Content uploaded by Navjot Sangwan
Author content
All content in this area was uploaded by Navjot Sangwan on Oct 28, 2019
Content may be subject to copyright.
1
3000 Years of Discrimination and Counting: How Caste Still
Matters in the Indian credit sector.
Navjot Sangwan*
Abstract
The caste system has dominated the social, political and economic lives of Indian people for
over three thousand years. Since independence, the Indian government has introduced a flood
of quotas, schemes and affirmative action to tackle caste discrimination. Can seventy years of
government policy reverse three thousand years of oppression? Taking a close look at the
country's credit system reveals that a new, more subtle, and less overt form of discrimination
appears to be emerging, and becoming more widespread. This paper examines whether caste-
based differences influence the amount of credit sanctioned to borrowers in India utilising data
from the India Human Development Survey collected in 2005 and 2011-12. Using the Blinder–
Oaxaca decomposition method, along with the Heckman procedure and the instrumental
variable approach to correct for selection and simultaneity bias, I find substantial credit
differentials between the general caste and other lower castes. I also show the evidence of caste
discrimination against the lower castes. The results of this research have been complemented
by qualitative data gathered from interviewing lower caste borrowers in North India to
understand the nature of discrimination and obstacles faced by them in the credit sector.
JEL: C21; J15; O11
Keywords: Caste Discrimination, Credit, Blinder-Oaxaca decomposition, Quantile
decomposition, Asia, India
* Durham University Business School, Durham DH1 3LB, UK, navjot.sangwan@durham.ac.uk.
Acknowledgements: I thank Joe Durham, Arjun Bedi, Anurag Banerjee, Kausik Chaudhuri, Muhamad Asali,
Vibhor Saxena and the participants at Development Studies Association (2018) Conference at the University of
Manchester for their helpful comments and discussions. Any remaining errors are mine.
2
1. Introduction
Discrimination based on caste is a well-established phenomenon in India. Not that long
ago, lower caste people were treated as untouchable, were consistently denied access to public
services, and were subject to exploitation, abuse, mistreatment and prejudice. Nowadays, the
scale and visibility have changed, but discrimination can still be seen in the form of relatively
subtle constraints and restrictions in different areas: education, housing, finance, and
employment. Although caste equality has been enshrined in the Indian constitution since 1950,
caste-based divisions have continued to dominate - in the economy (Deshpande, 2000; Kijima,
2006), in marriage (Ahuja and Ostermann, 2016), in employment (Agrawal, 2014; Thorat and
Attewell, 2007), in access to energy (Saxena and Bhattacharya, 2018), in education (Desai and
Kulkarni, 2008) and in general social interaction. These caste divisions are sometimes
reinforced through economic boycotts and physical violence (Narula, 1999; Thorat, 2005).
Despite various efforts by the government such as reservation policies in educational
institutions, and employment in the government and public sector, caste still remains an
important indicator of socioeconomic disadvantage (Kumar, 2016). Given the pervasive
presence of discrimination, it would be surprising if it did not have a significant influence on
credit outcomes.
Credit is one of the most critical constraints in economic development for the lower castes
(Thorat, 2009). Variations in access to credit constitute a major source of income inequality
(Demirgüç-Kunt and Levine, 2009). Previous research has shown that a lack of access to credit
constraints entrepreneurship (Banerjee, Breza, Duflo and Kinnan, 2012), poverty reduction
(Chowdhury, Ghosh and Wright, 2005), agricultural investment, and income growth (Kaboski
and Townsend, 2012), farm production (Kochar, 1997), and spending on education (Doan,
Gibson, and Holmes, 2004). These constraints are even higher for lower caste groups who
remain socially excluded from the mainstream and lack access to assets, public facilities and
opportunities to improve their plight (Thorat and Neuman, 2012).
I use the definition of discrimination proposed by Becker (1971) which states that
discrimination occurs when some individuals complete a market transaction at a higher cost or
under more stringent terms than others who share the same characteristics. In credit markets,
this translates into differences in loan outcome (approvals, amount, and interest rate) which are
based on differences in caste, race, or nationality between groups with otherwise similar human
3
and physical capital. Becker introduced the first model of discrimination which explains
discrimination by ‘taste for discrimination’. When applied to credit markets, this model implies
that lenders may discriminate against minority borrowers to avoid interacting with them,
regardless of the borrower’s ability to repay, and that they are willing to suffer a financial
penalty to do so. Another theory of discrimination, known as ‘statistical discrimination’ was
pioneered by Arrow (1973) and Phelps (1972). The premise of this model - when applied to
lending - is that the lenders have limited information about the circumstances of some
borrowers - particularly their ability to repay. This gives lenders an incentive to use easily
observable characteristics such as caste to assume the expected creditworthiness of borrowers
provided that these characteristics are correlated with creditworthiness.
