EXPLORING VIETNAMESE INEQUALITY USING A MICROSIMULATION FRAMEWORK

Transcription

1 EXPLORING VIETNAMESE INEQUALITY USING A MICROSIMULATION FRAMEWORK This Draft: February, 10, 2007 (Do not quote without permission) Rosaria Vega Pansini* ABSTRACT: Even tough Vietnam has experienced very good performance in terms of sustained growth rates driven by the reforms introduced by the Doi Moi and an impressive reduction in the poverty rate over the past fifteen years, the rising inequality gives reason for concern. Using a microsimulation technique to derive a counterfactual distribution of income, this research studies the evolution of Vietnamese inequality and its decomposition into four different effects: the change in rates of return to variables (price effect); the effect of a change in the socio-demographic structure of the population (population effect), a change in the residual dispersion of unobservables of wages (change in the unobservables effect) and the effect of a change in the occupational behaviour (occupational effect). The price effect and population effect appear to be the main unequalising forces though the effects differ when applied to urban and rural areas Keywords: microsimulation technique, inequality decomposition, Vietnam *Rosaria Vega Pansini, Catholic and Bocconi University, Milan, rosaria.pansini@unicatt.it, rosaria.pansini@unibocconi.it. The author would like to thank all the people whose help and contribution have been precious for the compilation of the paper. First, I thank Professor Sherman Robinson for his useful comments and suggestions. Second, my thank goes to Professor Francesco Timpano and Professor Renata Targetti Lenti for their constant and useful comments and supervision. Usual disclaimers apply. All remain errors are to be attributable to the author. 1

2 1. INTRODUCTION Of transition economies, Vietnam is surely one of the most successful examples. The economic reforms approved under the Doi Moi have put the country on a path of sustained growth and significant improvements in the standard of living of its population. Besides the more general objective of boosting the aggregate growth of the country, one of main concerns of Vietnamese authorities in trying to manage the consequences of the transition to a more market oriented economy, has been keeping the level of inequality as low as possible. Even though Vietnam has experienced very good performance in terms of proportion of people brought out of poverty 1 and continues to have a lower level of inequality than most other developing countries, the fact that inequality rose during the first period of reforms is worrying. Reasons for concern vary. First, the reduction in the rate of poverty brought about by economic growth could be diminished by the increase in inequality; second, the rate of future economic growth could be lowered by an increased disparity; and third, the rise in inequality may negatively affect the general evaluation of the impact of future economic policies. Accordingly to data published by the World Bank, the Gini coefficient calculated using per capita expenditures rose from in 1993 to in A more accurate analysis of the change in the inequality indexes reveals that the majority of the Vietnamese inequality can be explained by the gap between urban and rural areas and that this disparity has widened from 1993 to Per capita urban expenditure has, in fact, increased by 61% while that in rural areas only by half of that, i.e. by 30%. There is also a regional trend in the distribution of resources: 83% of the overall rise in inequality is attributable to widening difference between regions while the remaining 17% is due to rising inequality within each region. A first look at the dispersion in the distribution of expenditure between different areas of the country gives only a partial sight of the level of inequality. Even though in some cases data on per capita expenditures are preferable to capture the level of welfare especially in the context of developing countries (Deaton, 1997), it is useful to analyse which are the other main sources of household income and how accounting for them changes the analysis about the level of inequality. In the context of Vietnamese economy, it seems reasonable to account for 1 Accordingly to the WB 1999, the percentage of people felt from 58% in 1993 to 37% in 1998 using the national total poverty line. 2

3 differences in the distribution of three sources of earned income: wages, agricultural profits and self employment profits. 2 The first aim of this research will be to derive the level of household income, taking into account all its components and to explore which factors can influence its level and its distribution. The second objective will be the study of the evolution of household income over the period and the decomposition of inequality indexes into different components that may have affected the distribution of income. As mentioned before, understanding more deeply the sources of inequality helps to formulate a better judgement of the impact of policy measures adopted by a country and to indicate more efficient ways to reach a higher level of welfare with a more equalised distribution of resources. Among the different techniques available to understand the nature and components of inequality, microsimulation techniques allow for the construction of counterfactual distributions by changing the behaviour of individuals and markets, holding all other aspects constant. This methodology, applied to various case studies in Bourguignon et al (2001), allows for the distinction of four different forces that may influence the dynamics in the inequality of income: the effect of change in rates of return to variables at the individual and household level (treatment effect or price effect); the effect of a change in the sociodemographic structure of the population (endowment effect or population effect), a change in the residual dispersion of unobservable components of wages (change in the unobservables effect) and the effect of a change in the occupational behaviour (occupational effect). Both these objectives are reached based on the following structure. Section two presents some of the main approaches used in literature to address the problem of evaluating the impact of economic policy on the distribution of resources using microsimulation. In the same section, an overview on how to use microsimulation to derive disaggregated inequality indexes is presented. Section three contains a brief description of the data used in this research. The referred model and the richness in the availability of information about economic behaviour of Vietnamese households justify the choice of variables at different levels: individual, household and commune. In the same section, the methodology used for this study is 2 An accurate look at Vietnamese questionnaire, which the two datasets used in this research are based on, helped in selecting those indicated in the text as the main sources of household income for which data were available not only on profits but also on the relative proportion of people employed in all these three categories. 3

