Universidade Técnica de Lisboa Insiuo Superior de Economia e Gesão Mesrado em Economeria Aplicada e Previsão Labor Marke Flows afer he redesign of he Poruguese Labor Force Survey David Leie das Neves Orienação: Professor Douor Mário José Gomes de Freias Ceneno Júri: Presidene: Vogais: Professor Douor José Manuel de Maos Passos Professor Douor Mário José Gomes de Freias Ceneno Professor Douor Anónio José Franco Gomes de Menezes ISEG/UTL, Fevereiro de 2014
Labor Marke Flows afer he redesign of he Poruguese Labor Force Survey David Leie das Neves M.Sc.: Applied Economerics and Forecasing Supervisor: Mário José Gomes de Freias Ceneno Viva Voce Exam in: February 07, 2014. Absrac In he rs quarer of 2011, he Poruguese Labor Force Survey underwen a major redesign. The redesign subsanially changed he mehodology of collecing daa and conceps of employmen, unemploymen and inaciviy. Since hen, he labor marke ows became subsanially high compared wih hisorical sandards. This sudy documens he deniional changes and impacs of he redesign on labor marke ows daa. In addiion, i provides a sysemaic sudy abou he Poruguese labor marke ows for he period beween 1999 and 2012. LFS microdaa are used o compue labor marke ows in he usual hree-sae se-up, he ows disaggregaed by micro-characerisics of he respondens and he ows in a four sae se-up, wih employmen disaggregaed ino permanen and emporary, and inaciviy disaggregaed ino hose who wan a job and hose who do no wan a job. Then, we make use of regression disconinuiy mehods o invesigae he redesign eec in he ows series, conrolling heir cyclical and seasonal movemens. The resuls sugges ha he jump in he ows series is mosly explained by he redesign's eec, despie of he counercyclical paern found a some ows. Furhermore, he redesign eec seems o concenrae in specic groups, namely on he employed wih a xed-erm conrac, on he inacives who wan a job and on he older and less educaed respondens. On a second sage, we apply a se of cross-secional and longiudinal mehods o model he eec of he redesign in he probabiliy o move ou of employmen. The new mehodology appears o have increased he probabiliy o record ransiions ou of employmen. This appears o be more a resul of a redesign eec caused by he major mehodological changes han of dierences beween samples ha could posiively inuence he likelihood o observe a higher level of ransiions ou of employmen afer he redesign. Keywords: labor marke dynamics, labor marke ows, labor force survey. JEL codes: E24, E32, J60.
Os Fluxos do Mercado de Trabalho após as alerações ao Inquério ao Emprego David Leie das Neves M.Sc.: Economeria Aplicada e Previsão Orienação: Mário José Gomes de Freias Ceneno Provas concluídas em: 7 de Fevereiro de 2014. Resumo No primeiro rimesre de 2011, o Inquério ao Emprego Poruguês foi amplamene redesenhado, alerando-se subsancialmene a meodologia de recolha de dados, bem como os conceios esaísicos de emprego, desemprego e inacividade. Desde enão, os uxos enre esados do mercado de rabalho ornaram-se subsancialmene elevados, considerando o seu padrão hisórico. O presene esudo descreve as alerações concepuais que advieram do redesenho do inquério e preende analisar as suas implicações nas séries dos uxos. De forma complemenar, é feio um esudo sisemáico sobre os uxos enre esados do mercado de rabalho em Porugal durane o período de 1999 a 2012. A parir dos microdados do Inquério ao Emprego são calculados os uxos agregados enre os rês esados habiuais do mercado de rabalho, eses são poseriormene desagregados por caracerísicas micro dos inquiridos, e por úlimo calculam-se os uxos num modelo a quaro esados, onde o emprego é desagregado enre emporário e permanene, e a inacividade é desagregada enre inacivos que querem um emprego e inacivos que não querem um emprego. Em seguida, usamos méodos de regressão desconínua para invesigar os efeios das alerações ao inquério nas séries dos uxos, conrolando as suas uuações cíclicas e sazonais. Os resulados sugerem que o aumeno brusco dos uxos é fundamenalmene devido ao redesenho do Inquério ao Emprego, não obsane do comporameno conracíclico exibido por algumas séries. Adicionalmene, exise evidência que os uxos dos empregados com conracos de rabalho a ermo, dos inacivos que querem um emprego e dos indivíduos de idade superior e menor escolaridade, foram paricularmene aingidos pelas alerações meodológicas. Num fase poserior, são usados méodos seccionais e longiudinais para quanicar o efeio do redesenho na probabilidade de ransiar do emprego para o não-emprego. Os resulados sugerem que sob o novo inquério a probabilidade de se regisarem ransições do emprego para o não-emprego é superior, comparaivamene com o anerior méodo. Conudo, al parece resular mais das consideráveis alerações meodológicas, do que de alerações na composição da amosra do novo inquério que favoreçam a observação de ransições do emprego para o não-emprego. Palavras-chave: dinâmica do mercado de rabalho, uxos do mercado de rabalho, inquério ao emprego. Códigos JEL: E24, E32, J60.
Acknowledgmens I would like o express my graiude o Álvaro A. Novo and Mário Ceneno for heir guidance hroughou his sudy. I am also graeful o Lucena Vieira for he very useful ips she gave me abou he compuaion of worker ows. This sudy beneed from nancial suppor of Banco de Porugal, which is also graefully acknowledged.
