Labor Market Transitions in Peru

Transcription

1 DCUMNT D TRAVAL DT/2003/14 Labor Market Transtons n Peru Javer HRRRA Gerardo Davd RA HADY

2 LABR MARKT TRANTN N PRU Javer Herrera UR CPRÉ de l RD - N jherrera@ne.gob.pe Gerardo Davd Rosas hady nter Amercan Development Bank, Unversty of Pars 1 Panthéon-orbonne, DAL - UR CPRÉ de l RD davdro@consultant.adb.org Document de traval DAL / Unté de Recherche CPRÉ Novembre 2003 RÉUMÉ Les analyses tradtonnelles du marché du traval s avèrent ncapables d explquer le paradoxe apparent entre un taux de chômage relatvement modéré dans un pays tel que le Pérou (envron 10%, taux peu sensble aux fortes fluctuatons macro-économques) et la percepton d une grave crse de l emplo. Une explcaton possble pourrat résder dans le fat que cet ndcateur statque en coupe nstantanée ne mesure pas les flux élevés entre les stuatons d emplo et d nemplo. Pour analyser ces questons, l est nécessare de condure une analyse dynamque sur données de panel. Nous avons ans construt un panel natonal d ndvdus en âge de travaller pour la pérode à partr de l enquête péruvenne auprès des ménages (NAH). Comme d autres études réalsées dans des pays en développement, nous constatons qu l exste une mportante moblté de l emplo au Pérou. Nous trouvons également que la plupart des transtons ntervennent entre emplo et nactvté plutôt qu entre emplo et chômage. Le taux de chômage permanent apparaît très fable et le chômage serat donc essentellement un phénomène frctonnel. Pour aller plus lon, nous avons élaborés des profls de transton ncondtonnels, ncluant les caractérstques ndvduelles et du ménage, telles que le genre, l âge, et le nveau d éducaton, assocé avec chaque état de transton. Fnalement, après avor examné ces transtons sur le marché du traval et les bas de sélecton possbles, nous avons estmé un modèle logt multnomal. Ce modèle nous a perms d apprécer l ncdence (condtonnelle) des caractérstques ndvduelles et des ménages ans que des dfférents chocs sur les états de transton en matère d emplo. ABTRACT Tradtonal labor market analyss based solely on the net unemployment rate fals to explan the apparent paradox between a relatvely moderate unemployment rate n Peru (around 10%, wth a weak sensblty to wde macroeconomc fluctuatons), and the fact that unemployment s one of the major ssues n Peru. ne possble explanaton s that ths statc ndcator of cross secton net unemployment balance s compatble wth hgh flows n and out of employment states. To address these ssues we needed to conduct a dynamc analyss usng panel data. Usng the Peruvan natonal household survey (NAH), we constructed a panel of workng age ndvduals at the natonal level for the perod Lke prevous work n developng countres, we found that there s an mportant degree of job moblty n Peru. We also found that most of the transtons occur between employment and nactvty nstead of between employment and unemployment. We also showed that the rate of permanent unemployment s very low so that unemployment would be essentally a frctonal phenomenon. Further, consderng the dfferent transton states, we elaborated an uncondtonal transton profle, ncludng ndvdual and household characterstcs, lke gender, age and educaton levels for example, assocated wth each transton status. Fnally, after examnng these labor market transtons and the possble sample selecton bas, we estmated a multnomal logt model. Ths model allowed us to apprecate the (condtonal) ncdence of ndvdual and household characterstcs as well as the effects of dfferent shocks on the labor transton states. 2

3 Contents NTRDUCTN PRBLM TATMNT conomc performance and the labor market n the 90 s n Peru Man results of prevous studes of labor moblty n Peru DATA AND VARABL The NAH surveys and the panel The selecton bas ssue Varables LABR MBLTY N PRU The descrptve analyss bserved characterstcs of labor market moblty n Peru Labor moblty profle The determnants of labor market transtons The model Man regressons results CNCLUN APPNDC BBLGRAPHY Lst of tables Table 1: Descrptve statstcs for ndvdual n the panel and not n the panel, Table 2: Flows n the labor market durng the perod (%)... 9 Table 3: Labor market transtons n the urban sector, 1997/98, 1997/99 and 1998/99 (%) Table 4: Labor market transtons n the rural sector, 1997/98, 1997/99 and 1998/99 (%) Table 5: Urban labor market moblty between 1998 and 1999 by ndvdual characterstcs n Table 6: Rural labor market moblty between 1998 and 1999 by ndvdual characterstcs n Table 7: Urban Peru odds rato Table 8: Rural Peru odds rato Table 9: pecfcaton test of the dependent varable Table 10: pecfcaton test of the explanatory varables wth many modaltes Lst of fgures Fgure 1: Unemployment rates and macroeconomc fluctuatons, Peru Fgure 2: ntry and ext urban labor market flows Fgure 3: ntry and ext rural labor market flows Fgure 4: Urban odds rato Fgure 5: Rural odds rato

