THEORETICAL FRAMEWORK
Literature of migration and remittances
In recent decades, remittances have emerged as a significant form of international capital flow, acting as a vital channel of financial support following migration These funds play a crucial role in fostering economic growth in developing countries and serve as an essential development tool for labor-exporting nations The relationship between remittances and migration is well-documented, with research by Yoko Niimi et al (2006) highlighting the interconnectedness of these phenomena, as migration often directly correlates with the flow of remittances (IMF).
Standard economic theory posits that remittances are closely linked to international labor migration, primarily driven by income disparities between the migrant's home and destination countries (Holst et al., 2010) Hein de Haas (2007) emphasizes that remittances and migration are often intertwined, leading to the development of various theoretical models to explain this relationship.
Neoclassical theory: Macro and Micro Economics
The Neo-classical theory is the oldest framework for understanding remittance and migration, examining both macro and microeconomic perspectives From a macroeconomic standpoint, migration occurs in response to imbalances in the labor market and wage disparities, prompting workers from labor-abundant, low-wage regions to relocate to labor-scarce, high-wage areas This migration aims to achieve equilibrium in the labor market Subsequently, capital flows, including remittances, move in the opposite direction, transferring resources from capital-rich to capital-poor countries (Massey et al., 1993; Roel Jennissen, 2004 & 2007) The equilibrium mechanism is illustrated by Oberg (1995) in Figure 2.1.
Neo-classical theory in microeconomics, as articulated by Todaro (1969), explains migration as a decision driven by individual choice, where skilled laborers anticipate higher real wages in countries with greater labor productivity Prospective migrants weigh the benefits against the costs associated with relocation, which include learning a new language, moving expenses, job searching, and adapting to a different work environment Additionally, Massey et al (1993) highlight Borjas' (1990) perspective that migrants seek to maximize their net returns by moving to countries that offer better economic opportunities.
Figure 2 1 Neo-classical mechanism leading to equilibrium
High-wage region Low-wage region
Keynesian theory, as outlined by Roel Jennissen (2007), offers an alternative perspective on migration and remittances compared to Neo-classical theory According to Keynes, labor migration occurs due to disparities in unemployment rates across regions Workers from areas with high unemployment are likely to relocate to regions with lower unemployment in search of job opportunities This movement of labor is often accompanied by capital flows, as migrants send remittances back to their home countries.
The theory posits that significant capital transfer and industrialization in developing countries can drive rapid economic growth and modernization Proponents of this perspective, known as "migration optimists," argue that migration facilitates the transfer of investment capital from the North to the South, exposing traditional communities to liberal, rational, and democratic ideas, as well as modern knowledge and education Consequently, migration policies in developing nations have increasingly encouraged emigration, as return migrants are anticipated to bring back remittances, skills, knowledge, and experiences gained abroad, ultimately contributing to their home country's economic development (Hein de Haas, 2007).
The Dual Labor Market Theory, proposed by Piore (1979) and further developed by Massey et al (1993), divides the labor market into two distinct sectors: the primary sector, which relies on capital-intensive methods, and the secondary sector, which is characterized by labor-intensive methods This division leads to a bifurcated labor force, where workers in the primary sector enjoy stable, skilled positions equipped with advanced tools, while those in the secondary sector face precarious, unskilled jobs at the bottom of the labor market, often with the risk of being laid off without employer support.
Migration, whether international or domestic, serves as a crucial strategy for individuals and their families to combat unemployment and poverty Alongside the movement of labor across borders, migrant workers often send a portion of their earnings back home in the form of remittances This creates a significant link between migration and remittance, highlighting the economic impact of migrant labor on their families and countries of origin.
According to Dual Labor Market Theory, there is a clear connection between migration and remittances within the global economic framework The labor shortage in the secondary sector of developed countries drives the movement of workers from developing nations This migration not only facilitates the flow of labor but also encourages capital transfer in the form of remittances from developed countries back to developing ones.
The New Economics of Labor Migration (NELM)
The new economics of labor migration (NELM) presents an alternative perspective to the Neo-classical theory by focusing on the household as the primary decision-making unit rather than individual migrants This approach emphasizes the risk-sharing behavior of households, which diversify income sources to mitigate income risks and maximize earnings (Stark and Levhari, 1982) According to Lucas and Stark (1985) and Hein de Haas (2007), households strategically consider migration as a response to income risks, with migrant remittances acting as a form of income insurance for families in their home countries This theoretical framework elucidates the motivations for migration even in the absence of significant income disparities Additionally, migration serves as a crucial source of investment capital, particularly in developing countries where credit and insurance markets are often flawed.
The theory highlights the social role of remittances in the lives of migrants and their families, emphasizing that these financial transfers serve as resources exchanged within social networks A social network is characterized by ongoing associations among individuals connected through occupational, familial, cultural, or emotional ties When migrants send remittances, these resources are effectively integrated into their social networks, reinforcing relationships and support systems.
(i) Based on traditional or sentimental value, the migrant thought that they accumulate social obligation from the people to whom they remit for example childcare, sending goods
(ii) The migrant remitting, maybe, conforms to moral values learn as being a member of the group
(iii) Remittances increase their social visibility in the sending and receiving countries, in addition to avoiding the sanction by the social group if they do not remit.