Conens 1 Inroducion 1 2 Preliminary Conceps 2 2.1 Labor marke ows........................................ 2 2.2 Theoreical perspecives..................................... 2 2.3 Previous empirical evidence................................... 3 3 The Poruguese Labor Force Survey 5 3.1 The LFS redesign......................................... 5 3.1.1 General changes..................................... 5 3.1.2 Changes in labor marke saes deniion...................... 5 3.1.3 Oher changes....................................... 8 3.2 Consequences of he redesign.................................. 8 4 Worker Flows in he Poruguese Labor Marke 12 4.1 Consrucing worker ows.................................... 12 4.2 Aggregaed worker ows..................................... 13 4.2.1 Seasonaliy........................................ 13 4.2.2 Sae dependence..................................... 15 4.2.3 Regression disconinuiy approach.......................... 16 4.2.4 Resuls........................................... 17 4.3 Disaggregaed worker ows................................... 19 4.3.1 Flows by educaion................................... 19 4.3.2 Flows by age cohor................................... 21 4.4 A four-sae model of labor marke ows........................... 26 4.4.1 Disaggregaing employmen.............................. 26 4.4.2 Disaggregaing inaciviy................................ 30 5 Ous of Employmen 33 5.1 Cross-secional approach..................................... 33 5.1.1 Resuls........................................... 34 5.2 Longiudinal approach...................................... 39 5.2.1 Resuls........................................... 39 5.3 Ou-of-sample predicions.................................... 40 6 Conclusions 43 Appendix 45 A A noe on ransiion probabiliies B A noe on survey daa analysis C Esimaion Resuls References i ii iv xvi i
Lis of Figures 3.1 Average quarerly worker ows, 1999-2012.......................... 8 3.2 Quarerly GDP growh rae................................... 9 3.3 Means of populaion characerisics.............................. 11 4.1 Labor marke socks, gross ows and hazard raes..................... 14 4.2 Labor marke ows by educaion................................ 22 4.3 Hazard raes by educaion.................................... 23 4.4 Labor marke ows by age cohor............................... 24 4.5 Hazard raes by age cohor................................... 25 4.6 Labor marke ows: disaggregaing employmen...................... 28 4.7 Labor marke ows: disaggregaing inaciviy........................ 31 ii
Lis of Tables 3.1 Job search diligences by ype.................................. 7 4.1 Seasonaliy of labor marke ows since Spring 1999..................... 13 4.2 Average condiional ransiion probabiliies......................... 15 4.3 RD esimaes: Employmen, Unemploymen and Inaciviy ouows.......... 18 4.4 Average gross ows by educaion (% labor force)...................... 20 4.5 Average gross ows by age cohor (% labor force)..................... 21 4.6 Transiion marix, disaggregaing employmen (% per quarer).............. 27 4.7 RD esimaes: Employmen ouows, disaggregaing employmen........... 29 4.8 RD esimaes: Unemploymen and Inaciviy ouows, disaggregaing employmen. 29 4.9 Transiion marix, disaggregaing inaciviy (% per quarer)............... 30 4.10 RD esimaes: Employmen and Unemploymen ouows, disaggregaing inaciviy 32 4.11 RD esimaes: Inaciviy ouows, disaggregaing inaciviy............... 32 5.1 Probi model for ransiions ou of employmen...................... 35 5.2 Probi model for ransiions ou of employmen (Con.)................. 36 5.3 LPM for ransiions ou of employmen........................... 37 5.4 LPM for ransiions ou of employmen (Con.)...................... 38 5.5 Average Transiions........................................ 40 5.6 Longiudinal LPM for ransiions ou of employmen................... 41 5.7 Longiudinal LPM for ransiions ou of employmen (Con.).............. 42 C.1 Means by populaion characerisics.............................. v C.2 Means by indusry of he employed.............................. vi C.3 Means by occupaion of he non-employed.......................... vi C.4 RD esimaes: Employmen ouows............................. vii C.5 RD esimaes: Unemploymen ouows............................ vii C.6 RD esimaes: Inaciviy ouows............................... viii C.7 RD esimaes: Employmen ouows, conrolling for business cycle eecs...... viii C.8 RD esimaes: Unemploymen ouows, conrolling for business cycle eecs..... ix C.9 RD esimaes: Inaciviy ouows, conrolling for business cycle eecs........ ix C.10 RD esimaes: Employmen ouows by educaion..................... x C.11 RD esimaes: Unemploymen ouows by educaion................... xi C.12 RD esimaes: Inaciviy ouows by educaion....................... xii C.13 RD esimaes: Employmen ouows by age cohor.................... xiii C.14 RD esimaes: Unemploymen ouows by age cohor................... xiv C.15 RD esimaes: Inaciviy ouows by age cohor...................... xv iii
1 Inroducion The labor marke ows are crucial o our undersanding of labor marke dynamics. They drive movemens in aggregaed indicaors such as he employmen, unemploymen and inaciviy raes, heir size is frequenly aken as a proxy of he labor marke exibiliy and hey lie a he hear of sae-of-ar of search and maching models of unemploymen (Morensen and Pissarides, 1994). In he rs quarer of 2011, he Poruguese Labor Force Survey (LFS) was widely redesigned, which caused major changes eiher in he deniions of he labor marke saes or in he mehodology used o collec daa. Since hen, he labor marke ows became subsanially high compared wih hisorical sandards. The aim of his paper is o provide a sysemaic sudy of he deniional changes and impacs of he redesign on he labor marke ows. In addiion, i esablishes a number of key facs abou he Poruguese labor marke ows for he period beween 1999 and 2012. To address his issues, we follow wo main empirical sraegies. In he rs, regression disconinuiy mehods are used o quanify he redesign's eec on he aggregaed ow series, conrolling heir seasonal and cyclical properies. On a second sage, his exercise is repeaed for labor marke ows disaggregaed by age and educaion levels of he respondens. In order