Very Preliminary Result Labor Market Vulnerability in Indonesia: A Synthetic Cohort Panel Simulation Exercise. Vivi Alatas This Version September,2002 1. Introduction Poor workers faced many risks. Loosing job, changing jobs, being ill, economic downturn can lead to a decline in their well-being. Many workers while currently may have a good job are vulnerable to events that can easily put them into a lower pay job brackets. Variation in take home pays may be of a short-term nature or a long-term nature, and it may be affected by macro shocks or micro shocks that are specifics to particular individuals. Wages may varies due to seasonal factors as overtime and number of hours work may depend on the business cycle or it may varies due to misfortune and bad luck such as illness or being laid off. Since labor market segmentation may affect different degree of mobility between different groups, people are generally not free to advance in more secure jobs. Workers are vulnerable to shocks and this notion of workers vulnerability is not capture by just looking at a snapshot of who has low pay at a given point of time. This study examines Indonesia Labor Market Vulnerability during 1988-2001. It primarily concerns estimating vulnerability defined as the probability of having earnings below a threshold at least once over the next three years. There are several interesting empirical questions that can be raised. Do we see a different degree of vulnerability and mobility for young and old, for men and women and for different sector? Is vulnerability increasing over time? Do increase trade exposure translate into higher vulnerability for related sectors? A lot of things happened during the period of the study. The trade liberalization measures introduced in the mid 1980’s, the labor market distortions through establishing minimum wage law and other labor regulations concerning unions movement in early 1990s, and the financial crisis of 1997/98 directly affect the earnings mobility of workers. While trade liberalization may have been a positive factor in promoting growth and jobs creation in certain sectors, it may induce greater earning mobility through increased exposure to more and larger fluctuations of the world markets. This study is a part of a larger study that examines whether trade liberalization transmits more shocks into more exposed sectors of the domestic economy, making certain groups of workers more vulnerable. The paper is structured as follows. Section 2 presents and discusses the model and methodology used. It presents the method used in exploiting the synthetic cohort to provide estimates of vulnerability. Section 3 described the data sources. Section 4 examined the results that can give insight how vulnerability varies between groups and over the period. Section 5 summarized the conclusion and the need for further studies. 2. Model and Methodology A simple snapshot of cross-section picture hides the dynamic and mobility of earnings. The distribution of earnings may look similar in different years, while there may have been a great deal of dynamic movement. Ideally, studying vulnerability requires panel data that allow us to track individuals or households over time. We can then examine the dynamics of earnings and distinguish occupational choice shocks, earnings shocks and coping mechanism chosen such as increased participation, consumption smoothing, running down of assets, rely on other people to share risk with friends and kin. In the past there have been two attempts in directly measuring vulnerability in Indonesia using panel data. The first one is by Pritchett et. all using Panel of 100 village data and Panel of Susenas Mini Data. The second one is by Suryahadi et. all using Subham Chaudury’s method that utilizing cross-section data. There are some issues in using the above two methods. They assumed that the stochastic process generating consumption of future data is the same with current process. Each year is just another draw of the same micro shock process. The proposed synthetic cohorts methods comes from the following considerations. Micro shocks are a stochastic process; the micro shock this year may affect the micro shocks of next year. Assuming that each year is just another draw of the same micro shock process is too strong. Unfortunately panel data with national coverage is not available for recent years. However we have repeated cross section data on earnings and workers characteristics. How can we take advantage of the availability of such data? One possibility is through "pseudo-panel analysis" using synthetic cohort panel. This consists of following cohort of individuals with identical time invariant characteristics over time. We define