4 NTRDUCTN Unemployment s consdered to be one of the major ssues n Peru. However, the level of unemployment, estmated around 10%, s comparable to what s observed n other Latn Amercan countres and, most mportantly, s characterzed by a weak senstvty to wde macroeconomc fluctuatons. Ths apparent weak senstvty of unemployment rates to macro economc fluctuatons s possbly related to the level of labor moblty n Peru. Actually, some evdence exsts ndcatng that labor moblty n Peru s very hgh and that most of labor transtons occur between employment and nactvty. These flows n and out of the labor market cannot be captured by a tradtonal analyss based on the unemployment rate. Therefore, ths statc ndcator of cross-secton net unemployment balance fals to explan what really happens n the Peruvan labor market. However, to study labor moblty, we need panel data. Panel data allows us to follow the same ndvduals n the labor market durng a gven perod and to observe f they move or not from one labor state to another. The Peruvan natonal household survey (NAH) allowed us to construct a large panel of workng age ndvduals at the natonal level for the perod Thus, we could conduct a dynamc analyss to verfy f labor moblty s ndeed hgh n Peru and f permanent unemployment really exsts. Also, we have examned factors determnng labor moblty, focusng partcularly on ndvdual characterstcs assocated wth labor market transtons. The frst secton summarzes the labor market stuaton durng the nnetes showng the evoluton of the unemployment and the GDP growth rates. Ths secton also presents the prncpal results obtaned by prevous studes concernng labor moblty n Peru. The second secton gves some nformaton about the surveys used to construct the panel and how ths panel was constructed. Ths secton also presents statstcal tests n order to check for selecton bas n our panel. Fnally, the thrd secton analyzes determnants of labor market moblty n the urban and rural sectors n a descrptve and econometrc way. 1. PRBLM TATMNT 1.1. conomc performance and the labor market n the 90 s n Peru Durng the nnetes, the Peruvan Government mplemented a macroeconomc stablzaton program and an mportant set of structural reforms, especally a sgnfcant labor market lberalzaton. The economc outcome of these polces was a strong economc growth between 1992 and But, between 1997 and 2000, the economc actvty slowed down consderably (Fgure 1). Accordng to offcal fgures, the most dynamc sectors durng the frst perod were those of raw producton, constructon, fnancal sector and servces. 4

5 Fgure 1: Unemployment rates and macroeconomc fluctuatons, Peru % 15% 10% 10% 8% 5% U rate (%) 6% 0% GDP g (%) 4% -5% 2% -10% 0% Unemployment Anos rate GDP grow th -15% ource: N Note: Unemployment rates for Metropoltan Lma The labor market performance was only mldly affected by ths hghly contrasted economc evoluton. ven f the partcpaton rate grew consderably over the perod , the rate of unemployment dd not ncrease and the proporton of nactve people fell. The labor market was characterzed by a hgh job creaton. Accordng to offcal fgures (1998), the employment growth was most pronounced n small enterprses, fewer than ten workers, operatng n the servce sector. Durng the second perod ( ), economc actvty slowed down sharply and ths slowed down had an adverse mpact on the labor market and especally on employment growth. However, the performance of the labor market was also affected, partcularly durng the frst perod, by the radcal labor lberalzaton reform mplemented n Ths reform and, partcularly the reducton of job protecton and the creaton of new knds of job contracts, lke part tme and lmted tme contracts, mproved labor market flexblty and ncreased the rate of turnover. The consequences were a fall n the average employment duraton and a large ncrease n labor moblty durng ths perod (Chacaltana (1999), Daz and Maruyama (2001)) Man results of prevous studes of labor moblty n Peru Labor moblty has been rarely analyzed n Peru, the prncpal reason beng the lack of sutable data. Panel data over a contnuous perod has exsted n Peru only snce 1996, when the N mplemented a panel dmenson n ts NAH survey. Ths new survey allows us to construct a quarterly urban panel for 1996 and to follow up on people selected durng ths year. Currently, there are three mportant studes of labor moblty n Peru and all use the quarterly panel of The frst was undertaken by the Peruvan Mnstry of Labor n 1998 (MTP, 1998). Ths study was frst mproved by Chacaltana (1999), who also used an urban panel of , and then by Daz and Maruyama (2001). These studes confrmed that the mean duraton of unemployment n Peru s very short. Actually, permanent unemployment seems not to be a very mportant problem, less of 0.1% of unemployed people stay n unemployment more than one year. However, the authors found other nterestng results, manly that labor moblty s very mportant n urban Peru, more than 40% of the actve people changed labor market status durng the year, and the authors observed that the most mportant transtons n the labor market occur between employment and nactvty status, and vce versa. Moreover, the authors dentfed ndvdual characterstcs that have mportant effects on labor market moblty lke sex and age, females and young people are the most affected by transtons. 5