Theory of Remittances
Family affection creates intangible ties among members, influencing migrants' motives for remitting, primarily driven by altruism (Lucas and Stark, 1985 and 1988) Migrants feel obligated to send remittances due to their care for their families' well-being and the support received during their education Rapoport and Docquier (2005) also emphasize that family sentiment motivates migrants to remit As migrants' income increases, so do remittances, and vice versa Glytsos (2001) notes that the purpose of remittances varies with the duration of a migrant's stay abroad; permanent residents tend to remit for altruistic reasons, while temporary migrants often remit for investment and future consumption Rapoport and Docquier further support the connection between altruism and remittances, building on Funkhouser's (1995) model of remittance behavior.
“(i) Emigrants with higher earnings potential remit more;
(ii) Low-income household receiving more;
(iii) Remittances should increase with both the degree of proximity between the migrant, and the remaining household members and the migrant’s intentions to return; and
(iv) Remittances from a given migrant should decrease with the number of other emigrants from the same household;
(v) The time profile of remittances should depend on the comparison between the migrants’ time-discount factor and their earnings profile abroad”
Impure altruism, characterized by pure self-interest, contrasts with pure altruism Lucas and Stark (1985) identified three key reasons that explain the motivations behind migrants sending remittances to their families without altruistic intent.
Firstly, a migrant would send money home to increase their visibility hence eligible for inheritance, esteem or other resources in the community of origin
Secondly, migrants send remittances in order to reimburse the household for past expenditures such as schooling or the cost directly related to migration
In the context of self-interest, migrants often plan to return to their home countries, which encourages them to send remittances These funds are typically used to buy durable goods and invest in housing, land, livestock, or businesses back home.
Tempered Altruism or Enlightened Self-Interest
Lucas and Stark's (1985) theory of "tempered altruism," also known as "enlightened self-interest," was further developed by El Mouhoub Mouhoud et al (2008) and Jessica Hagen-Zanker et al (2007) This theory identifies three key motivations for remittances within family arrangements: exchange, insurance, and investment.
- Exchange motive derived from migrants’ expectation for members in their family to receive the better quality of welfare services (health, education, and so on) from their remittances
Insurance motives function as intra-family strategies that enable migrants and their families to mitigate income volatility in rural areas Rapoport and Docquier (2005) highlight that families often send some members abroad or to urban areas to secure their income against potential poor harvests, with all associated costs borne by the families This arrangement incentivizes migrants to send remittances back home to support their families.
Migrants often choose to transfer their savings back to their home countries when they perceive higher potential returns compared to their host countries This remittance decision is primarily influenced by the disparity in interest rates between the two nations, motivating them to invest where they can maximize their financial gains.
The relationship between migration and remittances is characterized by three key motives: pure altruism, pure self-interest, and tempered altruism Migrants often seek to send money back home to enhance the living conditions of their family members This dynamic is encapsulated in the research of Catalina Amuedo-Dorantes and Susan Pozo, which outlines the various motivations behind remittances.
Source: Catalina Amuedo-Dorantes and Susan Pozo (2006)
2 2 Literature review: Remittances, Income, Savings, Asset, Insurance and
Remittances sent by migrants to their families in their home countries are influenced by various theories that explain their significance At the microeconomic level, the impact and determinants of these remittances in recipient countries hinge on how households utilize them, whether for consumption, health, education, or other purposes This raises important questions about the effective use of remittances among receiving households The ongoing debates surrounding the effects of remittances highlight their complex role in shaping the economic well-being of recipient families, as noted by researchers such as Richard H Adams, Adriana Castaldo, and Barry Reilly (2008).
Remittances are recognized for their significant impact on household income and poverty reduction, as they serve as a vital source of financial support However, there is a contrasting perspective that suggests remittances may alter the behavior of recipient households, potentially diminishing the positive effects of income from other sources This duality of remittances highlights the complex dynamics of financial inflows and their varying influences on economic well-being.
Family-provided insurance (response of risks in host country)
Altruism (nope response of risks in host country)
Self-insurance (response of risks in host country)
(2003), Yéro Baldé (2010) gives judgments including (i) remittance is major spent for
Remittances play a significant role in household expenditure, with a notable portion allocated to various categories Research by Khawaja Mamun and Hiranya K Nath (2010) indicates that savings account for 3-7% of remittances, while loan repayments constitute 10-19% Additionally, health care expenses represent 0-4%, and food and clothing take up 20-36% of remittance spending Furthermore, the relationship between remittances and insurance highlights the impact of income volatility and wage risks, suggesting that migrants and their families may invest in insurance to mitigate these uncertainties (Catia Batista & Janis Umblijs, 2014; Dean Yang & HwaJung Choi, 2005).
Beside observations above, empirical studies as follows show the impacts of remittances on income, assets, borrowing, insurance, and savings in household receiving remittances
Relationship between Remittance and Income
Research indicates a significant relationship between remittances and household income in developing countries, including Egypt, Small Island Developing States (SIDS), Vietnam, and various Asian and Pacific nations.
Richard H Adams Jr and John Page (2005) conducted a study examining the relationship between remittances and poverty alleviation in 71 developing countries classified as low-income and middle-income.
The study analyzes data from 18 Sub-Saharan African countries and 53 developing nations, encompassing 184 observations related to income, poverty, and inequality Utilizing both instrumented and non-instrumented OLS estimates based on the growth-poverty model by Ravallion (1997) and Ravallion & Chen (1997), the findings indicate that an increase in per capita official remittances significantly reduces the number of individuals living on less than $1.00 per day Specifically, the instrumented OLS estimates reveal that a 10 percent rise in per capita remittances correlates with a 3.5 percent decline in extreme poverty, while non-instrumented OLS estimates show a 10 percent increase in remittances results in a 1.8 percent reduction in the same poverty measure.
