Labor Force Participation, Unemployment and the Poor

Cem Mete INDEX Labor Force Participation, Unemployment and the Poor 1. Introduction..84 2. Poverty, employment, and Unemployment 88 3. The unemployed: who are they, who are the most vulnerable among them? The characteristics of the unemployed..90 Vulnerability to becoming unemployed.92 Moving out of unemployment --- long term unemployment..92 More on the job-search process.93 4. The individual level determinants of employment and wages...95 Employment 96 Income from Employment 98 5. Conclusions....99 83

1. Introduction The end of planned economy era in Romania had significant implications for the labor market. Overall characteristics are shared by many other transition countries: emergence of open unemployment, reductions in employment rates, being stuck with a large industry sector (and loss making state-owned enterprises), a wage structure with little if any connection to productivity, an underdeveloped financial sector not capable of responding to the needs of the private sector that is supposed to flourish under such adverse conditions. Sometimes changes over time are impressive though, while at other times much needed reforms are only slowly implemented with mixed results. This section provides an overview of the labor market trends in Romania, to provide a context for the empirical analysis that follows. An attempt is made to discuss time trends when comparable data are available (figures based on Household Budget Surveys and Labor Force Surveys go back till 1995 and 1996 respectively). The chapter emphasizes poverty linkages for obvious reasons, and it further limits coverage to micro issues because the Country Economic Memorandum (under preparation) focuses on macro topics with an emphasis on Romania s integration with the EU. Employment rate of those who are 15 years and older steadily decreased from slightly over 60 percent in 1996 to less than 59 percent in 2001 (Figure 1). 1 The male-female gap in employment is decreasing over time, although not because of increases in female employment rates but because of decreases in male employment rates. The Roma are much less likely to be employed, but they are catching up with the rest of the population with employment rates rising from 39 percent in 1996 to 48 percent in 2001. 2 1 If one focuses on individuals aged 15 to 64, the employment percentages are 72 percent in 1995 and 68.5 percent in 2001 (slightly below the EU average of 69.2 percent). Figure 1 is based on the employment definition used by the Romanian Statistics Institute, which is all people aged 15 years and over, who have carried out an economic or social activity producing goods or services, with a duration of 1 hour at least during the reference period (one week), with a view to achieve certain incomes in form of salaries, in kind remuneration or other benefits. The 2002 employment estimate is not included in Figure 1 because the relevant survey question has changed in that year and thus it may not be comparable to the other years. 2 According to various INS surveys, the share of Roma in the population varies between 2 percent and 2.5 percent. This is believed to be an underestimation, partly due to individuals not revealing themselves as Roma. The under representation of Roma may or may not be a problem, depending on the objective of the analysis. For sample size purposes, in our case this is not a major issue: information about thousands of Roma exist in both the Household Budget Surveys and the Labor Force Surveys. The measurement error is not necessarily a problem if it stems from individuals not identifying themselves as Roma and if one is willing to focus attention on individuals who do identify themselves as Roma. The other major suspect in the underestimation is that the areas that are heavily dominated by Roma are less likely to be represented in the survey sample units. While the PA team s discussions with the INS officials did not produce support for this possibility, it can not be ruled out at this stage. Finally, note that the extent to which this underreporting results in misleading insights in multivariate analysis depends on the extent to which excluded Roma differ from the included Roma. 84

Figure 1. Employment by ethnicity and gender, 1996-2001 (World Bank staff calculations based on LFS) 70 65 60 Percentage 55 50 45 40 35 30 1996 1997 1998 1999 2000 2001 Year All Romanian Hungarian Roma Other Female Male The employed population is aging: the share of individuals aged 65 and over in total employed population increased from 8.2 percent in 1996 to about 10 percent in 2001. 3 One implication of this trend which is emphasized by the Government of Romania and European Commission Joint Assessment of Employment Priorities in Romania (October 2002) is that the elderly tend to be low-skilled compared to the rest of the population. Having said that, it is important to make the urban/rural distinction to better understand these figures, since about 90 percent of the employed elderly (defined to be ages 65 and over here) reside in rural areas making up almost 18 percent of the rural employment. The transition to market economy is underway, with a sizable shift in recent years from state to private sector employment (Figure 2). But the restructuring of state enterprises has been slower than originally intended, with large losses especially in the manufacturing sector. The layoffs routinely trigger protests (most recently protests took place to oppose layoffs in the steel plants, in the mining sector, and tractor & truck manufacturing plants). The unions negotiate severance payments with the government on a case basis. 4 3 Romanian population has been declining since 1990, with an estimated decrease of 1.8 million until 2020. The structure of the dependent population is changing, becoming more skewed towards the elderly as opposed to the children. This is primarily because of smaller-than-replacement total fertility rates at around 1.3 in recent years, but in the coming years it might be reinforced by increases in life expectancy life expectancy at birth is quite stable over time in Romania, which hit 69 in 1973 and since then recorded values between 69 and 70. 4 A large scale agreement in April was 20-24 average monthly wages to be paid to dismissed employees, in addition to two average wages payable at the time of dismissal. Furthermore, micro -credit programs for entrepreneurs are being launched in mining regions hit hardest by the restructuring. 85