4 presented. In section four, some of the main findings emerging from a first look at the data in terms of summary statistics are presented. Tables are reported presenting the descriptive distribution of the three components of household income and how they vary in terms of selected criteria (sex, gender, level of education, region and area of location). Section five presents the results of the empirical analysis applied to Vietnamese data. First, the basic estimation of the various components of the income generating model is presented in order to underline the determining factors. Second, the evolution of household income is sketched to evaluate the determination of income inequality between 1993 and Finally, a counterfactual distribution is derived and the adopted decomposition technique is applied to obtain the magnitude of the four different effects influencing the change in the distribution of income. Section six contains the main conclusions of the analysis highlighting some pros and cons and indicating directions of future research. 2. USING MICROSIMULATION TO DECOMPOSE DISTRIBUTIONAL CHANGES Understanding the different factors behind the change in the distribution of income has always been a key instrument to inform policy making about which direction interventions should take. This task is made difficult also by the fact that these forces are not independent but they tend to offset one another. Traditionally, the way in which these factors have been explored in the literature has been by using summary measures while ignoring the changes in the entire distribution. This methodology consists in deriving the change in the overall mean or change in the inequality measures as the aggregation of means of the socio demographic characteristics of various subgroups in which the population could have been divided. The need to take into account the full distribution of income leads to exploring new techniques of decomposition able to exceed the analysis based on the use at the mean of the distribution. These techniques belong to the use of the parametric representation of the household income and its relationship with some individual or household characteristics. 4

5 The remains of this section will be devoted to briefly present some of the approaches to decomposition based on summary statistics measures. After that, some example of the parametric approach to the study of distributional change will be presented. 2.1 SCALAR METHODS THE OAXACA-BLINDER DECOMPOSITION One of the most famous decomposition in distributional changes in the means of two populations is the one found independently by Oaxaca and Blinder in Assuming that the individual incomes of two subgroups A and B can be estimated using the following linear model: y = X ' ˆ A β + u A (2.1) A A y = X ' ˆ B β + u B (2.2) B B where the individual income depends on some observable mean characteristics error term u i X i and on an. If the components of the matrix X are some individual endowments, the coefficients β can be considered as returns to these characteristics, or price effects. Estimating the previous relations using the ordinary least squared estimator and considering the difference in the mean income between the two groups, the results is represented by: y = y y = ˆ β '( X ) ( ˆ A X B + β ˆ β ) ' X B (2.3) A B A A B The change in the mean income could be considered as the sum of two different effects: i) the endowment effect, i.e. the change in the mean characteristics at constant prices and ii) the treatment effect, i.e. change in the prices at constant mean of characteristics. This methodology can be applied to decompose the different in the mean income between two groups or the different in the mean income of the same group between two periods. Some points must be noticed about the Oaxaca- Blinder decomposition: first, it is path dependent, in the sense that there is no reason to think that the decomposition contained in (2.2) brings at the same results of a decomposition using as a staring point the mean income of the group B. 5

6 Second, the Oaxaca decomposition is a static methodology, as it analyzes the endowment and treatment effects at a given time. Juhn, Murphy and Pierce (1991) introduced a dynamic dimension of the decomposition of the first moment, as stated in Paci and Reilly (2004) DECOMPOSITION IN INEQUALITY MEASURES An alternative way of deriving an analysis about the decomposition of changes in the distribution of income is to use the decomposition not just for the first moment of the distribution but for higher moments. This in particular can be done using the General Entropy inequality measures that are characterised with very convenient properties of decomposition. Supposing that the population can be split into different subgroups and denoting with I g the inequality index of the group g and with I the overall inequality index, these measures satisfy the general property: G ( 1 1 ) ( I = I + I = I n, y ;...; n, y + I w n, m (2.4) b w G G g g g g = 1 ) Where n g and m g represent respectively the population and the income shares of group g and IW and I b stand for within and between components of the inequality measure. Due to the possibility of decomposing these indexes in the two components, also the change in the distribution can be easily decomposed in the change of the component between and within IB. In turn, both changes can be expressed as linear combinations of changes in the within group inequality measure I g and a change in the population and income shares, ng and mg inequality.. Here are many application of this methodology to the analysis of the evolution of I W One of the most famous is the one contained in the study f the evolution of distribution of income in the UK in Shorrocks and Mookherjee (1982). This kind of procedure is more appealing than the one contained in the Oaxaca- Blinder formula for two reasons: change in the socio demographic characteristics of the population are equivalent to change in the group s population weights and change in the group relative incomes play a similar role to 6

