Labour Economics 9 (2002) 513 – 529 www.elsevier.com/locate/econbase
Wage differentials and mobility in the urban labour market: a panel data analysis for Mexico Xiaodong Gong a,b, Arthur van Soest c,* a
Economics Programs, RSSS, Australian National University, Canberra, Australia b IZA, Germany c Tilburg University, P.O. Box 90153, 5000 LE, Tilburg, The Netherlands
Received 1 June 2001; received in revised form 4 January 2002; accepted 21 February 2002
Abstract We analyze wage differentials and transitions between the formal and informal sectors in urban Mexico. We use panel data on five quarterly waves from Mexico’s Urban Employment Survey. We develop a dynamic random effects panel data model consisting of separate wage equations for the two sectors and a logit part explaining the labour market state with wages included as explanatory variables. The model is estimated using simulated maximum likelihood. The estimates show that wage differentials increase with education level. The probability of formal sector employment strongly increases with the wage differential. For male workers, the choice between formal and informal sector is driven by wage differentials and unobserved heterogeneity, and true state dependence is not important. For women, non-participation is the most common labour market state, and true state dependence plays a much larger role. D 2002 Elsevier Science B.V. All rights reserved. JEL classification: C33; J23; J31; R23 Keywords: Wage differentials; Informal sector; Mobility; Panel data; Mexico
1. Introduction In countries with a large informal sector, the nature of the informal sector and its position compared to the formal sector are crucial for the functioning of the labour market and the overall economic structure. It affects income inequality and poverty, it has implications for the efficiency of the allocation of labour and for the distortions due to
*
Corresponding author. Tel.: +31-13-466-2028; fax: +31-13-466-3280. E-mail addresses:
[email protected] (X. Gong),
[email protected] (A. van Soest).
0927-5371/02/$ - see front matter D 2002 Elsevier Science B.V. All rights reserved. PII: S 0 9 2 7 - 5 3 7 1 ( 0 2 ) 0 0 0 4 5 - 3
514
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
taxes, social security, and labour market regulations imposed on the formal sector. It therefore also has major implications for economic policy. This explains why the role of the informal sector has been analyzed extensively during the last two decades. Two competing stylized views exist in the literature. The traditional staging hypothesis in the theoretical model of Fields (1975) is that formal sector employment is rationed. Those who cannot obtain a formal sector job and cannot afford to search from unemployment work in the informal sector. Informal sector jobs are secondary jobs and workers would be better off with a primary job in the formal sector. The informal sector is a buffer between not working and working in the formal sector. In the standard version of this model, the attractiveness of a job is determined by the wage only. The model implies wage dualism: in equilibrium, each individual’s wage in the informal sector is less than their potential wage in the formal sector. The other view sees the informal and formal sectors as symmetric and competitive. The two sectors are characterized by different production functions. Due to heterogeneity, some workers are more productive in the formal sector and others in the informal sector. Under the assumption that workers choose the sector where they are most productive and can earn the highest wage, Heckman and Sedlacek (1985) test this model using crosssection data on workers’ sector choice and wages. Magnac (1991) applies an extension of this model that also accounts for the state of not working to married women in urban Columbia. He does not reject the model and concludes that the Columbian labour market is in a ‘weakly competitive equilibrium’. Other empirical evidence on sector choice and wage differentials between the formal and the informal sectors is mixed. Strassmann (1987) found that 71% of home workers in Lima would require a considerable financial incentive to move to the formal sector (see also Thomas, 1992). Pradhan and van Soest (1995) estimated reduced form models for urban Bolivia, explaining the choice between formal sector, informal sector, and not working. They found that an ordered model performs better for men while an unordered model fits the data better for women. Using the same data and a structural labour supply model, Pradhan and van Soest (1997) found that wage differentials between formal and informal sector tend to be negative rather than positive, and that non-monetary job characteristics (job stability, social security, health care access, etc.) are needed to explain why so many people prefer formal sector jobs. Other studies on wage differentials for various countries give mixed results (see the overview in Pradhan and van Soest, 1995). All these studies are based on cross-section data. One of the few studies on this topic using panel data is Maloney (1999). He studies wage differentials and transition patterns of workers with at most high school education, using panel data on urban Mexico for 1991 – 1992. He concludes that ‘‘much of the informal sector is a desirable destination and the distinct modalities of work are relatively well integrated’’. His findings largely seem to support the competitive view, although he carefully points out the caveats of his reduced form models, which hamper an unambiguous conclusion. Another panel data study is Gong et al. (2000), who use the same data source as Maloney. With dynamic discrete choice panel data models, they investigate labour market mobility between three labour market states, emphasizing the role of education level. They do not explicitly incorporate wages, but find strong education effects that could reflect wage effects.