The reasons for credit differentials between castes may originate on both the supply and
demand sides of the credit market. On the supply side, some lenders may treat a loan
application differently based on whether it comes from a higher caste or a lower caste,
notwithstanding similar economic, household, and personal characteristics of the borrower -
simply because of preferences or cultural beliefs about castes. Other lenders may discriminate
against lower caste borrowers due to an expectation that lower caste clients lack the business
acumen to use a loan investment wisely. On the demand side, lower caste borrowers may
demonstrate traits such as a cultural reluctance to display entrepreneurship or initiative; a lack
of background in negotiation, or a cautious attitude to risk-taking - all of which could affect a
loan application. It could also be that a self-fulfilling prophecy was at work: the borrowers
themselves anticipated prejudice, felt that the lender would be unfair to them (high-interest
rates and unfair collateral requirements), and hence, did not seek large loans.
The credit differentials could also arise from lower debt repayment enforcement (as
shown by Rubin and Kuran, 2018) and profitability. Various government schemes and reforms
have passed in India to ensure a substantial flow of credit to lower caste groups at concessional
interest rates through priority lending. This hinders the enforcement of credit contracts, making
borrowers from these groups riskier and less profitable. In this setting where lending is biased
in favour of SC/ST, the lenders reduce the risk by providing them with less credit.
The qualitative interviews with lower caste borrowers in Northern India demonstrate that
modern-day discrimination is rarely in the overt form of denying all loans to the lower castes.
More subtle means are used: for example, giving a smaller loan amount, demanding higher
4
collateral, granting inadequate extensions on late repayment, imposing higher interest rates, or
denying marginal applications. The qualitative enquiries find evidence of petty discrimination
to discourage borrowers: long waiting times for opening bank accounts, lack of help with the
completion of paperwork, and intimidating inter-personal contact between higher caste lenders
and lower-caste borrowers.
One lower caste entrepreneur expressed his views on business lending from banks:
“….bank lending is not for the poor lower caste businesses. The banks never give us an
adequate loan. And it takes many weeks just to start the loan process - they give priority to the
higher castes.”
Another lower caste borrowers added about his experience in the informal credit market:
“….the loan terms are unfair to us. We get less loan for the same amount and quality of
land for collateral compared to upper caste. Lenders always treat it like they are doing us a
favour even though we pay such a high-interest rate”
The issue of caste discrimination in credit has largely been ignored in social science
research in India. We, therefore, have limited insight on the extent and nature of caste
discrimination in credit associated with group identity. This is one of the first studies to analyse
the discrimination against lower castes in India in the credit framework using qualitative and
decomposition methods.
Using nationally representative data from the India Human Development Survey (IHDS)
collected in 2005 and 2011-12 and qualitative interviews with lower caste borrowers, this paper
examines and compares loan amount differentials between castes in the Indian credit sector
over two periods. Using Blinder–Oaxaca decomposition, I demonstrate to what extent these
differences can be ‘explained’ due to the differences in observable characteristics of the
individuals and how much is ‘unexplained’ - which represents an indication of discrimination.
I further decompose the 'explained' component to identify the contribution of each specific
characteristic in generating the credit differences. In addition, I use the quantile decomposition
technique to analyse the caste gap across the entire credit distribution. Furthermore, I compare
5
the credit outcomes – loan application, approval rate and credit amount sanctioned – in lending
from banks, money lenders and social networks.
It is important to acknowledge at the outset that there are genuine statistical differences
between castes which affect loan outcomes, regardless of discrimination. There are particular
differences in observable characteristics between the general caste (GC) and lower castes -
Other Backward Castes (OBC); Schedule Castes (SC); and Schedule Tribes (ST). The former
is more urban, better educated, more likely to be self-employed or in regular salaried jobs, have
higher income and consumption levels. These disparities inevitably get reflected in the amount
of credit sanctioned - such that lower castes on average perform significantly poorly compared
to general castes. The differences, however, in the credit amount between the general caste and
other lower castes in India are not only because of lower quality attributes of lower caste (in
terms of education, income, assets etc) but also because these groups may be facing
discrimination in the credit sector.