cohort based on birth year (can be grouped into 5 year period of birth years), gender and skill. For instance, following all men born in 1970 with less than primary education between 1988 and 2001, one is actually observing each year between 1988 and 2001 a sample of the same population over time. One may then follow what happens to a representative individual of that cohort – i.e. the mean individual – but also to the heterogeneity of that group with respect to particular characteristics, in our case here the variance of the earnings residuals. The whole idea is to try to reconstitute stories about vulnerability for various labor market characteristics. Briefly, the method is as follows. Assume the earnings, wit, of individual i belonging to cohort group j at time t may be represented by the following equation: where Xit is a set of characteristics that are not used in the definition of the cohort group j. Supposed in addition that that the unobserved residual term ξ itj follows a process that is AR(1). By estimating (1) on each cross-section t for group j, one may get estimates of the variance of ξ itj , σ ξ2jt Then taking the variance of (2), one gets : σ ξ2jt = Ï? 2σ ξ2jt −1 + a j + bt + c age + stj 2 (3) By running this regression on the observed cohort variances, it is possible to get an estimates of Ï? , aj , bt and c. The term in c stands here for an age effect in the variance of the innovation, while the term aj depicts cohort effect and the term bt, represents year effect. Then, stj 2 is obtained as a residual of this equation and the variance of the innovation term, ε itj , in (2) may be obtained as the sum: a j + bt + ct + stj 2 (4) Once all this is done, we are very close to being able to simulate vulnerability. Combining (1) and (2) and the estimate (4) of the variance of ε itj , one can easily figure out what is the probability of any sequence of earning levels ( witj +1 , witj ) starting from the observation of witj and under the assumption that changes in the variables X in (1) – is there are indeed such time varying variables- are predictable. Then the probability that an individual observed as having low pay in year t be still poor in year t+1, or on the contrary above the threshold level may be easily evaluated assuming the innovation term is distributed as a normal variable. ( Pr witj +1 ≤ k X it , ξ it ) = ( Pr ln witj +1 ≤ ln k X it , ξ it ) = ( ˆ Pr ξ it +1 ≤ ln k − X it β X it , ξ it ) ξ it +1 can be easily simulated using the estimation of equation 3 and following the properties of conditional normal distribution. Similarly we can simulate ξ it + 2 and ξ it + 3 . Here we defined vulnerability as the probability of having earnings below the threshold at least once in the next three years. Without panel data but with repeated cross-sections, the preceding method is probably the best way of getting at some estimate of vulnerability and who is affected by it. A generalization of the preceding model would consist of considering at the same time changes in earnings for active individuals or income for households and changes in occupational status or family composition. This would require more complex econometric techniques. 3. Data and Sample Restriction The data used for this analysis is the Labor Force Surveys (Survei Angkatan Kerja Nasional ) or usually known as SAKERNAS, except for 1995 data where we use Inter- census survey (SUPAS). Because SAKERNAS was not collected during inter-census years (years that end with 5), we can instead make use SUPAS 95 as it contained the required questions for this study. The SAKERNAS survey is a nationally representative survey that covers approximately 35,000 households or 250,000 individuals each year from all provinces of Indonesia. It contains information on individual earnings and hours of work on the primary job, sector of primary as well as data on individual characteristics such as gender, age, and education level. Thus Sakernas will provide us necessary information that allows us to estimate earnings functions using Mincerian models. Between 1986 and 1988, the SAKERNAS survey provides only a rough classification for the economic sector of employment of individuals, consisting of only 5 codes for agriculture, manufacturing, trade, services and other industries. Although starting 1989, the number of codes for the sector of employment was expanded from 5 to 18, allowing a more disaggregate analysis, for the sake comparability we have to rely on aggregation of the 18 different codes into the 5 original classifications. We divide the data into 3 different sectors of employment, agriculture, manufacturing and versus non-tradable. Although Sakernas is quite a large data sets, it is still not large enough to allow us to have a refined disaggregation of education. Since we do not have enough observation to do synthetic cohort estimation for female workers with high education, we break their education into primary and secondary only. With larger number of observations, we can separately estimates the model for male workers with primary, secondary and high education for each sector. Sakernas provides us with labor force participation information for individual aged 10 years and above. However for this study, we restricted the sample only for 15 years old to 60 years old workers. 