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
515
In the current study, we construct a panel data model for sector choice and mobility as well as wages in which the effects of wage rates on sector choice are explicitly incorporated. For men, we consider the choice between formal and informal sector; for women, we also consider non-employment.1 The model consists of a dynamic binomial (men) or multinomial (women) logit model for panel data with random effects explaining the labour market state of each individual in each time period, together with two dynamic wage equations for potential wage rates in each sector. It is estimated on five quarterly panel waves drawn from Mexico’s Urban Employment Survey, the same data source as used by Gong et al. (2000) and Maloney (1999). Our first aim is to analyze wage formation and wage differentials controlling for selectivity bias due to correlated unobserved heterogeneity affecting wages and sector choice. This should lead to a better insight in the underlying differences in wage formation between sectors than the uncorrected wage differentials used by Maloney (1999). Second, allowing for direct effects of both sectors’ wages on the sector choice, we analyze the importance of wages for sector choice. We analyze the mobility patterns between the two sectors and, for women, non-employment, with emphasis on disentangling unobserved heterogeneity and true state dependence. The main novelty of our paper is that wages and sector choice are simultaneously addressed in one dynamic panel data model. Mexico is a particularly relevant country for studying the role of the informal sector. A typical feature is the low unemployment rate. Since the 1980s, the official urban unemployment rate decreased continuously to 2.6% in 1991. It remained below 4% until 1994 (see Fleck and Sorrentino, 1994), while Mexico’s labour force grew at an annual rate of about 2.9%. An explanation for the low unemployment rate is the existence of the informal sector where many individuals have some marginal job. While formal sector employment is subject to regulation, social premiums and taxation, with wages paid on a regular basis and explicit contracts between employers and employees, the informal sector consists of micro-firms that are not subject to institutional regulations. There are two explanations for a large informal sector in Mexico. First, Mexico’s formal sector labour market regulations are very extensive. Mexican Federal Labour Law governs virtually every aspect of labour relations, such as minimum wages, limits on working hours, overtime pay, profit sharing, etc. Many rules were especially designed to protect the individual employees’ job security (see Hollon, 1996; Zelek and de la Vega, 1992), such as severance payments for termination of employment. In addition, formal sector firms face health and safety requirements. All this would make hiring prohibitively costly for small firms. Many small firms thus avoid the requirements; they do not officially register and thus are in the informal sector. Second, Mexico has no system of unemployment compensation, so that individuals without a regular job are often forced into marginal activities like street vending, etc. Only those who can afford it search from unemployment (see Fleck and Sorrentino, 1994). Our main findings are as follows. The wage is the main factor driving the sector choice. Wage differentials between formal and informal sector are typically small for the lower educated, but increase strongly with education level. As a consequence, the probability of
1
The number of non-employed men is too small to get useful estimates of our structural model.
516
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
formal sector employment strongly rises with education level. For men, the probability of working in the informal sector decreases with the income of other family members, while for women, other family income increases the probability of not working. This confirms that only those who can afford it do not work, in line with the staging hypothesis. The remainder of the paper is organized as follows. Section 2 describes the data. The econometric model is discussed in Section 3. Results are presented in Section 4, with focus on wage differentials and transition probabilities. Section 5 concludes.
2. Data The data were drawn from Mexico’s Urban Employment Survey (Encuesta Nacional de Empleo Urbano), conducted by Instituto Nacional de Estadistica, Geografia e Informatica (INEGI, i.e. Mexican Statistical Institute). This is a rotating panel drawn in 32 Mexican cities, and it is the only quarterly household panel survey in Mexico. The survey provides detailed information on the economic activities of all household members older than 12, such as job characteristics, working hours and labour income, but no information on nonlabour income. The same survey was used by Fleck and Sorrentino (1994) to analyze unemployment and by Villagomez (1996, 1998), Caldero´n-Madrid (1999), and Maloney (1999) to study labour market segmentation and mobility. For our analysis, we use data on Mexico City, Guadalajara, Monterrey, Tijuana, and Ciudad-Juarez, covering 60% of urban employment in Mexico. In the border cities Tijuana and Ciudad-Juarez, there are many in-bond firms. These so-called Maquiladoras2, usually owned by multinationals or joint-ventures with multinationals, typically pay better than other (domestic) firms. Mexico City, Guadalajara and Monterrey represent about a quarter of the entire population of Mexico, and half of the population of cities with more than 100,000 inhabitants. Guadalajara is the city with the largest share of informal workers (see Villagomez, 1998). Our panel covers a period of economic growth: the first quarter of 1992 until the first quarter of 1993.3 The sample consists of about 2500 households in each wave. We created separate unbalanced panels of men and women, selecting only those individuals who are present in at least two consecutive quarters. Moreover, we only selected heads of households and their spouses who are younger than 65 years and are not full-time students. This leads to samples of 1691 males and 1907 females. We removed 269 males and 627 females because information on other family members’ income was missing. Finally, we removed 18 male and 82 female unpaid family workers, who could not be classified in either sector. The sample of women used for the estimations thus consists of 1198 observations. The number of male respondents who were not employed is so small that it would not be possible to estimate the model with non-working as a separate labour market state. Excluding non-working men from the sample for estimation leaves a sample of 1310 men. About 64% of the respondents are present in all the five waves, and about 12% in only two waves.
2
See Martin (1999) for some institutional background information.
3
The same analysis for the five panel waves drawn in 1994 – 1995 gave very similar results.