6 These results could be questoned on some ponts. Frst, the authors dd not take nto account the unemployment seasonalty durng the year. n addton, they dd not check the qualty of the quarterly panel used. These gaps could produce some bas n ther nterpretatons and conclusons. Moreover, these studes analyzed only urban households and were focused on the unemployment problem (manly on the duraton of unemployment). At present, a complete study about labor transtons n Peru does not exst. 2. DATA AND VARABL 2.1. The NAH surveys and the panel The NAH surveys have been developed by the N on a quarterly bass snce These surveys have a natonal coverage, ncludng both urban and rural areas, and deal wth all the permanent household resdents. The surveys prncpal objectve s to gve nformaton on the household lvng condtons and they consst of four questonnares. ne of these questonnares s the Questonaro general (general questonnare) that gves nformaton about the characterstcs, educaton, health and employment, of the dwellng and household member s and about expendtures and socal transfers. n ths paper, we used the NAH surveys of the last quarters of 1997, 1998 and The numbers of ndvduals n the samples were respectvely: 31,748; 33,325 and 18,786. These surveys have a panel dmenson that allowed us to follow some of the households and ndvduals durng these three years. n 1997, the N selected some dwellngs as panel dwellngs and the household and ndvduals lvng there were dentfed. A relatvely large number of ndvduals can thus be traced from one survey to the followng ones. We focused on workng age people, years old. To construct the panel, we used the ndvdual dentfcaton code and then nformaton about sex, age and names. Fnally, we obtaned a large panel of 6006 ndvduals for the perod The selecton bas ssue The ndvduals n the panel represent only 38% of ndvduals older than 14 years n 1997 (see table 1). The dfference s closely lnked to the panel attrton caused by a combnaton of dfferent factors-sample constructon, mgraton and mssng answers-that are not necessarly randomly dstrbuted. Therefore, the panel people are not completely representatve of the rest of the sample and we needed to check the qualty of our panel by comparng the characterstcs of ndvduals n the panel data aganst those not present n the panel n the 1997 survey. At frst glance, n 1997, the ndvduals present n our panel seemed to have the same characterstcs as those not present n the panel. But, tests carred out showed some sgnfcant dfference at 1%, 5% and 10% between the two samples. More precsely, n the panel sample, there were more people from Lma and less from the outh and Central erra. Moreover, we observed more household heads and more partners but fewer chldren and other relatves. n the panel, more ndvduals have prmary educaton and less have unversty educaton. We also observed a lower proporton of sklled people, a smaller number of hours worked durng the week, n the man and secondary jobs, and a hgher proporton of people who want to work more hours. Fnally, there were a hgher proporton of legal owners, a lower proporton of tenants or of owners wthout ttle and a hgher proporton of people wth workng assets. 6

7 Table 1: Descrptve statstcs for ndvdual n the panel and not n the panel, 1997 ndvduals characterstcs No panel Panel Age *** ex (%) - male female trata (%) - urban rural Geographcal regons (%) -North Coast Central Coast outh Coast North erra Central erra ** -outh erra *** -Jungle Lma *** Household head (%) *** Partner *** Chldren *** thers relatves *** ze of household Martal status (%) - lvng alone *** - lvng n couple *** ducaton (%) - wthout educaton * - prmary educaton *** - secondary educaton * - unversty and others *** tudent Human captal of the household Labor stuaton (%) - employed unemployment nactvty ectors of actvty (%) - prmary secondary tertary nsttutonal dvson (%) - publc ** - formal nformal klls (%) - sklled ** -unsklled ** Worked before (%) Hours worked durng the week * Wants to work more hours *** Wth a secondary job ncome - number of ncome earners dependency rate Dwellng ownershp status (%) - legal owner *** - owner wthout ttle ** - tenant and others *** Dwellng characterstcs (%) -wthout water, electr, wc confort/ confort/ wth water, electr, wc Dwellng wth sold walls (%) Assets (%) - luxury assets workng assets ** ample sze 12,168 6,606 ource: NAH Panel and NAH 1997, calculated by authors Notes:* Tests dfferences between the no panel and panel sample. * Dfference s sgnfcant at 10 % level, ** at 5% level and *** at 1% level. 7

8 2.3. Varables We used two knds of explanatory varables n ths paper: ndvdual, for example sex, age and educaton level, and household characterstcs, for example the level of human captal and the dependency rate. These varables were measured n two ways: the ntal characterstcs n 1997 and the change from 1997 to These varables were: ndvduals characterstcs Age Age groups ex (%) Household status Martal tatus (%) ducaton Level (%) Years of educaton tudent ndvdual labor market stuaton 14-24, 25-34, 35-44, or 55 and more years old Male or female Head, partner, chldren or others relatves Lvng alone or lvng n couple Wthout, prmary, secondary or unversty and others. tll studyng Labor market status nactve, unemployed or employed ectors of actvty (%) - prmary Agrculture - secondary Manufacture and constructon - tertary Commerce, transport, fnancal ntermedaton, etc. nsttutonal dvson (%) - Formal Workng n a Publc or prvate frm - nformal Workng n a frm wth less of 5 employees and where people don t have more than prmary educaton. Frm sze and more klls (%) Number of employees n the enterprse The varable was created usng the man occupaton type - sklled Professonals or techncal employees - unsklled ellers, farmers, blue collars workers, etc. Worked before (%) Had a job before Hours worked Durng the week n the man and secondary occupaton Wants work more hours Wants and can works more hours per week econdary occupaton Wth another occupaton Household Characterstcs ze of household Number of members Number of young chldren Number of chldren younger than 10 years old Human captal of the household (Years of educaton / age) for all of the household members ncome - number of ncome earners - dependency rate Number of ncome earners / household sze Dwellng ownershp status Legal owner, owner wthout ttle or tenant and others Dwellng characterstcs Wthout water, electrcty and w.c.,1, 2 or any of these three comforts Dwellng wth sold walls (%) Cement, brck, etc. Assets (%) Luxury assets or workng assets Varables of change (events) Change durng the perod change of household status For example: chldren n 1997 and household head n change of cvl status For example: lvng alone n 1997 and lvng n couple n change of sector of actvty For example: had a job n the prmary sector n 1997 and a job n the secondary or tertary sectors n change of the number of ncome earners ncreased or decreased n the number of ncome earners n the household 3. LABR MBLTY N PRU n contrast wth the other dynamc studes about labor moblty and specally those concernng unemployment phenomena, our study used a larger panel data set. Ths panel was not only characterzed by a longer tme perod of 3 years but was also larger n ts coverage; we used a natonal sample nstead of only an urban sample. Usng the last quarter of each year allowed us to analyze labor moblty wthout the nterference of seasonal effects The descrptve analyss Frst, we presented moblty transton matrces n order to grasp the mportance of labor transtons between the dfferent labor market status, employment, nactvty and unemployment, and then we examned the labor moblty profle. 8