Juthathip Jongwanich's 2007 working paper provides empirical evidence on the effects of workers' remittances on economic growth and poverty reduction in 17 Asian and Pacific countries Utilizing the Generalized Method of Moments (GMM) for analysis, the study examined panel data from 1993 to 2003, sourced from the World Bank and IMF’s Balance of Payments Yearbook The findings indicate that remittances significantly enhance the income of households that receive them.
Papers of Cuong Nguyen Viet (2008) and Wade Donald Pfau & Long Thanh Giang
A study conducted in 2010 utilized data from the Viet Nam Household Living Standard Surveys (VHLSS) spanning from 1992/93 to 2006 Employing impact evaluation methods and descriptive statistics, the research examined the correlation between international remittances and per capita income in Vietnam The findings revealed that international remittances play a significant role in enhancing the income of recipient households.
RESEARCH METHODOLOGY
Assumptions of PSM method
According to the foundational work of Rosebaum and Rubin (1983) and subsequent research by Marco Caliendo and Sabine Kopeinig (2005), Carolyn Heinrich et al (2010), and Shahidur R Khandker et al (2010), several key assumptions are necessary for accurately identifying the intervention effects between treated and control groups.
Assumption 1 – Conditional Independence Assumption (CIA)
There is a set of observed covariates X not affected by treatment, the potential outcomes Y are independent of treatment status T:
The assumption of unconfoundedness, also referred to as selection on observable characteristics, is crucial for accurately assessing the impact of an intervention The Conditional Independence Assumption (CIA) plays a key role in minimizing selection bias by ensuring that the differences between the treated group and the control group are reduced This is why a control group is essential for constructing a counterfactual for the treatment group, representing the unobserved outcomes of the treated subjects.
Assumption 2 – Common Support or Overlap Condition [Figure 3.1]
The expression indicates that the ratio of treated to untreated groups must exceed zero for the potential values of X The overlap condition, or common support, is essential to ensure that there is adequate similarity in the characteristics of both the treated and control groups for effective matching The Average Treatment Effect on the Treated (ATT) or Average Treatment Effect (ATE) is defined within the region of common support, and treatment assignment is considered strongly ignorable when the aforementioned assumptions are met.
Figure 3 1 Example of Common Support
Source: Shahidur R Khandker et al (2010)
Model Specification of Propensity Score Matching
Propensity score matching involves multiple steps to assess the effects of remittances on household income, savings, insurance, borrowing, and assets for both remittance-receiving and non-receiving households in the years 2006, 2008, and 2010 This process includes determining the region of common support to ensure accurate evaluations.
Step 1: Choosing Model and Variables for estimation of Propensity Score
According to Marco Caliendo and Sabine Kopeinig (2005), logit and probit models are favored over linear probability models in propensity score matching (PSM) regression due to their ability to maintain predicted probabilities within the [0,1] range While both models serve similar purposes, the logit model is often preferred because it is based on stronger assumptions compared to the probit model, which can be more complex and cumbersome in calculations.
The next step in the process is variable selection, which is crucial as omitting major variables can lead to increased bias in the resulting estimates To ensure accuracy, the selected variables must adhere to the Conditional Independence Assumption, meaning they should be independent of the treatment condition based on the propensity score and remain consistent over time For effective matching, these variables must also be independent of remittances and originate from the same source, thereby enhancing the precision of the propensity score Key variables included in the regression model are: household membership, gender and age of the household head, living area, ethnicity, proficiency in Vietnamese, marital status, internet access, total land area for cultivation or rent, toilet type, cooking energy source, housing type, and 11 dummy variables representing 12 provinces.
Command of PSM determining the propensity score and satisfy the balancing property in Stata platform is “pscore” command
The weighting for propensity scores is derived from the VARHS dataset, utilizing demographic information from the Statistical Yearbook of Vietnam for the years 2006, 2008, and 2010 This analysis compares the population data across twelve provinces with households surveyed in the same regions after integrating data from VARHS 2006, VARHS 2008, and VARHS 2010, as illustrated in Tables 3.2 and 3.3.
Step 2: Determining Region of Common Support and Balancing Tests
In accordance with Shahidur R Khandker et al (2010), the common support region has to be specified in the propensity score distribution for treatment and control group overlap
The balancing test assesses whether the average propensity score and the mean of variable X are equivalent between the treatment and control groups Following the application of matching balancing tests, the disparities in covariate means between the two groups significantly decreased.
Step 3: Matching Treated Group and Control Group
Research by Marco Caliendo et al (2005), Shahidur R.K et al (2010), and Carolyn H et al (2010) highlights various techniques for matching treated individuals with control individuals This study implements several of these techniques to ensure accurate comparisons and outcomes.
Nearest-neighbor matching (NNM) is a technique used to pair each treatment unit with a comparison unit based on the closest propensity score or similar observed characteristics There are two variants of NNM: "with replacement" and "without replacement." In the "with replacement" scenario, control group individuals can be reused as matches, which enhances the average quality of matching and reduces bias Conversely, "without replacement" limits each control unit to a single match, making the outcome dependent on the order of matching, necessitating randomization to ensure validity.