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
517
Attrition in the panel is substantial. Non-random attrition might bias the estimation results. One might expect that workers in the informal sector are more likely to exit from the panel either because they migrate more often to do seasonal work or because they might be more reluctant to disclose their income. To check this, we estimated probits explaining whether respondents leave the panel between waves t and t + 1, using dummies for the sector at time t as explanatory variables. For the various specifications we tried (controlling for different sets of background variables) and for both males and females, these dummies turned out to be insignificant, suggesting that sample attrition does not relate to labour market status. In Table A1 in the appendix, we present definitions and sample statistics of the independent variables used in the analysis. Here, we focus on the dependent variables. In Section 1, we have not precisely defined the distinction between formal and informal sector. Several definitions are used in the existing literature. Maloney’s (1999) definition is similar to that of the Mexican government, classifying firms with fewer than six workers as informal. Professionals (lawyers, doctors, etc.; about 5% of men and 0.5% of women) are formal sector workers, together with all those in firms with more than five workers. A definition that seems to be used for Mexico only is based upon whether social security premiums are paid or not (see Caldero´n-Madrid, 1999; Martin, 1999). In the international literature, the most common definition seems to be the one based upon survey information on the type of job. See, for example, Magnac (1991) and Pradhan and van Soest (1995, 1997). Piece-workers and those who work for their own account or manage a firm without employees are categorized as informal. Those who work for a fixed wage, cooperative workers, employers (with at least one employee) and independent professionals are categorized as formal. Unpaid family workers cannot be categorized according to either definition and were therefore deleted from the sample. Gong et al. (2000) compared the classifications according to the three definitions, and found that the transition patterns are similar. We did the full analysis for both the firm size and the job-type definition. Due to space limitations, we only present the results based upon the job type definition, which seems more in line with the paper’s aim to investigate whether the distinction between informal and formal sector corresponds to a distinction in primary and secondary jobs. Most of the results based upon the firm size definition are qualitatively similar (available upon request from the authors). Table 1 Labour market states Quarter
92.1
92.2
92.3
92.4
93.1
Men Formal sector Informal sector Not working
65.3 28.3 6.4
64.1 27.8 8.2
65.6 27.8 6.6
64.7 28.0 7.3
62.9 28.9 8.1
Women Formal sector Informal sector Not working
21.7 9.5 68.8
21.0 9.1 70.0
21.3 8.6 70.1
21.8 9.2 69.0
21.7 8.7 69.6
Percentages of complete samples of men and women.
518
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
Because of the small numbers of people classifying themselves as unemployed, we do not distinguish unemployment as a separate labour market state, but merge the unemployed with other non-workers. Table 1 shows the sample percentages of non-workers and formal and informal sector workers. For men, the formal and informal sector workers represent 65% and 28% of the labour force, respectively. Only about 7% are non-workers. On the
Fig. 1. Real log wages in formal and informal sector.
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
519
other hand, more than two-thirds of the women in our sample do not participate in the labour market. About 30% of working men and women have an informal sector job. Wages are observed for about 88.2% and 90.5% of working men and women, respectively. Fig. 1 compares means and standard deviations of log real wages in the two sectors,4 for those of the higher (at least intermediate) and lower (less than intermediate) education levels. For men with high education, the average log wage is substantially higher in the formal sector than in the informal sector, and wages in the formal sector are also more dispersed. The higher average wage in the formal sector can have two explanations: the formal sector has the more productive jobs and each individual can earn more in the formal than in the informal sector, or the high productivity workers select themselves into the formal sector. Controlling for observed and unobserved characteristics and selection effects in the econometric model will show to which extent these explanations are relevant. For men with low education level, a different picture emerges. The average ‘‘raw’’ log wage differential between formal and informal sector is negative in all time periods. The raw wage differentials thus provide some prima facie evidence that informal sector jobs are secondary jobs for highly educated men, but not for low educated men. For women, raw wage differentials between sectors are small and change sign. Here the most salient conclusion from the figures is the larger wage dispersion in the informal sector. It suggests that, at least for women, informal sector jobs are more heterogeneous than formal sector jobs. This result was found by Pradhan and van Soest (1995) for both men and women. In Table 2, we present some sample statistics on changes of individual log real wages between consecutive quarters. Along with means, we report medians, which are less sensitive to outliers, and the first and third quartiles. We do this separately for those who stay in the same sector and for those who change sector. The mean changes never significantly differ from zero. Median changes are negative—due to inflation—and significantly so for men and women who stay in the formal sector. The first and third quartile show that reported wages are very volatile, even for those who do not change sector.5 For example, for about 25% of all men who stay in the formal sector in two consecutive quarters, the real wage rate in the second wave is at least 28% (100(1 e 0.33)) lower than in the first wave. Although it is conceivable that wage mobility is larger in Mexico than in many other OECD countries, we think these large numbers are at least partly due to measurement errors in reported earnings. We will take this into account in the model (see Section 3). In Table 3, the sample transition rates among the three labour market states are presented. Male non-workers are excluded. The probabilities of remaining in the formal sector are larger than those of remaining in the informal sector. This does not necessarily mean that in the formal sector, jobs are more stable than in the informal sector; the difference in sector exit rates could be due to the larger size of the formal sector, making it more likely that someone who loses a job in the formal sector finds another job in the same sector. Since the 4 Nominal wages are computed as monthly income divided by actual working hours; real wages are obtained using the IMF CPI as the deflator (Source: Data Stream). Qualitatively similar results are obtained with monthly earnings instead of hourly wage rates. 5 This result remains unchanged if monthly earnings are considered instead of hourly wage rates.