9 bserved characterstcs of labor market moblty n Peru n Table 2, we examned flows nto and out of dfferent labor market status as well as those that reman n the same labor market status throughout the perod. Frst, we observed that, n Peru, labor moblty s very mportant, 27% of the workng age populaton, and permanent unemployment s nearly non-exstent and also permanent nactvty represent only 16% of observatons. The most mportant transtons n the labor market occurred between employment and nactvty, 16%. econdly, we observed that the level and characterstcs of labor moblty dffer between the urban and rural sectors. Labor moblty was hgher n the urban sector. n ths sector, transtons from employment to nactvty were predomnant whle the reverse was true n the rural sector, especally for women. Thrd, women seemed to be more moble than men, especally n the rural sector. These dfferences are related to ther producton and labor market characterstcs. n the urban sector there are more salary workers and the effects of the labor market reform were larger. Moreover, the reservaton wage for urban nactve people s hgher than n rural areas, especally for the young and for the women. Ths may explan why we have a hgher proporton of permanent nactve people, especally females, n the urban sector. n the rural sector, also famles are larger and the proporton of agrcultural producers s hgher. Therefore, t s hard to dfferentate between producton and domestc actvtes. Because, most of these actvtes are agrcultural, they are affected by seasonalty. ndvduals, especally the chldren and others relatves n the household move very easly from domestc to producton actvtes (and vce versa). Ths explans why the proporton of permanently nactve people s relatvely lower n ths sector. Table 2: Flows n the labor market durng the perod (%) Total Urban Rural Males Females Males Females mmoblty Always employed Always unemployed Always nactve Total mmoblty Moblty xt employment - to unemployment to nactvty xt unemployment - to employment to nactvty xt nactvty - to employment to unemployment Total moblty ource: NAH Panel , buld by the authors n Tables 3 and 4, we analyzed labor moblty durng three dfferent perods (1997/98, 1997/99 and 1998/99). The lnes refer to the labor status of the ndvduals n the frst year, whereas the columns refer to the status of the same ndvduals one and two years later. We observed that more than 70% of employees and 50% of nactve people dd not change ther labor market status n those years. The proporton of employees that transtoned drectly from employment to nactvty s hgher than the proporton of employees who entered or exted unemployment. We also confrmed that permanent unemployment was lower and that labor transtons were dfferent n the rural and urban sectors. For example, n the rural sector, a hgher proporton of unemployed and nactve people transted drectly to employment. These transtons are consstent wth the fact that, n the rural sector, permanent nactvty and permanent unemployment were relatvely lower, especally for males. 9

10 We also observed the effects of the economc recesson, whch started n 1997, on the labor moblty. Labor market moblty changed between and , especally n the urban sector. n ths sector, n the latter perod there were relatvely fewer permanent workers and more permanent nactve people. The economc recesson dd not ncrease transtons from employment to unemployment but ncreased transtons from employment to nactvty. n the rural sector, the changes were less mportant but the dfferences between males and females were more pronounced. For males, the proporton of permanent workers was lower. There were more permanent nactve people and the proporton of permanent unemployed ndvduals was zero. At the same tme, exts from employment ncreased whereas entres to employment decreased. The proporton of males who transtoned drectly from unemployment to nactvty ncreased nearly threefold. For females we observed the opposte stuaton; and n partcularly we observed an ncrease n the proporton of those who transtoned to employment. Table 3: Labor market transtons n the urban sector, 1997/98, 1997/99 and 1998/99 (%) Years U Total Total row U Total Total row Males U Total column U Total column Females U Total column U Total column ource: NAH Panel , buld by the authors Notes: = employed, U = unemployed and = nactve. 10

