Paper - Food and Agriculture Organization of the United Nations

advertisement
International Conference on Rural Finance Research:
Moving Results into Policies and Practice
FAO Headquarters
Rome, Italy
19-21 March 2007
Access to Credit and Borrowing Behaviour of Rural
Households in a Transition
by Cuong H. Nguyen
This paper was chosen through an open call for research in rural finance, whereby the
selected individuals were invited to Rome, Italy, to share their results during the
conference and to discuss key issues in shaping the rural finance research agenda as well
as ways of strengthening the ties between research, policy and practice.
Access to Credit and Borrowing Behaviour of
Rural Households in a Transition Economy
Cuong H. Nguyeny
Centre for Economic Reform and Transformation
Heriot-Watt University, Edinburgh, UK
January 2007
Abstract
This paper assesses the determinants of rural household credit activity
using data from Vietnam. We pay particular attention to identifying the
separate channels of credit demand and credit supply on the amount of credit
obtained by households. Credit amount per capita in the commune that
the household resides in and the distance from that commune to the nearest
bank branch provide the identifying variation for credit supply. The health
condition of household head helps to identify credit demand. To …nd the
impacts of household characteristics on credit demand and supply, we estimate
a bivariate probit with partial observability and a Heckman selection model.
We …nd that there is uniform access to formal credit aross rural communes
in Vietnam, although quantitative rationing does exist. The education level
of household head seems to have inverse u-shape a¤ects on formal credit participation: the least and the most educated households borrow least. The
problem of limited screening power and asymmetric information is revealed in
the result that the further the distance from a commune to a bank branch, the
less likely a formal bank will lend to residents of that commune. Prediction
of formal credit demand is estimated reducing over the years suggesting lack
of investment opportunity for rural households. With respect to policy implications, we suggest that the government could expand other programmes like
job creation or vocational training, and could introduce a policy calling for
more investment in far and distant rural areas.
JEL classi…cation: O12; O16; O17
Keywords: rural credit; credit access; credit rationing
Earlier versions of this paper was presented at the Annual Conference of the Scottish Graduate
Programme in Economics 2006, the 3rd Leicester PhD Conference on Economics 2006; and the
International Conference on Finance and Economics of Emerging Markets, Istanbul 2006. It has
also won the Prize for the Best Paper at the 3rd Leicester PhD Conference on Economics.
y
I am grateful to Mark Scha¤er and Julian Fennema for their guidance throughout this research.
I also bene…ted from useful comments by Gaia Garino and conference participants. Any remaining
errors in the paper are mine.
1
1
Introduction
The rural credit markets in developing countries are often described as repressed,
imperfect, and fragmented. It is a common that segments of borrowers have di¤erent
levels of access to certain types of loans and certain types of credit institutions (Ho¤
& Stiglitz 1990). In some markets, would-be borrowers may …nd themselves excluded
or dissuaded from formal sector. Possible reasons for these constraints have been
mentioned in various studies emphasizing the limited availability of formal credit,
the riskiness and uncertainty of rural credit market, the problem of asymmetric
information between lenders and rural borrowers, the weak enforcement system,
and the poorly developed economic infrastructure (see e.g. (Ho¤, Braverman &
Stiglitz 1993), (Aghion & Morduch 2005)). Formal credit applicants may then adjust
by turning to substitute, but possibly more expensive …nancing sources or may
modify their …rst best allocation plans in other ways .
In contrast to formal …nance, informal …nance provides a much easier conditions
to get loan. Borrowers and lenders often live in the same areas and easy to get to
know each others. Lenders usually do not require any collateral or documents, they
treat each other by informal laws. Borrowers are incentive to repay basing on the
relationship and trust, or sometime borrowers are threatened to repay by violence.
The di¤erence level of accessible to loan explains for the fact that informal money
lenders still exist and even dominated the market in some places, although they are
accused of taking their position to exploit the poor by charging high interest rate
and short term loan, which is sometimes pushing borrowers to state of unable to
repay (Aleem 1990)
Interaction and relationship of rural credit market agents, credit behaviours of
rural borrowers in the face of risk and uncertainty have been covered in a vast
literature which could be listed endlessly, e.g. (Aghion & Morduch 2005), (Bardhan
& Udry 1999), (Bell 1990), (Jain 1999), (Kochar 1997), (Morduch 1999), (Ray 1998),
(Townsend 1995), and (Ho¤ et al. 1993). While the overlap with earlier studies in
2
methodology and general results is inevitable, the contribution of this paper is to
provide more empirial evidences on determinants of rural household credit activity
using data from a transitional economy Vietnam. We pay particular attention to
identifying the separate channels of credit demand and credit supply on the amount
of credit obtained by households. Credit amount per capita in the commune that
the household resides in and the distance from that commune to the nearest bank
branch provide the identifying variation for credit supply. The health condition of
household head helps to identify credit demand. We employ a bivariate probit model
with partial observability and a Heckman selection model to disentangle a¤ects of
demand and supply.
The results are somewhat interesting. We …nd an uniform access to formal
credit aross rural communes in Vietnam, however quantitative rationing does exist.
Predicted level of rationing is relatively high although it seems to decrease over years.
The education level of household head appears to have inverse u-shape a¤ects on
formal credit participation: the least and the most educated households borrow
least. Distance to the nearest bank branch seems to reduce both accessibility and
amount of credit received of rural residents. Prediction of formal credit demand
is estimated to be large but decreased between the survey time suggesting lack of
investment opportunity in rural areas.
Structure of the paper is as follow. It …rst provides some general statistics of
Vietnam rural credit markets, characteristics of lenders and borrowers as well as the
credit attributions. Determinants of credit participation are analysed in section 4.
Section 5 studies impacts of household characteristics on credit amount obtained.
Conclusion and policy implications are presented in section 6.
3
2
The Data
Data for this study is the Vietnam Living Standard Survey (VLSS) conducted in
1992/93 and 1997/98 by the General Statistical O¢ ce of the Government of Vietnam, funded by United Nation Development Programme (UNDP) and the World
Bank. The surveys contain detailed information of 4,800 households from 150 communes in VLSS93 and 6,000 households from 194 communes in VLSS98. Samples
were weighted basing on the statistics of Vietnam Population Census in 1989 with
approximately 80% of Vietnamese households lived in rural areas. The communes
were randomly selected from a total of proximately 10,000 communes in 646 districts, and 64 provinces and cities in Vietnam, and then an average of 30 households
were randomly selected for interview in each commune. For the purpose of the paper, we select only households who are residing in rural areas at the time of the
surveys, which are including 3839 households in 120 communes for VLSS93, and
4882 households in 150 communes for VLSS98.