The success stories in privatization tend to involve relatively small enterprises. 5 The recent record is mixed, with some movement (e.g., Romtelecom, Tepro pipe maker) but many pending operations with failed attempts to privatize (most visible one being Petrom, the national oil company). Reasons for the slow progress are varied, in some cases having to do with the Government s desire to keep the controlling share or the requirements to retain a sizable portion of the employees after privatization. Also it has proven to be difficult to attract private investors for certain companies. Employment protection legislation of Romania is considered to be more restrictive than the OECD average but less restrictive than many other transition economies. 6 Having said that a labor code which became effective on March 2003 introduced further restrictions on employers when it comes to hiring and firing, and strengthened the role of unions. Within private sector employment a sizable portion is informal work (Figure 3). This is perhaps not surprising, given the high payroll taxes and social security contributions which are discussed below. Since 1996, the share of self-employed and unpaid family workers remained stable (if anything, increased slightly). Gender differences are visible, males being much more likely to be self-employed and females being much more likely to be unpaid family workers. Trends by urban versus rural residence (not shown in the graph) reveal significant but expected differences: in rural areas being an employee ( 26 percent versus 90 percent in urban) or employer (0.5 percent versus 2.1 percent in urban) is less likely while being self-employed (39 percent versus 6.1 percent in urban) or unpaid family worker (34 percent versus 2.2 percent in urban) is more likely. Figure 2. State Versus Private Sector Employment by gender, 1996-2002 (World Bank staff calculations based on LFS) Figure 3. Professional Status of those who worked, 1996 and 2002 (World Bank staff calculations based on LFS) 80 70 70 60 Percentage 60 50 40 30 20 10 Percentage 50 40 30 20 10 0 1996 all 2002 all 1996 male 2002 male 1996 female 2002 female 0 1996 all 2002 all 1996 male 2002 male 1996 female 2002 female State Private Other Employee Employer Self employed Unpaid family worker Other 5 Entrepreneurship and Enterprise Development. Romania. March 2002. OECD and EBRD. 6 See John Haltiwanger, Stefano Scarpetta, and Milan Vodopivec. 2003. How Institutions Affect Labor Market Outcomes: Evidence from Transition Countries. Manuscript, the World Bank; also on regulatory framework see Labour Market and Social Policies in Romania. 2000. OECD. 86

In the second half of 1990s, the private sector s share (added value) in agriculture has been more than 90 percent followed by construction at around 80 percent, services at around 70 percent and industry less than 60 percent (although a significant increase from less than 40 percent in 1995). 7 Access to finance is problematic for SMEs. For the large enterprises, the main sources of financing are internal-funds/retained-earnings (52 percent) and local commercial banks (23 percent), while for SMEs the main sources of financing are internal-funds/retained-earnings (65 percent) and family/friends (15 percent) with local commercial banks a distant third at slightly over 5 percent. 8 The relative contribution of services to GDP increased significantly over time, from 42.9 percent in 1995 to 56 percent in 2001. On the contrary, both agriculture and industry lost ground. 9 Interestingly though, the distribution of employment favored agriculture during this period (maybe in part due to a move towards subsistence agriculture): percentage of employed in agriculture/forestry/fishing increased from 34.4 percent to 41.4 percent; but decreased from 28.6 to 23.2 in industry and from 37 to 35.4 in services. Such a trend is quite disappointing since even in 1997 one of the structural reform priorities was re-deployment of labor to non-agricultural jobs. In recent years the unemployment rate is tends to be close to, but less than 10 percent (depending on the definition of unemployment). While 10 percent unemployment certainly deserves attention, among transition countries this figure is on the low side. 10 The registered unemployment rate was 9.5 in 1995; with some fluctuations in between it became 8.6 in 2001. 11 Both sizable unemployment and high share of informal sector workers among the employed can be linked to a labor market that is not flexible. Payroll taxes remain high at 52 percent of salaries, despite a slight decrease in January 2003 prior to that, the taxes made up 57 percent of salaries. 12 7 See the Romanian Government s National Action Plan for Employment (2002) for more on this. 8 EBRD/World Bank Business Environment and Enterprise Performance Survey, 1999. 9 The sector contributions to GDP in 2001 were 15 percent by agriculture/forestry/fishing, 29 percent by industry and 56 percent by services. For 1995, these percentages were 21,5; 35.6 and 42.9 in the same order. 10 Vodopivec, Worgotter and Raju (2003) calculated unemployment rates based on comparable labor force survey data for a number of transition countries. In 2000, Romania with an unemployment rate calculated as 7.7 percent fared better than Bulgaria (18.7 percent), Estonia (13.5 percent), Latvia (14.4 percent), Lithuania (15.9 percent), Poland (16.6 percent), and Slovakia (19.1 percent). The only countries in authors list with comparable or less unemployment to Romania were Czech Republic (8.8 percent), Hungary (6.6 percent) and Slovenia (7.1 percent). Milan Vodopivec, Andreas Worgotter and Dhushyanth Raju. 2003. Unemployment Benefit System in Central and Eastern Europe: A Review of the 1990s. Social Protection Discussion Paper Series No. 0310. The World Bank. 11 Some policies (such as requiring registration with unemployment office as a prerequisite for qualification for MIG benefits) may result in changes in registered unemployment rate but not in the ILO definition of unemployment rate or in the percentage of individuals who consider themselves unemployed. Indeed the ILO unemployment rate increased slowly but steadily from 6.7 percent in 1996 to 7.1 percent in 2000. 12 The current rates are distributed as follows: pensions (9.5 percent employee and 24.5 percent employer), health (6.5 percent employee and 7 percent employer) and unemployment fund (1 percent employee and 3.5 percent employer). 87