Table 3.1 Variables of Propensity score Matching
Variable Description Coding dmremit It is indication of household with remittance or without remittance
1= receiving-remittance household 0= receiving-no remittance household
Hhmem Total of members living in household shares lodging, income and expenditure for at least 6 in last 12 months Hhmem=1, 2, 3, …, 13
HeadGen Gender of Household Head 0 = Female, 1 = Male
Age of household Head In VARHS, age of Household head recoreded by year of birth, so we have to convert into age of year olds
Age in years Headstatus Marital status of household head 0 = Other, 1 = Married
Total area of a house in which total members of household lived including bedrooms, dining rooms, living rooms, study rooms
The article analyzes household characteristics across twelve provinces, focusing on ethnicity, language, and internet access It categorizes ethnicity as either "Other" or "Kinh," and examines whether Vietnamese is spoken by all household members Additionally, it assesses internet accessibility, indicating whether any household member can access internet services.
Wall The main construction material of the outside walls 0 = Other, 1 = Brick
Cooking The main source of energy for cooking in household 0 = Other, 1 = Electricity
Toilet With or without toilet in household 0 = No, 1 = Yes
Watersource The main source of cooking/drinking water of household 0 = Other, 1 = Tap water
Density of observations represented in comparison with population in each province The observations kept after merging VARHS 2006, VARHS 2008, and VARHS 2010
Tinh_1 Represent of Ha Tay (Ha Noi) province, dummy variable 0 = Other
Tinh_2 Represent of Lao Cai province, dummy variable 0 = Other
Tinh_3 Represent of Phu Tho province, dummy variable 0 = Other
Tinh_4 Represent of Lai Chau province, dummy variable 0 = Other
Tinh_5 Represent of Dien Bien province, dummy variable 0 = Other
Tinh_6 Represent of Nghe An province, dummy variable 0 = Other
Tinh_7 Represent of Quang Nam province, dummy variable 0 = Other
Tinh_8 Represent of Khanh Hoa province, dummy variable 0 = Other
Tinh_9 Represent of Dak Lak province, dummy variable 0 = Other
Tinh_10 Represent of Dak Nong province, dummy variable 0 = Other
1 = Dak Nong province Tinh_11 Represent of Lam Dong province, dummy variable 0 = Other
Table 3.2 Population in each province
No Province Households in each year
Table 3.3 Weight of Population in VARHS 2006, VARHS 2008 and VARHS 2010
Caliper or radius matching (RM) addresses the limitations of nearest neighbor (NN) matching, particularly the risk of poor matches due to significant differences in propensity scores between treated and control individuals By implementing a caliper, or a tolerance level for the maximum allowable distance in propensity scores, this technique helps ensure that matches are more closely aligned While this method is akin to matching with replacement within a defined propensity range, it may result in fewer available matches, which can subsequently increase the variance of the estimates.
Stratification and Interval Matching (SM) is a technique that divides the common support of the propensity score into distinct intervals, or strata, to assess the program's impact within each segment The program effect is determined by calculating the mean difference in outcomes between treated and control observations within these strata To obtain the overall program impact, a weighted average of the impact estimates from each interval is computed, using the proportion of treated individuals in each stratum as weights.
Kernel and Local Linear Matching (KM):
Kernel matching (KM) and local linear matching (LLM) are non-parametric matching estimators that utilize weighted averages from the entire control group to create counterfactual outcomes for treated individuals, offering the advantage of reduced variance by leveraging more data However, a significant drawback is the potential inclusion of poor matches, making the correct application of the common support condition crucial for the effectiveness of KM and LLM.
To ensure the consistency of findings, various matching methods can be utilized, as previously discussed Additionally, Shahidur R Khandker et al (2010) recommend employing direct nearest-neighbor matching, which can be executed in Stata using the command 'nnmatch'.
Software
To implement matching and estimate treatment affects, we use “pscore” program of Becker and Ichino (2002) for PSM estimators with Stata platform
3 2 Difference in Difference (DD) Method
The Difference-in-Differences (DD) method was utilized to assess the impact of remittances on households receiving remittances (treated group) compared to those not receiving them (control group) during the periods of 2006-2008 and 2008-2010.
Theory
The Difference-in-Differences (DD) method, as outlined by Shahidur R Khandker et al (2010), is employed when the parallel-trend assumption holds true This approach involves comparing a treatment group, which consists of participants, with a control group of nonparticipants, assessing both pre- and post-intervention outcomes The DD estimate reveals the changes in outcomes for the treatment group relative to the control group over time, starting from a pre-intervention baseline Ultimately, the DD method provides an estimate of the average impact of the intervention.
DD = E(Yt T – Yt-k T | T1 = 1) - E(Yt C – Yt-k C | T1 = 0) (1) Where:
+ t and (t-k) denote the time receiving remittance at t and (t-k) times;
+ Yt T and Yt C will be the outcomes of treatment and control groups at time t;
+ (Y1 C - Y0 C) will be the outcome changes of control group;
+ (Y1 T - Y0 T) will be the outcome changes of treatment group;
+ T1 = 1 denotes remittance-receiving households, and
+ T1 = 0 means no remittance-receiving households
In particular, estimative equation of DD presented by the following regression framework:
Yit= + βTi1 t + ρ Ti1 + γt + εit (2) Where:
+ Coefficient β: affect mutually between treatment variable Ti1 and time t with t 1…T, it is the average DD effect of remittance, and β = DD in (1)
+ Variables Ti1 and time t for any effect of time selected as well as the effect of remittance at (t-k) time in comparison with t time
Equation (1) is equivalent with equation (2) as follows:
E(Y1 C – Y0 C | T1 = 0) = ( + γ) – (4) Subtraction of equation (3) and equation (4) equals DD.