4. Results of Estimation The construction of a “synthetic-panelâ€? is done by dividing the sample into cohorts, groups that have a common characteristic. Here we defined cohorts as having the same 5- year period of birth years, gender, education and sector. For each cohort the estimation of equation (1) will provide us estimates of variance of earnings residuals. Using Sakernas 1988-2001 we will get a stacked data of estimated variance of earnings residuals by birth years, gender, education and sector. In considering earnings vulnerability, two thresholds for poverty are considered. The first one is absolute poverty, whose threshold is given by $1 adjusted PPP per day. And the other is relative poverty, by looking at quintile transition matrix. A dollar per day PPP threshold is widely used as international comparable poverty line. To have similar threshold for earnings we need to make some adjustment. If α is defined as ratio of wages to consumption then we will have that ∑w i wageearners =α consumption c * sizeh * α wi = h # wageearners with c h being per capita expenditure. To get an analog threshold for earnings, the PPP $ threshold need to be adjusted by a factor that take into account the ratio of wage earnings to expenditure and the relative size of wage earners in a household. 4.1 Estimates of Unobserved Heterogeneity Table 1 presented the estimates of Ï? 2 in equation (3) in the model. We see that the estimated Ï? 2 ranges from 0.06 to 0.47. All of these Ï? 2 are statistically significant (with the smallest t values being 3.5) except for female workers with secondary education in non-tradable sector. It supports the assumption of AR(1) process in earnings residual. The table also shows that the correlation of variance of residual for workers with primary education is pretty similar in agriculture and non-tradable sector, while primary educated workers in manufacturing sector have slightly smaller correlation. For agriculture and manufacturing sector, unlike non-tradable sector, we see that workers with secondary education portray similar degree of persistency of shocks in comparison with workers with primary education. Generally female workers have smaller correlation than their male counterpart except in manufacturing sector where their differences are not significant. Estimation of equation (3) will also provide us with decomposition of variance of earnings into age, gender and year effect. The appendix explains the method and results of this decomposition. 4.2 Mobility and Vulnerability in 2001 With lack of continuous time series data on effective protection, what we can do in analyzing the impact of trade liberalization on labor market vulnerability is to examine closely the evolution of the labor vulnerability rate during the period under study. However, before exploring the evolution of vulnerability rate, we need to look more closely who are currently the workers with ‘bad luck’, having earnings below the threshold 1 $ PPP/day or 2 $ PPP/day. As Sakernas 2001 is the most recent data1, we have to rely on this data to show us the most current result. In 2001, 10.8 % of the workers have earnings below the 1 $ PPP/ day. Table 2a and 2b revealed that these workers tend to have low education, work in agriculture, and live in rural areas. The data also indicates that female workers are three times more likely having earnings below the 1$/ day threshold. Not surprisingly, working in similar sector, with similar education, female workers have much larger tendency of earnings below the threshold. Sakernas 2001 shows that the proportion of male workers earnings below 1 $ / day threshold is 6.35 % while the proportion of female workers earnings below 1 $/day threshold is 20.7 %. The phenomena that female workers have more difficulty in securing good jobs appear in all sectors and level of educations. In some cases, for example primary educated workers working in manufacturing sector, the ratio of male to female proportion workers below the threshold is more than 500 %. If we change the threshold to 2 US $ the proportion of workers earning below this threshold jumps by three fold. With 