o consider he wo-ier srucure of he Poruguese labor marke and he dierenial behavior wihin he pool of inacives, we rs disaggregae employmen beween xed-erm and open-ended conracs, and hen inaciviy beween hose who wan a job and hose who do no wan a job. Overall, he redesign is found o explain a large par of he jump of he ows. In paricular, he disconinuiy in ows beween employmen and unemploymen seems o be mosly explained by he eec of he redesign in he older respondens wih a xed-erm conrac. Similarly, he disconinuiy in ows beween employmen and inaciviy appears o be due o he eec of he redesign in ows of he older and less educaed respondens moving beween a xed-erm conrac and inaciviy. In addiion, evidence suggess ha he jump in ows beween unemploymen and inaciviy is deermined by he redesign impac in ows beween he unemployed and he inacives who wan a job (he smalles group wihin he pool of inacives). Regarding he business cycle characerisics of he ows, one nds ha ows from employmen ino unemploymen are counercyclical, as well as ows beween inaciviy and unemploymen, and ows from permanen employmen ino emporary employmen. In conras, he ows from he inacives who wan a job ino unemploymen appear o be procyclical. On he oher hand, we nd ha he reacion of he ows series o recessions is no signicanly dieren beween recessionary quarers before and afer he redesign. Hence, i is essenially he redesign ha deermines he jump of he ows, even when hey exhibi a signican counercyclical behavior. In he second approach, we apply cross-secional and longiudinal mehods o model he probabiliy o exi employmen over he samples before and afer he redesign. Overall, he redesign appears o have increased he probabiliy o record ransiions ou of employmen. Moreover, he ou-of-sample ransiions compued over he sample before using he coeciens ed in he sample afer, sugges ha he mehodological changes may lie behind he higher level of employmen exis afer, despie he composiional dierences beween samples. The disseraion is organized as follows. Secion 2 conains preliminary conceps and convenions and provides an overview of he heoreical models and previous evidence on he cyclical properies of labor marke ows. Secion 3 describes he deniional changes brough by he redesign. Secion 4 quanies he impac of he redesign in he aggregaed ows, he ows disaggregaed by micro-characerisics and in he four-sae se-up, wih employmen and inaciviy spli as menioned above. Secion 5 models he redesign's eec in he probabiliy o exi employmen and compues he average ransiions before wih he coeciens ed afer. Secion 6 concludes. 1
2 Preliminary Conceps 2.1 Labor marke ows We begin by presening he fundamenal equaions ha describe he evoluion of he sock of employed E, he sock of unemployed U and he sock of inacive or ou of he labor force I, which sum he populaion W (i.e., W = E + U + I). Adding he rs wo pools i gives us he labor force L (i.e., L = E + U). The unemploymen rae is given by u = U L. Toal employmen evolves according o he following dierence equaion: where N XY E = E 1 + N UE + N IE N EU N EI, (2.1) is he labor marke ow from sae X ino sae Y, i.e., he number of people enering Y from X beween 1 and, where X, Y {E, U, I}. Alernaively eq. (2.1) may be wrien as a funcion of he hazard raes λ XY (i.e., he probabiliy o move from one sae o anoher): E = E 1 (λ EI + λ EU )E 1 + λ IE I 1 + λ UE U 1. (2.2) A similar decomposiion can be performed for unemploymen and inaciviy: U = U 1 + N EU + N IU N UE N UI, (2.3) I = I 1 + N UI Alernaively, focusing on he hazard raes: U = U 1 (λ UE + λ UI + N EI N IU N IE. (2.4) )U 1 + λ EU E 1 + λ IU 1 I 1, (2.5) I = I 1 (λ IE + λ IU )I 1 + λ EI E 1 + λ UI U 1. (2.6) Before proceeding, i is useful o ouline our noaion and clarify he precise meaning of some conceps of which we will make use. Research on labor marke ows considers wo varieies of ows: gross ows of workers and ows of jobs. Job ows measure wheher a new posiion has been creaed or desroyed by a rm, raher han changes in he labor saus of a worker. The gross worker ows measure ransiions in labor marke saus of workers. Throughou his paper we will focus only on gross worker ows, o which we indisincively refer as labor marke ows. We use he symbol o denoe a ow from one labor marke sae o anoher (for example, he ows from employmen ino unemploymen are abbreviaed as E U), we also use he symbol o denoe ows beween wo labor marke saes (e.g. E U will denoe E U and U E ows). Labor marke ows have been sudied hrough wo main approaches: he analysis of worker ows or he analysis of he hazard raes. Some auhors like Blanchard and Diamond (1990), Burda and Wyplosz (1994) or Davis e al. (2006) focus on worker ows, while more recen sudies of Fujia and Ramey (2009) or Shimer (2012) focus on hazard raes. The wo perspecives could be considered complemenary and numerous auhors explore boh in he analysis of labor marke. We will also explore boh, bu more emphasis will be given o he ows approach. 2.2 Theoreical perspecives Research by Blanchard and Diamond (1990) suggess a model of primary and secondary workers o jusify why ows beween employmen and unemploymen were markedly dieren from hose beween employmen and inaciviy. In heir model, primary workers have srong aachmen o he labor force, brief spells of unemploymen and only separae from jobs involunarily. In 2
conras, secondary workers have much weaker labor force aachmen and are likely o spend signican ime boh in unemploymen and inaciviy. Firms perceive hese workers dierenly and, as consequence, prefer o hire primary workers when available, and hey are mosly available during recessions when masses of hem suer involunary separaions. Secondary workers are hired only in booms, when primary workers are no available. Since secondary workers are ofen inacive his means ha ows from inaciviy o employmen are procyclical (ha is, hey go up in upurns and down in downurns). Laer, Blanchard and Diamond (1992) argue ha we migh expec movemens from employmen o unemploymen (or inaciviy) o be counercyclical, however ows from unemploymen o employmen are no necessarily procyclical, as we migh inuiively hink. In fac, heir proposed maching funcion implies ha, ceeris paribus, a larger sock of unemployed may lead o more hires, such ha ows from unemploymen o employmen may acually