The two surveys are similar in many respects which provide data in both household and commune level. The household questionnaire was based on the format of
the World Bank’s Living Standards Measurement Surveys with adaptation to Vietnamese characteristics. The surveys collected information of household as a whole
and also information of individuals within that household, such as personal pro…le,
education, health status, employment, agriculture production, consumption, credit
and saving activities. The commune questionnaire was only applied to rural areas,
collected basic data on commune geographical information, general economic conditions, transportation and infrastructure, and credit. For information on credit
activities, households are asked to report whether they borrow any money from
other individuals or organisations within the last 12 months, and details of loan
they taken inluding source of loan, amount, duration, interest rate and collateral.
General characteristics of households, household heads and communes are presented
in table 1.
4
3
The Vietnam Rural Credit Market
Similar to other developing countries, the Vietnam rural credit market is considered to be repressed, segmented and dual structured where formal and informal
…nance exist side by side (see, for example, (Do & Iyer 2003), (Huong, Thai, Thao,
Kovsted, Rand & Tarp 2002), (McCarty 2001)). The formal sector consists of the
government’s commercial banks, private banks and other organised credit institutions, led by the Vietnam Bank for Agriculture and Rural Development (VBARD)
and the Vietnam Bank for the Poor (restructured recently as Vietnam Bank for
Social Policies). The informal …nancial network includes private money lenders, relatives, friends, and local rotating savings and credit associations. Until 1993, the
rural market had been dominated by informal sector which accounted for more than
70 percent of the market share. However, parallel with economic reform and growth,
there is a remarkably increasing role of formal sector. Over 5 years span between
the two surveys, the formal credit expands steadily from 28 percent of market share
in 1993, to 46 percent in 1998 (see table 4 and table 5), and 70 percent by 2001
(McCarty 2001).
Facing hazard of limited information about loan applicants, one of the …rst solution for lenders is to require collateral for their loan (table 4, 5). However, it is not a
surprise to observe a large discrepancy of collateral requirement between formal and
informal lenders which is well analysed in the literature (see e.g. (Varghese 2005)).
In 1998, about 70 percent of formal loans required collateral while only 3 percent of
informal loans did so. The most usual forms of collateral are land ownership documents, house, …x assets or jewelry (table 6). A part from collateral requirement,
formal sector seems to overcome informal sector in term of loan size, loan duration
and interest rate. The average rate of interest charged by informal lenders in 1993
is 80% per annum which is double the rate of formal lenders (table 4). A similar
interest rate is found by (Aleem 1990) in his study on Parkistan rural market.
While there are noticable di¤erences between lenders, borrower and non-borrower
5
characteristics are quite similar (table 1). In general, borrowing household seems to
be bigger in size, better in education of household head, and holding a larger land
size. Most of borrowers are also working in agricultural production account for more
than 60% of borrowing samples.
4
Determinants of Credit Market Participation
As pointed out in the previous section, there are a massive movement of borrowers
from informal to formal credit during 1993 to 1998. The expansion of formal credit
sector is reported in both loan numbers and loan amount provided, and all sample
communes are serviced by at least one formal credit institution. However, even
in the case that formal credit is available to people of a commune, not everybody
can borrow, or at least there are people cannot borrow as much as they would
like to. If we assume that formal …nance is more desirable and most of people
would choose formal …nance as their …rst priority when they demand investment
capital, then the unsuccessful applicants could be considered as riskier or weaker
(in some aspects such as poorer or not targeted) than successful applicants. The
credit rationing mechanism and selection process of formal institutions will push
those failed applicants back to informal source.
This work is not an attempt to understand why formal credit institutions operate in some certain regions and not in others. It will, however, assess determination
factors of household participation in credit market to learn the household and credit
provider behaviour toward demand and supply of credit. The features of a transitional economy like Vietnam making our analysis particularly interesting because
of the changing in economic environment and legal regulations over the time of the
survey i.e. the Land Law that went to e¤ect by December 1993 which passed the …rst
survey while well precede the second one. We expect these changes in the economic
conditions will contribute to changes in credit behaviour of rural households.
6
4.1
The Model of Partial Observability
Let Dij and Sij 1 , respectively, be latent variables of notional credit demand and
credit supply of the household i in the commune j. Thus, the two variables will be
expressed in a simpli…ed form as
Dij = Hijd + uij
(1)
Sij = Hijs + vij
where H d ; H s are the vectors of exogenous household and commune variables;
and
the coe¢ cients to be estimated; and u and v the normal distributed errors
for demand and supply of household credit respectively.
Set Dij and Sij as dummy variables of demand and supply of household credit,
where
Dij = 1 if Dij > 0
(2)
= 0 if otherwise
and
Sij = 1 if Sij > 0
(3)
= 0 if otherwise
In case of perfect or full observation, we will be able to observe values of both demand and supply, and therefore observe the equilibrium of borrowing. Loan amount
that a household acquire is simply the intersection of demand and supply line i.e.
D = S: However, in our study, due to insu¢ cient information provided by the survey
1
Depending on the dependent variable of the equation to be analysed, we will understand D
and S as demand and supply of formal credit, or informal credit, or simply credit regardless of
source. In case of not being clearly stated, we should think of D and S as a general term of demand
and supply of credit.
7
data, we observe neither D or S. Instead we observe the value of Bij ; which is the
product of D and S; the dummy variable of credit participation2
Bij = 1 if household ij participates in a credit market (Dij = 1 and Sij = 1)(4)
= 0 if otherwise (Dij = 0 or Sij = 0)
In this simple model, following the literature ((Iqbal 1986), (Kochar 1997)),
we assume that formal credit participation of a household solely determined by
the formal institutions’s decision on access. In other words, households are all
having demand for formal credit and those who appear having no credit activities are
assumed being rationed out. The model turns into an analysis of credit supply and
its results will generally inform us how much characteristics of household a¤ect its
accessibility to formal credit markets. Hence, probability of being credit participant
is represented by a single access equation and easily estimated following univariate
probit model estimation
Pr(access) = Pr(Bij = 1jHij ) = Pr(Sij = 1jHij ) =
0
+
1 Hij
+ "ij
A possible reason for the popularity of this univariate probit model is its simpleness to implement and conformable results to theoretical predictions. However,
the model su¤ers from substantial drawback due to its unrealistic assumptions on
formal credit demand. In practice, non-borrowing household could simply have no
credit demand rather than be denied by formal credit lenders.
To relax the assumption on credit demand and address the problem of nonobservable credit demand and supply, we employ the model of partial observability.