2. Poverty, employment, and unemployment The linkage between poverty, type of employment and unemployment may seem trivial at first, but the strength of the relationship could vary. For example, in cases where unemployment is high for younger adults and not for others, one could expect a relatively weak relationship since unemployment is an individual level event and poverty is measured at the household level. Or if there is a substantial exchange of funds across households (e.g., through transfers from abroad), then unemployment may not necessarily lead to poverty which is a consumption based measure as defined in this study. Indeed, many policy makers in Romania seem to suspect a rather weak relationship between unemployment and poverty. 13 It is useful to illustrate the relationship between unemployment and poverty via two questions, with obvious yes or no answers but not so obvious magnitudes that go with them. Unemployed adults: Are they more likely to be poor? The 2002 household budget survey data suggest that 44.9 percent of unemployed adults of ages 15 to 64 are poor, as opposed to 25.8 percent of adults of the same age group who are either working or who are not in the labor force. 14 Interestingly, self-employed adults in agriculture are more likely to be poor than the unemployed: 55.6 percent of this group are poor. 15 Poor adults: Are they more likely to be unemployed? About 14 percent of poor adults reveal themselves to be unemployed, as opposed to 6.6 percent for the non-poor. 16 Once again selfemployment in agriculture comes into play: 29 percent of poor adults are in this group. 17 At the household level, the linkage between share of unemployed individuals in a household and consumption per adult equivalent is shown by Figure 4 (Quartic kernel, with a bandwidth of 0.25 is used to obtain the graph). There is a clear negative relationship between share of unemployed individuals and household consumption (correlation coefficient is 0.1), the function being 13 Based on discussions between the PA team and senior policy makers. Occasionally this issue enjoys direct or indirect coverage in newspapers, under titles false unemployment 2 billion USD entered Romania in one year only. 14 Similarly, 19.9 percent of unemployed adults are below the extreme poverty line as opposed to 9.4 for the remainder adult population. In this case, since we are using the Household Budget Survey data (with consumption information to define poverty), unemployment figures are not based on the ILO definition of unemployment but rather whether individuals reveal themselves to be unemployed. 15 A more general question is what is the relationship between occupational status and poverty? The least likely to be poor individuals were employers (2.5 percent, 0 percent extreme poor) and employees (11.2 percent, 2.1 percent extreme poor); followed by pensioners (20.5 percent, 5.6 percent extreme poor) and students (25.1 percent, 8.3 percent extreme poor); and finally housewives (39.2 percent, 18 percent extreme poor), self-employed non-agriculture (40.9 percent, 17.7 percent extreme poor), unemployed (44.9 percent, 19.9 percent extreme poor), the category other (dependent, military service etc.) (55.4 percent, 27.6 percent extreme poor), and self-employed agriculture (55.6 percent, 24.6 percent extreme poor) follow. 16 For the extreme poor this figure increases to 16.7 percent. 17 And the more general question is how does the occupational status of poor adults differ from the remainder of the adult population? The poor adults are less likely to be employees (14.6 percent versus 43.9 percent), less likely to be employers (0.05 percent versus 0.68 percent), more likely to be selfemployed non-agriculture (5.05 percent versus 2.76 percent), more likely to be self-employed agriculture (29 percent versus 8.8 percent), more likely to be unemployed (14 percent versus 6.6 percent), less likely to be pensioner (12.3 percent versus 18.2 percent), roughly as likely to be a student (10.4 percent versus 11.8 percent), and more likely to be a housewife (10.7 percent versus 6.3 percent). The percentages for the extreme poor are, in the same order, 7.4; 5.8; 34.2; 16.7; 8.9; 9.1; and 13.1. In each case the remainder responses fall into the other category. 88