increase during a recession, sill he associae hazard rae necessarily goes down. Pissarides (2000) also suggess ha we migh expec ows from inaciviy ino boh employmen and unemploymen o be procyclical, paricularly as labor marke ighness rises and as he employmen rae increases. We have been highlighing he predicions of heoreical models for he cyclicaliy of ows beween he dieren labor marke saes, however when discussing he cyclical properies of worker ows one also needs o ake some accoun wha Bleakley e al. (1999) called movemens a oher frequencies, i.e., oher relevan facors o explain worker ow ucuaions han business cycle phases. According o he auhors, hese movemens a oher frequencies occur a a macro level ( higher frequencies) and a a micro level (lower frequencies). A he higher frequencies are essenially movemens relaed wih seasonal facors (e.g., he movemens in and ou of he labor force relaed o he academic calendar) while a he lower frequencies are movemens relaed wih changes in characerisics of he populaion (age srucure, paricipaion of women, educaion and ohers). To summarize, we expec he ows ou of employmen o be counercyclical, ows from inaciviy ino boh employmen and unemploymen o be procyclical, and no clear paern for ows from unemploymen o inaciviy. The hazard rae from unemploymen o employmen should be procyclical, while he acual ow may be counercyclical. Finally, seasonal facors and changes in demographics also play a major role in he ucuaions of worker ows daa. 2.3 Previous empirical evidence There is a number of empirical sudies ha focus on individual European counries, examples are Bell and Smih (2002) or more recenly Gomes (2012) for Unied Kingdom, Schmid (1999) for Germany, Hairaul e al. (2012) for France, Balakrishnan (2001) or Silva and Vázquez-Grenno (2012) for Spain and Blanchard and Porugal (2001) or Ceneno and Novo (2013) for Porugal. Burda and Wyplosz (1994) buil a series of sylized facs for France, Germany, Spain and UK. The ndings of he cied sudies are broadly consisen wih he exisence of a common paern for he cyclical properies of labor marke ows across counries. Overall, ows from employmen o unemploymen are found o be counercyclical, as he associaed hazard rae. The reverse ow is also found o be counercyclical, while is hazard rae is procyclical. Flows beween inaciviy and employmen appear o be procyclical, while ows beween inaciviy and unemploymen are counercyclical. Conradicory evidence was found by Balakrishnan (2001) for Spain and by Gomes (2012) for UK, where unemploymen ouows end o be pro raher han counercyclical. In he case of he UK, no clear paern is found for ows from employmen o inaciviy (while for he reverse ow a procyclical paern emerge, consisenly wih he cross-counry paern). 3
Blanchard and Diamond (1990) and Bleakley e al. (1999) reached a number of common ndings abou he cyclicaliy and oher properies of gross worker ows in he US, even hough he wo sudies used dieren sample periods. 1 Flows beween unemploymen and employmen are found o be counercyclical, whereas ows beween inaciviy and employmen are found o be procyclical and no clear paern is found for ows beween inaciviy and unemploymen. Blanchard and Diamond explain his hrough heir model of primary and secondary workers discussed previously. Bleakley e al. (1999) go furher and nd ha he volailiy of employmen ouows is signicanly larger han employmen inows and conclude from his ha i is essenially job desrucion who drives he business cycle properies of worker ows in he US labor marke. More recenly, Kahn and McEnarfer (2013) use adminisraive daa o compue labor marke ows for US and conclude ha he evoluion of employmen over he business cycle is essenially driven by he hiring decisions of rms, namely he high-qualiy ones. Also using adminisraive sources, Ceneno and Novo (2013) reached a similar conclusion for Porugal: hirings have a larger conribuion o he business cycle han separaions, which is primarily explained by he behavior of he larges rms ha reduce heir workforce more srongly in recessions and lead hirings in upurns. We shall have more o say on his ndings laer in his paper. As a whole, he ndings on he paern of American labor marke ows are similar o hose found in he European counries. However, we should bear in mind some issues regarding heir comparabiliy. Firs, some sudies use daa from household surveys, while ohers use daa from adminisraive sources, and herefore hey are no direcly comparable. Second, hey are based on daa from dieren frequencies (in he US daa is colleced a a monhly basis, whereas in mos of he sudies for European counries daa comes from quarerly surveys) which rises he so-called muliple ransiions problem. A simple example illusraes his issue: if a worker moves from inaciviy o employmen via unemploymen in a shor period of ime, i will probably be regisered as an inaciviy o employmen ransiion in a quarerly survey, while a monhly survey would pick up he iniial I U followed by he U E ransiion. For example, Blanchard and Porugal (2001) comparing he US o Porugal, conclude ha on an annual basis he wo economies have similar worker ows, bu a a quarerly frequency Porugal has ows only of one-quarer of hose in he US. However, he auhors do no correc for ime aggregaion bias in he exrapolaion of ows a he dieren frequencies for which he surveys are carried, which migh generae serious biases and misleading conclusions. To summarize, he sudies oulined above provide evidence ha ows from employmen ino unemploymen exhibi a counercyclical paern, whereas he reverse ows have an heerogeneous behavior beween economies, being counercyclical in some and procyclical in ohers. In he ows beween inaciviy and employmen a procyclical paern is widely found, while in he ows beween inaciviy and unemploymen he cied sudies poin for a counercyclical behavior in Europe and idenify no clear paern in US. 1 Blanchard and Diamond (1990) consider Curren Populaion Survey (CPS) daa from 1968 o 1986, while Bleakley e al. (1999) use CPS daa from 1976 o 1999. 4