7 change in the price coefficients. Two differences with respect to the Oaxaca decomposition must be also noticed: first, this kind of decomposition is a non parametric one while the Oaxaca is based on the mean of the distribution; second, the decomposition contains a residual that sometimes is difficult to interpret especially when it is big. 2.3 PARAMETRIC METHODS FOR THE ENTIRE DISTRIBUTION The methodologies presented in the previous paragraph have been extensively used to study the change in the distribution of income. As noticed before, these methods are based on the analysis of one synthetic measure of the distribution without taking into account changes that may occur in all the parameters that characterized the distribution of income. The approach followed in this research is different principally because it studies the determinants of changes in the distribution of income when some or all the parameters are taken into account. In the specific case of the income generating model presented in the following section, the analysis is conducted simulating the change in all the parameters that characterised household income. In order to disentangle the effect of some of the characteristics contained in the matrix X, the analysis could also be conducted changing only the coefficient corresponding to these variables 3. This decomposition methodology based on the derivation of counterfactual distribution for one year using the parameters for the other years has been adapted to the specification of the country studied and to the availability of the data on the various components of household income, (Bourguignon et al 2001). Bourguignon, Fournier and Gurgand (2001) apply this methodology to explore the factors behind the stable household income distribution in Taiwan during a period, , of high growth rates. They found that far from being the result of an almost unchanged distribution of income, the stability in the inequality indexes is the results of numerous offsetting forces. Change in the population structure and changes in the female participation to the labour market appear to have been the most important unequalising forces behind the change in the distribution of household income. Grimm (2001) applies the same methodology to the decomposition of inequality and poverty changes in Cote d Ivoire. The referred context is interesting for our analysis of the 3 If we want to explore, for example, the effect on income inequality derived from a change in the rates of return to schooling, we should substitute only the parameters relative to the variables related to the level of schooling. 7

8 Vietnamese case because it is applied to a period of rapid macroeconomic adjustment. Grimm shows that the increase in income inequality was due principally to changes in the population structure and to increased dispersion in the unobservables determining the wage function. Alatas et al (2000) apply this kind of methodology to the study of evolution of inequality in Indonesia in the period They run the microsimulation exercise not only over the entire distribution of household income but also over its single components, like per example individual wages. They discover that while in the decomposition of inequality of individual earning the price effect represents an unequalizing force the overall price effect for the household income was negative, thus implying a decrease in the level of inequality. These results show how a particular factor can contribute differently to the sign and dynamic of individual ages and household income. Finally, the methodology based on the parametric derivation of counterfactual distribution has been applied not only to just a microeconomic context but also to models that link the micro level to the macro level. An example of this approach is contained in Bourguignon, Robilliard and Robinson (2003) that study the change in the inequality and poverty indexes linking the evolution of the labour market supply, the change in the level of earned wages and the level of self employment profits to their variation emerging at the macroeconomic level. This study is applied to the case of Indonesia before the Asian crisis and the findings show that taking into account also effects at the micro level and link them to the transformations emerging at the aggregate level contributes to a better understanding of the factors influencing the distribution of resources among the household and how they react to changes intervened at the macro level. This short view at the alternative approaches used to capture different effects on the evolution of the household income distribution has shown how the methodology chosen for this study on Vietnamese data has its own advantages in trying to capture different effects operating on the income inequality using the entire parameterisation of the household income The fact that it has been widely applied to different contexts shows that it is quite flexible to the specification that can be done in order to adapt the general model to the specific features of the country analysed. 8