2 $/day threshold almost all of female workers with primary education in agriculture and non-tradable sectors earn below the threshold. The data also revealed that workers in agriculture sectors have the highest proportion of earnings below the threshold2. Manufacturing sectors appears to have much lower proportion of workers with low pay jobs. While 18% of the primary educated male workers in agriculture have low pay job, we observe only around 5 % of the same group of workers in manufacturing earns below the threshold. Table 2a and 2b also indicates that although in some sectors especially for secondary and high education level we have quite a low number proportion of workers having earnings below 1$ PPP/day, however with 2 $/day threshold the proportion of workers having low pay job hugely increase enormously, almost seven times in the case female workers with secondary education working in manufacturing sectors. The huge differences in the proportion of workers having low pay jobs between US1$ PPP and US2$ PPP is a strong 1 Sakernas is conducted annually in August, at the moment Sakernas 2002 is being implemented in the field. 2 We do not have enough observations of male agriculture workers with high education indication that a large of the population is hanging around the threshold and are constantly vulnerable to falling into the low brackets. It is indeed the case. If we use the threshold of 1$ PPP adjusted/day, based on Sakernas 2001 the proportion of workers who are vulnerable to having low earnings is 25.7 %. With the ratio of vulnerability (25.7 %) to proportion earnings below the threshold (10.9 %) is more than double it clearly indicates that a lot of people who are having good luck nevertheless are at substantial risk of loosing their luck. Table 2c presents the profile vulnerability rate (based on 1 US $ PPP/ day) for different sectors, gender and education level. One salient pattern that we get from the table is indeed female workers are facing higher risk than male workers. Three out of four females with primary education who work in agriculture are vulnerable to low pay jobs, while the vulnerability rate for male workers in the group is around 44 %. In all sectors and all level of education, vulnerability rate of female workers is much higher than their male counterparts. The pattern of vulnerability rate among sectors is quite similar with the pattern of proportion of workers earning below the threshold that we observe in table 2a and table 2b. Although in magnitude agriculture workers portrays higher risk of falling into the low pay jobs, manufacturing workers have much higher ratio vulnerable workers to proportion of workers earning below the threshold as shown in table 3a. This indicate that in manufacturing sectors, we have a large bulk of workers who is currently have a ‘good job’ are nevertheless facing constant risk of loosing their luck. With the ratio of vulnerability to proportion earnings below the threshold is more than double it clearly indicates that who has ‘bad luck’ at any given point of time is not static, people entering and exiting the threshold. It is very fluid. We can try to look at the dynamic of labor market vulnerability by decompose the vulnerability into two probabilities. As we defined vulnerability as the probability of having earnings below the threshold at least once in the next three years (year t+1, t+2, t+3). Law of iterated expectations will give that this probability can be decompose into two sums of probabilities: Prob(Vulnerable at t) = Prob (Vulnerable at t and having earnings below the threshold at t) + Prob (Vulnerable at t and having earnings above the threshold at t). Decomposing the vulnerability rate into these two probabilities will give some insight about the persistency of the bad luck and also the fluidity of earnings. Table 3b and 3c presents the dynamics of labor market vulnerability. We observe in table 3b that people do escape from low job pays and the likelihood of escaping is greater with higher education. The persistency of bad luck is insignificant for male with high education. Only around 10% to 20 % of male with high education who have low job pay in 2001 are predicted of experiencing at least once episode of low pay again in the next three years. Again, the table shows persistency of gender disparities. Low skill female workers in agriculture and non-tradable sectors only have 10 % chances of escaping from low pay jobs. The tables show people exiting and entering the threshold. 