3 The Poruguese Labor Force Survey 3.1 The LFS redesign The LFS is a major source of informaion abou he Poruguese labor marke. In addiion o providing quarerly esimaes of he employmen, unemploymen and inaciviy, economiss and policymakers use daa from he LFS o examine broad socieal and cyclical changes in economic aciviy. I includes around 40 000 households quarerly ha are seleced o represen he populaion in he naion and in each region. 2 The probabiliy sample of housing unis is drawn using a mulisage sraicaion procedure, where he meropolian areas wihin each region are included and he remaining areas of a region are sampled on a probabiliy basis buil upon he populaion CENSUS, wih he probabiliy of selecion being proporionae o he populaion of he area. Households are inerviewed for six consecuive quarers, such ha each quarer 1/6 of he sample is roaed ou and 5/6 of he sample is reained, allowing us o observe he labor force saus in he quarer 1 and for 5/6 of he workers, and herefore compue labor marke ows or ransiion raes. We had access o he micro daa for he 1998-2012 period. In he rs quarer of 2011, he Poruguese Saisical Oce (Insiuo Nacional de Esaísica, henceforh INE) revised he quesionnaire and swiched o compuer-assised elephone inerviewing collecion procedure (CATI). In his secion we will describe he main changes. 3.1.1 General changes In he revised LFS he rs inerview in each of he six consecuive inerview quarers is conduced hrough a personal visi, in he subsequen quarers he inerviews are conduced over he phone, his mehod is known by compuer-assised elephone inerviewing (CATI). The respondens however have he opion o answer all he six inerviews hrough a personal inerview (CAPI) as in he old procedure. The roaion paern esablished prior o he redesign was mainained, as well as he mulisage sample scheme. 3.1.2 Changes in labor marke saes deniion Hereafer we will focus only on he pah of individuals wih 15 years old or older, since hose who are less han 15 are immediaely classied as inacive in boh surveys. In he previous LFS, he disincion beween employed and non-employed was done by asking he following muually exclusive quesions: Las week did you have a paid work, eiher occasional or for jus one hour?, Las week did you have a non-paid work for a relaive or for self-supply? and Regardless of no having done a paid or non-paid work, do you have any job or business from which you have been absen las week?. Those who answered armaively o one of he previous quesions were classied as employed, while a negaive answer implied o be classied as non-employed. Among he non-employed, he disincion beween unemployed and inacive was also sraighforward. Respondens were rs asked if hey acively search for a job, even for a par ime one or for self-employmen. Those who said yes and were available o sar working immediaely, or a leas in he nex wo weeks, were classied as unemployed. Those who said no were hen asked if hey expeced o be called for a job in he nex hree monhs and were available o sar working immediaely or a leas in he nex wo weeks (i sounds redundan, bu ha is how i was). If hey answered armaively o boh quesions, hey were classied as unemployed, oo. Consequenly, he non-employed individuals classied as inacive were hose who were neiher acively searching 2 More precisely he NUTS, he regions for saisical purposes. 5
for a job nor expeced o be called for a job in he nex hree monhs and/or were no available o sar working immediaely or a leas in he nex wo weeks. The classicaion algorihm in he old mehod was raher simple when compared o he one of he revised LFS. Le us begin by considering he following cases: 1. Worked las week and received compensaion. 2. Worked las week bu received no compensaion. 3. Did no work las week and did no have any job or business from which (s)he has been absen las week. 4. Did no work las week bu has a job or business from which (s)he has been absen las week. In cases 1 and 3 respondens are direcly classied as employed and non-employed, respecively. Whereas in case 2 respondens were hen asked a se of quesions in order o deermine wha kind of work hey had. If he non-paid work was due o an inernship or agriculure and sheries on own accoun and for self-supply, bu whose impac on he family budge is irrelevan, hen individual is re-allocaed o cases 3 or 4. In he remaining siuaions of case 2 he respondens are classied as employed. 3 In case 4 he responden's classicaion depends on he reasons for he absence. If maerniy/paerniy leave, vacaions or healh reasons are invoked, hen he responden is classied as employed. If he responden has a new job ha (s)he hasn' sared ye, (s)he is classied as non-employed. If oher reasons are repored, he responden is hen asked how long he absence is expeced o las. 4 In ha siuaion we have o consider he cases: 5. The absence is expeced o las hree or less han hree monhs. 6. The absence is expeced o las for more han hree monhs or for an unknown period. Individuals in case 5 are classied as employed, while for hose in case 6 furher renemen is done. If he respondens receive no compensaion or less han an half of heir wage, hey are hen classied as non-employed, whereas hose whose absence is ha of case 6 and are receiving an half or more of heir wage are classied as employed. Before proceeding, i is useful o poin ou how he probabiliy o be classied as non-employed increased wih he redesign, even for individuals in he same circumsances in he fourh quarer of 2010 and in he rs quarer of 2011. For example, an individual in cases 2 or 4 would have been immediaely classied as employed before he redesign, while in he revised survey he same individual has serious chances o be classied as non-employed. Having disenangled he rules underlying he classicaion of individuals as employed or nonemployed, now we aemp o disinguish hose ha are classied as unemployed or as inacive from he pool of non-employed. For his disincion, o declare availabiliy o sar working in a given period is crucial, hus respondens are asked: If you had found a job las week, would you be available o sar working las week or in he nex feen days?. Those who say no are direcly classied as inacive, whils he classicaion of he ohers depends on specic job search quesions. 3 Such as working for a relaive from whom he individual depends on and working in agriculure/sheries wih commercial purposes or for self-supply wih impac on he family budge. 4 Such as lay-o, seasonal work, leave wihou pay, srikes or oher labor conics, academic and professional raining and ohers. 6