Implementing DID
The DID approach utilizes panel data to analyze differences over time between participant and non-participant groups By examining this data at two distinct points, the study relies on the parallel-trend assumption, which posits that the gap between the two trends remains consistent throughout the observed period.
Figure 3.2 illustrates the true counterfactual outcomes as the lowest line, highlighting the distinction between the observed control outcomes and the actual counterfactual outcomes.
Figure 3 2 An example of Difference in Difference
Source: Shahidur R Khandker et al (2010)
Model of this study
This study operates under the parallel-trend assumption, which posits that the natural geographic and socio-economic conditions across provinces are uniform, leading to similar income trends for households over time It examines the impact of remittances on two groups: households receiving remittances (the treated group) and those not receiving remittances (the control group), to determine whether remittances have influenced the income levels of these households.
DD estimate of equation (1) regarding to impact of remittance on households receiving remittance and receiving no remittance is:
DD = E(Yt T – Yt-k T | T1 = 1) - E(Yt C – Yt-k C | T1 = 0) (5) Where:
+ (Yt T – Yt-1 T | T = 1): outcome changes of receiving remittances households in year t Receiving remittances households is called Treatment group
+ (Yt C – Yt-1 C | T = 0): outcome changes of receiving remittances households in year (t- k) Households receiving no remittance is called Control group
+ (t-k) : the year before year t, and k is positive integer and (k < t)
Based on equation (2), regression model of this study as follows - [Table 3.4]
Yit = 0 + 1Year + 2Dmremiti+ 3 (Year*Dmremiti) + i X i + i (6) Where
+ Yit: outcomes changing between treatment group and control group in year t and (t-k), it will be logarit form
+ Xi : characteristic variables for extension of this module (age of head household, gender of head household, …)
A study by Khawaja A Mamun et al (2011) indicates a significant relationship between borrowing and remittances in rural Bangladesh The research conducted in Matlab reveals that households receiving remittances experience a reduction in their borrowing needs, highlighting the positive impact of remittances on financial stability.
Research by Catalina A.D and Susan Pozo (2002) highlights a significant connection between remittances and insurance in Mexico Their study reveals that households receiving remittances utilize two primary forms of insurance: family-provided insurance, which guarantees a secure place within the family, and self-insurance, achieved through the accumulation of precautionary savings.
Furthermore, Una Okonkwo Osili (2005) found out the connection of remittance and saving through remittance help reducing poverty and providing saving for receiving remittance household in country exporting labor
Based on relationship of remittance and saving, borrowing and insurance, the other outcomes extended in this study are saving, loan, insurance, the regression models were presented as follows
Ln(Income) = 0 + 1Year + 2Dmremiti+ 3 (Year*Dmremiti) + i Xi + i [Model 1]
Ln(Saving) = 0 + 1Year + 2Dmremiti+ 3 (Year*Dmremiti) + i Xi + i [Model 2]
Ln(Asset) = 0 + 1Year + 2 Dmremiti + 3 (Year* Dmremiti) + i X i + i [Model 3]
Ln(Insurance) = 0 + 1Year + 2 Dmremiti + 3 (Year* Dmremiti) + i Xi + i [Model 4]
Ln(Borrow) = 0 + 1Year + 2 Dmremiti + 3 (Year* Dmremiti) + i X i + i [Model 5]
+ The difference of outcome changes for control group in 2006(2008) and 2008(2010) is (1)
+ The difference of outcome changes for treatment group in 2006(2008) and 2008(2010) is (1 + 3)
+ The difference of outcome changes between control group and treatment group is (3)
We have summarized in comparison control group with treatment group in [Table 3.5]
Table 3.4 Variables of Difference in Difference Method
Variable Description Coding dmremit It is indication of household with remittance or without remittance
1= receiving-remittance household 0= receiving-no remittance household
Year Represent of dataset investigated in 2006, 2008 and 2010 0 = 2006 (2008)
1 = 2008 (2010) Remittance defined in VARHS is money or goods the households received from persons who are relatives, friends or neighbours [Annex 1]
VND lnIncome Total income of households calculated by VND %
LnSav Total saving of households calculated by VND %
Total value of asset of households converted to VND, based on instruction of investigator and estimation of households
LnInsurance Total value of insurance bought by households, calculated
LnBorrowing Total value of loans of households, calculated by VND %
Hhmem Total of members living in household shares lodging, income and expenditure for at least 6 in last 12 months Hhmem=1, 2, 3, …, 13
HeadGen Gender of Household Head 0 = Female, 1 = Male
Age of household Head In VARHS, age of Household head recoreded by year of birth, so we have to convert into age of year olds
Age in years Headstatus Marital status of household head 0 = Other, 1 = Married
Total area of a house in which total members of household lived including bedrooms, dining rooms, living rooms, study rooms
The article analyzes the ethnic composition and language use in households across twelve provinces, focusing on the square meters of living space It categorizes ethnicity into two groups: 0 for Other and 1 for Kinh Additionally, it examines the prevalence of the Vietnamese language, indicating that a value of 0 means no members speak it, while 1 signifies that all members do Lastly, it assesses internet accessibility, where 0 indicates no access and 1 confirms that at least one member of the household has internet services.