520
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
Table 2 Real log wage changes by sector From
Formal sector
Informal sector
To
Formal
Informal
Formal
Informal
Men Mean Standard error of mean 25th Percentile Median Standard error of median 75th Percentile Number of observations
0.025 0.020 0.330 0.019 0.005 0.380 2580
0.028 0.059 0.495 0.087 0.056 0.603 326
0.008 0.060 0.537 0.026 0.038 0.479 315
0.019 0.035 0.542 0.022 0.012 0.556 898
Women Mean Standard error of mean 25th Percentile Median Standard error of median 75th Percentile Number of observations
0.033 0.034 0.314 0.019 0.009 0.269 674
0.186 0.131 0.242 0.197 0.144 0.753 47
0.062 0.131 0.327 0.066 0.048 0.346 44
0.063 0.084 0.586 0.026 0.079 0.674 191
data do not provide information on whether people change jobs or not, we can only look at sector mobility and cannot compare job mobility in the two sectors.
3. The model The model explains wages in the two sectors and the choice of labour market state of each individual in each quarter. We use a dynamic (binomial or multinomial) logit panel data model with random effects for the choices, and two linear dynamic random effect
Table 3 Sample transition rates t1
t=2
t=3
Form.
Infor.
89.2 24.9
10.8 75.1
Women Form. 75.0 Infor. 9.9 Noem. 3.5
4.8 52.7 5.0
Men Form. Infor.
Noem.
20.2 37.4 91.5
t=4
Form.
Infor.
88.6 29.8
11.4 70.2
81.4 7.3 4.2
4.5 51.0 4.7
Noem.
14.1 41.7 91.1
t=5
Form.
Infor.
88.9 27.3
11.1 72.7
77.2 17.6 4.9
7.1 52.7 4.4
Noem.
15.6 29.7 90.8
Form.
Infor.
88.4 21.5
11.6 78.5
80.6 13.2 4.5
5.3 50.5 3.9
Noem.
14.1 36.3 91.5
Number of transitions from labour market state in t 1 (row) to labour market state in t (column) as a percentage of number of people in labour market state in t 1. For example, 24.9% of all men who work in the informal sector at time of the first quarter of the 92 panel work in the formal sector 3 months later.
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
521
wage equations for the wages in the two sectors. Wages are included as explanatory variables in the choice equations. A respondent’s wage rate in a given sector and quarter is only observed if the respondent works in that sector during that quarter. This implies that the choice model cannot be estimated without the wage equations. The model we use generalizes the first-order Markov model of Heckman (1981a). It distinguishes between structural state dependence and unobserved heterogeneity by including lagged state dummies as explanatory variables and (random) individual effects to control for unobserved individual characteristics. The individual effects are assumed to follow a normal distribution, independent of the observed characteristics (and thus called random effects). The initial condition problem due to the small number of time periods in the panel is treated as in Heckman (1981b). The wage equations are specified as follows. lnwtj ¼ Mt Vmj þ Dt1 V sj þ kj þ ftj ;
j ¼ 1; 2:
ð1Þ
Here j = 1 and j = 2 denote the formal and informal sector, respectively. For notational convenience, we suppress the subscript for individuals. Mt is a vector of individual characteristics (including a constant term, educational dummies, regional dummies, time dummies and age). Dt 1 indicates the lagged labour market state. For men, this is one dummy for the formal sector, taking the informal sector as the reference state. For women, it is a vector with two dummies for the formal and the informal sector, with not working as the reference state. The kj ( j = 1, 2) are the random effects and the ftj are the idiosyncratic error terms. Both are assumed to be i.i.d. normal with mean zero and independent of each other and the exogenous variables. mj and sj ( j = 1, 2) are parameters to be estimated. We do not impose any constraints across the two wage equations so that wage formation is allowed to be completely sector specific. The ftj reflect both measurement error and genuine unsystematic wage variation over time in a given sector for a given individual. The two cannot be separately identified, but the descriptive statistics in Table 2 suggest that measurement error will be important. Consistent estimation of the sector choice part of the model requires including a wage variable cleaned for measurement error. To achieve this, we will use ‘‘predicted’’ log wages lnwtjp given by lnwptj ¼ Mt Vmj þ Dt1 V sj þ kj ¼ lnwtj ftj ;
j ¼ 1; 2:
ð2Þ
We put ‘‘predicted’’ in quotes, since this wage variable includes the random effect (kj), which is not estimated, but is assumed to be known by the individuals when they choose their labour market state. On the other hand, the respondents choose the state without taking account of ftj, either since this is measurement error or because this is not known to them when they make their choice. We can now define the choice part of the model. An individual can be in any of J labour market states at time t ( J = 2 for men and J = 3 for women): working in the formal sector ( j = 1), working in the informal sector ( j = 2), and not working ( j = 3). For notational convenience, we define lnwt3p = 0. The ‘‘utility’’ of state j ( j = 1, . . ., J) in time period t > 1 is given by V cj þ aj þ qlnwptj þ etj : V ðj; tÞ ¼ Xt Vbj þ Dt1
ð3Þ
522
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