This model is …rst introduced by (Poirier 1980), and developed and applied in various
empirical studies (see e.g. (Abowd & Farber 1982), (Feinstein 1989), and (Heywood
2
Generally, we de…ne B as household borrowing status. If we apply B by source of credit, then
we should also interprete D and S by credit source accordingly.
8
& Mohanty 1995)). Following the structure speci…ed in equations 1, 2, 3, and 4,
credit participation status (B: = 1) is only observed when there exist both demand
and supply (S: = 1 & D: = 1). The last three combinations of S:and D: (S: = 0
or D: = 0) are indistinguishable since we only learn that there is no credit activity
(i.e. B: = 0):According to Poirier (1980), the probability distribution of B:is driven
by a bivariate process representing binary choice of credit demander and supplier
concerning the level of observability of D: and S:
Pr(Bij = 1) = Pr(Dij = 1; Sij = 1)
(5)
(6)
= Pr(Dij > 0; Sij > 0)
= Pr(Hijd + uij > 0; Hijs + vij > 0)
= Pr( Hijd < uij ;
= 1
F ( Hijd ;
Hijs < vij )
Hijs )
= F (Hijd ; Hijs ; )
and
Pr(Bij = 0) = Pr(Dij = 0 or Sij = 0)
(7)
= 1
Pr(Dij = 1 & Sij = 1)
(8)
= 1
F (Hijd ; Hijs ; )
(9)
We then obtain the log likelihood function3 as
3
The di¤erence between full and partial observability likelihood function is in its information
matrix where the former has four observable outcomes of S. and D., and the later has only two.
Full observability information matrix: ' = S0 D0 + S1 D0 + S1 D0 + S1 D1
Partial observability information matrix: ' = B0 B1
9
Log L( ; ; ) =
n
X
Bij ln(F (Hijd ; Hijs ; )) + (1
Bij ) ln(1
F (Hijd ; Hijs ; ))
(10)
where is the correlation between uij and vij ; F (:) denotes the bivariate standard
normal distribution.
4.2
Identi…cation and Variable Selection
The identi…cation criteria for Poirier’s model is relatively weak. Under the general
principle (Rothenberg 1971), and as details in (Poirier 1980), the partially observed
bivariate probit models will be (locally) identi…ed i¤ the information matrix corresponding to 10 is non-singular. In other words, we need to distinguish demand from
supply equation, i.e. at least one variable included in the explanatory variable sets
of demand or supply equation (either Hijd or Hijs ) but not in both. The identi…cation variable sometimes simply stems from the method of sampling involved or from
exogenous facts a¤ecting one of the two equations. For example, variation in rain
fall across regions is often used as instrument for credit demand as lower rain fall
level leading to less agricultural production activities and hence lower demand for
investment credit (e.g. (Kochar 1997))
In our study, the credit demand for formal and informal sectors is identi…ed by
a proxy variable measuring number of illness days of household head and his/her
spouse in four weeks before the time of the survey. As mentioned, informal sector
is the main source of consumption credit while formal sector remains as investment
capital provider. It is likely that facing health problem and sudden money require,
households will initially search informal source for consumption borrowing and in the
mean time having less demand for investment credit. In addition, health condition
is often not observable by formal lenders hence plays no important role in credit
supply decision.
10
More di¢ cult, however, is the identi…cation of the formal credit supply equation 1. There are a number of variables have been used in literature which are
argued that having no impact on demand but on supply side, such as household’s
land holding size which formal institutions used as eligible criteria to lend (see e.g.
(Khandker 2005), (Pitt & Khandker 1998)), or weather condition in the studied
regions (see e.g. (Kochar 1997)). The VLSSs contain questions on land holding
size of each sample household, however there is no evidence that formal institutions
will solely base on this information to decide whether to lend to a household or use
this information as eligible condition. Hence, it is not possible for us to use this
variable as identi…cation. Fortunately, data at commune level provides measure of
distance from the nearest formal institution’s branch to the commune’s committee.
This variable highly a¤ects bank’s ability to observe household’s behaviours and
screening process. Consequently, the further distance from household’s commune
is, the more unwilling bank would like to lend to households in that commune. In
other word, distance and formal credit supply are negatively correlated while distance and demand present no relation. This allow us to employ distance variable as
identi…cation in the supply equation. Unfortunately, this measure is only available
in the VLSS98 but not in the VLSS93.
In addition, another possible identi…cation measure is commune credit per capita
which is proxy for the availability of formal funds. This variable is exogenous to
demand of credit from household but could be in‡uent on supply decision. Banks
and formal institutions may know more about the commune with high amount of
formal credit already lent there and as a result, it could be easier for applicants from
that commune to obtain formal credit.
Besides the identi…cation variables, Hijs and Hijd contain other information on
household and commune characteristics. In common with other studies on rural
credit markets (see e.g. (Bell, Srinivasan & Udry 1997), (Udry 1990), (Siamwalla,
Pinthong, Poapongsakorn, Satsanguan, Nettayarak, Mingmaneenakin & Tubpun
11
1990)), exogenous variables expected to have impact on credit participation of rural
household includes gender of household head, age, household size, education level,
house value, employment area, and land holding size. Other variables of commune
and loan characteristics consist of geographical dummy, average interest rate charged
by credit sectors in the region.
4.3
Estimation Results
The results from the univariate probit model estimation are reported in table 8.
For formal sector, household size and rate of working adults are found to have large
positive and signi…cant a¤ects on credit participation in both surveys. Given the
employment nature in rural Vietnam where agricultural production dominated, more
labour available in a house is clearly an advantage as agriculture projects are easier
to form and implement. Without hiring extra labour, small family does not have
motivation and capacity to expand the family business conducing to a less credit
demand and then credit participation.
Age of household head plays important role in formal credit access. The inclusion
of age_square in the regression reveals signi…cant and negative coe¢ cients which
imply that middle age household heads (aged 33 to 41 in VLSS93 and VLSS98
respectively) most likely to obtain loan from formal institutions. Household head
being male signi…cantly increases probability of receiving formal loan in 1993 while
this impact disappears in the second survey.
Household working in agriculture production is found signi…cantly correlated
to borrowing from formal source in VLSS98 while its coe¢ cients in VLSS93 are
positive but nonsigni…cant. This result re‡ects the course of economic development,
decollectivization of agriculture and a major change in land regulation in 1993 which
eventually create more incentives for households to invest and improve their living
conditions. Land holding size and house ownership are also found to have positive
impacts.
12
While education level of household head does not in‡uent formal credit access
in VLSS93, it shows inverse u-shape impacts in VLSS98. Household heads with secondary school education appear to borrow more frequent than household heads with
college or university education, or household heads with very low or no education. A
possible reasons for this relationship is that high education helps head of households
easier to …nd paid job and then do not have to work in farm or small self-business.