9 3. DATA AND METHODOLOGY 3.1 VLSS 1993 AND 1998 The present research uses data from two different household surveys representing the first two rounds of the Vietnam Living Standard Surveys, the VLSS 1992/93 and the VLSS 1997/98. Both surveys have been conducted by the General Statistical Office (GSO) under the supervision and technical support of the World Bank, UNDP and the Swedish International Development Authority (SIDA). Following the more general framework of the Living Standard Measurement Study (LSMS), a World Bank project about the measurement of living standards in the developing countries, both of the VLSS are based on a methodology of collecting data on households interviewed twice in the same round at a distance of two weeks. The first round of the VLSS was completed in October 1993 and is comprised of a sample of 4,800 self-weighted 4 households selected from the two stratification groups, urban and rural, according to the proportions of household in both groups emerged from the 1989 Population Census. 5 The aim of the second round of the VLSS undertaken in 1998 was to increase the number of households in the sample in order to be representative at the national level. In order to reach the target of 6,000 households, the 4,800 units of the first round were re-interviewed and 1,200 new households were selected from the sample of the Multi Purpose Household Survey (MPHS) conducted in the SELECTED VARIABLES Both the two rounds of household surveys contain two different kinds of questionnaire: the household and the commune questionnaire. The first contains several sections with information about the household demographic structure, education, health, employment, migration, characteristics of the household dwelling and fertility, agricultural and non- 4 Self weighting characteristic of the first round of the VLSS refers to the fact that each household in the sample had the same probability to be selected than other household. 5 According to the 1989 Population Census, the 20% of the total number of households lived in the urban areas. The design of the sample for the VLSS was then chosen in order to respect these proportions between the two stratifications: at the end, out of total 4,800 households, 3839 were selected from rural areas and 960 from urban areas. 6 The final composition of sample in the VLSS 1997/98 was 6,000 households, of which 1730 from urban areas and 4269 from rural areas. 9

10 agricultural activities and food and non- food expenses, remittances, saving and loans. The commune questionnaire contains information about socio-demographic characteristics of the commune, economic activities, infrastructures, information about educational and health facilities and finally, information about agricultural production at the community level INDIVIDUAL AND HOUSEHOLD LEVEL In order to perform the analysis about the evolution of household income inequality between 1993 and 1998, different variables have been selected both at the individual and household level (see Table A1). At the individual level, different variables have been selected indicating socio -demographic characteristics of the respondent like the age and sex, ethnic group, some measures of the quality of health 8, the level of schooling 9, the relative position in the household, the marital status and variables related to the labour market, like the sector of employment, the wage received, a dummy indicating whether or not the person receives a retirement pension and another dummy indicating whether the person works or the public sector. At the household level, variables indicating the demographic structure of the household have been selected like the size, the proportion of children and adult people and the dependency ratio. 10 Characteristics of the household dwelling like the type of house and of the water used are also included, like the area of the land owned by the household. In addition, regional dummies and four dummies indicating the quarter of the interview have been created to capture regional and seasonal effects, respectively. Variables indicating revenues and expenses in the agricultural sector and in the self employment sector have also been selected in order to calculate agricultural and self employment profit at the household level COMMUNE LEVEL 7 For a more detailed description of the characteristics of both questionnaires and of data, see World Bank (2000a, 2000b). 8 A dummy indicating to have suffered from an illness and a dummy to have been to the hospital in the last four months are variables used as proxies for the level of health. 9 The dummy variables indicate the level of schooling refer to the highest degree obtained. In both round of the VLSS analysed, there was no question indicating the total years of schooling. This is why the highest degree obtained was used as a proxy for the level of human capital. 10 People with less than 15 years old over the whole sample are considered children. The dependency ratio is constructed calculating the proportion of people with more than 60 years old relative to the household size. 10

11 The two rounds of the VLSS used in this analysis allow for the selection of variables also at the commune level. In order to explore which are the characteristics of the context in which the household operates and how they influence the analysis, variables indicating the different types and extension of land of the commune and dummies indicating the level of infrastructure within and accessing the commune have been taken into account METHODOLOGICAL FRAMEWORK DECOMPOSITION OF DISTRIBUTIONAL CHANGES USING MICROSIMULATION The methodology adopted in this research was first introduced by Juhn, Murphy and Pierce (1993) and further developed by Bourguignon, Fournier and Gurgand (1999, 2000) and Alatas, Bourguignon (2000). Different examples of the same methodology applied to study the evolution of income inequality can be found in Bourguignon, Ferreira and Lustig (2004). Let Y be a simple household income function, where the income of the household i at the time t depends on a set of five parameters: some observable socio-demographic characteristics of its members ( x it ); unobservable characteristics ( ε it ) remuneration rates observed ( β t ) and unobserved earning determinants ( t ), a vector of σ and a set of parameters defining the participation and occupational choice behaviour of its members ( λ t ) : y = Y( x, ε, β, σ, λ ) (3.1) it it it t t t The overall distribution of household income at time t is then obtained summing up all ( ) it y and some demographic characteristics possibly included ( x ), in one vector ( D ). can be written as a function H of the former parameters and the distribution of the observable and unobservable household characteristics at date t : it t Dt {, } Dt = H( x itεit, βt, σt, λ t ) (3.2) 11