11 Table 4: Labor market transtons n the rural sector, 1997/98, 1997/99 and 1998/99 (%) Years U Total Total row U Total Total row Males U Total column U Total column Females U Total column U Total column ource: NAH Panel , buld by the authors Notes: = employed, U = unemployed and = nactve. n the prevous sectons, we observed that unemployment n the urban and rural sectors was very low and that most labor market transtons occurred between employment and nactvty. The small number of unemployed people forced us, n what follows, to merge the nactve and unemployed people. Ths aggregaton can be also justfed by the Peruvan unemployment characterstcs, observed before, and especally by the fact that Peruvans do not have a partcular nterest n declarng themselves as unemployed or as nactve (there s no unemployment beneft system). Fgures 2 and 3 below show how complex labor market transtons can be n Peru and hghlght dfferences between urban and rural sectors. t s nterestng to note that nearly 21% of ndvduals n the urban sector but only 9% n the rural sector could be consdered as permanently not workng (nactve or unemployed) n each of the years of the observed perod. These ndvduals can be referred to as the hard core of permanent nactve people, representng respectvely over two-thrds and a half of all nactve ndvduals observed each year. n 1999, almost half of nactve ndvduals n the two sectors were n fact transent nactve -persons experencng some labor market transton durng the three-year perod. tatc unemployment rates had low senstvty to macroeconomc fluctuatons, n part because they were absorbed by transent nactves. The almost constant percentage of non-workng people observed each year s n fact the net result of compensatng nflows and outflows of workng and non-workng ndvduals. ndvduals permanently employed represented only 44% n the urban, and 62% n the rural workng age populaton. We also observed that the longer the ndvdual remans n the non workng status, the lower hs probablty of re-enterng the workng status. 11

12 Fgure 2: ntry and ext urban labor market flows Total 1999 not workng 26.4% not workng 20.9% not workng 36.4% workng 5.5% not workng 38.5% not workng 4.4% workng 10.0% workng 5.6% not workng 4.8% not workng 10.8% workng 6.0% workng 63.6% not workng 8.5% workng 61.5% workng 52.9% workng 44.4% ource: NAH Panel , buld by the authors Fgure 3: ntry and ext rural labor market flows Total 1999 not workng 14.1% not workng 9.15% not workng 24.7% workng 4.9% not workng 20.2% not workng 3.0% workng 10.6% workng 7.6% not workng 3.1% not workng 8.2% workng 5.1% workng 75.3% not workng 5.0% workng 79.8% workng 67.1% workng 62.1% ource: NAH Panel , buld by the authors. 12

13 Labor moblty profle We made dynamc labor market profles showng the ncdence of labor moblty accordng to ndvdual demographc and economc characterstcs. We obtaned a profle of moble people-.e. people who went out or nto employment- permanent nactve or unemployed people and those who are always employed. Ths exercse was partcularly useful to characterze these dfferent populatons, but t could not examne causalty between ndvdual characterstcs and the dfferent labor market transtons. The specfc effects of the dfferent varables were examned later n the econometrc part of the paper. n Tables 5 and 6 we presented these profles for urban and rural sectors and, n addton, we tested for means dfferences. For example, n the urban sector, the profle of ndvduals n a status of permanent nactvty (and unemployment) relatve to those n a status of permanent employment, corresponds on average to younger ndvduals and to a hgher proporton of women. Therefore, these ndvduals were less lkely to be heads of household or to lve as couples, but they were more lkely to be chldren. They lved n smaller households and n households where the number of chldren younger than 10 was smaller. Moreover, a smaller proporton of these people completed prmary educaton and a larger proporton were stll students. Regardng the labor market status of the permanent nactve people the year before, the proporton of these people who were already nactve or unemployed was hgher and the proporton that were employees, lower. Also, permanent nactve people were less sklled and worked fewer hours but were also less lkely to have a secondary job. Fnally, the permanently nactve people seemed to have a relatvely hgher standard of lvng. They lved n households wth a hgher number of ncome earners and most of them had workng assets. n the rural sector, the profle of permanent nactve people was qute dfferent. n Partcular, relatve to permanent workers, these ndvduals were more lkely to be chldren or partners and others relatve. The proporton that was already nactve or unemployed was very hgh. They were also lkely to work n the nformal sector. The profle of moble ndvduals was smlar to the profle of permanent nactve people. n partcular, both groups were relatvely young and ncluded a hgher proporton of women. But there were also some dfferences. n the urban sector, the proporton of moble people wth secondary educaton was hgher. ndvduals leavng the employment status () came from households wth a hgher level of human captal and a lower dependency rate. They had more luxury and workng assets and lved n better dwellngs. The proporton of those who were employed before was lower and the proporton of those who were nactve was hgher. Fnally, moble urban ndvduals had nformal jobs more often and worked n small enterprses. n the rural sector, the profle of moble and permanent nactve was almost the same. The only mportant dfference was that moble rural ndvduals were less lkely to have jobs n the secondary sector than permanent employees. Rather, they were more lkely to be tertary sector employees. We also observed some dfferences between ndvduals enterng the labor market compared to those leavng t. n the urban sector, labor market entrants () were lkely to be heads of household and less lkely to be partners. Therefore, they were more lkely to lve n households wth more chldren younger than 10 years old and n households wth a lower level of human captal. Moreover, they were more lkely to have completed ther studes and to be oblged to have a secondary job. Fnally, they seemed to have relatvely lower standards of lvng and fewer assets. n the rural sector, dfferences between the two knds of moble people were almost the same as n the urban sector. The only dfferences were that entrants had hgher probabltes of beng chldren than those leavng the employment status. Also, they were employed more often n the secondary sector and they worked relatvely more hours per week. 13