The reason for the quotes on ‘‘utility’’ is that demand side restrictions may play a role. A low value of V(1, t), for example, may mean that no formal job is available at time t, even though the respondent might prefer such a job. Xt is a vector of explanatory variables, including age, regional dummies, family composition, time dummies, etc. Educational dummies are not included in Xt but enter through Mt in the wage equations. Excluding educational dummies from Eq. (3) is necessary to identify the model. The vectors bj and cj are parameters to be estimated. aj are random effects reflecting time invariant unobserved individual heterogeneity and are assumed to be i.i.d. multivariate normal with mean zero, independent of Xt and Mt. By means of normalization, b1, c1, and a1 are set to 0. How attractive each labour market state is also depends on potential wages. We include the ‘‘predicted’’ wage rates as explanatory variables with coefficient q. The assumption that the effect of wages on utility is the same in the formal and informal sector implies that the utility difference between these two sectors depends on the log wage differential. The etj are i.i.d. error terms, assumed to be independent of the Xt, aj, and other errors in the model and following a Type I generalized extreme value (GEV I) distribution. We allow for non-zero correlation between random effects of different states, but not between random effects in the wage and state choice equations. The latter implies that correlation between random effects in choice and wage equations only enters through the structural part qlnwtjp in Eq. (3).6 Correlations between the idiosyncratic errors f and e are also ruled out. The assumptions imply that the conditional likelihood contribution Lt( j | Xt, Mt, Dt 1, AV, LV) for individual i in state j at time t>1 with observed wage wtj is given by expðXt Vbj þ Dt1 V cj þ aj þ qlnwpjt Þ fj ðlnwtj lnwptj Þ; J X expðXt Vbs þ Dt1 V cs þ as þ qlnwpst Þ
ð4Þ
s¼1
where fj is the density of ftj in Eq. (1). If no wage is observed (non-employed females or missings), the conditional likelihood contribution is the same but without the density factor. The initial condition problem due to the presence of the lagged dependent variables in Dt 1 is solved in the same way as in Heckman (1981b): for t = 1, Eq. (3) is replaced by a static multinomial logit with different parameters and not including Dt 1. The initial condition problem in the wage equations at t = 1 is treated similarly. The static equations can be seen as approximations to a reduced form, eliminating the lagged state dummies. Heckman’s Monte Carlos for the dynamic random effects probit model suggest that the bias involved with this procedure is small.7
6
We estimated the model allowing for correlation between aj and Ej but this did not lead to significant improvement. 7 As in Heckman’s model, the first order Markov assumption guarantees identification and there is no need to exclude variables in Eq. (5) from Eq. (3). Chay and Hyslop (2000) compare various ways of dealing with the initial conditions and find that the Heckman procedure works better than other procedures.
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
523
Given the random effects, the conditional likelihood contribution of individual i with observed states j1, . . ., jT would be given by Lða; lÞ ¼ PTt¼1 Lt ðjt j Xt ; Mt ; Dt1 ; aV; lVÞ:
ð5Þ
Since random effects are unobserved, the likelihood contribution is the expected value of Eq. (5) Z l Z l : : : L¼ Lða; lÞ/a ðaÞ/k ðkÞdadl ð6Þ l l |fflfflfflfflfflfflfflfflfflffl{zfflfflfflfflfflfflfflfflfflffl} 2j2
where /a and /k are the densities of A and L. Computation of the likelihood contribution in Eq. (6) involves 2J 2 dimensional integration. This can be done in several ways. We will use (Smooth) Simulated Maximum Likelihood. Since Eq. (6) is the expected value of Eq. (5), it can be approximated by a simulated mean. For each individual, R values of A and L are drawn and the average of the R likelihood values conditional on these draws are computed. The integral in Eq. (6) is thus replaced by LRi ¼
R 1X Li ðaqi ; lqi Þ: R q¼1
ð7Þ
The resulting estimator is consistent if R ! l with the number of observations (n). If n1/2/ R ! 0 and with independent draws across observations, the method is asymptotically equivalent to maximum likelihood (see, e.g., Hajivassiliou and Ruud, 1994). We present results for R = 30 (for R = 20, very similar results are obtained).
4. Results We only report the estimates of the dynamic equations. The estimates of the initial conditions equations are available on request from the authors. 4.1. Wages and wage differentials Table 4 presents the estimates of dynamic wage equations (Eq. (1)). Several findings are common for men and women. First, age has a significant effect on the formal sector wage, but not on the informal sector wage. The formal sector wage first rises and then falls with age. Second, returns to education are positive in both sectors, but much higher in the formal sector than in the informal sector. For example, in the formal sector, a man with high education level can earn almost 150% more than a man with the lowest education level (ceteris paribus). In the informal sector, this is only 44% more. An explanation is that in the larger formal sector firms, it is difficult for the employers to observe workers’ productivity directly, and experience and education level are used as signals. In the small informal sector firms on the other hand, employers have direct contacts with the
524
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
Table 4 Wage equations (t values in parentheses) Parameters
Men
Women
Formal mj
sj Rk
Rf
Const. Age Age2/100 Medu Hedu JuaTij Guada Mont. T3 T4 T5 Form. 1 Info. 1 r2 r3 q23 r2 r3
0.936 0.035 0.040 0.257 0.913 0.400 0.063 0.257 0.069 0.078 0.017 0.010
Informal (4.40) (3.57) ( 3.46) (7.60) (25.65) (11.03) ( 1.55) (7.43) ( 1.32) (1.47) ( 0.34) ( 0.18)
1.620 0.012 0.016 0.062 0.363 0.414 0.207 0.176 0.039 0.077 0.007 0.036
0.004 0.010 0.803 0.837 0.844
(1.25) (1.00) (0.80) (102.6) (48.82)
Formal (4.52) (0.67) ( 0.81) (1.42) (4.69) (6.41) ( 3.30) (2.62) ( 0.46) ( 0.97) (0.10) (0.61)
0.585 0.036 0.031 0.435 0.789 0.269 0.006 0.094 0.006 0.076 0.154 0.007 0.152
Informal (1.54) (2.00) ( 1.45) (7.04) (9.58) (4.01) (0.07) (1.49) ( 0.07) (0.79) (1.61) ( 0.08) (0.94) 0.010 0.004 0.645 0.748 0.971
1.063 0.029 0.037 0.076 0.209 0.232 0.110 0.306 0.113 0.115 0.157 0.163 0.119 (1.69) (0.28) (0.34) (45.35) (22.01)
(1.35) (0.75) ( 0.83) (1.00) (1.67) (1.26) ( 0.68) (1.92) (0.60) (0.65) (0.79) (0.71) ( 0.83)
rj: standard deviations of kj and fj, j = 2, 3; q23: the corresponding correlation coefficients. ‘‘Lowedu’’, ‘‘T1’’, and ‘‘Mex. City’’ are the omitted control group dummies.