12 Where {} refers to the distribution of the corresponding variables in the population. The difference between two distributions D and observed at two points in time can be t decomposed into four different effects: i) a change in the remuneration rates of the observed earnings determinants (price effect), ii) a change in the remuneration rates of the unobserved earnings determinants (effect of unobservable), iii) a change in the occupational choice behaviour (occupational effect), and iv) a change in the distribution of observed and unobserved earnings determinants (population effect). The decomposition of the change in the distribution of the household income can formally be written as: D t ' { } { } (): i B = H( x ε, β, σ, λ ) H( x ε, β, σ, λ ) tt ' it, it t ' t t it, it t t t { } { } ( ii): S = H ( x ε, β, σ, λ ) H ( x ε, β, σ, λ ) tt ' it, it t t ' t it, it t ' t t { } { } ( iii): L = H ( x ε, β, σ, λ ) H ( x ε, β, σ, λ ) tt ' it, it t t t ' it, it t ' t t { } { } ( iv): X = H ( x ε, β, σ, λ ) H ( x ε, β, σ, λ ) tt ' it ', it ' t t t it, it t t t (3.3) The microsimulation exercise consists in evaluating the impact of a change in the remuneration rates of the observed earnings determinants (price effect) by comparing the observed distribution at time t with the hypothetical distribution obtained by imposing the remuneration structure of observed earnings determinants at date on the population at time t. The change in the remuneration rates of the unobservable earnings determinants is measured by the change in the residual variance in the earnings functions, using a rank preserving transformation 11. In the same way, the population and the occupational effect can be calculated imposing the population and the parameters for occupational choice observed at time t ' on the observed distribution at the time t. After computing the price, occupational and effect of a change in the unobservable, the population effect can be computed as the residual of the other three effects. t ' The overall change in the distribution between t and identity: t ' can be expressed by the following 11 The residual variance at time t is changed in order to make it identical to the its distribution observed at time t using the following rank-preserving transformation: ε = it σ t ' ( ) εit σ t 12

13 C = B + S + L + X tt ' tt ' tt ' tt ' tt ' (3.4) In the case of the present research, the estimation and derivation of the household income will be performed considering two cross-sectional data sets, for 1993 and The decomposition of the overall change in the distribution of income will be calculated using 1993 as base year and applying to the distribution of observed earnings determinants in 1993, the remuneration rates of the unobservable earnings determinants and the population from HOUSEHOLD INCOME GENERATING MODEL In order to simulate the counterfactual distribution of household income, it is necessary to define how to calculate the household income according to (1). The household income generating model can be summarized by the following set of equations: L = ( x..., x, z..., z, υ..., υ, jt t t t t t t hi hi, j= 1, hi, j= J hi, j= 1, hi, j= J hi, j= 1, hi, j= J λ..., λ, λ..., λ ). (3.5) t t t t hi, j= 1, zi, j= J zi, j= 1, zi, j= J i=1 to kh h and j=w, A, SE i tj = w W ( t, t, t hi = w x hi u hi β ), i=1 to k h (3.6) h Π =Π( x, z, s, β, β ) tj = A, t t t t t hi hi hi hi x z, i=1 to k h (3.7) h kh kh kh t t t, j W t t, j A t t, j SE t h = = hi hi + hiπ = hi + hiπ = hi + y0 i= 1 i= 1 i= 1 h y L w L L (3.8) Equation (3.5) describes the labour supply of each household member i as a wage worker outside the family business (W), as a manager in an agricultural business (A) and as a 13

14 manager in a self employment business (SE). These functions express the labour supply of the household member i as a function of his/her characteristics x hi, some characteristics of the household and its environment and some characteristics at the community level, z hi. Equation (3.6) describes the wage equation as a function of proxies of human capital and other personal characteristics; equation (3.7) represents the profit function depending on a set of variables describing the individual characteristics of the person running the business and the characteristics concerning the environment in which the business is run. Both of the profit functions, agricultural and self employment, have the same specification but are estimated on a slightly different set of variables that could influence specifically the type of business to which they refer. Finally, equation (3.8) calculates the household income by aggregating the different sources of income over the household members. The term indicates income from other sources like transfers 12. Wage income is observed at the individual level, while profits, both agricultural and self employment, are observed at the household level according to both the rounds of the VLSS used in this study 13. All the coefficients of the model ( λ, λ, β, β ) and the standard deviations of the residual terms ( υ t, u t, s t ) are estimated t t t t x, j z, j x z over the two cross sections of data available for 1992/93 and 1997/98. Once that they have been estimated, they are used to simulate the effect on household income of a change in the parameters of the model valid for the year t with the ones estimated for the year y t 0h hi hi hi t ' 14. This kind of model enables to evaluate what would have been the income of household m at the time t if it had adopted the labour supply behaviour observed at time those observed at time t '. t ' or earning have been 4. THE EVOLUTION OF VIETNAMESE HOUSEHOLD INCOME FROM 1993 TO CHANGES IN THE OCCUPATIONAL CHOICE 12 As in Bourguignon et al. (2001), income from transfers and benefits is supposed exogenous. 13 Data on revenues and expenditures referred to agricultural and self employment activities are provided as mean values at the household level and are used to compute data on profits. 14 In the specific case of this research, the parameters will be estimated for both years but the microsimulation exercise will be performed apply only the parameters for 1998 to the population of