Hence, demand for credit is reduced. For those who never attend school or have little
education, it simply harder for them to get their loan application approved. Formal
banks are more likely to refuse applicants without appropriate level of education,
being literate to …ll application form at least.
Finally, we …nd no signi…cant correlation between formal borrowing activities
and distance to the nearest bank branches. Further more, in the 1998 survey, when
asked to report main formal credit institutions operating in the commune, every
commune is able to name at least one. Also, formal borrowers are found in all
communes included in the sample. These evidences together suggest an uniform
access to formal credit at the commune level in rural Vietnam.
The results for informal sector are not-surprisingly di¤erent. The coe¢ cients
on age of household head, education, gender, agricultural work, house owner and
land holding size are all found not signi…cant. Commune credit per capita, however,
signi…cantly reduce informal credit activities in VLSS98. This is in line with other
study results (see e.g. (Jain 1999)) and our previous observation of increasing roles
of formal sector and crowding out of informal sector in rural credit markets.
Results on health condition of household heads con…rm our prediction on behaviours of agents in informal markets. Poor health condition seems to push household
toward informal sector who is able to provide quick disbursement and consumption
loans.
The second model of partial observability bivariate probit generalises the univariate probit by relaxing the assumption that all households have positive demand
13
for credit. The probability of participation is now jointly determined by household’s
demand for credit and lender’s decision on access. Estimates of this model are
reported in table ?? and table 9, which only apply to formal sector.
The separation of demand and supply equation provides a mixture results. While
estimates using VLSS93 show no major di¤erence between the two models except
the signi…cant impact of education level on credit demand, results using VLSS98
yield considerable di¤erences in the access equation comparing to results from univariate probit. First, the signi…cant levels of variables used by formal institutions to
make supply decision di¤ers noticeably from those used by households to determine
credit demand. While most of coe¢ cients are signi…cant in supply equation, notably
maritial status, agricultural employment, house owner and land holding size, only
rate of working adult and health condition of household head have signi…cant coe¢ cients in the demand side. These di¤erences suggest a gap between real demand and
supply of credit and there are households self-rationed out of the formal markets.
The model identi…cation variables perform well as expected. Health condition
of household head signi…cantly decreases demand for formal credit. On the supply
side, credit per capita spurs formal sector accessibility while distance signi…cantly
reduces supply.
Finally, a major di¤erence of the two models is in their predictions of the extent
of formal credit rationing and formal credit demand. Relaxing the assumption of
all household demanding credit applied in univariate model, the bivariate model
estimates the probability of a household demanding formal credit at 81.66% in 1993
(table ??) and 57.63% in 1998 (table 9), and probability of access relatively low
at 30.72% and 39.75%. The separate predictions of demand and supply equations
suggest that access for formal sector do not meet demand for formal credit in both
years of the surveys, although there is improvement in probability of access in the
later year.
Particularly, the probability of participating in formal credit sector, given by
14
joint probability of demand and supply, are 29.81% and 66.62% in 1993 and 1998
respectively. In other words, there is a large reduction of credit rationing from
76.59% to 33.38% between the two surveys time span. The results imply expansion
and development of formal sector which improve formal credit accessibility for rural
households. Of households that demand formal credit, majority obtain loans in
1998. While credit rationing still exists, the rationing rate is not as high as 79.23%
in 1993 and 64.16% in 1998 as suggested by univariate model. The reason for this
discrepancy lies in assumption and prediction of formal credit demand.
The reduction of formal credit demand and low level of accessibility suggest important policy implications. First, the decreased demand of formal credit among
rural population from 1993 to 1998 re‡ects limitted investment opportunity and
strong competition of informal sector. This problem is also highlighted in recent micro…nance studies(see e.g. (Aghion & Morduch 2005)) which point out that credit is
not the …nal and more certainly not the only solution to help people in rural areas
of developing countries out of poverty. Hence, beside expanding formal sector to
provide cheaper and more ‡exible credit, or to extent the out reach of …nancial services, governments in developing countries should continue to improve the economic
infrastructure, health and education system, and most importantly the economic
and legal system. For example, land reform which consoles land rights (i.e. holding,
trading, exchanging, collateralising) may signi…cantly increase household’s investment incentives and demand for credit (see e.g. (Do & Iyer 2003), (de Soto 2000)).
5
Determinants of Credit Amount Obtains
Thus far, the focus has been on the determinant factors of overall probability of
credit market participation of rural households. In what follows, the focus will be
on the determinant factors of credit amount obtains. The VLSSs provide data on
every separate loans that a household held in the reference years including loans
15
with di¤erent lenders and di¤erent loans with the same lender. Hence the amount
of loan that a household received from the same source of credit could be derived.
Assume that a lender’s decisions on supply of di¤erent loans to the same household
are homogenous, then household characteristics will be judged identically on every
loan application and sum of loan amount obtained from the same lender could be
considered as one loan amount. Therefore, the unit of observation here is a loan
amount from each source of credit. Every household will only appear once in the
dataset whether it does not borrow, or borrow only one loan from one source, or
borrow more than one loan from more than one source of credit. Depending on
source of credit, a household may appear as non-formal borrower even though it
does take credit from informal source, and vice versa. Until now, all types of lenders
from the same sector are considered as homogenous and hence they will act in the
same way considering characteristics of credit applicants.
The reason for studying the determinants of loan amount obtains is due to the
fact that not every household could be able to borrow as much as they want to or
in other words there is a high possibility of quantitative credit rationing (Petrick
2005). In the previous section, the analysis has been on credit access/participation
of households which is referring to complete rationing where some households could
not borrow even the smallest loan amount. Those models do not aim to and also are
not able to explain what factor causing the di¤erences in loan size that households
had. The variation of average loan size by lenders can be found in table 4 and
table 5. For both reference years, loan size from government banks and relatives
are respectively higher than average of formal and informal sectors, and the loan
size actually increase more than three times for the 5 years gap between the two
surveys. These increases however are not likely equally distributed among borrowers.
Hence, the following analysis contributes to build up a broader picture of relationship
between lenders and borrowers in rural credit market.
16
5.1
Econometric Framework
The framework for our analysis of determinant factors of household’s borrowing
amount is the models of limited dependent variables, which …rst introduced by the
pioneering work of Tobin (1958)4 . Since 1970s, the Tobit model has become a
regularly applied model in various areas of economics due to the advance of computational capacity and technology as well as the increased availability of microdata.