4.3 The Evolution of Vulnerability Rate. Looking at the evolution of vulnerability rate since 1988 to 2001 (figure 1), it become apparent that labor market vulnerability pattern is similar to the pattern of poverty evolution and the pattern of proportion of workers having earnings below 1 $ PPP/day. The vulnerability rate fell sharply between 1988 and 1994, follow by a slight decrease up to 1997. During the crisis, the vulnerability rate has increased substantially – the peak in august 1998 is one and a half time the pre-crisis august 1997. These phenomena happened in all groups, male and female, all sectors, and all level of education. It confirms the general perceptions that the effect of the crisis in the labor market is through reducing wages instead of displacing workers out of labor force (Manning, 2000). It clearly suggests that the crisis has unambiguously increased the level of vulnerability in labor market. However more recent Sakernas data have indicated that since then vulnerability has fallen. We observe a hump shape curve post 1997 with the peak being in august 1998. The august 2000 vulnerability rate is comparable to the 1996 level. This phenomenon is similar to what happen to expenditure poverty (Alatas, 2002). In addition, Figure 1 shows that in 2001 the level of vulnerability and also proportion of workers earnings below 1$ PPP/day is already below the pre-crisis level. Although the vulnerability rate is generally lower among men, the time series pattern of the vulnerability rate is similar between men and women, as shown in figure 2. We clearly can recognize the same trend of fast reduction till 1994 follow by steady period till right before the crisis and then continue with a hump shaped pattern since the crisis. The steady period of 1994-1997 also exhibits signals of narrowing gap between male and female. Though, this indication was reversed during the early years of the crisis when we see a tendency of larger gap between male and female in comparison to the pre-crisis years. Again, 2001 shows as a promising year. Not only it exhibits a substantial decrease in vulnerability rate for both men and women, it also shows a narrower gap between men and women. 4.4 Vulnerability across Different Groups. Different groups faced different risks. An interesting empirical question is to examine how shocks vary among different groups. We examine the results for different sectors, gender and education level. The effect of Gender. The above discussion has shown persistent gender disparities in labor market. Female workers are three times more likely having earnings below the 1$/ day threshold. Working in similar sector, with similar education, female workers have much larger tendency of earnings below the threshold. Sakernas 2001 shows that the proportion of male workers earnings below 1 $ / day threshold is 6.35 % while the proportion of female workers earnings below 1 $/day threshold is 20.7 %. The phenomenon that female workers have more difficulty in securing good jobs appears in all sectors and level of educations. In some cases, for example primary educated workers working in manufacturing sector, the ratio of male to female proportion workers below the threshold is more than 500 %. Female workers also have a lower chance escaping bad luck comparing to male workers. Low skill female workers in agriculture and non-tradable sectors only have 10 % chances of escaping from low pay jobs, while their male counterpart have 30 % chances. Out of those who currently have earnings above the thresholds, female workers are more likely to loose their ‘good job’. As a result, we observe that female workers have greater rate of vulnerability. Education and Vulnerability. The analysis of vulnerability by education level in general gives expected results. The vulnerability and headcount rate are lower the higher the education for all sectors for both male and female. This is not surprising as high endowment of education increased the chances to secure a good job that offers more stability and higher wages. It is also not surprising with increasing function of return to education (Alatas and Bourguignon, 2000) that we have a negative correlation between education and the level of vulnerability. However, Table 3a indicates that the ratio of vulnerability to poverty rate is larger for workers with secondary education than the ratio for workers with primary education. Workers having high education have more earnings security, as indicates by low-level ratio, in addition to a low level proportion of having low pay at the first place. Figure 5 also shows that the crisis did not have significant effect on the earnings security for workers with high educations in non-tradable sectors. More precisely, it did not triggered massive shocks that push them down to low pay jobs or increase their risk of having such episodes in the near future. Although they did experience reduce real