Determinants on households’ partial credit rationing - An analysis from VARHS 2008

Due to it implication to the economic development in general, and to the effectiveness of Government credit for the poor program in particular, this paper will aim at examining the facto

UNIVERSITY OF ECONOMICS INSTITUTE OF SOCIAL STUDIES

VIETNAM-NETHERLANDS PROGRAMME FOR M.A IN DEVELOPMENT ECONOMICS

DETERMINANTS ON HOUSEHOLDS’ PARTIAL CREDIT RATIONING

AN ANALYSIS FROM VARHS 2008

By

NGUYEN VAN HOANG

Academic Supervisor:

Dr TRAN TIEN KHAI

HO CHI MINH CITY, NOVEMBER 2013

May thanks for Prof Dr Nguyen Trong Hoai, Dean of Vietnam – The Netherlands Programme who has provided necessary assistance and motivation for me to achieve this thesis

I would like give my special thanks for Dr Pham Khanh Nam, Academic Director of Vietnam – The Netherlands Programme Without his introduction for VARHS 2008, an important data set used in this research, this paper would be impossible to complete

Many thanks for Dr Truong Dang Thuy, Chair of the Department of Economics - Faculty of Development Economics, who has passionately cooperate with me to solve issues related to econometric techniques

Last but not least, I must express my most gratitude to my parents and my aunt’s family for providing comfortable condition during hard time, so that I can finish this thesis

3

List of Tables and Figures

L IST OF T ABLES

Table 1 - VARHS 2008 Survey Questions 27

Table 2 - Model Specification 44

Table 3 - Determinants of Credit Accessibility 54

Table 4 - Determinants of Partial Credit Rationing Probability 56

Table 5 - Determinants of Partial Credit Rationing Degree 60

L IST OF F IGURES Figure 1 – Credit Supplier Expected Return 15

Figure 2 - Rationing in Credit Market 17

Figure 3 - Identify Case of being Credit Rationed 20

Figure 4 - Survey Site Mapping for VARHS 2008 Source: IPSARD (2006-2008) 26

Figure 5 - Sample Distribution Source: Author Calculation from VARHS 2008 29

Figure 6 - Analytical Framework 38

Figure 8 - Credit Access & Credit Ration Source: Author’s calculation from VARHS 2008 45

Figure 9 - Credit Access & Household Head Age Source: Author’s calculation from VARHS 2008 46

Figure 10 - Household Head Age & Credit Ration Source: Author’s calculation from VARHS 2008 47

Figure 11 - Credit Access & Household Head Education Level Source: Author’s calculation from VARHS 2008 48

4

Figure 12 - Household Head Education Level & Credit Ration Source: Author’s calculation from VARHS 2008 49Figure 13 - Credit Access & Loan Purposes Source: Author’s calculation from VARHS 2008 50Figure 14 - Loan Purposes & Credit Ration Source: Author’s calculation from VARHS 2008 51Figure 15 - Credit Access & Credit Institutions Source: Author’s calculation from VARHS 2008 52Figure 16 - Partial Ration & Credit Institutions Source: Author’s calculation from VARHS 2008 53

5

Abstract

This study aim to identify key factors affected the partial credit ration’s probability and its degree in rural area of 12 provinces in Vietnam including Ha Tay, Nghe An, Khanh Hoa, Lam Dong, Phu Tho, Quang Nam, Long An, Dac Lac, Dac Nong, Lao Cai, Dien Bien, Lai Chau period 2006-2008 Based on VARHS 2008 data set, the research has employed Heckman sample selection bias model to investigate the determinants of partial ration’s degree, and bivariate probit with sample selection model to examine the determinants of partial ration’s probability Besides that, the impact of credit accessibility’s determinants, as a supplement outcome from the two regression models, were also revealed

The result showed that households who have following characteristics - Kinh ethnicity, large household size, high land value, suffering shock at household level (economic shock, illness, unemployment, etc.), holding social position (at least one member working for government, local authority unit) tend to have higher chance of credit access, while those who have high dependency ratio, and older household, tend to have negative correlation with credit accessibility

Formal credit institutions appeared to have higher rate of partial credit rationed than the informal sector, and those who requested a large size of loan were likely to be partial rationed as well In contrast, households who own larger house, borrowed for investment purposes (build/buying house, land and other assets) or holding social position had a lower chance of being partial rationed

The finding also uncovered the negative correlation between the degree of partial credit ration and following factors - Household head age, dependency ratio, house size and collateral value

On the contrary, household size, loan size applied, loan for consumption purposes negatively affect the degree of partial credit ration

The regression result also shown that unless treatments such as bivariate probit with sample selection bias or Heckman two stages regression are applied, the regression result might be bias due to inherent sample selection problem in the data set