Table 3.1: Job search diligences by ype Acive job search diligences Conacs wih privae job agencies Conacs direcly wih employers Conacs wih own social conecions or unions Publishes, replies or searches for job announcemens Has done job inerviews or recruimen ess Searches for land plos, faciliies or equipmens Passive job search diligences Sough licences or nancial resources Is waiing he resuls of a job applicaion Is waiing o be conaced by he Employmen Oce Is waiing he resuls of a public ender Again i is useful o ake ino accoun he following scenarios: 7. Have been looking for work las week or hree weeks before and: (a) Conaced an Employmen Oce (Cenro de Emprego) for he purpose of: i. Enroll as unemployed for he rs ime, ge informed abou a specic job oer received from he Employmen Oce or oher possible job oers. ii. Renew he enrollmen, apply o professional rainning (and ohers). (b) Have done acive job search diligences. (c) Have done passive job search diligences. 8. Have no been looking for work las week or hree weeks before. In he revised LFS, i is no enough o declare o be acively engaged in job search, in fac, for hose who sae o have been looking for work (case 7) are jus considered unemployed hose who have done, a leas, he diligences of cases 7(a)i and 7b (Table 3.1 summarizes wha i is mean by acive and passive diligences). In urn, individuals in case 7 whose conac wih Employmen Oce was due o he purposes of case 7(a)ii or whose job search diligences were hose of case 7c, are hen classied as inacive. In general, hose of case 8 do no mee he crieria o be classied as unemployed, however, here is one excepion: when he respondens have already found a job where hey will sar working in he nex hree monhs. To summarize, he LFS redesign brough major deniional changes in he conceps of employmen, unemploymen and inaciviy. In he revised LFS i is no sucien o repor a paid or non-paid work, or an absence from work o be classied as employed. Now, he specic naure of he non-paid work, he reasons for he absence and he individuals expecaions of how long he absence will las, play a major role in heir classicaion as employed or non-employed. In a similar way, he disincion beween unemployed and inacive is no merely based on he responden personal assessmen of his acive job search. The concep of unemployed incorporaes he specic job search diligences made by he responden. If hose were considered acive, he responden is classied as unemployed, oherwise (s)he joins he pool of inacive. I is no he case ha hose diligences were no asked before he redesign (hey were, even wih a raher similar wording), bu now hey are deerminan o he responden's classicaion. 5 5 In fac, before he redesign all he response caegories correspond o he curren acive diligences (wih he excepion of Sough licenses or nancial resources), he passive diligences were added o he revised LFS. 7
Figure 3.1: Average quarerly worker ows, 1999-2012 Source: Auhor's calculaions based on Poruguese LFS. Noe: The worker ows are expressed as a percenage of he labor force. The averages are compued over he samples: (a) 1999:1-2010:4 and (b) 2011:1-2012:4. 3.1.3 Oher changes There were oher changes eiher in he wording or in he srucure of he quesionnaire ha could be of ineres. One of hem has o do wih he quesion where he employed respondens were asked abou he ype of conrac hey have. Before he redesign, here were ve response caegories: (1) open-ended (permanen) conrac; (2) xed-erm (emporary) conrac; (3) selfemployed (recibos verdes); (4) seasonal work wihou formal labor conrac; and (5) odd jobs (biscaes). In he new survey he las wo caegories were dropped ou, while he rs hree were kep unchanged. Anoher change occurred in he quesion where he majoriy of unemployed provide heir reasons for unemploymen: he revised LFS no longer considers volunary quis as a response caegory. In he lieraure on labor marke ows, i is frequenly subjec of research (Anderson and Meyer, 1994; Bleakley e al., 1999; Gomes, 2012) he sensiiviy of hree classes during business cycle ucuaions: hose who are volunarily unemployed (volunary quis) and hose who are involunarily unemployed (involunary quis) and a broader caegory usually named ohers (ha is a cach-all caegory for a variey of dieren reasons). However, he ineres lies predominanly in he dierenial experience of he rs wo, which wih he referred mehodological change inroduces serious limiaions on daa o perform his analysis for he Poruguese labor marke. 3.2 Consequences of he redesign Poerba and Summers (1986) invesigaed he impacs of misclassicaion errors for he indicaors derived from he CPS. They found ha ambiguiies in he survey quesions, recording errors or simple misakes on he par of he respondens cause he measured ows o dramaically oversae acual movemens in he labor marke and, o a lesser degree, lead o biases in he surveyed characerisics of he respondens. Polivka and Miller (1998) assessed he impac of he CPS redesign for various aggregaed measures derived from i. They found ha he redesign had no signican eec on major saisics, such as he unemploymen rae, bu i grealy aeced some disaggregaed measures, as worker ows and he surveyed characerisics of he respondens. 6 6 The CPS redesign occurred in January 1994, he process had some similariies wih he redesign of he Poruguese LFS, namely he adopion of CATI and he inroducion of a ime hreshold for work absence in he classicaion as employed. The auhors esimaed adjusmen facors using wo parallel surveys: one prior o he 8