steeper at first followed by a more modest negative relationship. While the expected negative relationship is there, unemployment alone is not that good of a predictor for poverty which is consistent with the informal sector findings at the individual level. Remembering that poverty line used in this study is 1,535,370 Lei, only when around 75 percent of the household members are unemployed the household is predicted to fall into poverty. 18 2.6e+06 Figure 4. Household Consumption and Unemployment Per Adult Equivalent Consumption 2.4e+06 2.2e+06 2.0e+06 1.8e+06 1.6e+06 1.4e+06 0.2.4.6.8 Share of Unemployed Household Members The main message that comes out of this set of descriptives is that while there is a relationship between unemployment and poverty, the informal-employment and poverty relationship is as strong (see also the predictors of household consumption regression reported in the poverty profile section). The remainder of the paper is organized around these findings. The next section elaborates more on the unemployed (and transitions in and out of unemployment), and the final section turns to employment and wages by paying special attention to informal sector workers. 3. The unemployed: who are they, who are the most vulnerable among them? We don t find employment anywhere Often when you try to get a job, they look a little closer at you and they see you, and if they know whether you are a Roma or a Gypsy, as they say, they won t take you. (Low-income respondents, Roma, Alunis, rural. 2003) The state should help the poor, not those who are capable of finding employment, who are physically healthy but refuse to work. Why should the state help these individuals? But no one wants to hire these people anyway, especially if they see that they are Gypsy. There are plenty of other people, who are 40-50 years old, and who cannot find employment. (Average income respondents, Alunis, rural. 2003) No one wants to receive you or listen to you, anywhere you go to inquire about 18 (services For the kernel or jobs) estimate, no the one choice receives of the you, smoothing because parameter I am over (bandwidth) 45 years obviously old. Do has I some not have effect on how the the right graph to live, looks. do If I not this have parameter the right is too to large work, the even graph though is too I smooth am over in 45 that years there old? are few fluctuations and if the parameter is too small then there is a maze of up-and-down lines which are too accurate (Low income to interpret. respondent, The Roma, value Targu used Mures, here urban. is the first 2003) decimal value that moves out of the maze, yet still retains some fluctuation. 89

Identifying the losers and winners of the transition period is a difficult task. A starting point might be those who have stopped working. Excluding retirees (59.7 percent), the main reasons for stopping work in 2001 were fired or down-sizing of personnel (34.5 percent of non-retiree responses), end of temporary activity (26.3 percent of non-retiree responses) and sickness or invalidity (22.8 percent of non-retiree responses). 19 In rural areas end of temporary activity is much more common and individuals are less likely to be fired/downsized; and Roma are more likely to be cite end of temporary activity and fired or downsizing of personnel as reasons. Since open unemployment did not exist prior to the transition in Romania, it seems plausible to count discouraged workers (difficult to distinguish quantitatively from those who do not want to work), younger individuals who have not worked previously but are looking for work, those who face a higher risk of unemployment and those who are more likely to experience long-term unemployment among the losers. This section aims to provide some new insights by focusing on the characteristics of the unemployed and predictors of the duration of unemployment and also by identifying those who are not unemployed but face a higher risk of being unemployed. First basic descriptives are presented, a very brief summary of empirical findings based on longitudinal data from Labor Force Surveys follows. The characteristics of the unemployed The Roma are much more likely to be unemployed than the rest of the population. Even if the enormous gap between the Romanians (and Hungarians) and Roma in 1996 has closed over time, the Roma are still more than twice as likely to be unemployed (Figure 5). Females were slightly more likely to be unemployed prior to 1998, after that they are slightly less likely to be unemployed (Figure 6). In rural areas unemployment is lower and declining over time, in urban areas it is high and does not display a declining trend (also Figure 6). One could come up with competing stories to explain unemployment by school attainment graphs by speculating on reservation wage differences etc., here it is sufficient to highlight two key findings. Figure 7 shows that the unemployment rates are lowest among primary school graduates and among those who have completed high school (or more) there is a need to interpret the former trend jointly with low rural unemployment rates discussed previously though. The most important issue to note here is that since 1999, the graduates of professional, complementary or for apprentices schools are the most likely to be unemployed. There is increasing concern about the prevalence of vocational schools in the region 20, and the idea that these schools provide their graduates better labor market opportunities does not hold up against empirical scrutiny (see also the employment/wage analysis in the next section). Figure 8 shows that older individuals are much less likely to be unemployed compared to the young. Based on these trends, the need to deal with unemployment among the Roma (e.g., through making sure active and passive labor market programs reach Roma communities) seem to be immediately justified. Simple unemployment rates do not make the case for prioritizing 19 Based on the 2001 Labor Force Survey data. 20 For Romania, a World Bank Education Policy Note (October 2002, ECA Human Development Sector Unit) recommends (i) improving the match between the courses offered by vocational and technical education and the existing and prospective demand of labor markets, (ii) developing a rolling program for updating equipment used in technical and vocational schools, and (iii) developing an information dissemination program to ensure that there is a proper balance between students with academic background and students with technical and vocational skills. 90