The findings show three main results. First, there are significant differences in loan
amount between the general caste and other lower castes, and a considerable portion of these
differences remains unexplained. Second, the credit differentials have decreased between 2005
and 2011-12. Third, loan application and approval rates vary according to caste and lender.
Generally, lower castes have a higher loan application and approval rate in lending from
informal sector while general caste has a higher loan application and approval from banks.
Another contribution of this paper is highlighting the sticky floor
1
and glass ceiling
2
phenomenon in the credit gaps between the general caste and other lower castes. Using the
quantile regression-based decomposition method, I find that the credit gap between the general
caste and other lower castes varies across the credit distribution.
This paper is structured as follows: Section 2 draws on literature to give the background
of the Indian caste system. Section 3 presents data and uses descriptive evidence to highlight
1
Sticky floor refers to the scenario where the gap is higher at the bottom of the distribution and the lower caste at
the bottom are at a great disadvantage. In this particular case, it refers to the phenomenon of social rigidity in
which a certain group of lower castes fail to or are unable to take advantage of readily available options for
improving their social and economic status.
2
Glass ceiling refers to the scenario where the gap is higher at the bottom of the distribution and the lower caste
at the bottom are at a great disadvantage
6
caste differences in India. Section 4 sets out the methodology for the paper. Section 5 presents
the results from the selection equation, loan amount equation, decomposition of credit
differentials. Section 6 discusses and concludes the paper.
2. Background of the caste system in India
The Indian Constitution identifies three main categories of people for preferential
policies that reserve seats in legislatures, public sector enterprises, government jobs, and
educational institutions. These are OBC; SC also known as Dalits; and ST also known as
Adivasis. GC (also known as forward class) is a term used in India to classify communities
who do not qualify for any affirmative action schemes operated by the Indian government. By
default, 'general caste’ equates to the higher caste in more traditional categorisations.
In India, caste is associated with socio-economic status with a close relationship with
occupation and employment (Thorat and Attewell, 2007), income and expenditure (Deshpande,
2000), and capital (Kijima, 2006) - all which are of course helpful in accessing new lines of
credit. General caste groups usually have better economic outcomes than lower castes. There
is a great hierarchy among the OBC and generally, many OBC groups are closer to GC than to
SC or ST in terms of standard of living, income, education and other characteristics. The SC,
ST and OBC comprise about 19.5 percent, 8.6 percent, and 41 percent, respectively, of India's
population (National Sample Survey Office, 2011). But seven decades after Independence,
33.6 percent of SC, 44.8 percent of ST and 20.7 percent of OBC live below poverty line.
Table 1: Caste distribution according to population, poverty, expenditure and literacy
Caste
% of
population
% below
poverty
line
Average
Monthly
Expenditure
Rural (Rs)
Average
Monthly
Expenditure
Urban (Rs)
Literacy
Rate
General Castes
25
12.5
1281
2467
79%
Scheduled Castes
19.5
33.6
929
1444
58%
Scheduled Tribes
8.6
44.8
873
1797
50%
Other Backward
Classes
41.1
20.7
1036
1679
69%
7
Source: National Sample Survey Office, 2011. 1 Dollar = Rs 70 in May 2019
Dalits are the most oppressed and marginalised group in India. While Dalits make up
around 20% of the total population of India, their control over resources of the country is less
than 5% (National Campaign on Dalit Human Rights (NCDHR) report, 2009). Approximately
three-quarters of the Dalit workforce are landless or nearly landless agricultural labourers
(Census of India, 2011). According to an NCHDR report, the social conditions of Dalits are so
deplorable that more than half of the Dalit children are malnourished and less than 10% of
Dalit households can afford electricity, safe drinking water, and toilets.
The condition of ST households is no better than their SC counterparts. Even though ST
did not face exclusion in the form of untouchability, unlike SC, they have even poorer outcomes
in terms of health, education, jobs, and employment. Despite the reservation system, the share
of SC and ST in government jobs is 16.99% and 8.55 % respectively (Census, 2011). On the
whole, lower castes especially SC and ST perform badly in every development metric -
including credit.
