in he Unied Kingdom is much less flexible han in he Unied Saes. While 6.8% of he populaion change saus every quarer in he Unied Kingdom, in he Unied Saes 6.5% change saus every monh. In my opinion, a comparison beween hese values can be misleading. Firs, because here migh exis muliple ransiions wihin he quarer. Suppose someone is unemployed in he firs monh, hen moves o employmen in he second, and hen back o unemploymen. While a monhly survey would pick up all ransiions, he quarerly survey would no deec any. I is possible o overcome he problem of muliple ransiions by calculaing for he Unied Saes he quarerly probabiliies implied by he monhly raes. The resuls are shown in he second column of Table 1. 3 Afer correcing for muliple ransiions, he oal implied quarerly flows in he Unied Saes are around 15% of he working-age populaion, wice he value for he Unied Kingdom. A big par of he flows, however, is accouned by he flows beween inaciviy and employmen. Bu for his comparison o be correc, we are implicily assuming ha: firs, here is no hisory dependency and second, ha here is no heerogeneiy on he labour force in 3 A quarerly survey measures he probabiliy ha an individual in a given sae, is in a differen sae afer 3 monhs, so we canno simply muliply he monhly ransiion raes by hree. We need o compue he probabiliies of muliple ransiions. See Appendix for deails on he calculaions. 6

9 Table 1: Gross flows for he Unied Saes and he Unied Kingdom Unied Saes Unied Kingdom Monhly Quarerly Quarerly Annual (o) (ex-m) (o) (ex-a) (o) (ex-q) Employmen Unemploymen Employmen Inaciviy Unemploymen Employmen Unemploymen Inaciviy Inaciviy Employmen Inaciviy Unemploymen Toal Noe: gross flows in percenage of he working-age populaion. The values from he Unied Saes are aken from Bleakley, Ferris, and Fuhrer (1999). The columns wih (o) indicae he frequency of he original survey. The columns wih a * are exrapolaed from a differen frequency denoed by: monhly (m), quarerly (q) or annual (a), by allowing for muliple ransiions and assuming consan ransiion probabiliies. erms of ransiion probabiliies. However, in realiy hese assumpions do no seem o be mee. Ruhm (1991), for insance, finds ha displaced workers face higher unemploymen raes for a leas four years. Also, Sevens (1997) shows ha he effecs of displacemens in earnings are quie persisen, because of addiional job losses in he years following he displacemen. Shimer (2007) discusses in lengh he effec of heerogeneiy on he evaluaion of he relaive imporance of he job-finding rae. One way o es if hese wo assumpions hold is o look a he aggregae condiional ransiion probabiliies. Boh hisory dependence or heerogeneiy would reflec on differen aggregae condiional probabiliies. I compued he average condiional probabiliies in he LFS, based on hree-period flows (N hij ): λ ij E 2 = N ij E 2 N 1 Ei = N Eij, λ ij N 1 Ei U 2 = N ij U 2 N 1 Ui = N Uij, λ ij N 1 Ui I 2 = N ij I 2 N 1 Ii = N Iij. N 1 Ii We can see in Table 2 he subsanial differences in condiional probabiliies. The probabiliy of separaion from employmen o unemploymen is 1% if he person was previously employed, 11% if he was previously unemployed and 5% if he was inacive. The job-finding rae is 48% if wo quarers earlier he person was employed, 25% if he person was inacive and 20% if i was unemployed. In addiion, one inacive person is beween 3 o 6 imes more likely o reurn o he labour force, if i has only been inacive for one period. Whaever he cause of he differences of he aggregae condiional probabiliies is, heir exisence implies ha he exrapolaion of flows a a differen frequency han he one which he survey was carried, is biased. In order o show how possible misleading his can be, I compue he annual ransiion probabiliy for he LFS direcly, by looking a he flows beween he firs and he fifh wave, and he annual rae exrapolaed from he quarerly raes, by assuming equal condiional probabiliies. We can see from he las column of Table 7