wages during the crisis (Manning, 2000), it still above the cut off thresholds. Vulnerability among sectors. Figure 4-6 shows the evolution of vulnerability across sectors. These figures clearly show that the vulnerability rates are highest in agriculture for both male and female workers. However we need to take into account that the proportion of workers earning below the threshold in the based year is much larger for agriculture than for other sectors. The evolution of vulnerability rate for male worker with primary education is quite similar for manufacturing and non-tradable sector as shown in figure 4. Until 2000, their vulnerability rates for manufacture workers were always slightly above the corresponding rate for non-tradable sectors. It caught up in 2000, and actually became slightly below the non-tradable rate. We observe quite a large gap between agriculture and the other two sectors, especially for male workers. There is a slight tendency that this gap is narrowing in recent years, however still even in 2001 the vulnerability rate of male agriculture workers is more than double the rates in other sectors. Unlike the male workers, vulnerability rates for female manufacturing worker are consistently below the other sectors rates for all years. The hump shaped pattern is more apparent for female manufacturing workers. We observe much substantial increase of vulnerability for female manufacturing workers during the early years of the crisis, on the other hand the decrease of vulnerability rate since the peak in 1998 is also much faster for female manufacturing workers. Indeed, their vulnerability rate in 2001 is similar to the vulnerability rate of male agriculture workers. Figure 5 and figure 6 also exhibit similar pattern like previous graphs, decline in the rates follow by a hump shaped curve during the crisis, expect for high educated male workers in non-tradable sectors where the vulnerability rate is quite steady over the period under study. The evolution of vulnerability rate for male workers with secondary education is quite similar with the pattern observed for male workers with primary education. In the early years, the rates for manufacture workers were always slightly on top of the corresponding rate for non-tradable sectors. It caught up in 2000, and in 2001 vulnerability rate for manufacturing workers is actually below the non-tradable rate. 5. Conclusions and Further Studies (churching, I will need to work further on this) This paper shows that vulnerability in labor market is a serious problem. In a sample in which we have only 10.8 of the workers having earnings below the threshold, we have 25.7 % workers at risk of experiencing a period of low pay in the near future. We also observe significant differences of vulnerability rate across groups. The phenomena that female workers have more difficulty in securing good jobs appear in all sectors and level of educations. We also observed that the vulnerability and headcount rate are lower the higher the education for all sectors for both male and female. Examining the evolution of vulnerability rate over the period, one salient features emerged: labor market vulnerability pattern is similar to the pattern of poverty evolution. In all groups of workers, we clearly can recognize the same trend of fast reduction till 1994 follow by steady period till right before the crisis and then continue with a hump shaped pattern since the crisis. In 2001 the level of vulnerability and also proportion of workers earnings below 1$ PPP/day is already below the pre-crisis level. The method used assumption that there is a mean preserving process between each years of consumption, which most likely is not true. This assumes that household expenditures are expected to be the same in each period. For further studies, we can modify it based on our work on the macro forecast what will happen to the expected future growth (could be by sector or by type of households). We can also try incorporated forecasted macro shocks or forecasted labor market and household endowment projections. Appendix A: Decomposition of Variance of Earnings into Age, Gender and Year Effect Like the mean of wages, the variance of residual earnings may show characteristics life- cycle profiles. In addition young generation and older generation may have different profile of variance of earnings. Looking at a single cross section data will not help isolating the different effects as a cross section data blends together the age effect, the cohort effect and the year effect. Deaton (1997) shows how we can decomposed the cohort data to separately isolate the different effects. We followed deaton method in decomposing the cohort data. The results of