3. Data and sample characteristics
The data used in the paper comes from two rounds of the India Human Development Survey
(IHDS) - a nationally representative survey of 42,152 households in 2011-12 and 41,554
households in 2005 collected from 1,503 villages and 971 urban neighbourhoods across India.
The survey covers a range of questions relating to economic activity, income and consumption
expenditure, assets, social capital, education, health, marriage, and fertility. Realising that
quantitative secondary data is insufficient to capture most of the social reality of discrimination,
this study also makes use of qualitative data using semi-structured interviews and informal
discussions with several borrowers in three villages
3
in North India to understand the way caste
discrimination exists in the credit sector. 16 men (3 from OBC, 12 from SC, and 1 from ST)
and 8 (6 from SC and 2 from ST) women from lower caste communities were randomly
selected for the interviews
4
. The interviews followed a semi-structured approach, giving
participants the flexibility to discuss issues important to them. Informal discussions were held
3
The qualitative work was done in the months of January and February in 2018. The villages demographic are
representative of the states in the North India, however, it’s difficult to say that it represents nationwide trends
since India is such a diverse country. However, qualitative analysis are very much in line with the quantitative
analysis.
4
All the clients approached agreed to be interviewed for this study. The interviews were done with women at their
houses in the presence of a local social worker who helped with translation and conversations. Consent was taken
in verbal form.
8
with political activists and representative of Dalit communities, four junior employees of two
rural commercial banks
5
(3 GC and 1 OBC) and six local money lenders (belonging to general
caste) operating in the region.
Table 2 presents the descriptive of the variables of the four caste groups used in the
analysis. The proportion of caste groups used in IHDS 2011-12 is similar to National Sample
Survey Office (2006) where GC are 30 percent, OBC are 41.1, ST are 8.6 percent, SC are 19.5
percent. The primary dependent variable is the log of loan amount
6
. GC has the highest amount
of loan undertaken, followed by OBC, SC and ST. However, only 46 percent of the GC
participated in the credit market compared to 60 percent of the OBC, 44 percent of the ST and
56 percent of the SC in 2011-12. Similar trends can be seen in 2005.
Table 2: Summary Statistics
Variables description
2011-2012
2005
GC
OBC
ST
SC
GC
OBC
ST
SC
Proportion in the sample
28.57
41.10
8.78
21.74
32.48
39.19
8.28
20.05
Loan Details:
Loan amount (Rs)
64410
52949
24611
28483
24323
18748
7390
10970
Log of Loan
10.72
10.39
9.60
9.94
10.03
9.56
8.67
9.14
Dummy if loan taken
0.46
0.60
0.44
0.56
0.33
0.47
0.34
0.44
Number of loans taken
1.23
1.87
1.57
1.66
0.96
1.45
1.00
1.42
Household Characteristics:
Yearly income (Rs)
178309
114354
92998
99492
75420
47279
39268
38676
Yearly consumption (Rs)
148916
114919
83397
94903
68829
49863
32724
43027
Proportion have female
head
0.14
0.14
0.15
0.15
.09
.09
.10
.10
Age of the head
51.88
49.53
47.92
47.83
48.51
47.09
45.53
45.46
Education years of the head
7.45
5.46
3.91
4.39
7.30
5.22
3.63
3.99
Size of the Household
4.82
4.92
4.77
4.83
5.12
5.28
5.06
5.19
Amount of land in acres
11.94
11.60
16.52
4.78
22.28
12.97
15.54
3.42
Dummy if own land
0.46
0.46
0.59
0.35
0.41
0.45
0.55
0.33
Dummy if live in urban area
0.44
0.35
0.14
0.30
0.47
0.34
0.15
0.29
House Quality7
0.75
0.66
0.35
0.55
0.71
0.57
0.29
0.46
5
Punjab National Bank and State Bank of India.
6
Government policies have invariably seek to tackle the discrimination problem in the credit market by focusing
on the access to credit. The assumption is that - having gained access to credit market - the processes and controls
within that system will work to ensure equal treatment. This study does not focus on access, instead, it measures
how equitably the credit system treats different groups after they have gained access.. The key measure is therefore
the amount of credit - the size of the loan actually granted - to those who make applications.