14 Table 5: Urban labor market moblty between 1998 and 1999 by ndvdual characterstcs n 1997 No moblty Moblty ndvduals characterstcs Total Age *** 29.5*** 29.9*** 33.5 Age groups (%) *** 54.3*** 50.1*** *** 16.8*** *** 10.7*** 13.0*** ** 11.3*** 10.4*** and more * ex (%) - male *** 33.6*** 39.7*** female *** 55.3*** 60.3*** 52.3 Household head (%) *** 6.8*** 20.9*** Partner (%) *** Chldren (%) *** 50.1*** 46.8*** 36.7 thers relatves * ze of household ** 5.9** 6.0*** 5.7 Martal tatus (%) - lvng alone *** 59.9*** 57.1*** lvng as a couple *** 40.1*** 42.9*** 52.5 Number of chldren wth less than 10 years *** 0.64*** old +++ ducaton (%) - no educaton ** prmary educaton ** 20.1** secondary educaton *** unversty and others *** 19.7*** 26.2 tudent (%) *** 41.8*** 32.6*** Human captal of the household (rato) *** Labor market stuaton - employed *** 18.5*** 52.1*** unemployed ** 13.8*** 10.4*** nactve *** 67.7*** 37.5*** ectors of actvty (%) - prmary secondary tertary nsttutonal dvson (%) - publc *** 4.6*** formal nformal * 62.9*** 49.8 klls (%) - sklled ** 15.1*** 18.5*** unsklled *** 84.9* 81.5*** 73.9 Frm sze (number of employees) *** 75.3*** and more *** 7.3*** 20.1 Worked before (%) *** 74.1*** Hours worked *** 35.0*** 37.6*** 47.4 Wants to work more hours (%) ** 46.8 Has a secondary job *** 1.4*** 6.2*** ncome - number of ncome earners ** dependency rate *** 0.46*** 0.48 Dwellng ownershp status (%) - legal owner owner wthout ttle tenant and others Dwellng characterstcs (%) - no water, electr, wc *** 5.5** confort/ * confort/ ** has water, electr, wc *** Dwellng wth sold walls (%) ** Assets (%) - luxury assets *** 56.0** workng assets * 50.5** ource: NAH Panel , calculated by authors Notes: = always employed, = permanent nactve, = entry nto employment and = ext out of employment. * Tests dfferences between all categores wth respect to always employed and + Tests dfferences between exts out of employment wth to entres nto employment. * or + dfference s sgnfcant at 10 % level, ** or ++ at 5 % level and *** or +++ at 1 % level. 14

15 Table 6: Rural labor market moblty between 1998 and 1999 by ndvdual characterstcs n 1997 No moblty Moblty ndvduals characterstcs Total Age *** 27.6*** 28.5*** 33.8 Age groups (%) *** 53.6*** 46.3*** ** *** 10.7*** 15.8*** *** 10.3** 8.4*** and more *** 5.3*** 10.2 ex (%) - male *** 17.8*** 25.3*** female *** 82.2*** 74.7*** Household head (%) *** 1.5*** 4.9*** Partner (%) *** 40.7*** 44.2*** 30.6 Chldren (%) *** 52.3*** 43.0*** thers relatves ** 3.3* 7.9** 4.4 ze of household ** 6.5*** 6.3** 6.1 Martal tatus (%) - lvng alone *** 55.7*** 50.1*** lvng as a couple *** 44.3*** 49.9*** 61.4 Number of chldren wth less than years old ducaton (%) - no educaton ** prmary educaton ** 51.7* secondary educaton *** unversty and others tudent (%) *** 32.8*** 23.2*** Human captal of the household *** 0.30* 0.28 (rato) Labor market stuaton - employed *** 25.0*** 51.1*** unemployed *** nactve *** 65.4*** 44.9*** ectors of actvty (%) - prmary ** secondary *** tertary ** nsttutonal dvson (%) - publc ** formal ** nformal * klls (%) - sklled * unsklled * Worked before (%) *** 68.5** 73.1 Hours worked *** 29.0*** 35.5*** Wants to work more hours (%) ** 18.8*** 28.8** ncome - number of ncome earners dependency rate *** 0.32** 0.34 Dwellng wth sold walls (%) * 5.0 Assets (%) - luxury assets ** workng assets ** 29.8 ource: NAH Panel , calculated by authors Notes: = always employed, = permanent nactve, = entry nto employment and = ext out of employment. * Tests dfferences between all categores wth respect to always employed and + Tests dfferences between exts out of employment wth to entres nto employment. * or + dfference s sgnfcant at 10 % level, ** or ++ at 5 % level and *** or +++ at 1 % level. 15