An excellent literature review of theoretical development and empirical application
of Tobit and its generalisations models are provided in (Amamiya 1984). Following
the literature, our model is set up as below
Let Lij be the loan amount that household i in the commune j received by the
time of the survey, and as mentioned, this is a sum of all loan amount from the
same source. One will observe that Lij > 0 if household is a borrower and Lij = 0 if
not. The loan amount Lij is determined by households and regional characteristics
which can be expressed as
Lij = Hij + "ij if Hij + "ij > 0
(11)
= 0 otherwise
where Hij is the vector of household and regional characteristics,
the vector of
coe¢ cients to be estimated and "ij the residual. "ij is assumed to be normally and
independently distributed.
The censoring feature of Lij destroys the assumption of linear relationships between the loan amount obtained by household and its characteristics so that the
linear OLS regression is inappropriate. However, applying maximum likelihood estimation, the Tobit model can consistently derive the vector of ^ for the loan amount
4
The model used in (Tobin 1958) and its generalisations are popularly known among economists
as the Tobit models
17
equation 11 5 .
On the basis of signi…cance and potential contribution to the model’s performance, the following covariates are included in the vector Hij : gender of household
head, age, marital status, household size, rate of working adult, education, dummy
for agriculture employment, dummy for house owner, size of agricultural land holding, number of illness day of the most two important members, distance from commune to the nearest bank branch, and credit per capita in commune. The results of
Tobit model estimation will be presented in the next section ??.
Although the Tobit model has been popular in empirical economic research to
solve problem of censored data, it still faces limitations. As noted by (Cragg 1971),
among others, the standard Tobit speci…cation is quite restrictive because it imposes
that both participation decision and decision concerning the size of the loan depend
on the same factors. In other words, household could borrow as much as they want
and money lenders always provide credit if there is demand. In addition, the Tobit
model considers zero observation of Lij as an equilibrium and in its speci…ed world,
rationing in the credit market does not exist. These assumptions unlikely re‡ect the
facts in real world.
In further details, the standard Tobit model speci…ed in equation 11 is likely
to su¤er from sample selection bias due to non-random decision of household to
participate in credit market. In that case, characteristics of borrowing households
are systematically di¤erent from non-borrowing households’and Tobit estimations
result in biases. In addition, given the participation decision, households have to
face accessible constraints from credit suppliers. A non-borrowing households may
actually want to acquire credit but have been rejected or ceiled by lenders. If loan
amount that a borrower received is again non-random, the Tobit estimates are biased
and further correction need to be applied (Maddala 1983). This problem of selection
5
The standard
Y likelihood function
Y for the Tobit model (Type 1 model as called by (Amamiya
1
1984)) is: L =
[1
(Hij = )]
[(Lij Hij )= ] where and are the distribution and
0
1
density function respectively of the standard normal variable.
18
is commonly addressed by applying the Heckman model which employs an additional
regressor to correct for bias in the participation decision (Heckman 1979).
Following the logic of Heckman selection model (Heckman 1979), a household
musts pass a participation hurdle before he is observed with positive loan amount.
The reduced forms of participation and credit amount equations are described as
follow
Pij = Hijp + uij
Pij =
(12)
1 if Pij > 0
0 if Pij 0
and the credit amount equation
Lij = Hij + "ij if Pij = 1
(13)
= 0 if otherwise
where Pij is latent variable, Pij and Lij the dependent variables, Hijp and Hij
the vectors of characteristics,
and
the coe¢ cients to be estimated, and uij
and "ij the error terms for participation and credit amount obtained respectively
(uij ~N (0; 1); corr(uij ; "ij ) =
" ).
"ij is assumed to be normally distributed but
E("ij ) 6= 0 because of the truncation in Lij 6 : Pij = 1 and Pij = 0 indicate credit
market participation and non-participation respectively. Clearly, a positive loan
amount is only observed if Pij = 1:
Given the above structure, there are two distinguished cases. First, assume that
participation decision and loan amount obtained by households are independent,
i.e.
u"
= 0; then the selection or participation process do not have e¤ect on the
6
As discussed in (Wooldridge 2002), Section 17.4.1., the Heckman two-step estimation requires a weaker assumption on the distribution of u and ": For MLE, we need to assume that
(u; ")~N (0; 0; 2u ; 2" ; u" )
19
outcome equation of credit amount, or in other words, there is no sample selection
problem. Hence,
can be consistently estimated by OLS using the selected sample.
However, any correlation between the two errors, i.e.
u"
6= 0, will make the
regression results in bias and we need to take account of selection7 . (Heckman 1979)
suggests a two-step procedure to estimate this model. In the …rst step, the probit
equation (selection) 12 is estimated by MLE and sample selection correction term is
constructed. For each observation in the selected sample, we compute bij =
p
(Hij
^)
p
(Hij
^)
(the inverse Mills ratio). In the second step, 13 is then estimated by OLS including
the correction term bij as an additional regressor.
The loan amount equation 13 can now be re-written as
E(Lij jHij ; Pij = 1) = Hij +
5.2
^
u u" ij
(14)
Estimation Results
While there is no identi…cation required for the Tobit model, it is important to
discuss the identi…cation strategy for the Heckman model. In theory, the Heckman
selection model is formally identi…ed even if no exclusion restrictions applied, i.e.
Hij
Hijp ; due to the non-linearity of the inverse Mills ratio. However, this identi…-
cation bases on distributional assumptions on the error terms rather than variation
in the explanatory variables. It often results in collinearity between the estimated
inverse Mills ratio and the covariates in the outcome equation which will essentially
lead to large standard errors. This problem is well presented in literature (see,
for example, (Wooldridge 2002)) and the common solution is to introduce at least
one instrument variable which identi…es the selection equation from the outcome
equation, i.e. the variable that independent with explanatory variables, and a¤ects
selection process but not the outcome.
As discussed in the previous section (see ??), we have several options for in7
In other words, u and " are assumed to be bivariate normally distributed
20
struments. The foremost to mention is the variable capturing distance from the
commune that a household resides in to the nearest formal bank branch. This variable is obviously exogenous to household and commune characteristics that a¤ect
the credit demand of a household. However distance could be important for banks
to decide whether to agree on supply of credit due to problem of screening and high
lending cost. However, in addition to the a¤ect on formal credit access possibility,
distance can also a¤ect the amount that formal banks willing to lend to a household.
The correlation between distance and outcome of loan amount obtained is expected
to be negative i.e. the further distance a household resides from a formal bank, the
smaller loan he receives. Hence, distance should be contained in both participation
selection and loan size outcome equations.