10 Table 2: Condiional ransiion probabiliies, Labour Force Survey Uncondiional Condiional on: probabiliies E 2 U 2 I 2 Employmen Unemploymen Employmen Inaciviy Unemploymen Employmen Unemploymen Inaciviy Inaciviy Employmen Inaciviy Unemploymen Noe: averages beween 1996 and ha he resul is quie differen. The annual ransiions, calculaed hrough he quarerly raes end o overesimae he rue values, paricularly on he employmen-inaciviy ransiions. The oal gross flows, esimaed using he quarerly ransiion probabiliies, are 18.6% of working-age populaion as opposed o he consisenly calculaed value of 11.6%. When doing he reverse exercise exrapolaing he quarerly flows from he observed annual flows we underesimae he rue ransiions by roughly one half. This conclusion is in line wih he resuls of Elsby, Hobijn, and Sahin (2008). They esimae he job-finding and separaion raes for OECD counries, based on publicly available daa on unemploymen by duraion spell. They build upon he mehodology of Shimer (2007), using daa on he number unemployed by differen duraion spells: less han 1, 3, 6 or 12 monhs. They find ha, for he Unied Kingdom and he Unied Saes, he calculaed job-finding rae is quie differen depending on which unemploymen lengh was used, which hey inerpreed as evidence of duraion dependence. 4 Blanchard and Porugal (2001) also found a analogous resul, bu wih job flows. They find ha, a a quarerly frequency, job creaion and job desrucion in Porugal are subsanially lower han in he Unied Saes, bu a an annual frequency job creaion and desrucion are acually higher in Porugal. Because a survey compares he sae of an individual in wo poins in ime, differen frequencies aler he weighs pu on he differen ypes of unemployed or hisories. This migh no be relevan when we wan o evaluae he relaive imporance of he job-finding rae, as shown by Shimer (2007), bu i is imporan when we wan o compare he size of he gross flows beween counries. Having he ransiion probabiliies a a given frequency is no enough o characerise a labour marke. 4 For insance, for he Unied Saes, when using unemployed wih a spell shorer han 1 monh, he monhly job-finding rae is 60 percen, bu when using he unemployed wih a spell shorer han 12 monhs, he monhly job-finding was only 20 percen. 8

11 3.2 Evoluion of labour marke socks and flows The firs row in Figure 2 displays he evoluion of he employmen rae, unemploymen rae and inaciviy rae in he Unied Kingdom over he pas 30 years. The verical line signals he beginning of he flows sample. We can see ha he sample covers one complee business cycle, wih a slighly negaive rend in he unemploymen rae. I fell unil 2001, i was relaively sable unil 2005 and has increased since. I reached 8% by he end of he sample, sill below he peak value in he wo previous recessions. The inaciviy rae has a small downward rend, bu compared o hisorical sandards i can be considered relaively fla. The second and hird rows in Figure 2 show he flows beween he hree pools, as a percenage of he working-age populaion and as hazard raes. Mos of he acion has been driven by he flows ino and ou of unemploymen. For insance, a he beginning of he sample, 1.6% of he working-age populaion moved from unemploymen ino employmen Figure 2: Labour marke socks, gross flows and hazard raes Employmen rae (% working age populaion) Unemploymen rae (% labour force) Inaciviy rae (% working age populaion) Employmen rae Unemploymen rae (% labour force, ONS) Inaciviy rae q1 1985q1 1990q1 1995q1 2000q1 2005q1 2010q1 1980q1 1985q1 1990q1 1995q1 2000q1 2005q1 2010q1 1980q1 1985q1 1990q1 1995q1 2000q1 2005q1 2010q Employmen ouflows (% working age populaion) Unemploymen ouflows (% working age populaion) Inaciviy ouflows (% working age populaion) 1995q1 1998q1 2001q1 2004q1 2007q1 2010q1 Employmen o Unemploymen flows Employmen o Inaciviy flows 1995q1 1998q1 2001q1 2004q1 2007q1 2010q1 Unemploymen o Employmen flows Unemploymen o Inaciviy flows 1995q1 1998q1 2001q1 2004q1 2007q1 2010q1 Inaciviy o Employmen flows Inaciviy o Unemploymen flows Employmen ouflows hazard rae Unemploymen ouflows hazard rae Inaciviy ouflows hazard rae q1 1998q1 2001q1 2004q1 2007q1 2010q q1 1998q1 2001q1 2004q1 2007q1 2010q q1 1998q1 2001q1 2004q1 2007q1 2010q1 Employmen o Unemploymen hazard rae Unemploymen o Employmen hazard rae Inaciviy o Employmen hazard rae Employmen o Inaciviy hazard rae Unemploymen o Inaciviy hazard rae Inaciviy o Unemploymen hazard rae Noe: he flows series are in a four-quarer moving average o remove seasonaliy and high frequency movemens. 9

