UNIVERSITY OF ECONOMICS HOCHIMINH CITY UNIVERSITY OF ECONOMICS HOCHIMINH CITY VIETNAM INSTITUTE OF SOCIAL STUDIES THE HAGUE THE NETHERLANDS VIETNAM – NETHERLANDS PROGRAMME FOR M A IN DEVELOPMENT ECONO[.]
UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES HOCHIMINH CITY THE HAGUE VIETNAM THE NETHERLANDS VIETNAM – NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS AN ASSET-BASED GEOGRAPHIC TARGETING: EVIDENCE FROM RURAL VIETNAM A thesis submitted in partial fulfillment of the requirement for the degree of MASTER OF ARTS IN DEVELOPMENT ECONOMICS By PHAM THI NGOC AI Academic Supervisor: Dr PHAM KHANH NAM HO CHI MINH CITY, MAY 2014 ACKNOWLEDGEMENT The thesis would not have been finished without the kind assistance and fruitful guidance of many people who have the contributions of different aspects for accomplishing the thesis First of all, I am specially grateful to Dr Pham Khanh Nam who encourages me at the beginning of title and help my deep understanding on literature theory as well as thesis writing In addition, I would like to express the sincere gratitude to Dr Truong Dang Thuy for sharing his knowledge for the technique of the model and some valuable advices for the methodology I would like to give special thank for my boss and colleagues who create conditions and assist working in order that I have more time for the research Finally, my most gratitude is for my family, especially my parents and husband who have been always side by side with me during learning this program and researching process ABSTRACT The purpose of this paper is to find out which asset is the most suitable for a particular region through calculating marginal return to a range of assets and then creating a serial of maps The data are taken from Vietnam Living Standard Survey in 2006 The Weighted Least Squares is used for running the regression and combining with technique bootstrap and stepwise iterative deletion with the threshold of 5% All targetable assets are focused on calculating marginal benefit It gives the reasonable findings that have very heterogeneous average marginal benefit across areas The results give suggestion for choosing which assets are suitable for a particular region, thus it makes increases their efficacy However, the governors and donors should consider the existence of trade-off equity and efficacy TABLE OF CONTENTS CHAPTER I: INTRODUCTION 1.1 Problem statement 1.2 Research objective 1.3 Research questions 1.4 Research contributions 1.5 Organization of the paper CHAPTER II: LITERATURE REVIEW 2.1 Geographic targeting theory 2.2 Household welfare function 2.3 The small estimation method 2.4 Transfer in-kind 2.5 The linkage between household welfare and return to assets 2.6 Review of empirical studies 12 CHAPTER III: OVERVIEW OF HOUSEHOLD WELFARE IN VIETNAM AND METHODOLOGY 15 3.1 Overview of household welfare in Vietnam 15 3.2 Econometric models 19 3.3 Data 22 3.3.1 Independent variables 23 3.3.2 Dependent variable 28 CHAPTER IV: EMPIRICAL RESULTS 29 4.1 Descriptive statistics 29 4.2 Econometric results 33 4.2.1 Statistics and value of marginal return of assets at national level 34 4.2.2 Analysis for average of mean marginal return of assets at provincial level 36 4.2.3 Kinds of maps for Vietnam 42 CHAPTER V:CONCLUSION, POLICY IMPLICATION, LIMITATION AND FURTHER RESEARCH 49 5.1 Conclusion 49 5.2 Policy implication 50 5.3 Limitation of this study 51 5.4 Direction for Further research 52 REFERENCES 53 LIST OF CHARTS Graph 3.1:The quintiles of income in urban and rural of Vietnam 16 Graph 3.2: The quintiles of expenditure in urban and rural of Vietnam 16 Graph 3.3: The Quintiles of income in the eight regions of Vietnam 17 Graph 3.4: The quintiles of expenditure in the eight regions of Vietnam 18 Graph 3.5: Poverty rate at different level of region in Vietnam (Unit: %) 18 Graph 4.1: Proportion of literate for each region 29 Graph 4.2: Distribution of educational level for each region 30 Graph 4.3: Distribution of expenditure for each educational level and each region 30 Graph 4.4: Distribution of ethnic minorities across regions 31 Graph 4.5: Expenditure of some ethnics 32 Graph 4.6: Distribution of livestock for each region 32 Graph 4.7: Distribution of other assets across regions 33 LIST OF FIGURES Figure 4.1: Maps of AMB that is significantly greater than zero 43 Figure 4.2: Maps of proportion of positive AMB 44 Figure 4.3: Maps of maximum significant AMB 45 Figure 4.4: Maps of maximum proportion of positive 46 Figure 4.5: Example of choosing cattle transferred to provinces which meet three conditions: the magnitude of AMB at 0.035, 95% households have positive AMB and poverty rate 30% 48 LIST OF TABLES Table 3.1: Asset variables 26 Table 3.2: Control variables 270 Table 4.1: Average standard deviation of mean marginal return for