7
A binary variable distinguishing between dwellings that are designed to be solid and include cemented flooring
and strong roof compare to houses without a strong floor or roof. (Good = 1, Bad = 0)
9
Dummy if have a ration
card8
0.87
0.87
0.81
0.86
0.85
0.80
0.78
0.86
Market price of rice per kg
23.82
21.64
18.87
20.91
13.06
11.69
10.61
11.59
The general caste has better outcomes in terms of loan amount, income, consumption, and
education compared to the other castes. Income, consumption and loan amount increased by
more than double for all the castes between 2005 and 2012. Caste-based stratification translates
into low human capital for lower caste individuals. In 2005, the average number of education
years completed by ST (head of the households) in the sample were 3.63 years, 3.99 years for
SC, 5.22 years for OBC, followed by 7.30 years for the GC. Generally, the differences at lower
levels of education (primary) are less pronounced across social groups but start to diverge
widely by middle school and higher. For instance, only 3.5 percent of the SC heads of
household have achieved a graduate or post-graduate education compared to 14 percent of GC
individuals.
Although the GC are less likely to live in rural areas, they are more likely to own land
for agricultural purposes. In 2011-12, 56 percent of GC lived in the rural area, and 46 percent
owns land for agricultural purposes. Whereas 65 percent of OBC, 86 percent of ST and 70
percent of SC population live in the rural area, however, 46 of OBC, 59 percent of ST, and 35
percent of SC own land for agricultural purposes. The amount of land owned by SC/ST has
increased, while decreased for OBC and GC over this time. The general castes are more likely
to live in dwellings that are designed to be solid and include cemented flooring and strong roof
(also known as pukka houses).
The proportion of those taking a loan has also increased for all the groups. GC has taken
the lowest number of loans in the last five years; however, their loan amount is twice as big as
SC. However, SC/ST have significantly improved their credit outcomes between 2005 and
2011-12. Figure 1 below plots the kernel density distribution of log loan for all the castes. The
distribution of log loan of GC lies to the right of all the other lower castes.
8
Ration card is an identification document issued by state governments in India. It categorises household
according to their poverty level and allow the holder of below and extreme poverty households to obtain food and
other commodities at a subsidised price.
10
Figure 1: The kernel density distribution of log of amount of loan by various caste groups in 2011-2012 and 2005.
There is a clear distinction in the occupational structure of various castes (see Table 6 in
the Appendix). The major source of income for GC and OBC continued to be cultivation, non-
agricultural wage labour and salaried employment. Since a large proportion of ST own land, a
very significant portion of this caste group derives their income from agricultural activities.
The hierarchical nature of the caste system combined with low endowments of human and
physical capital implies that major portion of SC’s income continues to come from selling their
labour and a very small portion derives from cultivation.
The purpose of loans taken varies according to the caste group (see Table 7 in the
Appendix). In 2011, around 36.5 percent of the general caste loans are for productive purposes
such as buying land, agricultural, business, and education and the rest for non-productive
purposes such as marriage, consumption, educational, medical expenses etc. Around a third of
the loans by ST and OBC are for productive purposes which are in line with GC. SC mostly
comprising of wage labour has only 21 percent of loans for productive purposes and mostly
take loans for non-productive purposes.
Compared to 2005, the patterns within the group are more or less the same. Loan for the
non-productive purposes has increased for all the caste groups. Since consumption loans do
not generate any financial return and are deemed risky, increase in such loans for GCs and
OBC shows that lenders favour these castes over SC and ST. Loans for productive purposes
such as agriculture and business decreased for all lower castes but the decline in the SC was
the sharpest where it reduced to half.
11
Different castes tend to get their loans from different sources (see Table 8 in the
Appendix). The major source of finance for general castes comes from formal lenders such as
banks, while social networks and money lenders also play a significant role. OBC have
increased their share of lending from banks while reducing their reliance on money lenders
between 2005 and 2011-12. SC and ST are majorly dependent on informal sources for their
finance, however, these groups have greatly reduced their dependence on money lenders
between 2005 and 2011-12.
With the development of formal finance in India in the last decade, all the caste groups
have increased their reliance on formal sources such as banks, NGOs and credit groups in 2011-
12. With further development in financial services in India specially in microfinance and rural
banking, we may see a current trend of a diminishing role for money lenders in Indian society.
The data also show an increase in the share of loans from relatives and friends for all the caste
groups. Overall, we see a significant convergence of education, income, consumption, loan
amount of SC/STs toward non-SC/ST levels (also noted by Hnatkovska, Lahiri and Paul,
2012).