12 every quarer, bu by he end of he sample i was jus above 1.2% of he working-age populaion. Separaions from employmen o unemploymen have also fallen, from 1.4% o 0.9% of he working-age populaion in 2007 bu hey have picked up sharply since he beginning of he recession. During 2009 around wo million people have los heir jobs. By conras, flows beween employmen and inaciviy have remained broadly sable across he sample period, bu have somewha decline during he recession. Alhough he picure of he gross flows and hazard raes is very similar for employmen and inaciviy, his is no he case for unemploymen. While he acual number of people ha moved ou of unemploymen fell beween 1996 and 2005, shrinking he pool of unemployed, he probabiliy of moving ou of unemploymen increased in he same period. 3.3 Cyclical properies of labour marke flows The lieraure on worker flows defines he cyclicaliy of flows as heir correlaion wih he level of economic aciviy. I esimae i by running a ordinary leas square regression of he log of each variable on season dummy variables, a linear rend and he unemploymen rae. This follows Baker (1992), who underakes a similar procedure o analyse he cyclical movemens of unemploymen duraion. 5 The resuls are shown in Table 3. One should be aware ha he conclusions are based on only one economic cycle, which also presened a negaive srucural rend in he unemploymen rae. Inflows and ouflows of all pools are counercyclical. In economic downurns, as he labour marke ges looser, here are more movemens beween he hree saes. The acion occurs mosly in he unemploymen pool. More of he unemployed find a job or sop searching for one. Also, more of he inacive sar looking for a job and more workers lose heirs. The separaion rae from employmen o unemploymen, and he ransiion probabiliy from inaciviy o unemploymen are srongly counercyclical, while he job finding rae is srongly procyclical. In oher words, recessions are periods when i is harder for an unemployed individual o find a job, an employed person is more likely o lose heir job and an inacive person is more likely o sar looking for one. Transiion probabiliies beween employmen and inaciviy do no seem o have a cyclical componen. In erms of magniude, job-separaion rae seems o flucuae as much as he job-finding rae. Also quie responsive is he hazard rae beween inaciviy and unemploymen. I seems 5 I have also used simple correlaion coefficiens and he resuls are idenical wheher I use unemploymen or employmen rae as he cyclical indicaor. Addiionally, I used derended GDP or capaciy uilisaion bu, alhough he resuls poined in he same direcion, he correlaions were in mos cases no significan. 10

13 Table 3: Cyclical variaion of labour marke flows and hazard raes Gross flows Hazard raes Gross flows Hazard raes E (0.00) (-3.24) E U (10.12) (11.84) E (3.42) (4.85) E I (-1.59) (-0.79) U (14.24) (16.04) U E (4.69) (-15.51) U (10.61) (-24.00) U I (10.57) (-6.09) I (2.70) (2.67) I E (-3.62) (-3.46) I (1.03) (1.61) I U (7.29) (8.38) Noe: he cyclicaliy of he series is he coefficien on unemploymen rae in a regression of he series in logs on season quarer dummies, ime rend and unemploymen rae. T-saisics are in brackes.* denoes significan a 1% and ** significan a 5%. ha job-finding and separaion raes are equally imporan deerminans of unemploymen flucuaions. Neverheless, given he ongoing debae on he relaive imporance of each ransiion rae, I make a more careful analysis in he nex secion. 4 Wha drives unemploymen in Unied Kingdom? Wha dicaes he cyclical behaviour of unemploymen: hires or separaions? The seminal work on labour marke flows by Blanchard and Diamond (1990) and Davis and Haliwanger (1992) se he convenional wisdom ha recessions are mainly driven by high job loss raes. Recenly, Shimer (2007) and Hall (2006) have challenged his view by presening evidence ha cyclical unemploymen dynamics are largely driven by a ime-varying job-finding rae and ha he separaion rae is close o being acyclical. These wo papers had a very srong impac on he field. On he one hand, many researchers have used his evidence o develop models ha incorporae consan job desrucion raes (for insance Blanchard and Gali (2010) or Gerler and Trigari (2009)). On he oher hand, oher researchers pu forward more evidence ha opposes heir claims. Davis, Faberman, and Haliwanger (2006) provide new empirical evidence in suppor of he view ha a recession sars ou wih a wave of separaions. Fujia and Ramey (2009) and Elsby, Michaels, and Solon (2009) argue ha boh job-separaion and job-finding raes play a significan role in unemploymen flucuaions in he Unied Saes. The UK evidence is also conroversial. Pissarides (1986) finds ha, for he period beween 1967 and 1983, almos all changes in unemploymen can be accouned for by changes in he job-finding rae. In conras, Burgess and Turon (2005) claim ha beween 1967 and 1998 he unemploymen dynamics arise mosly from shocks o inflows. Perongolo and Pissarides (2008) find ha he job-separaion rae accouns for one hird of unemploymen volailiy. 11