national, regional and provincial level 34 Table 4.2: Mean of AMB and proportion of provinces with positive AMB at national level 36 Table 4.3: Values of AMB that are significantly greater than zero 38 Table 4.4: Data for proportion (%) of positive AMB of households in provinces 40 Table 4.5: Correlating between asset holdings and poverty with significant and proportion of positive AMB 47 CHAPTER I: INTRODUCTION 1.1 Problem statement Alleviating poverty is always the major targeting interested in by policy-makers in the developing countries There are many transfer programs made around the world from years to years However, increasing transfer program efficacy under the condition of scarce resources is the extremely important issue which governors and donors consider One of the methods used widely and popularly for researching and practicing is geographic targeting As it gives a visual and useful tool and performance with low cost and easy administration Thus, the important rule of the geographic targeting for poverty reduction is emphasized by Baker and Grosh(1994), Bigman and Foback (2000) Moreover, many papers have proven that the efficacy of transfer programs is higher when geographic units are smaller (Elbers et al 2007, Minot 2000, Bigman and Foback 2000) Poverty map is a tool of geographic targeting It displays poverty indicator across geography and answer the question where the poor people reside and who the poor people are (Elbers et al 2003, Minot and Baulch 2005) as well as why the area has high incidence poverty which is driven by natural resources (Szonyi et al 2010) However, the greatly advanced step of geographic targeting is targeting map with asset-based approach It answers the extremely important question that governors and donors should use in-kind transfer for a particular region to bring the highest benefit for the poor and might create motivation for them out of poverty Besides, it gives a visual and practical tool for policy-makers and donors to manage their transfer programs with budget limitation Benefit of transferring whether in-kind or cash is better, always consider by the researchers, donors and governors However, in many places in the world, they prefer transferring in-kind and in some cases, transfer in-kind is better (Hoffmann, Barrett and Just, 2009) Commented [PKN1]: I’ve made a new paragraph here for household head is in consistence with significantly higher household consumption; the different expenditure between having permanent and temporary house, semipermanent and temporary housing, respectively 23%, 14-15%; television ownership is positive and significant highly at 1% level; radio ownership is similar with television ownership, but at lower significant level At end, the poverty map which is built by poverty indicators shows that incidence of poverty is highest at mountainous areas like Northern Upland, but these areas are not high poverty density The lowest one is at delta areas like South-East and cities, however two deltas in which have highest poverty density For another research using geographic targeting method, Cuong et al (2010) also establishes poverty and inequality map in Vietnam They use the small estimation method by combining the Vietnam Living Standard Survey and Rural Agriculture and Fishery Census in 2006 The poverty and inequality indicators are calculated based on the household welfare function This function shows the relationship between logarithm of household consumption or income per capita and predictable variables of household characteristics like durables assets, condition on housing, water, electricity and toilet, commune variables like road, school, market and geographic variables like sunshine, temperature and rainfall The chosen model regression is the forward stepwise regression for each of eight regions The explanatory power of the model has the range from 0.43 to 0.7 The findings show that poverty indicator calculated by expenditure model and income model is similar almost, the rest is different for areas which have higher number of poverty However, poverty indicators determined from expenditure model is more strongly correlated with MOLISA poverty rate than income model The household expenditure reflects the household welfare better than household income The level of poverty strongly correlates with ethnic minority It means that return of endowments in household ethnic minorities are much lower than household Kinh/Hoa All durables assets and education for each region have different influence 14 on household welfare but always positively Besides, the housing condition is strongly correlated with household welfare Household will have more expenditure if housing condition is better 15 CHAPTER III: OVERVIEW OF HOUSEHOLD WELFARE IN VIETNAM AND METHODOLOGY 3.1 Overview of household welfare