The variable of health condition and commune credit per capita are other feasible
options. Due to limited observation and asymmetric information about characteristics of loan applicants, formal banks cannot observe health status of household
members applying for credit and cannot use this unobserved information for its decision on credit supply. Loan amount that a household received is, therefore, will
not be a¤ected by its head’s health status. However, it is reasonable to argue that
household head with bad health condition will not seek for formal credit to expand
production. Consequently, bad health condition has negative impacts on demand
for formal loan and then on formal credit participation. Similarly, variable captured commune credit per capita is independent with household demand for credit
but well associates with formal credit accessibility of households residing in that
commune. In other words, the higher commune credit per capita the higher formal
credit accessibility for its residents.
The results of the Tobit and the Heckman selection models are reported in table
10 and table 11 respectively. The Tobit model is applied on both formal and informal
loans while Heckman model only for formal sector.
Generally, Tobit results for the two surveys on formal credit are quite similar,
21
with some exceptions. Age of household head, household size, and working adult
rate are found to increase loan amount that household received in both years. It
is not a surprise to see the signi…cant and positive impacts of land holding size,
although house ownership and working in agriculture only signi…cant in 1998. This
is in line with the stylised facts of increasing collateral requirements from formal
credit institutions which are mostly …xed assets (see table 6).
The distance variable is again prove to negatively and signi…cantly correlate
with formal credit activities. In this case, the further distance from household to
bank branch, the less loan amount that household receives. From demand side,
households face higher transaction cost e.g. travelling time to formal branch and
expenses, hence formal credit demand would also decrease.
For the informal sector, the most notable results are coe¢ cents of credit per
capita and househol-head’s health condition. While the latter variable operates as
expected to increase informal loan amount, the former variable is found to significantly reduce loan size that informal borrower obtained. This result highlights
competition pressure of formal sector development on informal lenders. Also, better
access to formal credit allows household to acquire larger amount investment capital
from cheaper formal source leaving smaller amount consumption loan (but quicker
dispensation) for informal source. This contributes to the segmentation characteristic of rural credit market.
The Heckman model estimation results are in table 11. Both for VLSS93 and
VLSS98 the signi…cant of ^ and the well di¤erent from zero of rho suggest that
correction for participation selection is indeed necessary. The results, however, are
quite similar to the Tobit except that there is noticeable change in the signi…cant
level and magnitude of coe¢ cients. For VLSS93, while household size remains almost the same positive impacts as in previous estimates, land holding size becomes
non-signi…cant. In VLSS98, distance from bank branch continues to be signi…cant
constraint for formal credit participation and amount of loan household obtains.
22
Although working in agriculture still plays signi…cant role in formal credit participation as in the Tobit model, it does not help to increase loan size with Heckman
model estimate. The coe¢ cient of impact plunge down implying the less important
role of agricultural employment.
6
Conclusion and Policy Implications
The empirical analysis of this paper has drawn an overview picture of Vietnam rural
credit market during the period from 1993 to 1998. It highlights characterisitcs
and interactions of formal and informal sectors; characteristics of borrowers; and
its impacts on credit participation and credit amount obtained. The employed
econometric models view probability of participating in credit markets and credit
amount received as a joint determination of the function of household’s demand for
credit and the function of lender’s decision on supply.
Estimation results imply that education, health condition, …xed assets holding
and distance from household to formal bank branch are among the most important factors a¤ecting household’s credit activities. In addition, there is evidence
of uniform accessibility to formal sector across communes although level of credit
rationing is found to be signi…cant. Rather than solely concentrate on the structural
interpretations of the estimated coe¢ cients, we would like to draw attention to the
prediction of the models, particularly the estimates of the probability of demand and
supply of formal credit. The estimates of both univariate probit model and bivariate probit model with partial observability suggest a low participation probability.
The high demand for formal credit in 1993 is turn out to be high level of credit
rationing due to limitation of formal credit access. The improved formal access in
1998 however is anticipated by the reduction of formal demand. Given this result,
the role of formal credit in supporting rural development may be limited. This is
also suggested by other researches (see e.g. (Bell et al. 1997), (Kochar 1997))
23
The results of this paper imply some important policy implications. First, even
though formal credit network continues to expand greatly to cover most rural areas,
there is question on the outreach and ‡exibility of credit services. Credit scoring
system should be developed to reduce collaterals requirement which actually considered as one of the most constraints for formal credit access. Land regulations and
…xed asset legal entitlement should be reviewed together with improvement of administration e¢ ciency to enable households to use their …xed property as collateral
for credit. In addition, the decreased demand of formal credit from 1993 to 1998
re‡ects limitted investment opportunities and the impact of agricultural productivity. Hence, government should continue to improve the economic infrastructure
to faciliate agricultural product trade market, provide better health and education
system, and enhance legal enforcement power.
References
Abowd, J. & Farber, H. (1982), ‘Job queues and the union status of workers’, Industrial and Labour Relation Review 35(3), 354–367.
Aghion, B. & Morduch, J. (2005), The Economics of Micro…nance, The MIT Press.
Aleem, I. (1990), ‘Imperfect information, screening, and the costs of informal lending: A study of a rural credit market in pakistan’, The World Bank Economic
Review 4(3), 329–49.
Amamiya, T. (1984), ‘Tobit models: a survey’, Journal of Econometrics 24, 3–61.
Bardhan, P. & Udry, C. (1999), Development microeconomics, Oxford University
Press.
Bell, C. (1990), ‘Interactions between institutional and informal credit agencies in
rural india’, The World Bank Economic Review 4(3), 297–327.
24
Bell, C., Srinivasan, T. & Udry, C. (1997), ‘Rationing, spillover, and interlinking in
credit markets: the case of rural punjab’, Oxford Economic Papers 49, 557–585.
Cragg, J. (1971), ‘Some statistical models for limited dependant variables with applications to the demand for durable goods’, Econometrica 39(5), 829–844.
de Soto, H. (2000), The Mystery of capital, Basic Book.
Do, Q.-T. & Iyer, L. (2003), Land rights and economic development: Evidence from
vietnam, Technical report, Working Paper No. 3120, The World Bank. Working
paper.
Feinstein, J. (1989), ‘The safety regulations of u.s. nuclear power plants: Violations, inspections, and abnormal occurences’, Journal of Political Economy
97(1), 115–154.
Heckman, J. (1979), ‘Sample selection bias as a speci…cation error’, Econometrica
47(1), 153–162.
Heywood, J. & Mohanty, M. (1995), ‘Estimation of the us federal job queue in
presence of an endogenous union queue’, Economica 62, 479–493.
Ho¤, K., Braverman, A. & Stiglitz, J., eds (1993), The Economics of Rural Organisation, Oxford University Press.
Ho¤, K. & Stiglitz, J. (1990), ‘Imperfect information and rural credit markets–
puzzles and policy perspectives’, The World Bank Economic Review 4(3), 235–
250.