15 Accordingly to the model presented in section 3.3.2, one of the components used in the estimation of the household income is the occupational choice. Before presenting the results of the estimation it will help looking at the evolution of Vietnamese employment over the period considered by the two used surveys. Table 4.1 presents summary results about the structure of employment at the national level and distinguished using different criteria like gender and sectoral differentiation. The first feature to be noticed is the high rate of employment. Participation remains essentially constant over the years at the level of 1993 showing that the transformation of the Vietnamese economy has resulted more in a reorganization within the labour market instead of a real contraction of it. Employment in the rural sector is dominant, even in 1998 and varies with gender. A gender effect can be identified when the overall rate of employment is examined in a dynamic prospective. Male participation was 83% in 1993 decreasing by almost three percentage points after six years. In contrast, the overall female employment rate remains constant trough the years decreasing in the urban area and increasing by roughly the same proportion in the rural areas. This finding suggests that women are still dependent on employment opportunities in the rural sector while the urban labour market is relatively dominated by male workers. The differentiation of gender-specific participation rates between urban and rural areas could be brought about by the organisation of the labour market in the agriculture, self employment and wage labour. Female participation is still very high in the agricultural sector even if decreasing over time while female participation has remained low in the wage labour. These proportions show how female labour is still highly dependent on opportunities offered by the agricultural sector or by being involved in small, self-run businesses, mainly related to the manufacturing sector, though sometimes run together with labour supplied on the farm. From 1993 to 1998, economic growth influenced the sectoral changes in the labour market drawing people from the primary sector mainly to the service sector as it developed due to better economic performance. As shown in the following table 4.1, the percentage change in the number of people employed in the agricultural sector is more or less compensated by the increasing of the employment in the wage labour, mainly driven by the expansion of the service sector (ADB, 2000 and UNDP, 2001). 15

16 Table4.1: The Evolution of Employment Structure, Labour force Participation (%) Urban Rural Male urban rural Female urban rural Unemployment rate (%) Male urban rural Female urban rural Structure of employment Agriculture male female Self employment male female Wage employment male female Source: author s calculation based on VLSS 1993 and VLSS Note: as suggested in Pham and Reilly (2006), labour force has been calculated as those aged between 15 and 60 years old over the all sample; employment has been defined having a job over the past 7 days before the surveys while unemployment as being n the labour force, not working in the 7 days before the survey and not looking or a job; finally, employment sectors have been identified accordingly to the main occupation. Figures are calculated as percentage over the total of labour force. 4.2 DISTRIBUTION OF INDIVIDUAL WAGES Besides looking at the changes in the structure of employment, a deeper analysis of the distribution of individual wages is needed. Table A.2 in the appendixes presents results from the two surveys used in this research. The data on hourly wages seem to indicate that the economic growth experienced by Vietnam during nineties, principally driven by the reforms of the Doi Moi, brought about a general increase in the level of individual wages. This means that Vietnam s transition towards greater economic liberalization, as well as the structural 16

17 changes induced by these reforms, has been accompanied by a stable level of employment and by a higher level of wages. It is interesting to see how the figure changes moving from a national to regional analysis. Higher incomes are earned by people employed by the public sector instead of those that work for a private company. This is related to the fact that remuneration in the public sector is comprised of higher benefits, sometimes not earned by workers in the private sector. There is a clear gender pay gap in the Vietnamese labour market but what is interesting, is that it is narrowing with the time. The greater participation of women in wage employment also emerged from table 4.1 suggests that women have increasing access to better paid jobs in 1998 compared to Pham and Reilly (2006) show a general improvement in the relative position of women in the Vietnamese labour market during the transition both in terms of level of payment and in their participation. Not surprisingly, there is a positive correlation between the level of wages and the level of education. In 1993, only a person with a primary education earned less than the mean wage; having a university degree assured a level of earnings almost 70% higher than having a primary school diploma. From a dynamic perspective, the positive effect of education and wages has been maintained and increased in The interesting thing is that after five years also people with a lower secondary diploma do not reach the mean level of wages; otherwise people with a higher level of schooling have a level of earnings that is more than double the mean value. This means that the returns to higher level of education have increased while the returns to primary and lower secondary education have decreased. This picture could be related with the fact that the economic growth and the structural changes (the progressive shift of employment from the low paid agricultural sector to he higher paid wage labour) incurred in the Vietnamese economy have increased the opportunities available to better educated people and has awarded them with a higher level of payment. Another important factor determining 15 the level of individual wages is location. The urban- rural paid gap already wide in 1993, increased substantially by 1998, contributing to the observed widening of overall income inequality, as noted in Molini (2005). This unsurprising 15 As it also emerges from the results of the estimation of the Mincerian function contained in table