in Vietnam Vietnam is a country with 84 million people approximately, average income per capita is about 636.000 dong (GOS 2007) and GDP growth about 7% per year in 2006 It is considered as a rural country because 72% population resides in rural areas The whole country has 64 provinces and divides into eight regions including Red River Delta, North East, North West, North Central Coast, Central Highland, South Central Coast, South East and Mekong River Delta The regions in which converge the most urban residents are the Southeast and the second is Red River Delta, which contains Hanoi Capital Other regions such as North West and Central Highlands have the least urbanized To understand the various level of household welfare which belongs to different population groups at heterogeneous level of region, we divide population into quintiles In each quintile, there is 20% population equally in term of welfare elements Through Graph 3.1, Graph 3.2 and Graph 3.5, we can recognize that each quintile of income and expenditure always in rural areas is much lower than in urban one, especially the various level of quintiles increases for part of population with high incomes For example, the income and expenditure of Quintile which belongs to 20% population with lowest income, is respectively 15.640.000 dong, 13.263.000 dong in urban areas and 8.425.000 dong, 7.624.000 dong in rural areas For Quintile which belongs to 20% population with highest income, is respectively 111.450.000 dong, 90.486.000 dong in urban one and 63.368.000 dong, 49.729.000 dong in rural one Welfare in urban regions has twice in rural one approximately Poverty rate of urban 16 (7.7%) is lower twice than its rural (18%) We may suggest that increasing amount of urban residents is considered as making a move from the poorer regions to wealthier regions All figures are based on VHLSS 2006 Income 120,000 100,000 80,000 60,000 40,000 20,000 Quintile Quintile Quintile Urban Graph 3.1: Quintile Quintile Mean Rural The quintiles of income in urban and rural of Vietnam (Unit: thousand dong) Expenditure 100,000 90,000 80,000 70,000 60,000 50,000 40,000 30,000 20,000 10,000 Quintile Quintile Quintile Urban Quintile Quintile Mean Rural Graph 3.2: The quintiles of expenditure in urban and rural of Vietnam (Unit: thousand dong) 17 There are remarkable distances in welfare among the Vietnamese regions (Graph 3.3, Graph 3.4 and Graph 3.5) The highest welfare is in the South East (average income of 47.428.000 dong and average expenditure of 32.559.000 dong and poverty rate at 3.1%) The three regions in which have the lowest welfare indicators including North West, North Central Coast and North East The difference of welfare indicators in Quintile I seems similar for all the regions For instance of income, the lowest level in North West is 8.434.000 dong and the highest in South East is 14.385.000 dong But for Quintile 5, difference of welfare is by far across regions Moreover, richer regions have larger distance of the poor and the rich These things suggest that there is large inequality of welfare between the rich and the poor within each regions and increases more among different regions The difference of the rich and poor is remarkable Graph 3.3: The Quintiles of income in the eight regions of Vietnam (Unit: thousand dong) 18 Graph 3.4: The quintiles of expenditure in the eight regions of Vietnam (Unit: thousand dong) Similarly, poverty rate is the highest in North West and North West (27.5%) and the lowest in the South East (3.1%) Poverty rate (%) 30 25 20 15 10 Urban Rural Red River Delta Northern North Central midlands and and Central mountain coastal areas areas Central Highlands South East Graph 3.5: Poverty rate at different level of region in Vietnam (Unit: %) 19 Mekong River Delta Commented [PKN5]: Move this section to before the independence and dependence variables sections 3.2 Econometric models: To create targeting map with average marginal benefit of each asset for each region, we use the small area estimation method pioneered by Elbers et al (2002,2003) and combine with the function of welfare house defined by Naschold and Christopher (2011) and Lang et al (2013) Household welfare is function of asset holdings and place-particular expected asset returns shown the form of second order flexible function generally with explanatory variables includes: Aic private, targetable assets; Ac : location-specific means of Aic; Bc : public, targetable assets; Yic : private, nontargetable assets; Yc : location-specific means of Yic; Zc : public, non-targetable assets; and Xic : additional controls (Table 3.2) The details of these assets are shown as per Table 3.1 The general model function shows as follows Ln Expic = A’icRA(Aic,Ac,Bc,Yic,Zc) + B’c RB(Aic,Ac,Bc,Yic,Zc) + Y’ic RY(Aic,Ac,Bc,Yic,Zc) +δ’ Xic +εic (1) Where Expic: expenditure; R(k) is returns to asset type k: A,B,Y,Z The function of asset returns permit the expected return of asset to impacted by the inventory of each other asset For instance, pig’s return changes according to education of household head, average of pig existed at that area, rainfall or/and having a market or not For location-particular average has only the link with the same indicator at household level The function has the error term composed of a location-specific and a house-specific component: εic =βc (Mc) + δic where Mc= [Ac, Bc, Yc, Zc] Based on the above things, we can rewrite the detail for the functional form of household welfare: Lnexp= β0+β1mot +β2motedu +β3motroa +β4motban +β5motpop +β6moteth +β7motdis +β8motmar +β9mottem +β10motran +β11motdry +β12motmmo +β13bic +β14bicedu +β15bicroa +β16bicban +β17bicpop +β18biceth +β19bicdis +β20bicmar 20 +β21bictem +β22bicran +β23bicdry +β24bicmbi +β25wat +β26watedu +β27watroa +β28watban +β29watpop +β30wateth +β31watdis +β32watmar +β33wattem +β34watran +β35watdry +β36watmwa +β37lan +β38lanedu +β39lanroa +β40lanban +β41lanpop +β42laneth +β43landis +β44lanmar +β45lantem +β46lanran +β47landry +β48lanmla +β49cat +β50catedu +β51catroa +β52catban +β53catpop +β54cateth +β55catdis +β56catmar +β57cattem +β58catran +β59catdry +β60catmca +β61pig +β62pigedu +β63pigroa +β64pigban +β65pigpop +β66pigeth +β67pigdis +β68pigmar +β69pigtem +β70pigran +β71pigdry +β72pigmpi +β73chi +β74chiedu +β75chiroa +β76chiban +β77chipop +β78chieth +β79chidis +β80chimar +β81chitem +β82chiran +β83chidry +β84chimch +β85lit +β86litedu +β87litroa +β88litban +β89litpop +β90liteth +β91litdis +β92litmar +β93littem +β94litran +β95litdry +β96litmli +β97edu +β98eduroa +β99eduban +β100edupop +β101edueth +β102edudis +β103edumar +β104edutem +β105eduran +β106edudry +β107edumed +β109roapop +β110roaeth +β111roadis +β112roamar +β113roatem +β114roaran +β115roadry +β116ban +β117banpop +β118baneth +β119bandis +β120banmar +β121bantem +β122banran +β123bandry + β124fixphone + β125 tele + β126fridge + β127per_house + β128semiper_house + β129otherhouse + β130 tap_water + β131clean_water + β132 otherwater + β133flushtoilet + β134othertoilet + β135notoilet + β136elec + β137otherelec + β138mem_school There are many variables in the model This leads the probability of wrong estimation of relationship highly If we remove indicators which are not significant directly within the first running, we can delete important variables To solve both these issues, we perform bootstrap technique and combine with stepwise iterative deletion with the pvalue level at 5% at 200 times To find out expected marginal benefit of assets, we have to make two following steps: In the first step, we use the data of the household survey to determine the correlation between household expenditure and asset holdings This process is combined with 21 bootstrap technique and stepwise iterative deletion with the p-value level at 5% The model regression is Weighted Least Squares to reduce heterogeneity In the second step, the estimated coefficients which are taken from Eq(2), are imputed into the census data We calculate the expected marginal return of assets by taking derivatives Eq(2) for the all targetable assets The two above steps are performed repeatedly 200 times Thus, we have 200 the models after regression Each model will give the value of expected marginal return of each asset To take average all 200 values of each marginal return of each asset, we have the average expected marginal return of household to each asset Then, we combine all the households across the determined region and find out statistics fundamental for all the targetable assets To know the expected size of mean benefits correlated with particular asset for a determined region, we calculate mean and standard error of expected AMB in each particular region And we identify AMB that are statistically significantly greater than zero with the level of 0.05 To express the scale of benefits for the particular assets moved to the determined region, we compute the number of households which have positive expected AMB for each particular region To create a visual and useful tool, we display all calculated results on map There are three kinds of map One kind of map shows maximum of expected AMB of each asset (the most beneficial asset) or the highest proportion of households which have positive expected AMB for each spatial region Other kind of map displays proportion of household which have positive marginal return and expect marginal of returns that are greater than zero for a determined asset At end, we combine with poverty indicators which taken from the paper of Cuong et al 2011 to create the map that displays the data on poverty level, highest level of AMB of a give asset 22 3.3 Data: All the data of thesis are taken from the Vietnam Household Living Standard Surveys (VHLSS), the Rural Agriculture and Fishery Cencus (RAFC), website: http://fsiu.mard.gov.vn/data/khituong.htm and Yearbook of Statistics in 2006 The VHLSS’s data is representative at the region level It includes information of demographic and household characteristics, education, infrastructure, health, communes, income, expenditure, assets, housing, rural activities, migration The data set contains 39071 individuals of 9,189 households including 2,250 households at urban areas and 6,939 rural areas As for Rural Agriculture and Fishery Census, it still contains information on demographic and household characteristics, education, infrastructure, health, communes, assets, housing, rural activities, migration but has no information on income and expenditure but it can cover the data for small areas like communes, districts and provinces The General Statistics Office conducted Rural Agriculture and Fishery Census and Vietnam Household Living Standard Surveys data At the scope of thesis, we limit our attention for households in rural areas as the likelihood of expected asset return depends on the heterogeneous geography and natural capital in rural is higher than in urban We have to draw our attention when the process of merging two sets of data There are some different things between VHLSS and RAFC VHLSS only contains the individuals older than or equalivent to 15 years old and identify the head household, but RAFC is not Besides, the code for provinces, districts and communes is not the same between the survey and census The issue is solved by changing the figure code to exact name of this region when we want to merge the data 23 3.3.1 Independent variables: a) Category of asset: - Private and public goods: Fleisher et al., 1987 defines public goods as: “Goods and services are called public goods when users cannot easily be excluded from using them and when the good made available for one user become available to many without further cost.” Based on this definition, public goods have two characteristics: non-exclusiveness and non-rival The opposite of public goods are private goods - Targetable and non-targetable assets: According to capturing by Lang et al (2013), “targertable from non-targetable assets based on whether an asset’s quantity, quality or existence can be changed by an intervention” Based on two kinds of assets, we divide assets into four categories: Private targetable asset such as livestock holdings, literacy, and land holdings Private non-targetable assets such as education level of head household, gender of head household, ect Publictargetable assets such as source of potable drinking water, access to health clinics, and road access) Public non-targetable assets such as rainfall and temperature b) Criteria of choosing assets: To apply the small estimation method, we have to choose the common independent variables which exist in two datasets The main criteria is used for choosing as follows: Firstly, variables exist in VHLSS and RAFC Secondly, the definition of each variable in questionnaires is the same and summary statistics of two databases is similar (Both datasets is similar in size, amount or quality) 24 Thirdly, these variables have a significant impact on household welfare The assets are chosen as follows: -Cattle, pig and chicken are the livestock which are chosen to determine household welfare in Vietnam (Minot, 2000) and in Uganda (Lang et al, 2013) As these livestock is considered as substitutable asset as they can be sold for investing and as food for eating or establishing herd or creating non-farm income (Ellis and Freeman, 2006) They are measured as quantity of livestock per person and collected from file 6A question of VHLSS in 2006 -Own annual plant land: stipulated if household owns perennial crop garden It is quoted at 6A question This indicator is used by Minot (2000) and Cuong et al (2010) to consider its impact on household welfare in Vietnam - Water surface: stipulated if household owns aquacultural farms It is collected File 6A, question This indicator is used to calculate household expenditure by Cuong et al 2011 -Proportion of household literate: the ratio between members which are literacy and household size It is collected File 2A question Household welfare is improved if the literate of adult increases, then the ability of using clean water, toilet equipment and electricity is better (Dollar et al 1998) - Household head education: calculating the number of school years household head attains It is collected from File 2A question Dollar et al 1998 has proved that the household welfare is divided into different levels by different level of household head education Labor productivity and chance for taking part in economic