employees, and the signalling function of age and education will be smaller.8 Another explanation is that the formal sector is more capital or technology intensive and educational skills are a complement to capital and technology. In the two border cities Tijuana and Ciudad-Juarez, the wages in both sectors are higher than in Mexico City due to the in-bond firms that pay relatively high wages. There is no evidence that the lagged labour market state affects current earnings: all the lagged state variables are insignificant. Moreover, for men as well as women, we find no evidence of random effects in the wage equations: the estimated variances of the random effect terms are very small. On the other hand, the idiosyncratic errors are quite large, in line with the large standard deviations of wage changes in Table 2. We estimated the wage differentials for some individuals with given benchmark characteristics. The results are summarized in Table 5.9 The benchmark person is 40 years old, lives in Mexico City, worked in the formal sector at t 1, and has low education level. For the benchmark man, the wage differential is negative and significant at the 10% level. The point estimate is about 16%. Formal sector wages rise faster with age than informal sector wages, and the negative wage differential is particularly large (and 8 Stigler (1962) already suggested that small firms more often recognise and reward ability than formal schooling. 9 Wage differentials for the high educated are even larger if the size definition is used. Wage differentials for the low educated are then close to zero. This may relate to a firm size effect rather than to the difference between formal and informal sector. There is ample evidence for OECD countries that wages in small firms are smaller than in large firms (see, e.g., the survey in Polachek and Siebert, 1996).
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
525
Table 5 Log wage differentials for some benchmark persons Characteristics
Men
Mexico City; formal sector at t 1; low education level; age = 40 Mexico City; formal sector at t 1; low education level; age = 25 Mexico City; formal sector at t 1; low education level; age = 50 Mexico City; formal sector at t 1; intermediate education level; age = 40 Mexico City; formal sector at t 1; high education level; age = 40 Mexico City; informal sector at t 1; low education level; age = 40 Ciudad Juarez or Tijuana; formal sector at t 1; low education level; age = 40 Mexico City; not employed at t 1; low education level; age = 40
0.174 0.294 0.154 0.021
Women (0.101) (0.113) (0.101) (0.097)
0.268 0.433 0.142 0.091
(0.273) (0.298) (0.261) (0.275)
0.375 (0.113) 0.129 (0.104) 0.187 (0.101)
0.313 (0.278) 0.174 (0.258) 0.097 (0.209)
–
0.230 (0.285)
Standard errors in parentheses.
significant) at younger age levels. For high-educated men, the log wage differential between formal and informal sector is significantly positive. For someone 40 years old in Mexico City who worked in the formal sector the previous quarter, the estimated wage differential is about 45%. For women, the pattern is similar, but standard errors are much larger due to the small number of observed wages. As explained by Magnac (1991) and Maloney (1999), these wage differentials alone do not lead to unambiguous conclusions on the nature of formal sector vis-a-vis informal sector jobs, even though selection effects are now corrected for. The competitive model can still be valid for the high educated if either the majority of high educated workers (who can earn more in the formal sector) indeed work in the formal sector, or if non-wage job characteristics make the informal sector more attractive for many people. Still, the difference between high and low educated is quite salient. While the staging hypothesis may be valid for the high educated, this seems implausible for the low educated. 4.2. Mobility The estimates of the dynamic choice equations in Eq. (3) are presented in Table 6. A positive value of bj or cj ( j = 2, 3) means that the corresponding variable has a positive impact on the probability to be in state j compared to the probability to be in the reference state (the informal sector for men, non-employment for women). Wage effects are significantly positive for both men and women. For men, this means that a higher wage differential between formal and informal sector leads to a larger probability to work in the formal sector, given the other characteristics and the labour market state in the previous quarter.10 For women, it also means that a higher wage in the formal or in the informal sector increases the probability to participate.11 10
To illustrate the size of the wage coefficient: the formal sector work probabilities of a high- and a loweducated benchmark man are about 0.97 and 0.84; the difference is due to the log wage differential of about 0.55 (Table 5). 11 The probabilities of non-employment for the benchmark women with low and high education are about 0.92 and 0.47.