Wall The main construction material of the outside walls 0 = Other, 1 = Brick
Cooking The main source of energy for cooking in household 0 = Other, 1 = Electricity
Toilet With or without toilet in household 0 = No, 1 = Yes
Watersource The main source of cooking/drinking water of household 0 = Other, 1 = Tap water
Plotarea Total area of households’ possession Square meters
Table 3.5 DID estimation between treatment group and control group
C OMBINING PSM WITH DID METHODS
Shahidur R.K et al (2010) demonstrated the integration of the Difference-in-Differences (DID) method with Propensity Score Matching (PSM) to create a well-matched control group This process involves two key steps: first, utilizing baseline data for PSM to align the control group with the treatment group effectively The analysis employs data from the Vietnam Access to Resources Household Survey (VARHS) for the years 2006 and 2008.
In 2010, variables such as year and dmremit08 were generated to facilitate analysis The year variable was assigned a value of one for the year 2008 and zero for 2006, enabling the establishment of a common support region and conducting a balancing test using the 'pscore' command for both years The dmremit08 variable was created to maintain consistent values across these years, allowing for effective matching of treated and control pairs Once the balancing property of propensity score matching was confirmed, it indicated that households with identical propensity scores shared the same distribution of covariates across all blocks Subsequently, matched households from 2006 were merged with the 2006-2008 dataset, retaining only the matched households for analysis The Difference-in-Differences (DID) method was then applied to the matched sample, utilizing a sequence of regressions to address observable heterogeneity in the panel data This combined approach yielded superior results compared to using the Propensity Score Matching (PSM) or DID methods in isolation, and a similar methodology was applied to the 2008-2010 dataset.
Utilizing the Propensity Score Matching (PSM) method, we calculate the propensity score to ensure the balancing property between treatment and matched control groups sharing identical characteristics This matched sample enhances the outcomes of the Difference-in-Differences (DD) method Additionally, we employed fixed-effects regression and ordinary least squares (OLS) to effectively measure the double differences.
D ATA
This study utilizes panel data from the Vietnam Access to Resources Household Survey (VARHS) conducted in 2006, 2008, and 2010 to analyze the effects of remittances on households that receive them compared to those that do not The research focuses on 12 provinces in rural Vietnam, aiming to understand the socio-economic implications of remittance income on these households.
The study employs the Propensity Score Matching (PSM) approach and the Difference in Difference (DD) method to analyze the impact of remittances By utilizing the PSM method, we identify matching characteristics between households that receive remittances and those that do not for the year 2006.
This study analyzes the impact of remittances on income, savings, borrowing, and assets between 2006 and 2010 Utilizing a double differences method, we examined the changes in outcomes for households receiving remittances compared to those not receiving them during the periods of 2006-2008 and 2008-2010 By integrating Propensity Score Matching (PSM) and Difference-in-Differences (DID) estimations, we assessed the effects of these two methodologies on income, borrowing, savings, and assets across the specified timeframes.
General information of VARHS Data
The VARHS data sets, conducted by the Institute of Labor and Social Sciences, a subsidiary of the Ministry of Labor, Invalids and Social Affairs, utilize PSM and DD methods to derive participant and non-participant data from the same source For this analysis, we selected the VARHS data sets from 2006, 2008, and 2010, which were implemented across 12 rural provinces in Vietnam, including Ha Tay (Ha Noi), Lao Cai, Phu Tho, Lai Chau, Dien Bien, Nghe An, Quang Nam, Khanh Hoa, Dak Lak, Dak Nong, Lam Dong, and Long An, with each province coded separately.
The provinces of Ha Tay, Lao Cai, Phu Tho, Lai Chau, Dien Bien, Nghe An, Quang Nam, Khanh Hoa, Dak Lak, Dak Nong, Lam Dong, and Long An are represented by the codes 105, 205, 217, 301, 302, 403, 503, 511, 605, 606, 607, and 801 These codes are utilized for calculating weighting in Propensity Score Matching (PSM) and Difference-in-Differences (DID) analyses.
According to the Ministry of Labor, Invalids and Social Affairs, the average annual outflow of Vietnamese migrant workers from 2003 to 2008 was 77,500 In this study, the twelve provinces accounted for 16,361 migrant workers, while the remaining 51 provinces contributed 61,139 workers This results in an average of 1,363 workers per province in the twelve provinces, compared to 1,199 workers per province in the others Thus, these twelve provinces serve as a representative sample for analyzing migrant labor trends across Vietnam.
The 2008 survey was designed to build upon the findings from the 2006 survey, expanding its reach to additional households Investigators were tasked with updating information for households that had relocated since the 2006 survey and continued this practice for the 2010 survey As a result, the data collected in VARHS 2006, VARHS 2008, and VARHS 2010 is largely consistent and comparable The investigation was conducted systematically to ensure accurate and comprehensive data collection.
The interviewer will be given a list of households to be interviewed All the households have previously been interviewed by GSO and/or ILSSA in previous year
If a household has moved or split up, or the household head has died, the interviewer should follow these guidelines:
In cases of divorce between the household head and their spouse, it is essential to conduct interviews with individuals currently residing at the previous household location where the last interview took place.
- If the household head has died, do the interview with those who live at the location where the household was living when they were previously interviewed
In cases where a household has divided, such as when a son marries and establishes a new household, interviews should be conducted with the individuals residing in the original family home.
- If the household has moved within the commune, they should be found and interviewed
When all members of a household have relocated outside the commune, it is essential to administer the "absent households" questionnaire This survey can be conducted with former neighbors or individuals familiar with the household's circumstances It is permissible to seek assistance from multiple sources to ensure accurate completion of the absent household questionnaire.