Huong, V. N., Thai, L. V., Thao, N. M., Kovsted, J., Rand, J. & Tarp, F. (2002),
Financial sector reforms in vietnam: Selected issues and problems, Technical
report, Central Institute for Economic Management, Vietnam.
Iqbal, F. (1986), The Demand and Supply of Funds among Agricultural Households
in India, in I. Singh, L. Squire & J. Strauss, Agricultural Household Models:
25
Extensions, Apllications and Policy, Johns Hopkin University Press, chapter 6,
pp. 183–205.
Jain, S. (1999), ‘Symbiosis vs. crowding-out: the interaction of formal and informal
credit markets in developing countries’, Journal of Development Economics
59, 419–444.
Khandker, S. (2005), ‘Micro…nance and poverty: Evidence using panel data from
bangladesh’, The World Bank Economic Review 19, 263–286.
Kochar, A. (1997), ‘An empirical investigation of rationing constraints in rural markets in india’, Journal of Development Economics 53, 339–371.
Maddala, G. (1983), Limited-dependant and Qualitative Variables in Econometrics,
Cambridge University Press.
McCarty, A. (2001), Micro…nance in vietnam: A survey of schemes and issues,
Technical report, The State Bank of Vietnam and DFID.
Morduch, J. (1999), ‘The micro…nance promise’, Journal of Economic Literature
37, 1569–1614.
Petrick, M. (2005), ‘Empirical measurement of credit rationing in agriculture: a
½
methodological survey’, Agricultural Economics 33(2), 191U203.
Pitt, M. & Khandker, S. (1998), ‘The impact of group-based credit programs on poor
households in bangladesh: Does the gender of participants matter?’, Journal
of Political Economy 106(5), 958–996.
Poirier, D. (1980), ‘Partial observability in bivariate probit models’, Journal of
Econometrics 12(2), 209–217.
Ray, D. (1998), Development economics, Princeton University Press.
26
Rothenberg, T. (1971), ‘Identi…cation in parametric models’, Econometrica 39, 577–
91.
Siamwalla, A., Pinthong, C., Poapongsakorn, N., Satsanguan, P., Nettayarak, P.,
Mingmaneenakin, W. & Tubpun, Y. (1990), ‘The thai rual credit system: Public subsidies, private information, and segmented markets’, The World Bank
Economic Review 4(3), 271–95.
Tobin, J. (1958), ‘Estimation of relationships for limited dependent variables’,
Econometrica 26, 24–36.
Townsend, R. (1995), ‘Consumption insurance: an evaluation of risk-bearing systems
in low income economies’, The Journal of Economic Perspectives 9, 83–102.
Udry, C. (1990), ‘Credit markets in northern nigeria: Credit as insurance in a rural
economy’, The World Bank Economic Review 4(3), 251–69.
Varghese, A. (2005), ‘Bank-moneylender linkage as an alternative to bank competi½
tion in rural credit markets’, Oxford Economic Papers 57, 315U335.
Wooldridge, J. (2002), Econometric Analysis of Cross Section and Panel Data, The
MIT Press.
27
A
Figures and Tables
28
29
Table 1: General household characteristics
VLSS93
VLSS98
Borrower
All
Borrower
All
Male head of household (Male=1)
.79
.77
.80
.76
Age
43.07
44.85
45.24
47.52
Marrital status (Married = 1)
.84
.82
.85
.82
Household size
5.22
4.97
5.13
4.81
Education level
1.32
1.33
3.21
3.14
Years in school
6.89
6.64
Work in agriculture (Yes=1)
.83
.82
.72
.69
House ownership (Yes=1)
.97
.96
.97
.96
House area (m2)
59.59
61.85
67.56
69.67
House value (VND mil.)
7.25
7.82
27.17
24.53
Annual-crop-land holding size (m2)
953.05
918.78
2657.28
2447.98
Commune’s nonfarm employment (Yes=1) .44
.46
.57
.59
Total consumption (VND mil.)
5.48
5.35
11.91
11.76
Consumption per capita (VND mil.)
1.07
1.12
2.42
2.59
N
1985 (51.71%) 3839
2584 (52.95%) 4882
Note: (1) Education highest diploma ranging from 1-6 including pre-school, primary school, lower secondary school, upper secondary school,
vocational training, university. (2) House ownership: 1=whole property, 2=part of property, 3=rent. (3) Number of loans are calculated as
average among borrowing households only (n=2585). The land holding size is calculation of land for agricultural production purpose only.
Standard deviations are in parentheses.
Variable
Table 2: Distribution of borrowers by region
Region
Coastal
Inland Delta
Hills/Midlands
Low Mountains
High Mountains
N
a) As percentage of total
Formal sourcea Informal sourcea
6.82
9.09
47.72
51.57
6.42
7.41
21.40
18.87
17.64
13.06
1,729
1,309
borrowing households from the same source
Table 3: Distribution of borrowing households by emplyment of household head
Employment
All samplea Formal Borrowersb Informal Borrowersb
Working in Agriculture 63.83
70.34
64.14
Self employment
13.62
11.61
14.63
Paid employment
13.05
11.90
13.65
Unemployed
9.50
6.15
7.58
N
4,883
1,740
1,319
a) As percentage of total borrowing households. b) As percentage of total borrowing
households from the same source
30
31
a
b
VLSS93
Loan number
Loan amount Loan durationc Interest rated Colaterale
Formal source
28.17
1260.41
9.21 (14.22)
40.36 (10.40) 58.96 (26.01)
- Bank for the poor
- Other government banks
57.53
1482.71
6.98
39.51
77.8
- Private banks
14.77
1118.07
9.87
63.27
4.09
- Other credit organisations 27.70
875.45
16.98
29.93
49.06
Informal source
71.83
1402.72
10.72 (13.14)
90.83 (17.63) 4.04 (3.98)
- Money lenders
22.14
1371.09
8.54
108.35
9.52
- Relatives
50.68
1768.30
13.33
73.43
.24
- ROSCA
- Other individuals
27.18
746.73
8.84
108.99
6.68
a) As percentage of total loan number; b) Loan amount in thounsand VND; c) Loan duration measured in month;
d) Interest rate is average
P
Lr