14 The ongoing debae gives emphasis o he use of differen mehodological approaches and daa sources across researchers. To evaluae he conribuion of job-finding and jobseparaion raes in he Unied Kingdom I use wo decomposiions of unemploymen ha have been proposed in he recen lieraure: Shimer (2007) and Fujia and Ramey (2009). I also correc he daa for ime aggregaion bias, by applying he coninuous correcion mehod proposed by Shimer (2007). 4.1 Unemploymen decomposiions The saring poin for all unemploymen decomposiions is he equaion of he seady-sae unemploymen u ss. Wih hree saes, he equilibrium unemploymen is a funcion of all six ransiion probabiliies: 6 u ss = f(λ EU, λ EI, λ UE, λ UI, λ IE, λ IU ). (7) Shimer (2007) isolaes he effec of he each ransiion rae by consrucing a counerfacual unemploymen rae if all oher ransiion probabiliies were always o be a heir sample average ( λ ij ). For insance, if we focus of he ransiion probabiliy beween employmen and unemploymen, he counerfacual unemploymen would be: u EU = f(λ EU, λ EI, λ UE, λ UI, λ IE, λ IU ). (8) If we ignore he flows in and ou of inaciviy, and focus only on wo saes, he equilibrium unemploymen is given by: 7 u ss = λ EU λ EU + λ UE and he wo counerfacual unemploymen raes are: u EU = λ EU λ EU + λ UE, uue =, (9) λ EU λ EU + λ UE. (10) Shimer s decomposiion has faced some criicism because he seady-sae approximaion is non-linear in he wo hazard raes. In his sense, if we chose differen values for insead of he sample average we could ge differen answers. I access he robusness of he mehod by applying he wo-sae decomposiion of Fujia and Ramey (2009). By linearising 6 See appendix for he exac formula. 7 While he hree-saes equilibrium unemploymen racks he acual unemploymen rae very well, he wo-saes equilibrium unemploymen is one percenage poin below acual unemploymen. However, he correlaion beween equilibrium and acual unemploymen is around 0.93 for boh mehods. λ ij 12

15 he seady-sae unemploymen around he previous period seady-sae u ss 1, we ge he following expression: u ss u ss 1 u ss 1 = (1 u ss 1) λeu λ EU 1 λ EU 1 (1 u ss 1) λue 1 λ UE 1 λ UE, (11) which is simply breaking down he percenage change of he seady-sae unemploymen rae ino percenage changes of boh job-finding and job-separaion raes. We can resae his expression as du ss du ss uss u ss u ss 1 = du f + du s, where 1, du f (1 u ss 1) λue 1 λ UE 1 λ UE, du s (1 u ss 1) λeu 1 λ EU 1 λ EU. (12) Anoher way o assess he robusness of he resuls is o repea he exercise using daa generaed a a monhly frequency. I use daa on he claiman coun unemploymen ouflows and inflows o calculae a proxy for job-finding and job-separaion raes. This daa, provided by ONS, covers he unemployed ha are claiming unemploymen benefis. I is a proxy for wo reasons. Firs, people regisered in he claiman coun are only a subse of he unemployed. Second, despie consiuing he large majoriy, claiman accoun flows include no only flows beween unemploymen and employmen bu also include flows beween unemploymen and inaciviy. Wih his daa we can go back o 1989, which allows us o also capure he early 1990s recession. The correlaions beween he LFS and he claiman coun series are quie high: 0.99 beween he unemploymen raes, 0.90 beween he job-separaion raes and 0.91 beween he job-finding raes. Table 4 displays he imporance of each ransiion probabiliy using he LFS and claiman coun daa. When we use he hree-sae decomposiion, we find ha slighly more han 20 percen of he flucuaions in unemploymen can be aribued o flows beween inaciviy and he labour force. From he remaining, job-finding rae is slighly more imporan han he job-separaion rae (60-40 spli). When we do a wo-sae decomposiion, he jobseparaion rae is somewha more imporan, accouning for more han 50% of he volailiy of unemploymen, boh using he LFS (40-60 spli) and claiman coun daa (50-50 spli). There are lile differences beween Shimer s and Fujia and Ramey decomposiions. These values are in line wih he ones repored by Perongolo and Pissarides (2008). Using an alernaive decomposiion, hey find ha job-separaion rae has he same conribuion o unemploymen volailiy as he job-finding rae wih he LFS daa, while for he claiman coun since 1967, he conribuion of job-separaion rae has varied beween 25 o 45 percen. 8 8 One difference relaive o Perongolo and Pissarides (2008) is ha while hey compue he ransiion 13