6

Table of Contents

Chapter 1 Introduction 9

1.1 Research Context 9

1.2 Research Problem 10

1.3 Research Objectives 11

1.4 Research Questions 11

1.5 Scope of Study 11

1.6 Thesis Structure 12

Chapter 2 Literature Review 12

2.1 Rural credit 12

2.1.1 Definition 12

2.1.2 Characteristics of rural credit market 12

2.1.3 Types of rural credit 13

2.2 Asymmetric Information and Credit Rationing 14

2.2.1 Asymmetric information 14

2.2.2 Problems of lenders in context of asymmetric information 15

2.2.3 Screening mechanism in lending 17

2.3 Credit Rationing 18

2.3.1 Types of Credit Rationing 18

2.3.2 Identify Credit Rationing 19

2.3.3 Impact of Credit Rationing in Rural Area 21

2.4 Empirical Studies 21

2.4.1 Factors of Credit Demand 22

7

2.4.2 Factor of Credit Supply 23

Chapter 3 Methodology 25

3.1 Data Source and Features 25

3.2 Issue of Data Bias (Sample Selection Problem) 28

3.3 Heckman Two-Stages Model 30

3.3.1 Sample Selection Bias vs Omitted Variables Bias 30

3.3.2 Heckman Two Stages Procedures 32

3.3.3 Application to study & Model Specification 33

3.4 Bivariate Probit with Sample Selection Model 34

3.4.1 Model Review 34

3.4.2 Application to study & Model Specification 36

3.5 Multicollinearity Test 37

3.6 Analytical Framework 38

3.7 Hypothesis 40

3.7.1 Hypothesis for the probability and degree of partial credit rationing 40

3.7.2 Hypothesis for the probability of access to credit 42

3.8 Model Specification 44

Chapter 4 Results and Discussion 45

4.1 Characteristics of Borrowers by Credit Rationing 45

4.2 Determinants of Credit Accessibility 54

4.3 Determinants of Partial Credit Rationing Probability 56

4.4 Determinants of Partial Credit Rationing Degree 58

4.5 Multicollinearity Test 61

8

Chapter 5 Conclusions and Policy Implications 62

5.1 Conclusions 62

5.1.1 Findings, answers for research questions 62

5.1.2 Conclusions on degree of solving research objectives 62

5.1.3 Limitation of the study 63

5.2 Policy implications 63

5.2.1 Policy implications 63

5.2.2 Research perspectives 65

an important component The Government has launched many credit programs supporting the development of rural area such as preferential credit for the poor, agriculture forestry and fishery encouragement through special state own banks or government agencies The credit program has some significant impact to economic development of Vietnam rural areas; however there still exist some issues such credit rationing in the program When credit rationing occurs, credit suppliers ignore to offer loan to some borrowers to avoid the risk of default, thus it limit the credit accessibility of the poor Due to it implication to the economic development in general, and to the effectiveness of Government credit for the poor program in particular, this paper will aim at examining the factors that affect to credit rationing, particularly focus on the cases of partial credit ration, with the hope of revealing some potential implications for policy makers

10

1.2 Research Problem

Rural credit market is an importance factor that helps to foster the economic development in rural areas, thus improving the poor living standard and supporting poverty alleviation One of its function is funding household’s credit demand

However, the degree to which rural credit impacts on the rural area welfare depends on how well the rural credit market operates, however the problem of credit rationing could have negative effect on the performance of rural credit market Credit rationing could be described as the cases

in which credit lenders refuse to offer loan to borrowers, or offer an amount of loan that is less than borrower’s request, even though the borrowers willing to accept higher level of interest rate

to help the lender to cover the default risk (Barham, Boucher, & Cater, 1996; Buchenrieder, 1996; Heidhues & Schrieder, 1998; Zeller, 1993) The higher the probability of credit rationing, the more difficulties for household to satisfy their credit demand In other word, if credit rationing is a common practice in the area, it may lead to the inefficiency of the credit market in the area as a consequence

As Stiglitz, J and A Weiss (1981) pointed out, asymmetric information is an explanation for the problem of the credit rationing In rural credit market, which is characterized by numbers of poor households and the difficulties to evaluate their credit worthiness, is concealed by a fog of asymmetry information between lenders and borrowers In other word, lenders are reluctant to lend as they are uncertain about the loan repayment probability To overcome this problem, lenders require different kinds of information about their borrowers such as household’s dependency ratio, household size, land value, social position, etc., to assess their repayment ability and make a basis for lending decision

As such kinds of information may affect to the probability and the degree to which a borrower be credit rationed, a question has been raised by many researchers is that what factors determined lenders’ decision of credit ration The answers are different depending on the period and location under examination of a study For this paper, the research aim to examine the determinants of credit rationing – especially for the case of partial credit ration, concentrating on 12 provinces Ha

The general objective could be archived by meeting following sub-objectives:

 Identify key factors that affects the household’s credit accessibility

 Identify key factors that affects the degree of partial credit rationing

 Identify key determinants on the probability of partial credit rationing

 Suggest policy implication to reduce partial credit rationing practice

1.4 Research Questions

The research aims at answering the following questions:

 What are the key determinants on the credit accessibility of rural households?

 What are the key determinants on the partial credit ration probability of rural households?

 What are the key determinants on the degree of partial credit rationing of rural households?

1.5 Scope of Study

This study focuses on the issue of partial credit rationing (the case in which credit lenders constraint the amount of loan, thus borrowers do not fully satisfy their credit demand) at household level for 12 provinces in Vietnam including Ha Tay, Nghe An, Khanh Hoa, Lam Dong, Phu Tho, Quang Nam, Long An, Dac Lac, Dac Nong, Lao Cai, Dien Bien, Lai Chau period 2006-2008 using Vietnam Access to Resources Household Survey (VARHS) 2008 data set

12

1.6 Thesis Structure

The remainder of this paper is organized into 4 chapters Chapter two is literature review which aims to provide a basis for rural credit understanding, theoretical and empirical framework for the study Chapter three describes VARHS 2008 data source, discusses about the issue of sample selection of the data set and methodologies, which including Heckman two-stage probit model and Bivariate probit with sample selection model, applied in this study Chapter four presents major finding revealed by methods of descriptive statistics and econometric regression, as well

as providing results discussion Chapter five, the last one, is for concluding remarks, limitation of the study and policy implication

Chapter 2 Literature Review

The section firstly discusses concepts related to rural credit such as what is rural credit, characteristics of rural credit, problem of asymmetric information in rural credit, and credit rationing as well as theoretical background of credit rationing, to provide basic understanding about the study

Secondly, the determinants of credit supply and credit demand are summarized from earlier empirical researches, to provide empirical framework for the study of credit rationing