women and elderly, but duration of unemployment and vulnerability to becoming unemployed need to be considered (which is done next). Even if women and elderly do not emerge as higher risk groups, it could be that even small and short duration unemployment among these groups is socially undesirable. Thus once can argue the need to target women and elderly regardless of risk status, but it is important not to put these individuals in the same category with the younger individuals and Roma: effective policy interventions are likely to differ widely between these groups simply because of the differences in the target group size and geographical distribution. Figure 5. Unemployment by Ethnicity, 1996-2001 (World Bank staff calculations based on LFS) Figure 6. Unemployment by gender and urban/rural residence, 1996-2001 (World Bank staff calculations based on LFS) 25 14 Percent Unemployed 20 15 10 5 Percent Unemployed 12 10 8 6 4 2 0 0 1996 1997 1998 1999 2000 2001 1996 1997 1998 1999 2000 2001 Year Year All Romanian Hungarian Roma Other All Female Male Rural Urban Figure 7. Unemployment by schooling attainment, 1996-2001 Figure 8. Unemployment by age group, 1996-2001 (World Bank staff calculations based on LFS) (World Bank staff calculations based on LFS) 12 25 10 20 Percent Unemployed 8 6 4 Percent Unemployed 15 10 2 5 0 0 1996 1997 1998 1999 2000 2001 1996 1997 1998 1999 2000 2001 Year Year All Primary High School (first 2 years) High school No schooling Middle Professional, complementary or for apprentices More than high school All Ages 15-24 Ages 24-34 Ages 35-44 Ages 45-54 Ages 55-64 Ages 65 and over 91

Vulnerability to becoming unemployed The Labor Force Surveys (quarterly) contacted individuals of ages 15 and over four times. The second survey took place three months after the first one, another follow-up nine months later that, and a final follow-up three months later. The empirical analysis reported in this section relies on survey data from 2000, 2001 and 2002. The sample sizes are large, with around 140,000 observations per year. Benefiting from such large surveys, it is possible to estimate the probability of becoming unemployed (at least once) during the period under observation. One benefit of relying on an analysis of unemployment that occurred while under observation is that one can then include individual-status information from wave 1 (i.e., while not-unemployed) in the model. 21 Separate Probit models are estimated for individuals aged 15 to 23 and others (because for the former group broader school attainment indicators are needed since some of these individuals are still attending school). About 4.7 percent of the 15 to 23 year olds became unemployed while under observation. The results described here are based on specification 2 reported in Table 2. When all explanatory variables are at their mean values, the marginal effects on unemployment are as follows. Males are 1 percent more likely to be unemployed. Marriage reduces chances of unemployment by 2.3 percent. Schooling reduces the likelihood of unemployment but the estimated coefficients are not statistically significant at 10 percent level and marginal effects are small. Relative to the other category, those who were working or who were students at the time of the first survey are less likely to become unemployed. And finally, the urban/rural residence does not have a statistically significant effect on unemployment for this age group. Table 3 (specification 2) reports the estimates for those who are 24 years or older. While under observation, 2.3 percent of this group became unemployed. After controlling for other explanatory variables the elderly are less likely to be unemployed, probability of unemployment is higher for males (0.7 percent), and for those who were students when they were first surveyed (0.7 percent). Marriage reduces chances of unemployment (by 0.8 percent), so does working at the time of first wave survey (0.5 percent). While schooling reduces the probability of unemployment, effects are statistically significant only for high school and higher education, and the magnitude of the effect is around 1 percent (relative to those with no schooling). In an alternative specification, type of employment (public, private or other) at the time of first-wave survey information is also captured, which shows that unemployment probability is higher by about 0.3 percent for those who were employed in the private sector compared to those who were employed in the public sector. Thus despite restructuring in the public sector, the probability of falling into unemployment is still significantly higher in the private sector. Moving out of unemployment --- long term unemployment In recent years, about 60 percent of the unemployed are estimated to be in this situation for over 9 months. The unemployment duration exceeded 12 months for slightly less than 50 percent of the 21 Probit models are used for estimating predictors of becoming unemployed. Since some individuals were under observation longer than others, and because the value of the dependent variable depends on this duration (if an individual is observed longer, he or she will be more likely to be recorded as unemployed), each model includes months under observation as an explanatory variable. Not surprisingly in each model this variable has a positive and statistically significant effect on unemployment. 92