annual interest rate (%) weighted by loan amount aross households borrowing from the same source: r = P Li i i ; e) As percentage of total
loan amount. Standard deviation in parentheses
Variable
Table 4: Mean characteristics of loan by source
32
a
VLSS98
Loan number Loan amount Loan duration Interest rate Colateral
Formal source
46.31
4335.75
18.40 (17.42)
13.76(3.15)
68.19
- Bank for the poor
15.6
1926.48
21.98
12.84
48.61
- Other government banks
60.97
5319.33
16.55
15.13
90.43
- Private banks
.29
1391.04
18.67
20.97
30.61
- Other credit organisations 23.14
3391.69
21.39
10.67
23.23
Informal source
53.69
4004.18
9.58 (14.12)
36.90(14.2)
3.74
- Money lenders
18.81
3536.41
9.57
46.2
9.04
- Relatives
43.17
3778.66
11.32
27.93
2.71
- ROSCA
1.15
2095.92
18.27
150.82
0.00
- Other individuals
36.87
4566.42
7.92
38.09
2.35
a) As percentage of total loan number; b) Loan amount in thounsand VND; c) Loan duration measured in month;
d) Interest rate is average
P
Li ri
P
; e) As percentage of total
annual interest rate (%) weighted by loan amount aross households borrowing from the same source: r =
Li
loan amount. Standard deviation in parentheses
Variable
Table 5: Mean characteristics of loan by source
Table 6: Distribution of collateralised loan by collateral assets - VLSS 1997/98
Variable
Land House Furniture/Fixed-assets
Formal source
58.59
31.64
4.77
- Bank for the poor
44.81
42.21
10.39
- Other government banks
62.20
27.87
4.05
- Private banks
100.00
- Other credit organisations 45.31
50.00
3.91
Informal source
8.02
90.38
1.60
- Money lenders
29.17
68.75
2.08
- Relatives
3.77
95.60
0.63
- ROSCA
- Other individuals
4.95
92.08
2.97
Note: All numbers are percentage
33
Others
5.00
2.60
5.88
0.78
-
-
34
Variable
VLSS93a
VLSS98a
Investment Consumptionb Relend
Investment Consumptionb
F ormal source
71.24
28.67
.09
79.15
20.80
- Bank for the poor
74.31
25.69
- Other government banks
86.24
13.59
.17
83.57
16.35
- Private banks
51.63
48.37
50.00
50.00
- Other credit organisations
50.52
49.48
71.13
28.87
Inf ormal source
43.45
56.32
.23
40.16
59.59
- Money lenders
59.83
39.49
.68
60.83
39.17
- Relatives
37.57
62.36
.07
27.93
71.69
- ROSCA
35.71
64.29
- Other individuals
41.09
58.77
.14
44.08
55.69
Note: a) Numbers are in percentage; b) Including repay other loan
Table 7: Distribution of loan by bowworing purpose
Relend
.05
.08
.25
.38
.22
Table 8: Probit result - VLSS98
Variables
VLSS93
VLSS98
Formal Informal
Formal Informal
Gender
.15*
.02
.03
.06
Age
.02*
-.01
.04***
-.02
Age^2
-.03**
.00
-.05***
.003
Marritial status
-.09
-.13*
.04
-.11
Household size
.04*** .06***
.09***
.05***
Working adult rate
.21**
.17**
.18**
.12*
Education level
.02
-.03
.03
-.04
Schooling year
.04*
.01
Schooling year square -.003*** .00
Farm
.09
.05
.26***
-.04
House owner
.17
.21*
.33***
-.05
Land-holding-size
.07**
-.02
.01**
-.002
Health status
.01*
.01**
.003
.014***
Credit per capita
.11
-.22
0.14**
-.20***
Distance
-.003
-.000
Pr(participation)
.2077
.3932
.3584
.2694
N
3839
4724
Results with robust standard errors adjusted for intracommune correlation. * signi…cant
at 10 percent, ** signi…cant at 5 percent, *** signi…cant at 1 percent.
35
Table 9: Partial Observability Bivariate Probit - VLSS98
Variable
Demand
Coe¢ cient
-.43
-.06
-.11
.12**
.34***
Standard Error
.28***
-.00
.02
.41***
.07*
Supply
Coe¢ cient
.02
.02
-.13
1.10***
-.05
.002
.07
.37
-.01
-.002
-.01*
Standard Error
.04
.05***
.25***
-.35
.07**
-.005***
.19***
.29**
.05***
.01
.25***
-.01***
-2.45***
Gender
Age
Maritial Status
Working adult rate
Schooling year
Schooling year square
Farm
.74**
.03
House owner
Land-holding-size
.24**
.31***
Commune population .04***
.003***
Health status
.03**
Credit per capita
.79***
Distance
Constant
-2.34***
-2.31***
-2.28
Pr(Demand)
.8166
.5763
Pr(Supply)
.3072
.3975
Pr(SupplyjDemand)
.2981
.6662
rho
0.6163
0.9433
N
3839
4724
* signi…cant at 10 percent, ** signi…cant at 5 percent, *** signi…cant at 1 percent.
Table 10: Tobit results
Variables
VLSS93
VLSS98
Formal Informal
Formal Informal
Gender
.44**
.03
.53
.48
Age
.04
-.02
.34***
-.13
Age^2
-.05
.001***
-.40*** .01
Marritial status
-.29
-.12
.25
-1.45*
Household size
.17*** .26***
.96***
.67***
Working adult rate
.73**
.26
1.34**
1.18
Education level
.07
-.04
.08
-.21
Schooling year
.40**
-.03
Schooling year square -.02*** .01
Farm
-.001
-.20
1.37*** -.38
House owner
.56
.82*
3.34*** .61
Land-holding-size
.12**
-.16*
.09**
-.02
Health status
.02*
.01
.02
.12***
Commune population .003
-.00
.07
.04
Credit per capita
.07
-.25*
.81**
-1.54***
Distance
-.09*** -.01
N
3839
4724
* signi…cant at 10 percent, ** signi…cant at 5 percent, *** signi…cant at 1 percent.
36
Table 11: The Heckman Selection Model Estimation Results
Variables
VLSS93
Participation Loan amount
Gender
.15**
.47
Age
.02*
.02
Age^2
-.0003**
-.00
Maritial status
-.10
-.31
Household size
.05***
.25***
Working adult rate
.26**
.45
Education level
.02
.05
Schooling year
Schooling year square Farm
.09
-.56*
House owner
.18
.72
Land-holding-size
.07***
-.01
Health status
.01*
Credit per capita
.01
Distance
rho
.78
^
2.20*
N
3839
* signi…cant at 10 percent, ** signi…cant at 5 percent, ***
37
VLSS98
Participation Loan amount
.01
.81
.04***
.29**
-.0004***
-.003**
.05
.18
.08***
1.03***
.18**
.95
.03
-.03
.05**
.37*
-.003***
-.02*
.22***
.20
.29**
4.19***
.01**
.08**
-.01**
.16***
-.01***
-.11***
.73
6.78***
4724
signi…cant at 1 percent.
Download