16 Thus, the analyss of transton matrces showed us that the labor moblty n Peru s hgh and permanent unemployment dd not really exst, especally n the rural sector. Moreover, we found that the most mportant labor market transtons occurred between nactvty and employment and that the labor market moblty dffered greatly between rural and urban sectors, moblty was relatvely hgher n the later one, and across perods of tme. Fnally, we observed that age, sex, educaton level and lvng condtons seemed to have mportant effects on labor market moblty The determnants of labor market transtons n the next secton, we expanded on the knowledge of the prncpal factors that determne labor market transtons n Peru. n commentng on the labor transton profle, we have examned the uncondtonal rsk that ndvduals wth gven characterstcs may experence any of the labor market transtons. For a more analytcal purpose, we consdered the relatve rsks condtonal on the other factors that determne labor market transtons. We estmated the determnng factors of dfferent forms of labor moblty between 1998 and 1999 usng a multnomal unordered logt model, because our dependent varable s a categorcal varable wth four values correspondng to each of the labor market transtons, always employed (), permanent nactve or unemployed (), ext out of employment () and enter nto employment () The model Ths model was desgned to estmate the mpact of the dfferent explcatve varables on each of the forms of labor moblty. The model predcted the probablty that an ndvdual wth gven characterstcs wll experence one of the four labor market transtons. n order to dentfy the model one of the labor market transtons was taken as the baselne case. Dfferent sets of coeffcents were obtaned for each state. We frst commented on the statstcal sgnfcance of the regresson coeffcents of the logts. n accordance wth Long s (1997) graphcal presentaton, we studed the mpact of dscrete changes n explanatory varables on the probablty of endng n one of the four categores (,, or ), n terms of odds rato (relatve rsk rato) gven that we were nterested n labor market dynamcs. n others words, we were nterested n knowng how each varable affects the odds of a person beng permanent nactve, gong nto employment or gong out of employment relatve to beng always employed (the base case). The multnomal logt s: ( 1) Pr( y = m x ) = exp( x β ) j=,,, m exp( x β ) j Where Y s the dependent varable wth m nomnal outcomes and ( y = m ) observng outcome m gven x. Pr the probablty of x To dentfy the model we decded that β = 0 (the base case s always employed). Because exp( x β ) exp( x 0) = 1, the model s commonly wrtten as: = exp( x β m ) ( 2) Pr( y = m x ) = 1+ exp( x β j ),, pour m 16

17 We expressed the model n terms of the odds. The odds of outcome m (m=, et ) relatve to the base case outcome () gven x, reads: ( 3) Pr Pr ( y = m x ) ( y = x ) = exp( x β ) j=,, j=,, m exp( x β ) 1 exp( x β ) j j = exp( x β ), avec m=,, et β = 0. m Therefore exp( x β m ) represented the relatve probablty of beng or relatve to beng for a unt change n x. The nterpretaton became easer because, as Long notes, the value of the factor change n the odds does not depend on the value of the level of the varable consdered or on the level of the other varables, as n the case of the margnal mpact (Long 1997: 169). Most of the explanatory varables used n the estmatons of the model were dchotomous. However, there were some contnuous varables, e.g. age, and some categorcal varables, e.g. age group. The nterpretaton of the coeffcents for these varables was also easer: for the former, we had to nterpret the coeffcents relatve to the average and for the latter we had to nterpret the coeffcents to the omtted category. We complemented the nterpretaton of results usng odds fgures (see appendces). As Long (1997) explans, The large number of coeffcents makes t dffcult to see patterns n the results. f you also keep track of whch coeffcents are statstcally sgnfcant, the dffculty ncreases. And odds rato plots make t smple to fnd patterns among the coeffcents Man regressons results Because, labor moblty dffers n the urban and rural sectors we estmated models separately for each sector: n the urban sample, lke n the descrptve analyss, sex and age had mportant effects on labor moblty. However, n ths case the relatve probablty of beng permanent nactve relatve to beng always employed ncreased wth age. Moreover, no dfferentated mpact of age was found on outcome (ext out of employment) and (entry nto employment). Women had hgher probabltes of beng permanently nactve or moble, especally of beng n, relatve to beng always employed. No dfference was found n the sex varable for, or. Logcally, we observed the opposte stuaton for the household heads (most of them are males) relatve to ther partners. Household heads have lower probabltes of beng permanently nactve but are more lkely to be n category relatve to (ths result could be related to the hgher degree of labor moblty n the urban sector). Hgher levels of educaton seemed to protect aganst permanent nactvty. The mpact of educaton was not sgnfcantly dfferent for and. tudents, who were relatvely younger, were more lkely to be permanent nactve or moble relatvely to always employed. Labor market varables, lke n the descrptve analyss, had hgh and sgnfcant effects on labor moblty. n the one hand, the odds of beng moble and of beng permanent nactve were hgher for people who were nactve durng the prevous year. n the other hand, work experence and sklls seemed to protect aganst permanent nactvty. Moreover, people who worked n the prmary or secondary sectors, relatve to the tertary sector, had hgher probabltes of beng permanent nactve than of beng always employed. Lkewse, people wth a publc job relatve to people wth nformal jobs had lower probabltes of leavng employment or of beng permanent nactve. Fnally, the 17