Figure 3.2: Quarerly GDP growh rae 2 1 0 1 2 3 Source: OECD. Noe: Growh rae compared o previous quarer, seasonally adjused. Shadings indicae recessions. The verical line signals he revised survey. We can face he LFS redesign as a kind of deerminisic misclassicaion, since here are deliberaed changes in he labor marke saes' deniion ha migh cause he same individual in he same circumsances o be classied dierenly. For example, le's consider he case of a responden who repors o be absen from work or o be in a non-paid work (cases 2 and 4). Before he redesign, he would be direcly classied as employed, while in he revised LFS he could be considered non-employed. In a very similar way, a non-employed individual who saes o be acively searching for work could be classied as inacive if he repored job search diligences do no mach he required ones o be classied as unemployed, whereas before he redesign he same individual would be considered unemployed. Thus ransiions in and ou of employmen, unemploymen and inaciviy migh show-up in he revised LFS, while before he redesign none would acually occur. Figure 3.1 illusraes he average quarerly worker ows as a percenage of he labor force over wo subsamples: (a) he period before he redesign and (b) he period afer he redesign. One can observe ha he labor marke ows suered a dramaic change afer he redesign. In almos all cases more han doubles he averages aken before he redesign. However, as Figure 3.2 displays, a deep recession sared almos a he same ime as he LFS redesign. So, he quesion of wheher hese numbers are solely explained by mehodological changes emerges. For example, i seems reasonable o ask wha mosly explains he huge increase in he employmen ouows: he redesign or heir counercyclical behavior? As oulined in secions 2.2 and 2.3, aside from business cycle ucuaions and seasonal movemens (he higher frequencies in he words of Bleakley e al., 1999), he characerisics of he populaion (he lower frequencies) also inuence he likelihood o obain cerain labor marke oucomes. Therefore, several facors mus be aken ino accoun in our aemp o explain he changes in he worker ows since he redesign. We will focus now on he lower frequencies and compare he surveyed populaion characerisics before and afer he redesign, in paricular beween he fourh quarer of 2010 and he fourh quarer of 2012. For ha we compue he means of he variables frequenly used in he lieraure o explain individual's labor marke mobiliy. redesign using he new collecion procedures and anoher afer he redesign using he old mehodology. 9
Figure 3.3 gives us a picure of he changes, bu more rigorous informaion can be found in Table C.1. 7 Taking as reference he usual 5 percen level, Table C.1 reveals ha he new mehodology migh have signicanly decreased he proporion of individuals: (a) married or living as married, (b) in he age cohor of 25 o 34 years old and (c) living in Alenejo or Azores. However, he mos noiceable change occurs in he educaional levels: he proporion of individuals wih none or lile educaion (Educ 1 and Educ2) signicanly decreased, 6 and 9.7 percenage poins, respecively. On he oher hand, he proporion of individuals wih nine years of educaion (Educ3), an high school degree (Educ4) or a college degree (Educ5) signicanly increased by 0.9 p.p., 1.8 p.p. and 3 p.p., respecively. In addiion o variaions in he surveyed populaion characerisics, i migh be also useful o invesigae wheher he respondens' occupaions by indusry signicanly changed wih he redesign. As he occupaion is no a generic populaion characerisic, we have o disinguish he employed respondens from he non-employed ones, ha's wha is done in Figure 3.3 and again in Tables C.2 and C.3. The proporion of respondens working in Services signicanly increased by 3.2 p.p., while i is observed a signican decline in he proporion of hose employed in Agriculure, Manufacuring and Consrucion. If here was no redesign eec, one migh aribue his decline o he business cycle eecs (as he recession became more severe beween he wo sample periods) and hus expec a coheren raise in he proporion of non-employed previously working in Agriculure, Manufacuring or Consrucion. Sill, he proporion of non-employed by previous occupaion signicanly declines in all indusries, wih he excepions of Consrucion (whose decline is no saisically signican) and Services, which increase by 6.8 p.p. (Table C.3). These resuls sugges ha he new mehod migh have signicanly lowered he respondens in Agriculure and Manufacuring and raised he respondens in Services, in paricular for he non-employed ones. In shor, he average labor marke ows over he sample of he new mehod are subsanially higher han hose compued under he previous survey. Three facors compee o explain his srucural break: (1) he redesign iself, (2) business cycle ucuaions and (3) possible changes in he characerisics of he populaion. Signican dierences are found in he laer, however, hese migh also be due o he fac ha hey are derived under dieren mehodologies, which is in line wih previous sudies focused on he impacs of misclassicaion and survey redesign. In he following secions, we will propose a sraegy o es he redesign eec on he worker ows series, conrolling for he business cycle. We will also ry o gauge how changes in micro characerisics migh aec individual ransiions, conrolling for possible redesign eecs on he surveyed characerisics by ineracing hem wih a variable ha indicaes he survey daa source. 7 The naure of survey daa should be incorporaed in esimaion procedures whenever ha is possible. The saisical package Saa hrough i's module svyse, is one of he mos complee and widely used packages for survey daa analysis. Neverheless, a number of rouine procedures are no available wih svyse, among hese are he usual do graphs of proporions like hose of Figure 3.3. This is why he values in he graphs of Figure 3.3 may dier from hose repored in Tables C.1 hrough C.3 of Appendix C. See Appendix B for deails on esimaion of proporions wih survey daa. 10
Figure 3.3: Means of populaion characerisics Means by demographic characerisics Means by educaion Means by age cohor Before Male Married Before Educ1 Educ2 Educ3 Educ4 Educ5 Before <15 15 24 25 34 35 44 45 54 55 64 >65 Afer Male Married Afer Educ1 Educ2 Educ3 Educ4 Educ5 Afer <15 15 24 25 34 35 44 45 54 55 64 >65.46.48.5.52.54.05.1.15.2.25.3.35.4.45.5.08.13.18.23 11 Means by region Means by indusry: employed Means by indusry: unemployed Before Azores Alenejo Algarve Cener Lisboa Madeira Norh Before Agriculure Exracive Manufacuring Elecriciy Consrucion Public Adminisraion Before Agriculure Exracive Manufacuring Elecriciy Consrucion Public Adminisraion Afer Azores Alenejo Algarve Cener Lisboa Madeira Norh Afer Agriculure Exracive Manufacuring Elecriciy Consrucion Public Adminisraion Afer Agriculure Exracive Manufacuring Elecriciy Consrucion Public Adminisraion.05.1.15.2.25.3 0.02.04.06 0.02.04.06.08 Source: Auhor's calculaions based on LFS. Noe: Here Before indicaes 2010q4 and Afer indicaes 2012q4. Means compued wihou sampling weighs. The "Services" were dropped for scaling purposes.