2.1 Rural credit

2.1.1 Definition

Rural Credit is referred to the credit offered to farmers to fund their agricultural and other rural relating activities It is estimated that 90 percent of rural finance activities is rural credit (Pham, T., 2010)

2.1.2 Characteristics of rural credit market

Rural credit market has some distinct characteristics:

13

High transaction costs:

Transaction costs in rural credit market is high due to several reasons: the dispersion of local users, wide segments of the farming community, small loan value, the value of time lost, travel costs, and other noninterest costs in getting and repaying loans and making deposits, high information and marketing costs due to low developed infrastructure for transport and communication

High risks:

Another characteristic of rural credit market is high risk, for the reasons of vulnerability due to unfavorable climate and weather, low return on investment of agricultural activities, need of household consumption, chain effect due to concentrated in small geographic rural areas, price changes causing further variability in farmers' income and the related repayment capacity, high probability of default, little acceptable loan collateral, property rights to mortgaged land may be uncertain and hard to enforce, the weak legal system and the ineffective reinforcement arrangements

2.1.3 Types of rural credit

Rural credit providers can be categorized into three sectors – formal, semi-formal and informal credit suppliers

Formal credit providers

Including commercial banks, branches of foreign banks, joint-stock banks and joint venture banks Some examples are Vietnam bank of agriculture and rural development (VBARD) is the major commercial source of credit for rural households Bank of social policy (BSP) is government-owned and non-profit bank, providing credit mainly to poor, ethnic minority, social policy households

Semi-formal credit providers

Providing loans through socio-political unions in rural areas and the level of activity of this sector in a region is related to priority programs of the government, consignment services of the

14

bank and the activity of unions, for example: PCF, Women’s Union, Farmers’ Association etc Their loan is often low interest rates, small loan amounts, short-term

Informal credit providers

Informal sources have been traditional providers of credit in rural areas and are the result of an underdevelopment formal credit market Four forms of informal credit sources: mutual lending among friends and neighbors, rotating savings and credit associations, specialized moneylenders including pawnbrokers, and traders Loan from informal sector is often high variety in interest rates and loan amounts

2.2 Asymmetric Information and Credit Rationing

2.2.1 Asymmetric information

Asymmetric information is a common problem in rural credit market The issue could be understood as a situation in which one party has more information than the other in a transaction For instance, in credit market, borrowers may know their credit worthiness better than their lenders, as the information such as income for repayment or loan use is on hands of the borrowers

Two main problems related to information asymmetry:

1 Adverse selection- immoral behavior that takes advantage of asymmetric information before a transaction For example, credit borrowers know their project is high risk and high return, so they may readily accept a high level of interest rate that the lenders offer them

2 Moral hazard - immoral behavior that takes advantage of asymmetric information after a transaction For instance, when the loan has been disbursed to borrowers, lenders may have difficulty in monitoring on how their lending money is used, and the borrowers may use the loan for purposes other than the one they stated in the loan contract The purposes may earn more return to the borrowers, but riskier for the lenders

15

2.2.2 Problems of lenders in context of asymmetric information

To understand the behaviour of credit lenders in context of asymmetric information, it is necessary firstly to understand credit lenders expected return function According to Jafee, D & Stiglitz, J.E (1990), the expected return to the bank is a function of quoted interest rate, graphically represented by an concave curve

Figure 1 – Credit Supplier Expected Return

Credit lenders reach the highest level of expected return when it charges the loan at interst rate r*

At the level of interst rate higher or lower than r*, the expected return to the lenders will fall, thus

it is reluctant for them to variate their interst rate away from the optimum level of r*

An important question arises is that what is the basis underlying the assumption of the concave curve of the credit lenders’ expected return, which is the key building block of the whole Jafee & Stiglitz’s (1990) explanation for the behavior of credit lenders In other words, why the increasing in level of interest rate may lead to a fall in the expected return to the lenders The answer is imperfect information problem in credit market, in particular it is the result of adverse selection and adverse incentive effects

r* Quoted interest rate Expected return to

lender

16

Adverse selection effect

The adverse selection effect could shape credit suppliers’ expected return curve into concave form in which as interest rate raises above the optimum level, the expected return begins to fall

As the interest rate is increased, the lending portfolio of lenders will change adversely, along with the risk of default of the portfolio That is due to the fact that safe potential borrowes who need credit to undertake the projects with low risk - low return, are unable to pay the high interst rate and consequently drop out of the market In constrast, the number of risky borrowers, who need fund to finance their high risk – high return projects, will increasingly take place in the lending portfolio Consequently, risk of default in their lending will increase, thus decreasing their expected return (Stiglitz, J & A Weiss, 1981)

Adverse incentive effect

Adverse incentive (or moral hazard) is another effect that could shape credit suppliers’ expected return curve into concave form In this case, the action of borrowers tends to change in response

to high interest rate after getting lending contracts approval That is the applicants do not follow their commissions in the lending contract in term of undertaking riskier projects rather than the ones stated in the contract, so that they can seek higher return to offset the high interest However, it is in turn increasing the risk of lending portfolio in an unexpected way, and therefore lowering lenders’ expected return Although it is the function of lenders’ monitoring practices to keep borrowers on track with their contract obligation, it is costly and never perfect

Thus, for those reasons of information asymetry, it is reluctant for credit suppliers to variate their interst rate away from the optimum level r*; and it is this rational behavior of lenders that lead to

a situation that although the high demand for credit may lead to a raise in the level of interest rate from credit suppliers due to the law of supply and demand, the equilibrium interest rate in the market do not adapt to change away from the credit lenders’ optimum level of interst rate r* In this case a proportion of credit borrowers, who willing to pay a high level of interest rate in order

to satisfy their credit demand, will be unable to raise the amount of credit supply in the market, and therefore getting unapproval for their loan application In other word, they are credit rationed

by the credit suppliers

Rate

18

Direct screening

In context of asymmetric information, credit lenders can adopt direct screening mechanism to decide whether to approve a loan by ensuring their clients’ repayment probability Credit suppliers can control the risk of default by three following approaches:

Firstly, collecting and evaluate necessary information about risk of their clients such as income, education level, ages, etc In this type of screening, the lenders can directly ration borrowers when they do not have enough required information to evaluate risk of the loan

Secondly, credit suppliers can enforce borrowers to inter-linkages with other markets such as input and output market to ensure the loan is used for right purposes Or limiting the range of lending in a particular geography and kinship group residents in a given region, or individuals with whom they trade

Finally, using collateral such as land, livestock, or other kinds of asset to back the risk of default

is often required by lenders If the collaterals are not secured enough to back the loan, the borrowers could be evaluated as unqualified to get loan approval

2.3 Credit Rationing

According to Hoff and Stiglitz (1990) – “Credit rationing is broadly defined as a situation in which there exists an excess demand for loans because quoted loan rates are below the Walrasian market clearing level” In other words, when credit borrowers are credit rationed, loan demand of those credit borrowers in the market cannot be fulfilled as credit lenders limit their lending to them even though the borrowers willing to accept a higher level of interest rate than the one that credit lenders set

2.3.1 Types of Credit Rationing

There could be various types of credit rationing depend on how the term - “excess demand for loan” is defined

It could be excess demand in term of a borrower receives a smaller loan size than the one requested at a given loan rate; and the borrower has to accept a higher rate in order to obtain a

Redlining is also a type of credit rationing in which “given the risk classification, a lender will refuse to grant credit to a borrower when the lender cannot obtain its required return at any interest rate Moreover, loans which are viable at one required rate of return (as determined by the deposit rate) may no longer be viable when the required return rises.” (Jafee, D & Stiglitz, J.E., 1990)

Pure Credit Rationing: This is a form of credit rationing that arises as an effect of imperfect information problem In this instance, there is discrimination in the credit lenders’ loan approval decision between two apparently identical groups of borrowers, although they have precisely the same terms in loan contracts One group is accepted for loan, while the other one do not According to Jafee and Stiglitz (1990) when it is the case, “changes in the availability of credit, not change in the interest rate, may determine the extent of borrowing”

2.3.2 Identify Credit Rationing

According to Barham, Boucher and Cater (1996), Buchenrieder (1996), Heidhues and Schrieder (1998), Zeller (1993), the case of being credit rationed can fall into three situations:

Completely Credit Ration

For those who has demand for credit and make loan applications, but their applications are fully rejected by credit lenders, so they cannot get any amount of credit that they requested

Partially Credit Ration

In this case, loan applications from credit borrowers are accepted, but loan size is not fully granted In other words, the credit borrowers receive an amount of credit less than the one they

21

requested Petric (2003) said that - “farm household are credit rationed by formal lenders in the sense that they cannot borrow as much as needed to finance inputs, investment and indispensable consumption expenditure.”

2.3.3 Impact of Credit Rationing in Rural Area

Many studies have investigated the impact of credit rationing in rural area welfare and showed that credit rationing has been negatively affecting the efficiency function of credit in term boosting the economic performance supporting social welfare in rural areas

Firstly, efficient credit market can improve the productivities in rural area Pham and Izumida (2002) found that credit had a considerable impact on household production For the case of rural area in Ethiopia, it was estimated to increase agricultural productivity in high potential, favorable condition producing areas by 11% if credit constraint was eliminated (Ali, D., & Deininger, K., 2012)

Secondly, a well-functioning rural credit market may help reduce poverty and contribute to rural household income growth As Józwiak (2001) shown, in general higher income growth and a greater extent of increasing family labors used tend to be the case of farmers who could get borrowing Krandker and Faruqe (2003) also gave proof on the contribution of credit on the farm welfare improvement

In contrast, in a reasearch of Li et al (2013), credit rationing was found to cause a loss in net income of 15.7% and a loss in consumption expenditure of 18.2% for households in China rural areas Feder et al (1990) also shown negative impact of credit rationing on farm profitability as well

2.4 Empirical Studies

To understand behavior of lenders and borrower such as credit rationing or credit accessibility in credit market, it is essential to examine the forces of credit demand and credit supply This section aims to review earlier studies about determinants of credit demand and credit supply, providing an empirical framework for the factors of partial credit rationing

22

2.4.1 Factors of Credit Demand

For various studies, age is essentially a factor that may affect credit demand Mpuga (2004) showed that the youngs have higher demand for credit than the olds for the reason that the youngs are more active engaging in doing business and need credit as a source of funding, while the old are more rely on their saving However, Tang et al (2010) revealed a contradict finding

in which the old farmers, with wide social network and social capital, are more likely to get borrow than the youngs Okurut et al (2005) also found the same result with Tang et al (2010)

in Uganda

Credit demand may depend on which gender of borrower is Women in rural area are oftern seen

as responsible for housework rather than market-oriented activities, thus their demand for credit

is not much necessary as man’s (Nwaru, 2011)

Education is realized as having effect on credit demand Tang et al (2010) study indicated that highly educated individuals are more likelly to borrow, especially in formal credit sectors However, it may not be the case at higher education level such as four year universtiy level, as upper level educated people tend to rely more on their high income rather than credit (Chen & Chiivakul, 2008)

Labor structure of household may affect their credit demand For instance, number of adults normally positively related with loan amount borrowed (Barslund & Tarp, 2008) Pham and Izumida (2002) argued that the need for expanding production in households with large number

of adults leads them find credit market as source of funding Higher dependency households may also demand more for credit (Pham & Izumida, 2002; Okurut et al., 2005), for the reason that household seek more credit to smooth economic burden bore by large number of dependents in their family