16 Table 4: Unemploymen decomposiions LFS (1993:2-2010:1) Claiman coun Three Saes Two Saes (1989:1-2010:1) Shimer Shimer F & R Shimer F & R Employmen Unemploymen Employmen Inaciviy 0.01 Unemploymen Employmen Unemploymen Inaciviy 0.07 Inaciviy Employmen 0.07 Inaciviy Unemploymen 0.05 Noe: for Shimer, he value is he raio beween he covariance beween u ss and u ij and he variance of u ss. For Fujia and Ramey, he value correspond o raio beween he covariance beween du ss and du s and he variance of du ss. The series are previously seasonally adjused using he X12 Census programme and correced for ime aggregaion bias using he mehodology applied by Shimer (2007). For Shimer, he series are previously linearly derended. 5 Oher perspecives on he UK labour marke 5.1 Job-o-job flows Many economiss hink ha on-he-job search and job-o-job flows are imporan elemens of business cycles. For insance, Krause and Lubik (2007), building on he Pissarides (1994) on-he-job search model, concluded ha on-he-job search and job-o-job ransiions grealy amplify shocks o he economy. One advanage of he LFS, relaive o he US surveys, is ha i allows us o calculae job-o-job flows. I asks he responden wha year and monh i sared he curren job, making i possible o compue he lengh of he curren job enure. I coun as job-o-job ransiions he cases where an individual is employed in he firs quarer and employed in he second quarer bu wih a job enure of less han hree monhs. We should bear in mind ha his measure of job-o-job flows includes people ha changed job direcly as well as individuals ha had non-measured spells of unemploymen or inaciviy. In oher words, i includes individuals ha moved ou of employmen and back ino employmen wihin he quarer. The firs graph of Figure 3 plos he job-o-job flows as a percenage of he working populaion. Job-o-job flows increased from 1996 o 2001, bu have fallen since. As one expecs, here are subsanial differences in he ransiion probabiliies among employees probabiliies for all people above 15, I exclude people ou of reiremen age (60 plus for women and 65 plus for men). Because hey include hem, heir ransiion probabiliies are lower hen mine, paricularly ou of inaciviy. The cyclical properies do no seem o be affeced. 14

17 Job o job flows (% working age populaion) Figure 3: Job-o-job flows Job o job hazard rae by search saus Share of employed searching for a differen job Job o job flows Searching q1 1998q1 2001q1 2004q1 2007q1 2010q No searching Share of employed searching q1 1998q1 2001q1 2004q1 2007q1 2010q1 Searching No searching 1995q1 1998q1 2001q1 2004q1 2007q1 2010q1 Noe: he series are in a four-quarer moving average o remove seasonaliy and high frequency movemens. engaged in on-he-job search and he ones ha are no searching. If a worker is searching for a job, he probabiliy of changing job in any given quarer is, on average, 14%. If he is no searching, he probabiliy is only 2%. Each quarer, on average, 6.2% of workers are searching for a differen job. This is higher han he value of 5.2% found by Pissarides and Wadsworh (1994). All in all, roughly one hird of all he job changers were previously searching for a job. Evidence from he Unied Saes by Fallick and Fleischman (2004) suggess ha job-ojob flows are procyclical. We observe he same paern in he Unied Kingdom as we can see in Table 5. Job-o-job ransiion probabiliies are srongly procyclical as well as acual job-o-job flows. Some on-he-job search heories have differen predicions for he cyclicaliy of he number of employees searching for a differen job. For insance, he sylised model presened in Pissarides (2000) predics ha increasing produciviy leads o more people searching for jobs. Conversely, Nagypál (2008) argues ha workers underake less on-he-job search when hey face lower unemploymen risk, as is he case of expansions. In he Unied Kingdom, hese effecs seem o cancel ou such ha he number of employees searching for a differen job is no relaed o he business cycle. Table 5: Cyclical variaion of job-o-job flows Job-o-job flows (-10.12) Job-o-job hazard rae if searching (-9.97) Job-o-job hazard rae if no searching (-10.33) Employees searching (1.13) Noe: he cyclicaliy of he series is he coefficien on unemploymen rae in a regression of he series in logs on season quarer dummies, ime rend and unemploymen rae. T-saisics are in brackes.* denoes significan a 1% and ** significan a 5%. 15