unemployed. Such a range is typical of transition countries. 22 Not much else is known about the transition into and out-of unemployment, however. In particular gender, age, schooling and ethnicity differentials in the duration of unemployment are not well understood. The interval regression estimates of the duration of unemployment (months) also use the longitudinal labor force surveys, years 2000 to 2002. 23 The key findings are as follows, based on a number of alternative specifications reported in Tables 4 and 5. The set of explanatory variables included in the models is identical to that used in the previous section. Although separate models are estimated for those who are between ages 15 23 and others, the directions of the estimated coefficients are identical. The duration of unemployment is longest for males and for those who live in urban areas. Notice that previously we have shown that urban/rural residence is not a statistically significant predictor of becoming unemployed, but residence comes in strongly when duration is considered. Interestingly marriage, schooling and status at the time of the first wave survey (employed, student or other) do not have statistically significant influences on the duration of unemployment yet the previous section showed that (with the exception of schooling) these are key determinants of whether one becomes unemployed. More on the job-search process The Social Protection chapter of the Poverty Assessment has shown that unemployment benefits are progressive although it is important to recognize that the main objective of unemployment benefits is not poverty reduction but easing transition between jobs. This section adds to that discussion. Unemployment benefits are administered by the National Agency for Employment (NAE). The benefits are paid for a maximum duration of 12 months (depending on the duration of contribution to the unemployment fund) at 75 percent of minimum wage. The unemployment benefits are also paid new graduates without jobs and to those who have completed military service and could not find employment for a six month period, the monthly benefit being set at 50 percent of the minimum wage. The resources come from the unemployment insurance fund, although adjustments (in the form of loans from Treasury) have occurred in the past. The fund experienced surpluses in 2001 and 2002. The NAE also administers the active measures program which includes counseling services, vocational training, job search assistance, job fairs, 22 Milan Vodopivec, Andreas Worgotter and Dhushyanth Raju. 2003. Unemployment Benefit Systems in Central and Eastern Europe: A Review of the 1990s. Social Protection Discussion Paper No. 0310. The World Bank. 23 Three key methodological issues are the following. First, similar to the analysis of predictors of falling into unemployment, the estimation sample is limited to those individuals who became unemployed during the period under observation. The surveys do ask about unemployment duration to the individuals, but inclusion of those who entered the survey as unemployed would be still problematic because then one has to deal with selection bias (the sample would over represent the long-term unemployed). Second, with the exception of the 2002 questionnaire, the labor force surveys did not inquire about the time of starting employment for those individuals who were unemployed at the time of the previous survey but found employment sometime after that (but before the follow-up survey that they revealed themselves as employed). Thus, hazard rate models that require precise information on the timing of transitions are not applicable. Here interval regression is utilized for estimation instead. Finally, the individuals who became unemployed while under observation but who were still unemployed at the time of the last follow-up survey contribute information to the maximum likelihood function as observations with upper-censoring. 93

temporary employment in public works etc. The spending on active programs tend to be small, representing about 2.5% of the unemployment fund budget in recent years. 24 Some insights on the effectiveness of active labor programs in Romania arise from a quasiexperimental evaluation conducted in 2002, which focused on the impact of four active labor market programs: training and retraining (TR), small business consultancy and assistance (SB), public works community job creation (PW), employment and relocation (ER). 25 The findings include the positive impact of TR, SB and ER on participants employment outcomes; and the positive impact of SB and ER on participants monthly earnings. The PW was not associated with positive outcomes, if anything the study revealed some negative impact on likelihood of future employment which is not easy to justify (this is likely to be due to unobserved characteristics of the program participants versus the characteristics of the control group which was constructed with some assumptions). The finding that public works programs do not work is not new. 26 Even for the programs that the above study found positive impact, the cost-benefit ratio is not known. Thus available evidence raises questions about the payoffs to active labor programs in general and in public works in particular. In the Romanian context, pubic works programs are believed to be especially difficult to implement in rural areas. The NAE identifies young people, women, long-term unemployed, disabled persons, over 45- year old persons, Roma and unemployed from other disadvantaged categories as target groups for the active measures program. 27 This is a comprehensive list, maybe too comprehensive in that essentially only the middle aged males are excluded from the targeting scheme. In practice, such a targeting scheme means there is no targeting. For example, while the Roma can benefit from the employment office functions, special activities towards Roma consist of rarely held job fairs. Same goes for women: once or twice a year job fairs for women are unlikely to have much of an impact. The lack of a functioning targeting scheme is not necessarily a problem for all the groups listed above. Given the need to prioritize, some of these individuals should not be targeted to start with. For the younger individuals, the main intervention is the payments to employers for 12 months, a monthly payment equal to the minimum wage for hiring new graduates with open ended contracts. The employers have to keep these employees for at least three years, otherwise monetary reimbursement with interest is required. Wage subsidy programs are also suspect when it comes to their effectiveness: as emphasized by the previous World Bank Poverty Assessment for Romania (1997), the firms might respond to incentives for hiring new graduates not by increasing overall number of employees but by substituting the cheaper labor for more expensive labor. 24 For more information, see ; Joint Assessment of Employment Priorities in Romania, October 2002, Ministry of Labour and Social Solidarity and EU. Based on information provided by the National Statistics Institute, active measures expenditures in 2002 were as follows in billions of lei: qualification/requalification programs (41), graduates hiring payments (201), hiring within the unemployment benefit period (118), bonuses for labor force mobility (16), encouraging employers to hire persons from disadvantaged categories (62), public works expenditures (118), credits (1000). 25 Impact of Active Labor Market Programs in Romania. 2002. Abt Associates Inc. Manuscript prepared for the Ministry of Labor and Social Protection. 26 More generally, a review of studies which evaluate the effectiveness of active-labor-market-programs in a number of developed and developing countries reveals that in many cases cost-effectiveness of such programs are disappointing (Dar and Tzannatos, 1999). 27 National Strategy for Employment for 2002-2004. October 2001. Ministry of Labour and Social Solidarity, National Agency for Employment. 94