Household assets such as livestock, farming area were found to have positive impact on credit demand (Pham & Izumida, 2002) This is due to the need of working capital to raise livestock or farming The finding was also confirmed in the work of Hussain and Khan (2011) However, study of (Dulflo et al., 2008) found a contradict result that households who had large number of

2.4.2 Factor of Credit Supply

Empirical studies showed that major credit rationing determinants are demographic characteristic, i.e individual characteristics, borrowers’ skill, credit history, household head’s reputation, dependency ratio, gender, education, and collateral (Petric, 2003; Bester, 1987; Diamond, 1989; Pham & Izumida, 2002; Craigwell, 1992; and McKee, 1989) Borrowing purposes and loan size also affect the chance of being credit ration of households (Pham & Izumida, 2002) Political and social network was found to have affect credit rationing behavior (Ali & Deininger, 2012) Land holding, livestock and durable goods possession, and village infrastructure level may also determine whether household being credit rationed or not (Chaudhuri & Cherical, 2012)

Collateral plays an important role in rural credit market It acts as a signaling mechanism in which low risk borrowers are identified as those who willing to secure their loan contracts with a high amount of collateral Moreover, rationing due to problem of moral hazard is limited as higher degree of collateralization can induce investment in safer projects In either case, rationing occurs in case of lacking collateral (Bester, 1987) Land, livestock, and asset are normally supposed relating the credit constraint Land is conventional collateral used in credit market, and the investigation often concentrated on this kind of collateral (Petric, 2003; Barslund

& Tarp, 2008; Vuong et al., 2012; Aguilera, 1990; Ali & Deininger, 2012; Li, Huang & Zhu, 2013; Ping, Heidhues & Zeller, 2010) Land was confirmed as a significant factor of credit rationing probability in various researches According to Petric (2003), household whose farming with more rented land would be likely to be credit rationed; as rented land, unlike owned land, was not qualified collateral to secure the loans Livestock was also considered as collateral in rural credit market (Okurut, 2005; Barslund, & Tarp, 2008)

24

Regarding to human capital, the effect of education or farming experience on credit constraint were normally examined (Petric, 2003; Chaudhur & Cherical, 2012; Pham & Izumida, 2002; Vuong et al, 2012; Barslund & Tarp, 2008; Zeller, 1994, Ping, Heidhues & Zeller, 2010) As education was supposed to be related to productivity of a person and educated people tend to get respect from society, higher education level means higher loan repayment ability and higher credit worthiness

Concerning to household characteristics, Petric (2003) found that gender had a significant influence on the chance of being credit rationed As women tended to be responsible more for housework rather than income generating activities, credit repayment ability was low in those households with more women and credit was more constraint to them as a consequence Oppose

to the result of Petric (2003), Chaudhur and Cherical (2012) showed that female head households had a higher chance of loan approval However, larger familiy size reduce the probability of receiving in case of loans from banks

Kereta (2007) found that young and old people in Ethiopia are less likely to access credit than the middle age Chaudhur and Cherical (2012) showed that age factor was positive relate to the chance of getting credit approval while the research of Pham and Izumida (2002) resulted in negative relationship; however those two researches are not showed a significant impact of age

on lenders’ rationing decision

For various researches, dependency ratio, which measures the ratio between numbers of dependent over household size, has appeared to be a key determinant of ration decision by credit lenders Higher dependent ratio may lead to higher rate of credit ration (Pham & Izumida, 2002) The interpretation was that households with large number of dependents are normally poor as the more dependents means the more economic burden for the households

Lenders’ rationing decision may also be influenced by the factor of household reputation Pham and Izumida (2002) found that those households with low reputation are likely to be rationed Another argument of Diamond (1989) that reputation has effect on interest rate between lenders and borrowers Political and social network was found to have affect credit rationing behavior as well in the study of Ali and Deininger (2012)

3.1 Data Source and Features

The research will employ Vietnam Access to Resources Household Survey (VARHS 2008) as primary source for its sample data set to econometrically conduct the investigation on determinants of rural credit rationing in Vietnam Started in 2002, Vietnam Access to Resources Household Survey (VARHS), began to survey with around 1000 households in 4 provinces of

Ha Tay, Phu Tho, Quang Nam and Long An Then, the survey is repeated each 2 years with expanded surveyed samples In 2006 VARHS implemented in 12 provinces with 2,324 households and in 2008 VARHS implemented in 12 provinces with 3,223 households The next survey will be implementing in 2010 in the framework of this project

VARHS has been supported by Vietnamese Government to conduct nationwide investigation in order to provide detail information on situation of rural household access to resources such as land, credit, S&T, market information as well as other material resources for economic and livelihood development Started in 2002, Vietnam Access to Resources Household Survey (VARHS) implemented a survey with around 1000 households in 4 provinces of Ha Tay, Phu Tho, Quang Nam and Long An; and was repeated every 2 years with expanded the scope of survey In 2006, VARHS conducted the investigation in 12 provinces with 2,324 households and increased survey sample to 3,223 households in 2008 The 12 provinces in VARHS 2008 was area including Ha Tay, Nghe An, Khanh Hoa, Lam Dong, Phu Tho, Quang Nam, Long An, Dac Lac, Dac Nong, Lao Cai, Dien Bien and Lai Chau

For the purpose of the research, the data from VARHS is selectively extracted as follow:

27

Table 1 - VARHS 2008 Survey Questions

Cover page Q1 Is the household classified as rural or urban Dummy Variable Rural/Urban

Q2 What is the ethnicity of this household Identify Ethnicity variable 1A - Roster 1 Q2 What is the relationship of [NAME] with the Identify Household head