18 5.2 Ouflows from employmen by reason Are separaions from employmen driven by firms or workers? The LFS allows us o spli he cause of employmen separaions ino hree caegories: involunary separaions, resignaions, or oher reasons. The firs caegory includes dismissals, erminaion of emporary employmen conracs or redundancies, which are involunary from he worker s poin of view. Resignaions include cases where he worker resigned, and also where hey ook volunary redundancy. Finally, oher reasons encompasses giving up work for healh, family or personal reasons or aking early reiremen. Roughly half of oal separaions from employmen are due o oher reasons and he oher half is due o resignaions and involunary separaions in equal shares. The graphs in Figure 4 plo, for he hree ypes of job separaions, he share caused by each reason. The flows from employmen o unemploymen are dominaed by involunary separaions. On average, hey accoun around 45% of oal employmen o unemploymen flows, bu a he peak of he recession hey accouned for 60% of he flows. In he beginning of he sample resignaions only accouned for 20% of oal employmen o unemploymen flows, in 2007 ha value was close o 30% bu i has fallen since. As expeced, oher reasons accouns for 70% of he employmen o inaciviy flows, wih he remaining being spli equally beween involunary separaions and resignaions. Finally, only a minoriy of he job-o-job flows are caused by involunary separaions. Around 50% of job-o-job flows are due o oher reasons and 35% are due o resignaions. Table 6 shows ha, as expeced, involunary separaions are srongly counercyclical, while resignaions are very procyclical. In economic slowdowns less people qui, which parially counerbalances he fac he more people lose heir jobs. Separaions by oher reasons are slighly procyclical which is consisen wih he incidence of personal reasons having a weaker relaionship wih he business cycle. Figure 4: Employmen ouflows by reason Employmen o Unemploymen flows by reason (% of oal) Employmen o Inaciviy flows by reason (% of oal) q1 1998q1 2001q1 2004q1 2007q1 2010q1 1995q1 1998q1 2001q1 2004q1 2007q1 2010q1 Job o job flows by reason (% of oal) 1996q4 1999q4 2002q4 2005q4 2008q4 Involunary separaions Resignaions Oher reasons Involunary separaions Resignaions Oher reasons Involunary separaions Resignaions Oher reasons Noe: he series are in a four-quarer moving average o remove seasonaliy and high frequency movemens. 16

19 Table 6: Cyclical variaion of employmen ouflows hazard raes by reason Toal Employmen o Employmen Job-o-job separaions unemploymen o inaciviy Involunary separaions (8.06) (10.39) (3.16) (-1.78) Resignaions (-12.23) (1.25) (-6.14) (-13.90) Oher reasons (-3.50) (2.14) (-2.46) (-5.11) Noe: he cyclicaliy of he series is he coefficien on unemploymen rae in a regression of he series in logs on season quarer dummies, ime rend and unemploymen rae. T-saisics are in brackes.* denoes significan a 1% and ** significan a 5%. When we disaggregae he separaions even more, we see ha he couner-cyclicaliy of he involunary separaions is mainly driven by he employmen-unemploymen flows while he procyclicaliy of resignaions is much sronger in he job-o-job ransiions. 5.3 Disaggregaing inaciviy Given he magniude of he flows in and ou of inaciviy, researchers have asked if some of he inacive should be considered as unemployed. Flinn and Heckman (1983) analysed condiional and uncondiional ransiion probabiliies beween he wo saes and concluded ha hey are essenially differen. In he Unied Kingdom, however, Joyce, Jones, and Thomas (2003) found ha many subgroups of he inacive have he same ransiion probabiliy o employmen as he unemployed. Blanchard and Diamond (1990) disaggregae he pool of inacive ino wo sub-groups: hose ha wan a job and hose ha do no wan a job. The inacive ha wan a job can be considered marginally close o he labour marke, and consequenly more likely o go ino he labour force. The LFS also allows for his disincion. The firs graph in Figure 5 shows he wo series as a percenage of he working-age populaion. On average a quarer of all he inacive wan a job. The relaive size of he wo groups has changed over he sample. While he pool of inacive ha wan a job had Figure 5: Inaciviy by sub-groups Inaciviy rae by objecive (% of working age populaion) Hazard raes from Labour force o Inaciviy by objecive Hazard raes from Inaciviy o Labour force by objecive Inacive wan a job q1 1998q1 2001q1 2004q1 2007q1 2010q1 ime Inacive do no wan a job q1 1998q1 2001q1 2004q1 2007q1 2010q q1 1998q1 2001q1 2004q1 2007q1 2010q1 Inacive wan a job Inacive do no wan a job Labour force o Inacive (wan) Labour force o Inacive (ou) Inacive (wan) o Labour force Inacive (ou) o Labour force Noe: he series are in a four-quarer moving average o remove seasonaliy and high frequency movemens. 17

20 a similar rend as he unemploymen rae, he pool of inacive ha do no wan a job has increased over he sample period. Table 7 repors he ransiion probabiliies beween he four groups. The inacive ha wan a job are wice as likely o join he labour force, and almos four imes more likely o join he pool of unemployed han he inacive ha do no wan a job. Addiionally, every quarer, 11% of he unemployed move ino inaciviy bu sill wan a job while only 6% move o inaciviy and do no wan a job. There are also relaively high ransiion probabiliies beween he wo groups of inacive. Around 21% of he inacive ha wan a job abandon heir inenions by he following quarer. I seems ha his sae is a limbo beween inaciviy and he labour force. Table 7: Transiion marix, Labour Force Survey (% per quarer) From: Employmen Unemploymen Inacive Inacive To: (wan) (ou) Employmen Unemploymen Inacive (wan) Inacive (ou) Noe: averages beween 1993:2 o 2010:1. Table 8 exhibis he cyclical properies of he gross flows and hazard raes beween he wo groups of he inacive and he labour force. The flows beween he inacive (ou) and he labour force are no relaed o he cycle, while all he flows beween he inacive (wan) and he labour force are counercyclical. In recessions, more people leave he labour force bu sill wan a job. Taking he evidence as a whole, here seems o exis a closer link beween he pool of he inacive ha wan a job and he labour force, paricularly unemploymen. Table 8: Cyclical variaion of flows in and ou of inaciviy Gross flows Hazard raes Labour force Inacive (wan) (10.20) (6.80) Labour force Inacive (ou) (-1.61) (-1.69) Inacive (wan) Labour force (3.91) (2.26) Inacive (ou) Labour force (-0.98) (0.17) Inacive (wan) Inacive (ou) (2.73) (1.01) Inacive (ou) Inacive (wan) (4.16) (5.27) Noe: he cyclicaliy of he series is he coefficien on unemploymen rae in a regression of he series in logs on season quarer dummies, ime rend and unemploymen rae. T-saisics are in brackes.* denoes significan a 1% and ** significan a 5%. 18