The investigation of job search methods in Romania, the role of employment agencies and the characteristics of those who benefit from those services also show that there is room for improving targeting. For those who are unemployed and looking for a job in 2001 (ages 15 to 64), the leading methods used for finding a job or for a supplementary work to the present job --- in the last 4 weeks is appeal to friends/relatives/colleagues/trade-unions (62.2 percent), direct approach to employers (54.7 percent), registering to the agencies for employment and vocational training (45.9 percent), answering to advertisements (22.7 percent), and publishing advertisements (7.7 percent). Each of the following categories had less than 3 percent of responses: made arrangements for a self-employment activity, registering at private job agencies, and other. Only 34.4 percent of these individuals used one single method for job search, and only 19 percent of those who were registered to the agencies for employment used this as the sole job search method. The women are more likely to mention using employment agency in the job search process (49 percent as opposed to 45.9 percent of all unemployed adults), so are those who are between ages 45 and 64 (52.6 percent). The younger unemployed adults (ages 15 to 23) are slightly less likely to mention the employment agency (43.1 percent). But, the Roma unemployed are significantly less likely to mention using employment agency in the job search process: Only 11.9 percent of Roma revealed using employment agency during job search (as opposed to 45.9 percent of all unemployed adults). Those who are estimated to be poor are also less likely to mention registering to the employment agency as a job search method (41.2 percent versus 50 percent). 28 The other differences between the poor and the non-poor are minor, the poor being slightly more likely to rely on direct approach to employers and appeal to friends, relatives and less likely to respond to advertisements etc. The calculations for the poor are rough estimates, since poverty status is estimated with the help of HBS data. Having said that, even if there is a large error margin, clearly the poor unemployed are not more likely to utilize employment agency as part of their job search process. 4. The individual level determinants of employment and wages The analysis of employment and wages in Romania has direct implications for poverty. First, as shown previously, unemployment is a rather weak predictor of poverty: those who work in the informal sector are also likely to be poor, one question that comes to mind is do informal sector participants earn less even after controlling for their observed characteristics such as schooling attainment?. Other predictors of employment and wages include non-income dimensions of poverty (in particular health and schooling). 28 Since Labor Force Surveys do not contain consumption information, those who are poor are estimated as follows. First, for those who are unemployed and between ages 15 and 64, probability of being poor is estimated (via probit model) using 2002 Household Budget Survey data. The explanatory variables are age, agesquared, gender, marital status, household size, schooling, ethnicity and urban/rural residence indicator. Then, the estimated coefficients are used to predict probability of being poor in the 2002 Labor Force Survey data. Since about 45 percent of unemployed adults are found to be poor in the HBS, 45 percent of those with highest probability of being poor are designated as estimated poor in the LFS. 95

Using the 2002 Romania Living Conditions Survey, this section investigates the determinants of being employed as well as determinants of income from employment conditional on employment status. 29 Only the key findings will be highlighted below. Employment The determinants of employment is investigated by a number of alternative models, reported in Tables 7 to 11. The basic set of explanatory variables are: potential experience (and experience squared), gender, marital status, schooling attainment, ethnicity and urban/rural residence. Variations on this framework include additions of health status indicators and estimating separate models for males and females as well as for those who reside in urban and rural areas. Probit model is used for estimation, where the dependent variable takes the value 1 to indicate employment and 0 otherwise. Probability of employment increases by age, although the paste of increase declines as an individual gets older. Marriage increases the probability of employment by 9 percent, when all other explanatory variables are at their mean values (first column of Table 7). This is a common finding in the literature, and it can be explained by both the selection argument (i.e., the individuals with desirable labor market characteristics are more likely to be married) or by the beneficial causal effects of marriage argument. Schooling significantly increases the probability of employment with the highest schooling category increasing employment chances by 43 percent (relative to no schooling). Ethnicity variables do not have statistically significant effects on employment (with the exception of category other which is the residual category after identifying major ethnic groups). Finally, urban residence decreases probability of employment by 8 percent. Further insights can be gained by expanding on this reduced form specification. Note that there are other interesting points that can be made by comparing different models reported in Tables 7 to 11, the discussion here is very selective. When separate models are estimated for urban and rural areas, we find that schooling have a much larger influence on probability of employment in urban areas compared to rural areas (Tables 8 and 9). Marriage increases the probability of employment of males in urban and rural areas, but it decreases probability of employment of females (by 14 percent) in rural areas the effect is not statistically significant for urban females. A key finding emerges when an indicator or reporting chronic illness is included in the model (Table 10). Even after controlling for other factors, those reporting chronic illness are 30 percent less likely to work. The effect is larger for males at 34 percent, and for females at 25 percent. This finding supports the responses to direct questions on reasons for stopping work, and the magnitudes are quite startling. 30 The health-work relationship is further confirmed by models that use self evaluated health in place of chronic illness variable (Table 11). A further look at the predictors of chronic illness reveal that compared to Romanian ethnicity, the Roma are 8 percent more likely to report chronic illness and Hungarians are 4 percent more likely to 29 It might be beneficial to estimate wage equations by taking into account employment-selection via Heckman type models. Finding valid instrumental variables for the first-stage employment equation is the challenge in that case. 30 In an employment equation, health indicators are endogenous and thus the causation is open to question. But it seems unlikely that if health is treated as an endogenous variable such a sizable correlation disappears altogether. 96