The analysis utilized data from the Vietnam Access to Resources Household Survey (VARHS), comprising 2,324 households in 2006, 3,269 households in 2008, and 3,208 households in 2010 After merging these datasets, the study focused on 1,059 households that were consistently observed across all three years, ensuring a robust longitudinal analysis.
Information of VARHS data sets collected by questionnaire including ten sections as follows
Section 1 Cover page, Household roster, general characteristics and identification of household members
Section 2 Land use, General information about plots, Plots owned and operated, land rented in or borrowed, land law Section 3 Agricultural land and crop agriculture
Section 4 Livestock, forestry, aquaculture, agricultural services and access to markets Section 5 Occupation, time use and other sources of income
Section 6 Training and supports in agricultural production
Section 7 Food expenditures, other expenses, saving and household durable goods
Section 9 Shocks and risk coping
Section 10 Social capital and networks
This study examines the effects of remittances on households that receive them, focusing on key variables such as remittances, assets, borrowing, saving, insurance, total income, household members, age and gender of the household head, and total plot area Data for this research was gathered from multiple sections, specifically Section 1, Section 2, Section 5, Section 7, Section 8, Section 9, and Section 10, as outlined in Appendix B.
EMPIRICAL ANALYSIS
4 1 Descriptive analysis of the sample
In the earlier mention, data sets of VARHS applied in this study will be VARHS
The VARHS data collected in 2006, 2008, and 2010 included 2,324, 3,269, and 3,208 households, respectively After merging and selecting variables for the Propensity Score Matching (PSM) and Difference-in-Differences (DID) models, a total of 3,177 households were retained for analysis, resulting in an average of 1,059 households per year.
The VARHS data reveals significant changes in household remittance status over the years In 2006, there were 745 households receiving remittances compared to 314 that did not By 2008, the number of remittance-receiving households decreased to 351, while those not receiving remittances increased to 708, totaling 1,059 households In 2010, the trend shifted again, with 624 households receiving remittances and 435 without, maintaining the total of 1,059 households Table 4.1 provides a summary of these findings, highlighting the dynamics of remittance flows among households over the specified years.
Table 4 1 Number of HHs receiving remittance and receiving no remittance
No VARHS 2006 VARHS 2008 VARHS 2010 Total of HHs
(**) No: No remittances-receiving household
Number of households with remittance reduced from 745 in VARHS 2006 to 351 in VARHS 2008, and increased to 624 of households in VARHS 2010, it was accounted for 70.34% in 2006, 33.14% in 2008, and 58.9% in 2010 respectively
In 2006, average remittance received by each household is 3.019 million VND, and
As of 2008, half of the households receiving remittances in Vietnam reported amounts under 1.5 million VND, with an average remittance of 7.5 million VND per household By 2010, the average remittance increased slightly to 7.9 million VND, while the threshold for the lower 50 percent of remittance-receiving households rose to less than 2 million VND Notably, 50 percent of households receiving remittances still received less than 500,000 VND, highlighting the continued disparity in remittance distribution.
Therefore, these shown that amount of remittance in 2008 increased in comparison with amount of remittance in 2006 and 2010
Besides average remittance received in each households in years of 2006, 2008 and
In 2010, data summarized in Table 4.2 indicates that households receiving remittances had higher income, savings, and assets compared to those not receiving remittances, with differences ranging from 24 to 25 percent However, in 2008, the gap narrowed significantly, with households receiving remittances showing only 4 to 11 percent higher income, savings, and assets than their non-remittance counterparts.
Table 4 2 Summary of Households with and without remittance
Year of 2006 Year of 2008 Year of 2010
Households without remittance show varying financial metrics, with total incomes ranging from 29,000,000 VND to 109,000,000 VND Savings average between 12,000,000 VND and 34,800,000 VND, while insurance contributions fluctuate from 305,000 VND to 2,400,000 VND Debt levels are significant, with amounts owed varying from 8,050,000 VND to 19,500,000 VND Additionally, asset values range from 15,100,000 VND to 37,800,000 VND, highlighting the financial diversity among these households.
Remittances 3,019,000 VND 0 VND 7,500,000 VND 0 VND 7,900,000 VND 0 VND
The share of remittances in total household income varied between 8.38% and 13.63% during the years 2006, 2008, and 2010 This fluctuation is illustrated in [Table 4.3], which highlights the anticipated trends in the PSM method for these three years.