the decomposition are presented in figure A1-figure A6. Table 1: Correlation of Variance of Residuals Agriculture Manufacturing Non-Tradable Sector Education Male Female Male Female Male Female Primary 0.4716 0.4083 0.3548 0.3746 0.4583 0.4098 Secondary 0.4333 0.3954 0.3547 0.3549 0.1841 0.0617 High 0.3525 N.A 0.4124 N.A 0.2165 N.A Notes: For female workers we combine the secondary and high education as not enough observations prevent us from doing synthetic cohort analysis for female with high education separately. Table 2a: Proportion of Workers Earning below 1US$ PPP/Day Adjusted Threshold by Sector, Skill and Gender, 2001 Agriculture Manufacturing Non-Tradable Sector Education Male Female Male Female Male Female Primary 0.1791 0.4658 0.0493 0.2409 0.0960 0.4135 Secondary 0.0583 0.2000 0.0123 0.0565 0.0353 0.0923 High 0.0044 0.0265 Table 2b: Proportion of Workers Earning below 2 US$ PPP/Day Adjusted Threshold by Sector, Skill and Gender, 2001 Agriculture Manufacturing Non-Tradable Sector Education Male Female Male Female Male Female Primary 0.6033 0.8913 0.4067 0.7003 0.4106 0.9221 Secondary 0.2945 0.6222 0.1997 0.3497 0.2013 0.2909 High 0.0439 0.0577 Table 2c: Mean of Probability being Vulnerable (based on 1 US$ PPP/day) by Sector, Skill and Gender, 2001 Agriculture Manufacturing Non-Tradable Sector Education Male Female Male Female Male Female Primary 0.4424 0.7571 0.2239 0.4849 0.2797 0.7596 Secondary 0.2136 0.4946 0.0969 0.1863 0.1189 0.2267 High 0.0054 0.0313 Table 3a: Ratio vulnerable workers to proportion of Workers Earning below 1 US$/Day adjusted Threshold by Sector, Skill and Gender, 2001 Agriculture Manufacturing Non-Tradable Sector Education Male Female Male Female Male Female Primary 2.4699 1.6252 4.5444 2.0129 2.9142 1.8368 Secondary 3.6669 2.4727 7.8835 3.2980 3.3625 2.4573 High 1.23 1.1824 Table 3b: Probability of being vulnerable and Earning below 1 US$/Day adjusted Threshold by Sector, Skill and Gender, 2001 Agriculture Manufacturing Non-Tradable Sector Education Male Female Male Female Male Female Primary 0.1358 0.4157 0.0268 0.1751 0.0650 0.3520 Secondary 0.0366 0.1746 0.0079 0.0310 0.0142 0.0490 High 0.0004 0.0053 Table 3c: Probability of being vulnerable and Earning above 1 US$/Day adjusted Threshold by Sector, Skill and Gender, 2001 Agriculture Manufacturing Non-Tradable Sector Education Male Female Male Female Male Female Primary 0.3066 0.3415 0.1971 0.3098 0.2147 0.4076 Secondary 0.1770 0.3200 0.0890 0.1552 0.1047 0.1777 High 0.0050 0.0261 Table 4: Mobility Matrix for Male Workers based on Sakernas 2001 Male Workers with Primary Male Workers with Primary Low Skill with Primary Education in Agriculture Sector Education in Manufacturing Education in Non-Tradable Sector Sector Low Pay High Pay Low Pay High Pay Low Pay High Pay Low Pay 0.6714 0.3286 Low Pay 0.5026 0.4974 Low Pay 0.6406 0.3594 High Pay 0.1301 0.8699 High Pay 0.0663 0.9337 High Pay 0.0687 0.9313 Male Workers with Secondary Male Workers with Secondary Male Workers with Secondary Education in Agriculture Sector Education in Manufacturing Education in Non-Tradable Sector Sector Low Pay High Pay Low Pay High Pay Low Pay High Pay Low Pay 0.5831 0.4169 Low Pay 0.6050 0.3950 Low Pay 0.3424 0.6576 High Pay 0.0597 0.9403 High Pay 0.0247 0.9753 High Pay 0.0338 0.9662 Table 5: Mobility Matrix for Female Workers based on Sakernas 2001 Mobility Matrix Mobility Matrix Mobility Matrix Low Skill Female Workers in Low Skill Female Workers in Low Skill Female Workers in Agriculture Sector Manufacturing Sector Non-Tradable Sector Low Pay High Pay Low Pay High Pay Low Pay High Pay Low Pay 0.7918 0.2082 Low Pay 0.6331 0.3669 Low Pay 0.7500 0.2500 High Pay 0.2877 0.7123 High Pay 0.1634 0.8366 High Pay 0.2952 0.7048 Mobility Matrix Mobility Matrix Mobility Matrix High Skill Female Workers in High Skill Female Workers in High Skill Female Workers in Agriculture Sector Manufacturing Sector Non-Tradable Sector Low Pay High Pay Low Pay High Pay Low Pay High Pay Low Pay 0.8122 0.1878 Low Pay 0.5043 0.4957 Low Pay 0.3528 0.6472 High Pay 0.1790 0.8210 High Pay 0.0551 0.9449 High Pay 0.0772 0.9228 Figure 1: The Evolution of Labor Vulnerability Rate 1 Vulnerability Rate 0.8 Proportion Workers below 1 0.6 $/day threshold 0.4 0.2 0 1988 1990 1992 1994 1996 1998 2000 2002 Year Figure 2: Vulnerability Rate 1 0.8 0.6 Male Female 0.4 0.2 0 1988 1990 1992 1994 1996 1998 2000 2002 Year Figure 3: Proportion Workers Below 1 $/day Threshold 1 0.8 Male 0.6 Female 0.4 0.2 0 1988 1990 1992 1994 1996 1998 2000 2002 Year Table 6: Decomposition of Vulnerability Rate Prob(Vulnerable Prob(Vulnerable Prob(Low Prob(Vulnerable Prob(Vulnerable & Prob( High Pay) Prob(Vulnerable) |High Pay) |Low Pay) Pay) & High Pay) Low Pay) 1988 