21 5.4 Flows by educaion Previous sudies on labour marke flows have paid relaively lile aenion o differences by levels of educaion. To explore such differences, I divide he working-age populaion ino hree groups depending on he level of educaion: higher educaion (Educaion 1 ), A-levels and GCSE or equivalen (Educaion 2 ) and below GCSE (Educaion 3 ). There has been a significan change in he UK economy over he pas decade, wih he share of working-age populaion wih higher educaion increasing from 20% in 1997 o above 35% in Over he same period, he share of he working-age populaion in he lowes educaion caegory fell from around 30% o below 20%. There are sriking differences across he hree educaion groups wih respec o employmen, unemploymen and inaciviy raes, as one can see in he firs row of Figure 6. The average employmen rae among he mos educaed is 87%, as opposed o 55% for individuals in he lowes educaion caegory. Boh he unemploymen rae and he inaciviy rae Figure 6: Labour marke socks and hazard raes by educaion Employmen rae by educaion 1998q1 2001q1 2004q1 2007q1 2010q Unemploymen rae by educaion 1998q1 2001q1 2004q1 2007q1 2010q Inaciviy rae by educaion 1998q1 2001q1 2004q1 2007q1 2010q1 Educaion 1 Educaion 2 Educaion 3 Educaion 1 Educaion 2 Educaion 3 Educaion 1 Educaion 2 Educaion Employmen o unemploymen hazard rae by educaion 1998q1 2001q1 2004q1 2007q1 2010q Employmen o inaciviy hazard rae by educaion 1998q1 2001q1 2004q1 2007q1 2010q Unemploymen o employmen hazard rae by educaion 1998q1 2001q1 2004q1 2007q1 2010q1 Educaion 1 Educaion 2 Educaion 3 Educaion 1 Educaion 2 Educaion 3 Educaion 1 Educaion 2 Educaion Unemploymen o inaciviy hazard rae by educaion 1998q1 2001q1 2004q1 2007q1 2010q Inaciviy o employmen hazard rae by educaion 1998q1 2001q1 2004q1 2007q1 2010q Inaciviy o unemploymen hazard rae by educaion 1998q1 2001q1 2004q1 2007q1 2010q1 Educaion 1 Educaion 2 Educaion 3 Educaion 1 Educaion 2 Educaion 3 Educaion 1 Educaion 2 Educaion 3 Noe: he series are in a four-quarer moving average o remove seasonaliy and high frequency movemens. 19

22 are monoonically decreasing in he level of educaion. Individuals in he lowes educaion caegory face an average unemploymen rae of 11%, almos four imes higher han he average unemploymen rae of hose wih higher educaion. The average inaciviy raes are, in increasing order of level of educaion, 38%, 19% and 11%. The difference beween educaion caegories exends o ransiion probabiliies, as shown in he remaining figures of Figure 6. For example, he average separaion raes from employmen o unemploymen are 0.9%, 1.4% and 1.9% as we go down he educaion ranking. The job-finding rae also presens significan differences. Individuals in he highes educaion caegory are wice as likely o find a job han individuals in he lowes educaion group. Table 9 presens he coefficien of he regression of each ransiion probabiliy wih he unemploymen rae. The cyclical properies of mos ransiion probabiliies are quie robus across levels of educaion. The job-finding rae is highly procyclical and he separaion rae from employmen o unemploymen is counercyclical. The probabiliy of moving from inaciviy o unemploymen is counercyclical a all levels of educaion. From he magniude of he coefficiens we can see ha individuals wih higher educaion face less cyclical flucuaions in ransiion probabiliies, han he individuals wih lower educaion. Table 9: Cyclical variaion of labour marke hazard raes by educaion Educaion 1 Educaion 2 Educaion 3 Employmen Unemploymen (5.40) (9.38) (5.70) Employmen Inaciviy (0.23) (-0.44) (-3.36) Unemploymen Employmen (-6.79) (-12.14) (-8.30) Unemploymen Inaciviy (-2.01) (-3.43) (-4.14) Inaciviy Employmen (-3.05) (-4.20) (-5.55) Inaciviy Unemploymen (3.90) (7.08) (5.31) Noe: he cyclicaliy of he series is he coefficien on unemploymen rae in a regression of he series in logs on season quarer dummies, ime rend and he unemploymen rae of he respecive caegory. T-saisics are in brackes.* denoes significan a 1% and ** significan a 5%. 6 Conclusion The objecive of his paper was o describe he main developmens in, and esablish a number of key facs abou, he recen hisory of he UK worker gross flows. I provided a picure of a wide range of informaion abou worker gross flows from differen angles, which is essenial o undersand he UK labour marke. The findings of his paper can be summarised as follows: 20