For DID method, the expected signs of variables are in [Table 4.4], when applying regression models (from [Model 1] to [Model 5]) for year pairs of 2006-2008 and 2008-2010
Table 4.3 Expected sign in PSM Model
Variables Description Unit Expected Sign
This is dummy variable, it indicates Households (HHs) receiving remittance or not receiving remittance dmremit = 1 or dmremit = 0 Lnincome
Amount of Income in each HH was investigated continuously in three years of 2006, 2008 and
Amount of Saving in each HH was investigated continuously in three years of 2006, 2008 and
Amount of asset in each HH was investigated continuously in three years of 2006, 2008 and
Amount of Insurance in each HH was investigated countinuously in three years of 2006,
Amount of Borrowing in each HH was investigated continuously in three years of 2006,
(*) Expected sign (+) for Lnborrowing variable explain that receiving-remittance households received the support loans of banking for expanding household business; or
Expected sign (-) for Lnborrowing variable explain that received remittance paid for owes of households
Table 4.4 Expected sign of variables in DD model
Variable Description Symbol Expected Signs
Year This is dummy variable, it indicates year of
Year =0 if nam 06 Year =1 if nam 08 (+) dmremit
This is dummy variable, it indicates HHs receiving remittances or receiving no remittances dmremit = 1 dmremit = 0 (+)
Year*dremit Difference of remittance was received between receiving-remittance households and receiving- no remittance households
Year This is dummy variable, it indicates year of
Year =0 if nam 08 Year =1 if nam 10 (+) dmremit
This is dummy variable, it indicates HHs receiving remittances or receiving no remittances dmremit=0 dmremit=1 (+)
Year*dmremit Difference of remittance was received between receiving-remittance households and receiving- no remittance households
4 2.1 Propensity score matching (PSM) model
The algorithm developed by Sascha O Becker and Andrea Ichino (2002) employs Propensity Score Matching (PSM) to mitigate bias in treatment effect estimation by controlling for confounding factors It operates on the principle that bias is minimized when comparing outcomes between treated and control subjects The output from the "pscore" command indicates that the average propensity scores for treated and control groups differ across all blocks until they converge, which in this thesis occurs across five blocks Subsequently, the "pscore" command tests the balancing property for each covariate, identifying a common support region of [.066641, 99769314] that ensures the mean propensity scores of treated and control groups are comparable within each block The common support condition, enforced by the "comsup" option, results in the omission of block identifiers for five control observations outside this common support, reducing the total number of observations from 1,059 to 1,054 (see Appendix C).
This study analyzes the impact of remittances on wealth accumulation in households by comparing those receiving remittances to those without in the years 2006, 2008, and 2010 To estimate the effects, various matching techniques are employed, as detailed in Chapter III, with results summarized in Table 4.5 Additionally, Appendix B provides further insights into the Propensity Score Matching (PSM) method used in the analysis.
In the 2006 income regression analysis, the null hypothesis was rejected, as evidenced by the t-statistic values across various matching techniques: 2.005 for Nearest-neighbor matching, 2.036 for Stratification matching, 2.056 for Radius matching, and 2.210 for Kernel matching, all exceeding the critical value of 1.645 This indicates a significant impact of remittances on household income, with total income increasing by 14-26%, significant at the ten percent level Furthermore, when direct Nearest-neighbor matching was applied, the results were consistent, showing an Average Treatment Effect on the Treated (ATT) of 23.8% with a p-value significant at the five percent level.
In the 2006 insurance model, the null hypothesis was rejected, indicating significant findings The matching techniques revealed t-statistic values for the Average Treatment Effect on the Treated (ATT) that exceeded the critical t-value of the student distribution, with values of 1.692, 2.505, 1.709, and 2.228, all surpassing the threshold of 1.645 Furthermore, these results were validated through robustness checks at a 5% significance level, demonstrating that remittances significantly influenced the insurance purchasing behavior of households receiving them.
In the 2006 Savings model, the null hypothesis is accepted, indicating no significant relationship between the dependent and independent variables This is evidenced by the t-statistic values from various matching techniques: 0.634 for Nearest-neighbor matching, 1.000 for Stratification matching, 0.949 for Radius matching, and 0.865 for Kernel matching, all of which fall below the t-critical value of the student distribution Therefore, there is no discernible impact of remittances on the savings of households receiving them, and these results remain consistent upon robustness checks.
In the regression analysis conducted for the year 2006, the null hypothesis was accepted, indicating that remittances do not significantly impact the assets, insurance, and borrowing behaviors of households receiving remittances The findings revealed that the t-statistic values for the Average Treatment Effect (ATT) did not meet the t-critical value of the Student's t-distribution at a ten percent significance level, and this outcome remained consistent upon testing for robustness.
The these findings were verified by omitting dummy variables such as Tinh_1, Tinh_2, Tinh_3, Tinh_4, Tinh_5, Tinh_6, Tinh_7, Tinh_8, Tinh_9, Tinh_10, Tinh_11
The identified region of common support is [.09676315, 90167309], with seven control observations lacking block identifiers outside this range, resulting in a total of 1,052 observations instead of 1,059 Ultimately, there are six blocks in the analysis The T-statistic of the Average Treatment Effect on the Treated (ATT) values, estimated using various matching methods including Nearest-neighbor, Stratification, Radius, and Kernel matching, along with robustness checks, indicates that remittances do not have a significant effect on the accumulated wealth of households receiving remittances.
The PSM method was applied to the VARHS 2008 and VARHS 2010 datasets to estimate the effect of remittances on the behavior of receiving-remittance households, following the sequence used for the VARHS 2006 dataset The analysis revealed that the t-statistic of ATT values and robustness checks indicated no significant impact of remittances on income, insurance, savings, assets, or borrowing, consistent with the findings from 2006.
Table 4.5 Summary of Average Treatment Effect of on Treated Group
Income Saving Asset Insurance Borrowing
The PSM method reveals no significant relationship between remittances and household wealth accumulation, including savings, borrowing, insurance, and assets This lack of connection suggests that remittances are primarily utilized for consumption or educational purposes rather than wealth building Consequently, the behavior of households receiving remittances is influenced by these financial inflows.
Table 4.6 Summary the impacts of remittance by PSM method
Income Saving Asset Insurance Borrowing
Year 2006 Positive impact No impact No impact Positive impact
Year 2008 No impact No impact No impact No impact No impact
Year 2010 No impact No impact No impact No impact No impact