0.384 0.873 0.579 0.421 0.223 0.368 0.590 1989 0.395 0.878 0.606 0.394 0.239 0.346 0.585 1990 0.350 0.864 0.635 0.365 0.222 0.316 0.538 1991 0.350 0.849 0.678 0.322 0.237 0.273 0.510 1992 0.346 0.857 0.715 0.285 0.248 0.244 0.492 1993 0.310 0.819 0.683 0.317 0.212 0.260 0.471 1994 0.274 0.775 0.784 0.216 0.215 0.167 0.382 1995 0.253 0.647 0.799 0.201 0.202 0.130 0.332 1996 0.246 0.763 0.814 0.186 0.200 0.142 0.342 1997 0.235 0.751 0.842 0.158 0.198 0.118 0.317 1998 0.348 0.821 0.752 0.248 0.262 0.203 0.465 1999 0.333 0.787 0.750 0.250 0.249 0.197 0.446 2000 0.276 0.776 0.824 0.176 0.228 0.137 0.364 2001 0.205 0.686 0.891 0.109 0.183 0.075 0.257 Figure 4: Labor Market Vulnerability Rate by Sectors Workers with Primary Education 1.0 0.8 Agriculture, Male Vulnerability Rate 0.6 Agriculture, Female Manufacture, Male 0.4 Manufacture, Female Non-tradeable, Male 0.2 Non-tradeable, Female 0.0 1988 1990 1992 1994 1996 1998 2000 Year Figure 5: Labor Markets Vulnerability Rate by Sectors Male Workers with Secondary and High Education 1.0 0.8 Agriculture, Secondary Education Vulnerability Rate 0.6 Manufacture, Secondary Education 0.4 Non-tradeable, Secondary Education Manufacture, High Education 0.2 Non-tradeable, High Education 0.0 1988 1990 1992 1994 1996 1998 2000 Year Figure 6: Labor Market Vulnerability by Sectors Female with Secondary Education 1.0 0.8 Vunerability Rate 0.6 Agriculture Manufacture 0.4 Non-Tradeable 0.2 0.0 1988 1990 1992 1994 1996 1998 2000 Year Figure A1: Age Effect Decomposition of Variance of Earnings Residual Male Workers Low Skill Agriculture . Low Skill Manufacturing . Low Skill Non-Tradable .2 3 3 3 .2 .2 8 ag 2 ag ag e e e .2 eff .2 eff 5 eff .2 6 ec 1 ec ec . .2 2 4 . .1 2 .2 9 0 2 4 6 0 2 4 6 2 0 2 4 6 0 age 0 0 0 age 0 0 0 age 0 0 . . 6 High Skill Agriculture . High Skill Manufacturing 6 High Skill Non-Tradable 5 . 5 ag . ag .4 ag 5 e e 5 e eff eff . eff 4 ec ec . ec . 4 4 . 3 .3 . 5 3 . 0 2 4 6 0 2 4 6 2 0 2 4 6 0 age 0 0 0 age 0 0 0 age 0 0 Figure A2: Age Effect Decomposition of Variance of Earnings Residual Female Workers Low Skill Agriculture Low Skill Manufacturing Low Skill Non-Tradable . . .2 2 3 6 .2 ag 0 ag 4 ag .2 5 e e e eff eff .2 eff 2 ec ec ec - . .2 2 . 2 .1 .1 - 8 5 .4 0 2 4 6 0 2 4 6 0 2 4 6 0 age 0 0 0 age 0 0 0 age 0 0 . 7 High Skill Agriculture . 3 High Skill Manufacturing .4 5 High Skill Non-Tradable . 6 ag ag ag . e e e 4 eff . eff .2 eff 5 5 ec ec ec .3 5 . 4 . . . 3 3 0 2 4 6 2 0 2 4 6 0 2 4 6 0 age 0 0 0 age 0 0 0 age 0 0 Figure A3: Cohort Effect Decomposition of Variance of Earnings Residual Male Workers 0 0 0 Low Skill Agriculture Low Skill Manufacturing Low Skill Non-Tradable co co co ho ho ho rt rt - rt - - eff .02 eff .05 eff .05 ec ec ec - - .1 - .04 0 2 4 6 0 2 4 6 .1 0 2 4 6 0 cohort 0 0 0 cohort 0 0 0 cohort 0 0 0 0 0 High Skill Agriculture High Skill Manufacturing High Skill Non-Tradable co co co - - ho ho .1 ho .1 rt rt rt - eff .2 eff eff ec ec - ec .2 - .2 - - .3 - .4 0 2 4 6 0 2 4 6 .3 0 2 4 6 0 cohort 0 0 0 cohort 0 0 0 cohort 0 0 Figure A4: Cohort Effect Decomposition of Variance of Earnings Residual Female Workers . 6 Low Skill Agriculture . 1 Low Skill Manufacturing . 1 Low Skill Non-Tradable co co co . .0 .0 ho 4 ho 5 ho 5 rt rt rt eff eff eff ec ec ec . 0 0 2 0 - - .05 .05 0 2 4 6 0 2 4 6 0 2 4 6 0 cohort 0 0 0 cohort 0 0 0 cohort 0 0 0 . 0 1 High Skill Agriculture High Skill Manufacturing co co co - 0 ho .2 ho ho rt rt rt eff eff - eff .05 ec ec ec - - .4 .1 - - - .6 0 2 4 6 .1 0 2 4 6 .2 0 2 4 6 0 cohort 0 0 0 cohort 0 0 0 cohort 0 0 Figure A5: Year Effect Decomposition of Variance of Earnings Residual Male Workers Low Skill Agriculture Low Skill Manufacturing Low Skill Non-Tradable .0 .0 2 2 .0 2 ye ye 0 ye ar ar ar eff 0 eff eff 0 ec ec ec t t - t .02 - .02 - - .02 198 199 199 200 .04 198 199 199 200 198 199 199 200 5 0 year 5 0 5 0 year 5 0 5 0 year 5 0 . 1 High Skill Agriculture .0 4 High Skill Manufacturing .0 4 High Skill Non-Tradable .0 ye .0 ye 2 ye .0 5 ar ar ar 2 eff eff 0 eff ec ec ec 0 t t t 0 - .02 - .05 - - 198 199 199 200 .04 198 199 199 200 .02 198 199 199 200 5 0 year 5 0 5 0 year 5 0 5 0 year 5 0 Figure A6: Year Effect Decomposition of Variance of Earnings Residual Female Workers Low Skill Agriculture Low Skill Manufacturing Low Skill Non-Tradable . 2 . .0 1 5 ye . ye .0 ye 1 5 ar ar ar eff eff eff 0 ec ec ec 0 0 t t t - - .1 .05 - .05 198 199 199 200 198 199 199 200 198 199 199 200 5 0 year 5 0 5 0 year 5 0 5 0 year 5 0 . . . 1 High Skill Agriculture 1 High Skill Manufacturing 1 High Skill Non-Tradable .0 5 ye ye .0 ye .0 5 5 ar ar ar eff 0 eff eff ec ec ec t t 0 t 0 - .05 - - - .1 .05 .05 198 199 199 200 198 199 199 200 198 199 199 200 5 0 year 5 0 5 0 year 5 0 5 0 year 5 0