23 In each quarer, 7% of he working-age populaion change saus beween inaciviy, employmen and unemploymen and 2.1% of he working-age populaion change heir employer. Gross flows in and ou of he hree pools are counercyclical. In expansions, as he labour marke becomes igher here are fewer movemens beween he hree saes. Employmen o unemploymen flows are counercyclical, as well as he job-separaion rae. Unemploymen o employmen flows are counercyclical oo, bu he job-finding rae is procyclical. There are differences in aggregae condiional ransiion probabiliies. For example, he job-separaion probabiliy is 1% if he person was previously employed, 5% if inacive and 11% if unemployed. The job-finding rae is 20% if he person has been unemployed for wo periods, bu i is 48% if he person was previously employed. This suggess he presence of heerogeneiy or hisory dependence. The job-finding rae and job-separaion rae are equally imporan deerminans of unemploymen flucuaions. Every quarer, 6% of all employees are searching for a differen job. They are seven imes more likely o change jobs han he ones no searching. Job-o-job ransiion probabiliy is srongly procyclical, bu he number of employees searching for a differen job is acyclical. Resignaions are srongly procyclical while involunary separaions (layoffs) are counercyclical. Involunary separaions dominae he employmen o unemploymen flows while 70% of all employmen o inaciviy flows occur because of oher reasons. Only 15% of he job-o-job flows are driven by involunary separaions. The inacive ha wan a job are wice as likely o move ino he labour force and four imes more likely o move ino unemploymen han he inacive ha do no wan a job. There are subsanial differences in he employmen, unemploymen and inaciviy raes of differen educaion caegories, as well as in ransiion probabiliies. Individuals in he lowes educaion caegory face a hree imes higher unemploymen and inaciviy rae, wice as high separaion rae and half he job-finding rae, han individuals in he highes educaion caegory. 21

24 In addiion o hese findings, i is also relevan o menion ha, eiher due o he presence of heerogeneiy or hisory dependence, esimaing annual gross flows or ransiion probabiliies based on he quarerly ransiion probabiliies overesimaes he acual ones. This suggess ha one should be cauious when comparing resuls from surveys carried ou a differen frequencies, which ofen happens beween he Unied Saes, Unied Kingdom and oher European economies. References Baker, M. (1992): Unemploymen Duraion: Composiional Effecs and Cyclical Variabiliy, American Economic Review, 82(1), Bell, B., and J. Smih (2002): On gross worker flows in he Unied Kingdom: evidence from he Labour Force Survey, Bank of England working papers 160. Blanchard, O., and J. Gali (2010): Labor Markes and Moneary Policy: A New Keynesian Model wih Unemploymen, American Economic Journal: Macroeconomics, 2(2), Blanchard, O., and P. Porugal (2001): Wha hides behind an unemploymen rae: comparing poruguese and U.S. labor markes, American Economic Review, 91(1), Blanchard, O. J., and P. Diamond (1990): The cyclical behavior of he gross flows of U.S. workers, Brookings Papers on Economic Aciviy, 21(2), Bleakley, H., A. E. Ferris, and J. C. Fuhrer (1999): New daa on worker flows during business cycles, New England Economic Review, (July), Burgess, S., and H. Turon (2005): Unemploymen dynamics in Briain, Economic Journal, 115(503), Clarke, P., and P. Tae (2000): Mehodological issues in he producion and analysis of longiudinal daa from he labour force survey, GSS Mehodology Series 17, Office for Naional Saisics. Davis, S. J. (2006): Commens on Job Loss, Job Finding, and Unemploymen in he U.S. Economy over he Pas Fify s by Rober E. Hall, in NBER Macroeconomics Annual 2005, ed. by M. Gerler, and K. Rogoff, vol. 1, pp The MIT Press. 22