report chronic illness (Table 6). Other parameter estimates reveal patterns that are similar to those observed in other countries: the elderly report more chronic illness, males report less, married report less, schooling attainment has a mixed story and reporting chronic illness is more likely in urban areas. Probably the last two findings are in part due to the fact that better educated individuals and those residing in urban areas have better access to health information and health services (thus conditional on having a chronic illness which we do not observe they may be more likely to report chronic illness). Among those who have chronic illness, the educated are less likely to report that the illness limits their ability to work or affect daily activities (percent reporting very much as limitation steadily declines with schooling from 62 percent for uneducated to 26 for those who had more-than-high-school education). Two possible explanations for this association is (i) the work and daily activities of the uneducated (who are much more likely to be poor) are more challenging, so having a disease has a more catastrophic effect on their life, and (ii) the educated are more likely to have effective treatment for the chronic diseases that they report, so for the educated the restrictiveness of diseases is minimized. Future research should attempt to distinguish between these two explanations, since policy implications differ depending on the dominating effect. If the first explanation is the explanation, then the policy interventions might focus on improving work/life conditions of the poor etc. If the second explanation is more important, then one may need to focus on improving the access to quality care of the uneducated/poor individuals. Everyone, everyone goes to work for food, as daily laborers. We find employment, day by day (Low income respondents, Alunis, rural. 2003) The payroll tax increases starting from 1998 can be expected to result in more unemployment and more informal sector work, since labor becomes more expensive in the formal sector. Figure 9 shows that as payroll taxes increase, percentage of employees decrease and percent self-employed & unemployed increase. This finding is verified by multivariate regressions at the individual level which control for changes in the characteristics of adult population over time (not reported). Figure 9. Payroll Taxes, Employees, and Self-Employment/Unemployment 1995-2002, Ages 15-64 (World Bank staff calculations based on HBS) 60 55 50 45 Percent 40 35 30 25 20 1995 1996 1997 1998 1999 2000 2001 2002 Year Payroll tax Employee All self employed & unemployed 97

Income from employment As shown by IMF (2003), the economy-wide average gross wage in Romania is low compared to other transition economies (at around US$ 140 in 2001), which strengthens Romania s competitiveness in labor intensive industries. 31 Until recently minimum wage influenced a small portion of the labor force: in 2002 it was 1,750,000 lei (gross), which corresponded roughly to US$ 52. But in 2003 gross minimum wage became 2,500,000 lei (or US$ 75). The predictors of earnings are discussed in the remainder of the document from a poverty perspective, recognizing that the CEM will include a more comprehensive analysis of wage setting policies. Various documents observe that real wages are higher in public (44 percent higher in 2002) than in the private sector. 32 It seems evident that wages in long-term loss-making large public enterprises are not linked to productivity, but what additional insights can one obtain? For example, while it is known that wages are higher in the public sector, to what extent is it because public sector employees have higher qualifications? In other words, does the wage differentia l between public and private sectors remain after taking into account schooling and other characteristics of individuals? For brevity, this discussion will focus on the direction of effects (magnitudes can be derived from the estimated coefficients reported in Tables 12 to 15). The basic insights from the baseline income regression (using the same set of variables used for the baseline employment model described previously) are that males earn more than females, increased schooling has a large and consistently improving effect on wages, and those residing in urban areas earn more. The only ethnicity variable that is statistically significant is the identifier for Hungarians, who earn more than Romanians after controlling for other variables in the model. The Roma earn slightly less (as always, after controlling for other variables such as schooling etc.) but the effect is not statistically significant. When public and private sector employment indicators are added to the model, the Roma are found to earn more (first column of Table 15). Second and third columns of the same table report estimation results separately for those who are employed in the public sector and for those who are employed in the private sector. These estimates show that the Roma who have private sector employment earn more than others who report private sector employment, and in the public sector employment regression the Roma ethnicity dummy is not statistically significant (but negative). The male-female wage differential is robust to alternative specifications not only in models that control for schooling etc. but even after taking into account sector (public/private) and type of employment (employee/employer/self-employed). 33 In fact, the gender gap prevails even if separate regressions are run for public sector and private sector (admittedly the gender gap in wages is smaller in the public sector). On the contrary, after controlling for schooling etc., the Roma earn as much as others do. In other words, there is no evidence of ethnic discrimination on 31 Romania: Selected Issues and Statistical Appendix. January 2003. IMF Country Report No. 03/12. 32 See, for example, Monthly Statistical Bulletin, No 1/2002, NIS; Joint Assessment of Employment Priorities in Romania, October 2002, Ministry of Labour and Social Solidarity and EU. 33 Making the full-time versus part-time employment distinction also does not influence the findings. In 2002, 89 percent of males and 87 percent of females considered their work as full-time. 98