526
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
Table 6 Estimates of the dynamic choice Parameters
Men
Women
Formal
Formal
Informal
bj Const. Age Age2/100 JuaTij Guada Mont. Child Adults Othinc Nmar T3 T4 T5 lwage
5.450 0.120 0.066 0.270 2.178 0.068 0.082 0.065 0.080 0.386 0.240 0.990 0.080 6.247
(1.77) ( 0.80) (0.38) ( 0.49) ( 3.50) ( 0.12) ( 0.64) (0.90) (3.18) ( 0.93) (0.38) ( 1.49) ( 0.14) (4.87)
8.787 0.093 0.230 1.268 0.163 0.139 0.725 0.046 0.059 2.579 0.231 0.065 0.241
( 3.77) (0.86) ( 1.78) ( 3.21) (0.34) ( 0.35) ( 4.13) ( 0.67) ( 1.41) (7.65) (0.59) (0.16) ( 0.56) 3.555
11.592 0.146 0.178 1.318 0.758 0.985 0.215 0.040 0.070 1.977 0.410 0.293 0.624 (4.86)
( 3.55) (0.93) ( 0.98) ( 1.82) (1.18) ( 1.39) ( 1.43) (0.55) ( 1.64) (5.39) ( 0.61) ( 0.44) ( 0.83)
cj Form. 1 Info. 1
0.456 –
(0.86)
1.905 0.200
(4.70) (0.28)
0.409 1.592
( 0.48) (2.78)
3.550
(12.14)
2.713 2.350 0.735
(8.43) (4.05) (4.44)
Ra r2 r3 q23
t-Values in parentheses. Reference states: informal sector (men), not employed (women). Omitted control group dummies: Lowedu, T2, Mex. City, and noem. 1 (women) or infor. 1 (men).
The variances of the random effects are substantial and their confidence intervals do not include zero. The random effects in the state equations play a much larger role than the random effects in the wage equations and contribute more to explaining the state choice than the idiosyncratic errors (which have variance p2/6). For women, the two terms are positively correlated, suggesting a common factor reflecting preferences for work versus non-work. Given the wage rates, sector choice is not significantly affected by age. This corresponds to the finding of Maloney (1999) that the effects of experience on mobility are not clear. He interprets this as evidence against the staging hypothesis, which would imply that the more experienced who have queued longer, have a larger probability to work in the formal sector. Somewhat surprisingly, men’s sector choice is not affected by their labour market state in the previous quarter, implying that there is no genuine state dependence. The fact that men tend to stay in the same sector is explained by observed and unobserved heterogeneity and is not a structural effect. Thus, there is no evidence of costs of entry into the formal sector, which would be one of the ingredients of the staging hypothesis.
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
527
For women, the estimates indicate that being in either of the three states increases the probability of being in that same state in the next quarter. There are no cross-effects: women who are not employed have the same probability to find a formal (informal) sector job as women in the informal (formal) sector with the same characteristics. This is not in line with the notion that people prefer unemployment to informal sector work because it is easier to find a formal sector (primary) job from unemployment than from informal sector employment, which is one of the features of the Fields (1975) model. A plausible explanation for this is that many non-working women are non-participants who are not looking for work. Men in families with higher income of other family member have a larger probability to work in the formal sector. For women, income of other family members reduces the probability of working in either sector by about the same extent, showing that leisure is a normal good. Having younger children reduces women’s probability to work, particularly in the formal sector. This is in line with common findings in the literature. Living in one of the border cities increases women’s probability of non-employment, given wages and other characteristics. The in-bond industries pay relatively high wages, but apparently, women either do not like the nature of the jobs (given the wages), or find it hard to get access to them.
5. Conclusions We have investigated labour market segmentation in urban Mexico by studying the wage differentials between the formal and informal sectors and the transition patterns between the labour market states formal sector employment, informal sector employment, and (for women only) non-employment. Mobility between these states is very large compared to other OECD countries. To explain the wage differentials between the sectors and the labour market state of each individual in each quarter, a dynamic multinomial logit model with random effects for the choice and two linear dynamic random effect equations for the wages in the two sectors was developed. We estimated this model for a Mexican panel data set covering the first quarter of 1992 to the first quarter of 1993. In line with the existing literature, we have found that wages in both sectors increase with education, with much stronger education effects in the formal sector. We found large positive wage differentials between formal and informal sector for the higher educated, and small or even negative differentials for medium and low education levels. These differentials also vary across cities and, particularly for females, they increase with age. We have not found any evidence of unobserved heterogeneity or an effect of the previous labour market state on the current wage. Our estimates show that the probability of formal sector employment strongly increases with the wage differential. For men, the probability of working in the informal sector decreases with the level of income of other family members. For women, other family income mainly increases the probability of not working. Simulated transition probabilities were used to show that the main reason why men tend to stay in the same state is unobserved heterogeneity. For women on the other hand, we also find a substantial structural impact of the lagged labour market state. The high wage differ-
528
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
entials for workers with high education level imply that the high educated more often work in the formal sector. High wages also induce more (female) non-workers with high education level to work. For higher educated workers, rates of formal sector turnover are low, while, due to the wage differentials, informal workers move to the formal sector quite often. The main goal of the paper was to contribute to the debate in the literature on the role of the informal sector. For the reasons well documented in Maloney (1999), we cannot formally test whether either the staging hypothesis or the competitive labour market view is a realistic description of the actual labour markets in Mexico. Still, many of our findings strongly suggest that for the lower educated workers, the dualistic view of the labour market is not a good description. The same conclusion was also drawn by Maloney (1999). For the higher educated on the other hand, the urban labour markets in Mexico have important dualistic features.