18 gap could be related to the fact that the rural labour market is composed mainly of people occupied in the agricultural sector that has, on average, a lower level of payment than the wage sector dominant in the urban areas (UNDP, 2001). Moving from the north to the south of the country, wages go from below to above the national mean, even if this is just part of the story. As shown in table A.2, wages are higher on the North Central Coast and in the Southeast region in In 1998, the Southeast becomes the region with the highest level of wages 16. Finally, wages increases with the age of the worker. Not surprisingly, as one considers the age as a proxy of the experience, higher wages are registered as the individual becomes older but until a certain level of age. In 1993 the highest level of income is observed for people aged between 50 and 59; while in 1998 the highest remuneration is registered for younger people, between 40 and 49 years old. The fact than people with more than 60 years received the lowest payment could be related to the fact that after that age an individual receive a state pension and so potentially are out of the labour market DISTRIBUTION OF AGRICULTURAL AD SELF EMPLOYMENT PROFITS Tables A.3 and A.4 present summary results about the evolution of agricultural and self employment household profits 18. Mean agricultural household profits increased between 1993 to This could be related to the fact that economic expansion helped the agricultural sector in terms of higher revenues especially in the export oriented cultivation, like coffee and rice (Minot et al 2001). Not surprisingly, there is a gender gap also in the distribution of earnings from agriculture. 16 The region of Southeast is also the richest in the country, where there has been the greater improvements in the economic conditions of the population bringing the highest proportion of people out of poverty in the country, WB(1999). In addition, the Southeast region counts the highest proportion of people living in urban areas, another factor that contributes increase the average level of ages in the region. 17 In the preceding paragraphs about the evolution of the employment, people more than 60 years old are not considered in the labour force. The fact that they are considered as recipients of earned income is related to the fact that could continue to work in the agricultural sector. 18 It should be noticed that in both the used surveys data about agricultural and self employment revenues and expenditures are reported as mean values at the household level instead of individual level as in the case of wages. Moreover, there is a high number of missing values in the reported data and this could a reason why they are not considered reliable as measures of household welfare. In the case of the present research they have been used in conjunction with more accurate data on occupational choice and individual wages in order to give an indicative picture of the different sources of income at the household level and how they are distributive among the Vietnamese population. 18

19 Households with a male head appear to earn on average a profit higher than those whose head is a female. Not surprisingly, there does not appear to be a strong correlation between the level of education of the household head and agricultural profits. Returns to schooling are not strong in the case of agricultural employment and the figure for 1998 confirms that profits for households with a primary school educated head earned more than those with a head with a secondary school diploma. Looking also to the evolution of employment and the relative increase in wage labour participation against the agricultural sector, this effect could be related in general to the presence of more households heads with a primary level of education in the agricultural sector with respect to those with better educated household heads. Agricultural profits are also differentiated between different regions. The highest level is registered going to the south of the country and this trend is maintained also in This could be explained with the fact that southern regions are those with the highest productivity and dedicated to the production of rice, for with Vietnam is one of the main exporter. With regard to the profits derived from a self employment activity, there is still an increasing trend that confirms the positive economic performance of Vietnam during the nineties. The gender gap in the case of self employment profits is the reverse of that identified in the agricultural sector. Households with a female head have, on average, higher income. This will be confirmed also by the results of the estimation of the determinants of occupational choice in the self employment sector and it could be explained by the fact that women are more often involved in some small businesses devoted to manufacturing. Male head households are more successful in agricultural activities or in the wage labour, as noted above. Not surprisingly, and in contrast with agriculture, both the level of education and the age of the household head are positively correlated to the level of profits in the self employment sector. More experience and a higher level of education allow the household s head to be involved in more profitable activities. Finally, as expected, the southern regions of the country remain the most profitable. 5. EMPIRICAL RESULTS USING MICROSIMULATION 5.1 ESTIMATION OF HOUSEHOLD INCOME OCCUPATIONAL CHOICE As stated in the model described in chapter 4, in order to derive the household income, we first need to estimate the occupational choice made by each member of the household. The 19