Q3 What is the gender of [NAME]? Identify Household head's gender variable Q4 What year was [NAME] born ? Identify Household head's age, Identify

number of adult member, dependency ratio 1A - Roster 2 Q16 What is the highest diploma [NAME] has Identify Household head's education variable 1B - Housing Q3 How many square meters does your household

occupy, including bedrooms, dining rooms, living Identify House's area variableQ7 Does your household own this dwelling? Identify household's asset variable 2A - Land 5 Q2 Do you have a red book for this land? Identify household's asset variable

Q3 Whose name(s) are in the red book? Identify household's asset variable

2 - Land 2 Q3 What is the total area of the plot? Identify Land's are variable

5H - Household income Q10 Total Net Income Identify Household's income

7C - Saving Q3 What was the money value of this saving/asset 12

months ago ? Identify household's asset variable8A - Credit 1 Q1 Has your household borrowed money or goods

(including seeds or fertilizer) from any source

Identify Credit demand of household (have demand or not)

Q3 Which member(s) of the household applied for

the loan?

Identify borrowers and their characteristics that may affect to credit rationing possibility (age, education, …)

Q4 How much did this person apply for? Identify Partial credit rationing cases (amount

of credit receive is less than the amount Q5 How much did this person receive (cash or cash

equivalent)?

Identify Partial credit rationing cases (amount

of credit receive is less than the amount Q9 From which institution or individual was the loan

of the guarantor to the member of the household Dummy guarantor variable8A - Credit 3 Q20 How many times, if any, has the person

responsible for this loan failed to make a due Identify credit worthiness variableQ22 How many times, since 1 July 2006 have you had a

loan rejected? Identify the case of being credit rationedQ23 What were the three main reasons for this? If these were clearly state, do we need to run

regression to examine the other variables??? 10E - Political Q1

Does any member of your household hold any office or other positions of public responsibility

in the Commune, or higher levels of

Identify social capital variable 10A - Groups1 Q1 Are you a member of any groups, organizations,

or associations? Identify social capital variable

28

3.2 Issue of Data Bias (Sample Selection Problem)

Statistically, the sample collection design should ensure the objective representation of the larger population’s characteristics; otherwise the inferences extracted from the sample may not valid for the population

However in some circumstances, the sample does not objectively reflect the population due to selection issues, i.e systematically selecting a sample base on particular criteria, or in other words - non-random selection on the dependent variable

The selection issues arise in the VARHS-2008 data set as the dependent variable which indicates cases of credit rationing was not randomly selected from the population In detail, the partial credit rationing households are identified only for those observations that recorded as having credit, while it is un-identifiable for those who do not have credit In other word, the population

is divided into two groups (those households who have credit and those who do not in which the sample is representative for one group only rather than the whole population

No Credit Access (Credit Balance = 0)

30

3.3 Heckman Two-Stages Model

This section aims to present Heckman (1979)’s theory about sample selection problem and his suggested treatment This model is appropriate for the case in which the dependent variable is in outcome equation is in continuous form, while the other in selection equation is dichotomous The determinants on degree of partial credit rationing will be examined via this model

3.3.1 Sample Selection Bias vs Omitted Variables Bias

According to Heckman (1979), if the sample used in the regression model only represents specific groups and does not objectively reflect the whole population (sample selection), then biases will arise in the estimation result and may lead to error inference for the whole population

In detail the logic between sample selection and bias estimation result can be demonstrated as follow:

Assume the Degree of Partial Credit Rationing Function as:

Where 𝑌1𝑖 is the continuous variable, defined as the difference between loan size received and loan size applied of households (i.e loan size apply – loan size received) If the value of 𝑌1𝑖 is larger, that means the more extreme the household got partial credit rationed In addition, 𝑌1𝑖 is observed only for those household who were recorded as got credit between year 2006 and 2008

in VARHS 2008 data set

𝑋1𝑖 is a vector of observed variables relating to the ith household’s characteristics such as credit institutions, loan size, loan purposes etc

Sample Selection Problem:

The population regression function for the equation (1) can be presented as:

𝐸(𝑌1𝑖| 𝑋1𝑖) = 𝛽1𝑋1𝑖While the regression function for the subsample of available data, i.e the data of 𝑌1𝑖 and 𝑋1𝑖 that are only observed when 𝑌2𝑖∗>0

3.3.2 Heckman Two Stages Procedures

The problem of sample selection bias was mathematically treated as omitted variable problem, so Heckman proposed to extract the inverse Mill ratio from equation (2) and insert it to equation (1)

as if it represents for the omitted regressor

To extract the inverse Mill ratio, Heckman conducted the estimation as follow:

√𝜎𝑢𝑢

𝜙(−𝑋2𝑖𝛽2

√𝜎𝑢𝑢)Φ(𝑋2𝑖𝛽2

𝜙(−𝑋2𝑖𝛽2

√𝜎𝑢𝑢)Φ(𝑋2𝑖𝛽2

√𝜎𝑢𝑢

)

=𝜙(−𝑋2𝑖𝛽2)Φ(𝑋2𝑖𝛽2)

And the function can be expressed shortly as:

𝐸(𝑌1𝑖| 𝑋1𝑖, 𝑌2𝑖∗ > 0) == 𝛽1𝑋1𝑖+ 𝜎𝜀𝑢𝜙(−𝑋2𝑖𝛽2)

Φ(𝑋2𝑖𝛽2)

So, if one knew 𝑍𝑖and hence 𝜆𝑖, one can enter 𝜆𝑖as a regressor in the equation:

𝑌1𝑖 = 𝐸(𝑌1𝑖| 𝑋1𝑖, 𝑌2𝑖∗ > 0) + 𝑉1𝑖 in which 𝑉1𝑖is an error term

However, as one does not know 𝛽2 , the value of 𝑍𝑖 is also unknown Heckman (1979) suggested that:

Firstly, one can use the known value of 𝑋2𝑖 and 𝑌2𝑖 to estimate 𝛽2 using probit regression in the equation 𝑌2𝑖 = 𝛽2𝑋2𝑖+ 𝑢𝑖 with 𝑌2𝑖 = 1 𝑖𝑓 𝑌2𝑖∗ > 0 and 𝑌2𝑖 = 0 otherwise, then with the

estimator of 𝛽2 – i.e 𝛽̂, one can estimate 𝑍2 𝑖 and hence compute 𝜆̂ – the estimator of 𝜆𝑖 𝑖

Secondly, estimate 𝛽1 and 𝜎𝜀𝑢 by OLS method basing on the data of 𝑌1𝑖, 𝑋1𝑖, and 𝜆̂ 𝑖

3.3.3 Application to study & Model Specification

First step – Probit model:

Probit (𝑌2 = 1|𝑋 ) = Φ(𝛽 𝑋 )

34

= 𝛽1𝑒𝑡ℎ𝑛𝑖𝑐𝑖𝑡𝑦 + 𝛽2ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝑠𝑖𝑧𝑒 + 𝛽3ℎ𝑒𝑎𝑑 𝑎𝑔𝑒 + 𝛽4𝑛𝑢𝑚𝑏𝑒𝑟 𝑎𝑑𝑢𝑙𝑡

+ 𝛽5𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑦 𝑟𝑎𝑡𝑖𝑜 + 𝛽6ℎ𝑒𝑎𝑑 𝑒𝑑𝑢 + 𝛽7ℎ𝑒𝑎𝑑 𝑔𝑒𝑛𝑑 + 𝛽8ℎ𝑜𝑢𝑠𝑒 𝑠𝑖𝑧𝑒+ 𝛽9𝑙𝑎𝑛𝑑 𝑣𝑎𝑙𝑢𝑒 + 𝛽10𝑙𝑖𝑣𝑒𝑠𝑡𝑜𝑐𝑘 𝑣𝑎𝑙𝑢𝑒 + 𝛽11𝑖𝑛𝑐𝑜𝑚𝑒 + 𝛽12𝑠ℎ𝑜𝑐𝑘

+ 𝛽13𝑠𝑜𝑐𝑖𝑎𝑙 𝑝𝑜𝑠𝑖𝑡𝑖𝑜𝑛 + 𝑢 Second step – OLS model:

𝑌1𝑖 = 𝐸(𝑌1𝑖|𝑋1𝑖, 𝑌2𝑖 > 0) + 𝑉1𝑖 = 𝛽𝑋𝑖+ 𝜎𝜀𝑢𝜙(−𝑋2𝑖𝛽2)

Φ(𝑋2𝑖𝛽2) + 𝑉1𝑖

= 𝛽1ℎ𝑜𝑢𝑠𝑒ℎ𝑜𝑙𝑑 𝑠𝑖𝑧𝑒 + 𝛽2ℎ𝑒𝑎𝑑 𝑎𝑔𝑒 + 𝛽3𝑑𝑒𝑝𝑒𝑛𝑑𝑒𝑛𝑐𝑦 𝑟𝑎𝑡𝑖𝑜 + 𝛽4ℎ𝑒𝑎𝑑 𝑒𝑑𝑢 + 𝛽5ℎ𝑒𝑎𝑑 𝑔𝑒𝑛𝑑

+ 𝛽6ℎ𝑜𝑢𝑠𝑒 𝑠𝑖𝑧𝑒 + 𝛽7𝑙𝑎𝑛𝑑 𝑣𝑎𝑙𝑢𝑒 + 𝛽8𝑙𝑖𝑣𝑒𝑠𝑡𝑜𝑐𝑘 𝑣𝑎𝑙𝑢𝑒 + 𝛽9𝑖𝑛𝑐𝑜𝑚𝑒+ 𝛽10𝑙𝑜𝑎𝑛 𝑠𝑧𝑖𝑒 𝑎𝑝𝑝𝑙𝑖𝑒𝑑 + 𝛽11𝑙𝑜𝑎𝑛 𝑓𝑜𝑟 𝑝𝑟𝑜𝑑𝑢𝑐𝑡𝑖𝑜𝑛

+ 𝛽12𝑙𝑜𝑎𝑛 𝑓𝑜𝑟 𝑐𝑜𝑛𝑠𝑢𝑚𝑝𝑡𝑖𝑜𝑛 + 𝛽13𝑙𝑜𝑎𝑛 𝑓𝑜𝑟 𝑖𝑛𝑣𝑒𝑠𝑡𝑚𝑒𝑛𝑡+ 𝛽14𝑐𝑜𝑙𝑙𝑎𝑡𝑒𝑟𝑎𝑙 𝑣𝑎𝑙𝑢𝑒 + 𝛽15𝑠𝑜𝑐𝑖𝑎𝑙 𝑝𝑜𝑠𝑖𝑡𝑖𝑜𝑛 + 𝛽16𝑐𝑟𝑒𝑑𝑖𝑡 𝑖𝑛𝑠𝑡𝑖𝑡𝑢𝑡𝑖𝑜𝑛 + 𝜎𝜀𝑢𝜆𝑖+ 𝑉1𝑖

3.4 Bivariate Probit with Sample Selection Model

In case of the two dependent variables in outcome equation and selection equation are both dichotomous, the bivariate probit with sample selection model is appropriate This model is similar to Heckman’s sample selection model in term of correcting the problem of non-random sample However, rather than dealing with one probit regression (for selection equation) and one OLS regression (for outcome equation) in Heckman’s model, the bivariate probit with sample selection deals with the two equations both regressing in probit model (Nicoletti & Peracchi, 2001) The model is applied to examine the determinants on probability of partial credit rationing