18 ndvduals wth hgher probabltes of beng permanent nactve or of enterng employment were those who had the worst jobs. They were the ones who wanted to work more hours per week or to have a secondary job. Most of the varables lnked to lvng condtons, e.g. the knd of dwellng, were not sgnfcant. However, the probablty of beng permanent nactve relatve to beng always employed ncreased wth the level of human captal of the household (ncome effect). The dependency rate had the same effects on the relatve probablty of beng permanent nactve as t does on that of beng moble. The varables related to events showed nterestng results. For example, havng prevously exted from an economc sector apparently decreased the probablty of beng permanent nactve but ncreased the probablty of leavng employment (relatve to beng always employed). Changes n the number of ncome earners had dfferentated effects on and (ncome effect) they ncreased the probablty of beng n but decreased the probablty of ext employment. n the rural sample, varables were less sgnfcant but the results and the coeffcents were somewhat dfferent from the varables n the urban sample. Age affected the probablty of enterng nto employment. Ths probablty ncreased wth the age for all categores relatvely to beng always employed. The effect of sex was stronger than n the urban sector (ths s consstent wth the descrptve analyss). The effect of beng a student was also stronger. n the other hand, the effect of beng sklled was dfferent. klled ndvduals had relatve hgher probabltes of enterng nto employment. The effects of been prevously nactve and the effect of the level of household human captal were not as strong as n the urban sector. Fnally, two varables that were nsgnfcant n the urban sector were sgnfcant here. Dwellng qualty, represented by a dummy for lvng n a dwellng wth sold walls, and a dummy for havng workng assets both ncreased the probablty of beng n relatve to beng always employed. Fnally, we conducted three knds of Wald tests to verfy the robustness of our estmatons. The frst one (last column n tables 7 & 8) ndcated that most of our explanatory varables had sgnfcant effects n all the categores of the dependent varable. The others test (Table 9 and 10) confrmed that the constructon of our dependent varable and of the explanatory varables wth many modaltes, e.g. sectors of actvty, were correct. 18

19 Table 7: Urban Peru odds rato Transtons Ch2 ndvdual characterstcs n 1997 Age *** *** ex (woman = 1) 1.483** 1.506*** 1.740*** *** tatus n the household (reference: partner) - head 0.499*** 0.276*** *** - chldren others relatves 0.397** * Lvng as a couple 0.614** Years of educaton *** *** tudent 1.859*** 2.623*** 2.320*** *** nactve or unemployed 2.621*** *** 6.006*** *** ectors of actvty (reference: tertary) - prmary * secondary ** ** nsttutonal dvson (reference: nformal sector) - publc ** 0.525* 6.649* - formal klls (reference: unsklled) 0.405*** 0.550** 0.655* *** Worked before *** *** Wants and can work more hours per week 0.390* 1.457** 1.343* 9.474** Wth a secondary occupaton 0.651*** 0.433*** *** Household characterstcs n 1997 Household sze Number of chldren wth less than 10 years old Human captal of the household *** *** Dependency rate 1.879** 1.952** 1.890** 8.050** Dwellng ownershp status (reference: legal owner) - owner wthout ttle tenant and others ** ** Dwellng wth sold walls Luxury assets Workng assets Varables of change (97/98) - change of the head of the household change of place n the household change of cvl status change of sector of actvty 1.835*** 0.601** *** - change of skll 1.771*** 1.613** 3.263*** *** - varaton of the number of ncome earners 0,674*** *** *** ource: NAH Panel , buld by authors. Notes: Number of observatons: 3807 Log lkelhood = Pseudo R2 = = always employed, = always nactve, = entry nto employment and = ext out of employment. * Tests dfferences between all categores wth respect to always employed. * dfference s sgnfcant at 10% level, ** at 5% level and *** at 1% level. The last column shows a Wald test performed to verfy f an ndependent varable has a sgnfcant effect for all of the categores of the dependent varable

20 Table 8: Rural Peru odds rato Transtons Ch2 ndvdual characterstcs n 1997 Age ** 6.377* ex (woman = 1) 2.623** 4.847*** 2.809*** *** tatus n the household (reference: partner) - head 0.397*** 0.782*** 0.196*** *** - chldren others relatves Lvng as a couple Years of educaton tudent 3.063*** 2.422*** *** nactve or unemployed *** 2.796*** *** ectors of actvty (reference: tertary) - prmary secondary klls (reference: unsklled) ** 1.133** 9.233** Worked before Wants and can work more hours per week * Wth a secondary occupaton 0.606* Household characterstcs n 1997 Household sze ** Number of chldren wth less than 10 years old Human captal of the household * Dependency rate Dwellng ownershp status (reference: legal owner) - owner wthout ttle tenant and others Dwellng wth sold walls * Luxury assets Workng assets ** Varables of change (97/98) - change of place n the household change of cvl status change of sector of actvty *** 0.159*** *** - change of skll 1.862*** 11.11*** *** *** - varaton of the number of ncome earners 0,695*** *** *** ource: NAH Panel , buld by authors. Notes: Number of observatons: 2628 Log lkelhood = Pseudo R2 = = always employed, = always nactve, = entry nto employment and = ext out of employment. * Tests dfferences between all categores wth respect to always employed. * dfference s sgnfcant at 10% level, ** at 5% level and *** at 1% level. The last column shows a Wald test performed to verfy f an ndependent varable has a sgnfcant effect for all of the categores of the dependent varable. Table 9: pecfcaton test of the dependent varable Results Urban Rural ch2 P>ch2 ch2 P>ch Notes: Ho: The categores of the dependent varable can be collapsed Table 10: pecfcaton test of the explanatory varables wth many modaltes Results Rural Urban ch2 P>ch2 ch2 P>ch2 ectors of work Dwellng ownershp status Notes: Ho: The categores of an ndependent varable can be collapsed. 20