20 aim will be to investigate factors influencing the probability of attachment to one of three main employment categories: 1) paid labour; 2) agricultural sector and 3) self employment labour. Given that multiple occupational choices have been selected, a multinomial logit model (MNL) (McFadden, 1973, 1984) will be used to estimate the relationship between the probability of attachment to one of the three sectors and some characteristics at the individual, household and commune level 19. The model has been estimated for both surveys 1992/93 and 1997/98 and separately for different members of the household. Accordingly to Grimm (2001), it is reasonable to assume that occupational choices within the household are interdependent. This effect is taken into account estimating the MNL model in a sequential decision process, i.e. estimating first the decision of the household head and afterwards that of the household s head spouse and of the other members of the household conditional to the decision taken by the head 20. Before presenting the results of the estimation of the model, some observations need to be done about the specification of the model. First, the Theil normalization 21 for the MNL model has been applied to the category with the highest frequency of employed people, in this case the agricultural sector. Following this specification, all the coefficients would be interpreted as the contribution to the probability to be involved in one occupational category relative to the one chosen as base reference. Second, consistency with the hypothesis of Independence of Irrelevant Alternatives (IIA), which states that coefficients remain unchanged if there are changes in the set of alternative outcomes, has to be verified. Results for the Hausman, the Small-Hsiao and the Wald test have been reported in table A2 in the appendixes. There are Y = 19 th th Letting 1if the individual chooses the j employment outcome and 0 otherwise (where ij i j=1,2,3), the probability that an individual i chooses the alternative j can be expressed (following the suggested specification for Vietnam contained in Hung Pham (2006:10) as: Pr ob( Y = j) = i j exp α + β jx + λ X + γ X 3 = 1 ind hh com i i j j i j exp α + β jx + λ X + γ X ind hh com i i j j i j Where Pr ob( Y = 1), Pr ob( Y = 2), Pr ob( Y = 3) represent respectively the probability that the individual Y = is employed as a wage worker, n the agricultural sector or is a self employed person, and where ij ind X i, hh X i and com X i are matrices of individual, household and commune characteristics, and β j, λ j and γ j are vectors of coefficients. 20 The sequential feature of the occupational choices of the different household members is capture inserting in the estimation of the model for the spouse and or other household members the predicted probability to find the household head as a wage worker, as working in the agricultural sector and as a self employed worker, respectively. 21 The Theil normalization implies that one category is chosen as a base and is coefficients are then set to zero. 20

21 some cases in which the IIA hypothesis tested using the Small-Hsiao test 22 is not respected and this can be due to the fact that there are some observations for people reported in more than one occupational category. This phenomenon can be attributed, for example, to the fact that people can report to work as a wage workers and also for the household in some agricultural business. This may further suggest the presence of an additional occupational category that needs some more investigation, the one of occupational multiplicity 23 for of those people working or the household. Results from the estimation of the multinomial logit models for different household members are reported in the appendixes. The derived marginal effects for infinitesimal changes in the continuous variables or the impact effects for dummy variables (indicated with a d in brackets) on the probability to choose one of three occupational outcomes, evaluated at the sample mean of the dependent variable and ceteris paribus, are reported bellow in tables 5.1, 5.2, Analysing the factors influencing the occupational choice of the household head at the individual level over the two surveyed years, while the ethnicity doesn t seem to be have a strongly determined 25 impact on occupational choice, the first significant effect at the individual level is represented by the sex of the respondent capturing the gender effect on occupational choice. Men are more likely to be wage workers and less likely to be selfemployed with respect o being 22 Results are reported for the three tests performed on the model specification. As one of the drawbacks of the Hausman test is that it could produced negative values for the statistics, like in the case of our model as showed in the table A2, the results of the Small-Hsiao test are preferred, as noted also in Small and Hsiao (1985). 23 The phenomenon of occupational multiplicity, i.e. of people working in more than one sector, or of people with more than one job in the Vietnamese labour market is reported also in Haughton et al.(2001). 24 The estimated coefficients for the multinomial logit are exported in the appendixes. 25 It should be noted that the marginal effect associated to the coefficient of the dummy for belonging or not to the major Vietnamese ethnic group (Kihn) is significant only at the 5% level only for the category of wage labour for 1998: on average and ceteris paribus, belonging to the Kihn majority increases by 7.2 percentage points the probability to be self employed with respect to be employed in the agricultural sector in This ethnic effect can be due to the fact that other ethnic minorities are concentrated in the remote rural areas of Vietnam making then difficult for them to be a self employed person. 21