Acknowledgements We are grateful to Elizabeth Villagomez for her help and for providing the data, and to Simon Burgess and two anonymous referees for useful comments.
Appendix Table A1 Variable definitions, sample means and standard deviations Variable
Explanation
Child
Number of children younger than 6 years old Number of family members older than 11 years Age of the individual 0 – 6 years of schooling 7 – 12 years of schooling >12 years of schooling Income of other family members (in 1995 pesos) Living in Ciudad Juarez or Tijuana Living in Guadalajara Living in Monterrey Living in Mexico City Single or divorced Working in the formal sector Working in the informal sector Not employed Time dummies for the five quarters
Adults Age Lowedu Medu Hedu Othinc JuaTij Guada Mont. Mex. City Nmar Form. Infor. Noem. T1 – T5
Men
Women
0.750
(0.87)
0.660
(0.85)
3.312
(1.70)
3.287
(1.76)
38.844 0.441 0.370 0.189 950.7
(10.86)
(1643)
0.291 0.181 0.208 0.321 0.052
Variables from Lowedu to Noem. are dummies with value of 1 if condition is satisfied. Standard deviations in parentheses. See Table 1 for descriptives of Form., Infor., and Noem.
38.670 0.554 0.375 0.071 2373.1 0.331 0.147 0.211 0.311 0.170
(12.13)
(2501)
X. Gong, A. van Soest / Labour Economics 9 (2002) 513–529
529
References Caldero´n-Madrid, A., 1999. Job Stability and Labour Mobility in Mexico During the 1990s, working paper, Centro de Estudios Econo´micos, Mexico. Chay, K., Hyslop, D., 2000. Identification and Estimation of Dynamic Binary Response Panel Data Models: Empirical Evidence using Alternative Approaches University of California, Berkeley, mimeo. Fields, G., 1975. Rural-Urban migration, urban unemployment and underemployment, and job search activity in LDCs. Journal of Development Economics 2, 165 – 187. Fleck, S., Sorrentino, C., 1994. Employment and unemployment in Mexico’s labour force. Monthly Labour Review 117 (11), 3 – 31. Gong, X., van Soest, A., Villagomez, E., 2000. Mobility in the Urban Labour Market: a panel data analysis for Mexico, Center Discussion Paper, Tilburg University. Hajivassiliou, V.A., Ruud, P.A., 1994. Classical estimation methods for LDV models using simulation. In: Engle, R.F., McFadden, D.L. (Eds.), Handbook of Econometrics, vol. IV. North-Holland, New York, pp. 2384 – 2443. Heckman, J., 1981a. Statistical models for discrete panel data. In: Manski, C., McFadden, D. (Eds.), Structural Analysis of Discrete Data with Econometric Applications. MIT Press, London, pp. 114 – 178. Heckman, J., 1981b. The incidental parameters problem and the problem of initial conditions in estimating a discrete time-discrete data stochastic process. In: Manski, C., McFadden, D. (Eds.), Structural Analysis of Discrete Data with Econometric Applications. MIT Press, London, pp. 179 – 195. Heckman, J., Sedlacek, G., 1985. Heterogeneity, aggregation and market wage functions: an empirical model of self-selection in the labour market. Journal of Political Economy 93, 1077 – 1125. Hollon, C., 1996. Individual employee employment security under Mexican federal labour law. Labour Law Journal, 648 – 655, October. Magnac, Th., 1991. Segmented or competitive labour markets. Econometrica 59, 165 – 187. Maloney, W., 1999. Does informality imply segmentation in urban labour markets? Evidence from sectoral transitions in Mexico. The World Bank Economic Review 13 (2), 275 – 302. Martin, P., 1999. Trade and Migration: The Mexican – US Experience University of California, Davis mimeo. Polachek, S., Siebert, W., 1996. The Economics of Earnings. Cambridge Univ. Press, Cambridge. Pradhan, M., van Soest, A., 1995. Formal and informal sector employment in urban areas of Bolivia. Labour Economics 2, 275 – 297. Pradhan, M., van Soest, A., 1997. Household labour supply in urban areas of Bolivia. Review of Economics and Statistics 79, 300 – 310. Stigler, G., 1962. Information in the labour market. Journal of Political Economy 70, 94 – 105. Strassmann, W., 1987. Home-based enterprises in cities in developing countries. Economic Development and Cultural Change 36, 121 – 144. Thomas, J., 1992. Informal economic activity. LSE Handbook in Economics. Harvester Wheatsheaf, Hertfordshire. Villagomez, E., 1996. Informal Markets: A Model of Household Labour Supply, working paper, Universidad de Alcala de Henares. Villagomez, E., 1998. Mobility between the formal and informal sector: A panel data analysis for five Mexican cities, working paper, Universidad de Alcala de Henares. Zelek, M., de la Vega, O., 1992. An outline of Mexican labour law. Labour Law Journal, 466 – 470, July.