4 Worker Flows in he Poruguese Labor Marke 4.1 Consrucing worker ows Numerous issues regarding he esimaion of worker ows have promped researchers o sugges a variey of ways o esimae hem. Abowd and Zellner (1985) discuss a lengh he issues surrounding he esimaion of worker ows on he basis of survey daa and propose a se of ex-pos adjusmens o eliminae he sources of spurious ransiions. A comprehensive discussion of survey's mehodology and he eecs of non-response bias on he ows' esimaion can also be found in Clark and Tae (2000). We will discuss here he issues regarding he esimaion of he ows in he conex of he Poruguese LFS, namely he mehods adoped o mach individuals in consecuive quarers, o adjus and disaggregae he ows series. There are wo bases for compuing ows, eiher a change in saus oday relaive o las quarer (backward maching) or change nex quarer relaive o oday (forward maching). In he rs case, quarers 2, 4 and 6 are mached wih heir counerpars looking backward (i.e., 1, 3 and 5), while in he second quarers 1, 3 and 5 are mached wih heir counerpars forward. I is worh o noe ha hese forward or backward maching mehods are no mere convenions. In fac, he resulan ows will be dieren according o he adoped mehod, because he sampling weighs vary wih he quarers. We use he backward maching approach in his sudy, and hus he ows are he populaion-weighed sums of all workers who change labor marke saus in quarer relaively o quarer 1. In consrucing worker ows, we rack he individuals' labor marke saes over he six quarers hey are in he panel, hese are in urn deermined by he algorihms presened in secion 3.1.2. Before he redesign, here was no individual's unique idenier, so we mach individuals based on a core group of variables. 8 The new mehod inroduced a variable ha idenies each individual, making he maching procedure simpler. Sill, furher adjusmens are done. A each pair of adjacen quarers, we verify if a given ransiion, mached wih he referred ideniers, is consisen wih wo observable characerisics: age and sex. The inconsisen observaions are removed from he sample, which in pracice makes he fracion of he mached respondens lower han he 5/6 referred in secion 3. Afer consrucing our mached sample, we cross-abulae he labor force saus in each quarer, which gives us a marix describing he worker ows for ha quarer. Repeaing his process for each pair of adjacen quarers in he sample generaes he series of aggregaed worker ows. Neverheless, here are wo main obsacles o consruc a coninuous se of worker ows from 1998 o 2012: (i) beween he fourh quarer of 1998 and he rs of 1999 he households ideniers were scrambled by he INE; and (ii) beween he fourh quarer of 2010 and he rs of 2011 here was he menioned change in he individuals' ideniers. As a consequence of his wo issues, we were unable o follow individuals and esimae he corresponden ows. In he laer case, we adoped an impuaion procedure in order o keep he seasonal paern of he ows and he expeced srucural break due o he redesign, while in he former we rimmed he rs four quarers of he sample. 9 A hos of dicul daa consrucions issues surrounds he use of he survey and we ouch on a few key issues here. Sill, we believe ha our eors reec a good compromise given he available daa. 8 Accommodaion ID (Iem cua); locaion ID (Iem seccao); household idenier inside he accommodaion (Iem num_familia); individual idenier inside he household (Iem num_individuo) 9 Based on a four quarer moving-average. See Heeringa e al. (2010) pp. 345-359 for a complee overview on impuaion models for survey daa. 12
Table 4.1: Seasonaliy of labor marke ows since Spring 1999. E I I E E U U E I U U I Winer 1.693 1.630 1.290 1.329 1.296 1.155 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Spring 1.523 1.581 1.062 1.396 1.190 1.202 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Summer 1.594 1.568 1.185 1.275 1.424 1.207 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Auumn 1.798 1.539 1.413 1.337 1.423 1.301 (0.000) (0.000) (0.000) (0.000) (0.000) (0.000) Observaions 55 55 55 55 55 55 R 2 0.699 0.702 0.795 0.869 0.889 0.907 F-Saisic 29.649 30.064 49.531 84.685 101.726 124.562 Source: Auhor's calculaions based on LFS. Noe: p-values in parenheses. 4.2 Aggregaed worker ows The aim of his secion is o analyze he basic facs of he Poruguese labor marke ows from 1999 o 2012. We will look a some labor marke descripive ndings and hen he ows disconinuiy and heir cyclical properies will be assessed wihin a regression disconinuiy framework. Figure 4.1 shows he labor marke socks, ows and hazard raes; he shadings indicae recessions while he verical line signals he beginning of he new survey. I becomes clear how he redesign aeced dierenly labor marke socks and ows: in all he ows series one observes a dramaic shif in he ows derived under he revised LFS, while, for insance, here is no such a break in he sharply upward rend of he unemploymen rae. Eyeballing Figure 4.1 one can also observe he following feaures: (a) There is no a clear rend in boh employmen and inaciviy ouows, conrarily o unemploymen ouows, which exhibi an upward rend; (b) The ows series conain periodic spikes relaed o seasonal movemens, as well as ucuaions relaed o he economic cycle, wih employmen separaions (and somewha inaciviy ouows) peaking during recessions; (c) The picure of worker ows and hazard raes is very similar for employmen and inaciviy ouows. This is no he case of he unemploymen ouows hazard rae, where one can see a sharply downward rend. Alhough, here is a quie logical reason for his: as he pool of unemployed enlarges due o much less hires (Ceneno and Novo, 2013) he raio beween he unemploymen ouows and he sock of unemployed shrinks, decreasing he probabiliy o move ou of unemploymen. 4.2.1 Seasonaliy Examinaion of Figure 4.1 made us suspec ha worker ows are srongly inuenced by seasonal facors. In order o conrm his, we regress each of he ows series on a se of seasonal dummies. Table 4.1 shows ha all coeciens are saisically signican a any usual level. As shown by he value of he R 2, seasonaliy is mos imporan in U I ows, followed by E U and E I. This moivaed us o seasonally adjus he ow series using Census Bureau X-12 (CB X-12). We used his procedure because i is a sandard and readily available package for seasonal adjusmen and also because previous sudies, such as hose of Blanchard and Diamond (1990) and Bell and Smih (2002) chose is predecessor CB X-11 for seasonally adjus worker ow daa. 10 10 There are no reasons o believe ha he redesign change he seasonal paern of he ows. Sill, in order o invesigae his, we repeaed his exercise using only he rs subsample and he resuls didn' change. 13
Figure 4.1: Labor marke socks, gross ows and hazard raes 50 52 54 56 58 60 Employmen rae 0 5 10 15 20 Unemploymen rae 37 37.5 38 38.5 39 39.5 Inaciviy rae Employmen ouflows (% labor force) Unemploymen ouflows (% labor force) Inaciviy ouflows (% labor force) 14 0 1 2 3 4 5.5 1 1.5 2 2.5 3 1 2 3 4 5 Employmen o inaciviy flows Employmen o unemploymen flows Unemploymen o employmen flows Unemploymen o inaciviy flows Inaciviy o employmen flows Inaciviy o unemploymen flows Employmen ouflows hazard rae Unemploymen ouflows hazard rae Inaciviy ouflows hazard rae percenage 0 1 2 3 4 percenage 10 12 14 16 18 percenage.5 1 1.5 2 2.5 3 Employmen o inaciviy hazard rae Employmen o unemploymen hazard rae Unemploymen o employmen hazard rae Unemploymen o inaciviy hazard rae Inaciviy o employmen hazard rae Inaciviy o unemploymen hazard rae Source: Auhor's calculaions based on LFS. The sock series are he INE ocial raes. Noe: Shadings indicae recessions. The verical line signals he revised survey.