activities are improved through education (Kam et al 2005) - Motorbike ownership: stipulated if household has motorbike or car It is quoted from File 6A question This asset is used for the function of household expenditure (Cuong et al 2011, Lang et al, 2013) 25 - Bicycle ownership: stipulated if household has bicycle It is quoted from File 6A question This asset is rarely used for consider the household welfare However, Lang et al 2013 use as proxy to compare to motorbike benefit when he determines expected marginal return to asset - Infrastructure variables like road access, existence to market and distance to urban areas is significantly correlated with household welfare These indicators are better, it leads to be lower transaction cost, more easily access to market and increasing livelihoods choice and at end, the household welfare may increase(Okwi et al (2007) For Vietnamese researches, these variables are also used to determine the household expenditure function (Cuong 2010, 2011) + Road access : stipulated if the community has car road and is quoted from File 5, question Tải FULL (73 trang): https://bit.ly/3BWebii Dự phòng: fb.com/TaiHo123doc.net + Existence to market: stipulated if the community has market and quoted from File 5, question 24 + Distance to urban area: stipulated as distance from community to province (km) and collected File 5, question 25 - Population density: is the number of population divided by square km of land province taken from Yearbook of statistics in 2006 Population density is negative impact significantly on poverty rate Where population density is high, poverty rate at there is low This thing is explained by impacting on intensity of rural labor of production such as choosing technology, commodities, and land control for production process Moreover, people tend to move the place where they can improve their welfare (Okwi et al 2007) - Ethnic diversity: It is measured by matching randomly two people with different ethnic group and the origin of data is taken from File ttchung of VHLSS is negatively impacted on welfare as it has the strong correlation with low infrastructure, low 26 education, rent-seeking policies, and other elements related to slow economic growth (Easterly and Levine, 1997) - Climatic variables such as temperature (oC), rainfall (mm/month) and rainfall in driest month (mm/month) (Cuong et al 2010, 2011, Lang et al 2013) They are taken from website http://fsiu.mard.gov.vn/data/khituong.htm Table 3.1: Asset variables Tải FULL (73 trang): https://bit.ly/3BWebii Dự phòng: fb.com/TaiHo123doc.net Independent Variables Abbreviation Type Private targetable assets Cattle ( head) cat Continuous Pig pig Continuous Chicken (head) chi Continuous Own annual plant land (1=yes) lad Binary Water surface (1=yes) wat Binary Motorbike ownership mot Binary Bicycle ownership bic Binary Proportion of household literate lit Continuous (head) Binary Public, targetable assets Road access (yes=1) rod Binary Bank (yes=1) mar Binary hed Discrete Population density (person/sq km) pop Continuous Ethnic diversity eth Continuous Average distance to province (km) dis Continuous Temperature (oC) tem Continuous Private non targetable assets Household head education (years) Public, non-targetable assets 27 Existence to market (yes=1) mar Binary Rainfall (mm/month) ran Continuous Rainfall in the driest month (mm/month) dry Continuous Table 3.2: Control variables: Control variables Abbreviation Type Per_house Binary Semiper_house Binary temporary house Otherhouse Binary tap water Tap_water Binary Notap_water clean_water Binary other water Otherwater Binary flush toilet Flushtoilet Binary other toilet Othertoilet Binary Notoilet Binary Tele Binary have fixed phone Fixphone Binary have refrigerator Fridge Binary Mem_shool Discrete Elec Binary Otherelec Binary Permanent house Semi-permanent house no totilet have color television Number members in school electricity for light Other resources for light 3.3.2: Dependent variable: Household income and expenditure are used as proxies for monetary dimension of household welfare In developing countries, especially in Vietnam, household expenditure is usually used due to some following reasons: 28 6674636 ... +β37lan +β38lanedu +β39lanroa +β40lanban +β41lanpop +β42laneth +β43landis +β44lanmar +β45lantem +β46lanran +β47landry +β48lanmla +β49cat +β50catedu +β51catroa +β52catban +β53catpop +β54cateth +β55catdis... +β111roadis +β112roamar +β113roatem +β114roaran +β115roadry +β116ban +β117banpop +β118baneth +β119bandis +β120banmar +β121bantem +β122banran +β123bandry + β124fixphone + β125 tele + β126fridge... greatly advanced step of geographic targeting is targeting map with asset-based approach It answers the extremely important question that governors and donors should use in-kind transfer for