3.4.1 Model Review

The model of bivariate probit with sample selection can be illustrated as follow:

Suppose:

Outcome equation: 𝑌1𝑖∗ = 𝛽1𝑋1𝑖+ 𝑢1𝑖

35

Where 𝑌1𝑖∗ is the latent variables represent for the difference between loan amount received and loan amount applied, i.e amount applied – amount received (or the degree of partial credit ration) that credit borrowers got 𝑋1𝑖 is the vector of explanatory variables for 𝑌1𝑖∗, 𝛽1 is vector of explanatory parameters, and 𝑢1𝑖 is error term

And 𝑌2𝑖 = 1 𝑖𝑓 𝑌2𝑖∗ > 0 𝑎𝑛𝑑 𝑌1𝑖 = 0 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒

That means cases of credit access (𝑌2𝑖 = 1) are identified if loan amount borrowers got larger than 0

Assume: 𝑢1𝑖, and 𝑢2𝑖 are independent and identical standard normally distributed, i.e

𝑢1𝑖, 𝑢2𝑖 ~ 𝑖𝑖𝑑 𝑁(0,0,1,1), and 𝑐𝑜𝑟𝑟(𝑢1𝑖, 𝑢2𝑖) = 𝜌, their joint probability distribution function (pdf) will be:

36

However, in case of sample selection, there is likely to exist a relationship between two error terms, i.e 𝑐𝑜𝑟𝑟(𝑢1𝑖, 𝑢2𝑖) = 𝜌 ≠ 0; and the estimation of 𝛽1 cannot simply base on 𝑃𝑟(𝑌1𝑖 =1|𝑋1𝑖), but rather 𝑃𝑟(𝑌1𝑖 = 1|𝑋1𝑖, 𝑌2𝑖 = 1), 𝑃𝑟(𝑌1𝑖 = 0|𝑋1𝑖, 𝑌2𝑖 = 1), and 𝑃𝑟(𝑌2𝑖 = 0)

The probabilities for three types of observations in a sample are as follow:

In order to estimate the vector of model parameters is 𝜃 = (𝛽1, 𝛽2, 𝜌), the Maximum likelihood

method is applied, in which the following sample log-likelihood derived from (6) is maximized:

𝐿(𝜃) = ∑{𝑌1𝑖

𝑁

𝑖=1

𝑌2𝑖𝑙𝑛𝜙(𝑋1 𝛽1, 𝑋2 𝛽2; 𝜌) + 𝑌2𝑖(1 − 𝑌1𝑖) ln[𝜙(𝑋2 𝛽2) − 𝜙(𝑋1 𝛽1, 𝑋2 𝛽2; 𝜌)]+ (1 − 𝑌2𝑖)𝑙𝑛𝜙(−𝑋2 𝛽2)}

3.4.2 Application to study & Model Specification

The log-likelihood model:

𝐿(𝜃) = ∑{𝑌1𝑖

𝑁

𝑖=1

𝑌2𝑖𝑙𝑛𝜙(𝑋1 𝛽1, 𝑋2 𝛽2; 𝜌) + 𝑌2𝑖(1 − 𝑌1𝑖) ln[𝜙(𝑋2 𝛽2) − 𝜙(𝑋1 𝛽1, 𝑋2 𝛽2; 𝜌)]+ (1 − 𝑌2𝑖)𝑙𝑛𝜙(−𝑋2 𝛽2)}

𝑋1 is a vector of explanatory variables for the probability of credit access, including:

Ethnicity, household size, head age, head gender, head education, number adult, dependency ratio, house size, land value, livestock value, income, shock, social position

𝑋2 is a vector of explanatory variables for the probability of partial credit ration, including: household size, head age, head gender, head education, dependency ratio, house size, land value, livestock value, income, loan size applied, loan purpose for production, loan purpose for consumption, loan purpose for investment, collateral value, social position, credit institution

3.5 Multicollinearity Test

Multicollinearity is a problem that the correlation between explanatory variables with each other may bias the regression result such as larger variances, unstable parameter estimates, sign of coefficient bias (Menard, 2002) Thus multicollinearity test should be an essential step in analyzing regression result, and should be conducted as an initial step in multiple regression analysis (Mansfield & Helms, 1982)

In this study, the issue of multicollinearity will be detected via a correlation matrix which represents the correlation value between independent variables The correlation has min and max value of 0 and 1; and the indication for multicollinearity is that the correlation gets value above 0.8 The solution for multicollinearity is to drop the one of the two variables that has high correlation value (above 0.8) (Grewal, Cole & Baumgartner, 2004)

38

3.6 Analytical Framework

Figure 6 - Analytical Framework

Explanatory Variable Description:

 Ethnicity – dummy variable that takes value 1 in case of household ith is Kinh ethnicity and 0 otherwise

 Household Size – variable records the number of members in a household

 Household Head Age – variable records the age of households’ head

 Household Head Gender – dummy variable identifies the households’ head gender The variable takes value of 1 if household head is male and 0 otherwise

 Household Head Education – variable records the education level of households’ head, ranging from lowest level 1 to the highest level 6

 Number of Adult – variable records the number of adult member in a household The adult member is identified if household member is older than 18

Head Age Head Gender Head Education

Land Value

Household Size

Ethnicity

House Size Dependency

Livestock Value

Income Household Shock Social Position

Loan Size Demand

Loan Purposes Collateral Value

Credit Institution Number Adults

Probability of Partial Credit Ration

Degree of Partial Credit Ration

Probability of

Credit Access

Credit Supply

Credit Demand