The survey done in 2011-2012 also has information on the breakdown of loan approval
and rejection of households from banks, money lenders, and social network (see Table 9 in the
Appendix). There is no clear pattern, however, general castes are more likely to borrow from
banks, whereas other castes are more likely to borrow from informal sources such as money
lenders and friends.
4. Methodology
This paper presents estimates of the mean caste loan amount gap in the Indian credit sector and
the extent to which this differential can be explained by differences in observable
characteristics or ‘endowments’ of clients across caste groups. The amount of credit the
borrower has arises from the following equation:
(1)
12
Where
is the natural logarithm of loan amount of ith individual in jth social group ranging
from GC, SC, ST and OBC.
is a vector of observed characters, and
is a coefficient vector
to be estimated for each caste type, and
is assumed to be a normally distributed error term
with mean zero and positive variance.
The Blinder–Oaxaca decomposition is employed to decompose the credit amount gap in
outcomes between various castes
9
. Oaxaca (1973) and Blinder (1973) developed a regression-
based decomposition to divide the gap in an outcome of interest between two groups into an
‘explained’ and an ‘unexplained’ portion. The ‘explained’ portion of the gap is the actual
difference between the mean values of two castes which could be explained by differences in
endowments and personal attributes. The ‘unexplained’ portion of the gap arises from group
differences in the effects of the independent variables (Sen, 2014). This is also known as
discrimination function or unexplained residual – a part that cannot be accounted for by
differences in characteristics. While the unexplained component is often used as a measure for
discrimination, it is very likely that the residual also includes the effects of unobservable or
unmeasurable characteristics (Deshpande and Sharma, 2014). All decomposition analyses are
subject to this caveat given that it is generally very difficult to control for all the borrower’s
characteristics that may affect creditworthiness
10
.
The difference in the credit amount arise from following equation:
(2)
EXPLAINED UNEXPLAINED
Where
is the natural logarithm of loan amount, g and l subscripts stand for general
caste and lower castes (SC, ST and OBC) respectively.
is a vector of observed characters
for general caste,
is a vector of observed characters for various lower castes, and
is a
coefficient vector to be estimated for general caste,
is a coefficient vector to be estimated
9
Blinder–Oaxaca decomposition has been used to measure differences between castes in health outcomes
(Maity, 2018), labour market (Hnatkovska, Lahiri and Paul, 2012), poverty (Borooah, 2005), school enrolment
(Borooah and Iyer, 2005), access to energy (Saxena and Bhattacharya, 2018) in Indian context.
10
It is also possible that pre-market discrimination affects the development of characteristics, and thus, the
explained component could also constitute the effects of past discrimination. Considering this, the estimates of
the unexplained components should not be taken as precise measurement of discrimination but as rough estimates
of its scale (Deshpande and Sharma, 2014).
13
for lower caste and
is the estimate of the non-discriminatory credit coefficient and can be
written as:
The non-discriminatory credit coefficient
, can be estimated using the coefficients from
the higher caste where (D=1) or the lower caste as the reference coefficients (D=0). However,
there is no particular reason to assume that the coefficients of any of the groups are non-
discriminating (Jann, 2008). It has been claimed that the undervaluation of one group comes
along with an overvaluation of the other (Cotton, 1988). Considering this, I use the method
proposed by Neumark (1988) using the coefficients from a pooled regression over both groups
as an estimate for
4.1 Selectivity and simultaneity bias
Another methodological problem faced in analysing the caste gap is the existence of
endogeneity which can be caused by self-selection and simultaneity bias. Selection bias could
occur when individuals with similar characteristics (education, assets or consumption level)
have different levels of entrepreneurship, perseverance and ability, which may lead to different
probabilities of their participating in the credit market. The self-selection is corrected by using
the Heckman two-step procedure in the analysis.
Using the Blinder-Oaxaca decomposition, the observed earnings differential can be further
decomposed into:
(3)
EXPLAINED UNEXPLAINED
where is the coefficient of the inverse Mills ratio ().
Simultaneity bias could be caused by the presence of endogenous variable such as
consumption expenditure which may cause reverse causality. To remove the simultaneity bias,
we require an instrument for consumption expenditure – an exogenous variable that is
correlated with consumption expenditure but is not otherwise associated with the loan amount.
In this case, the market price of one-kilogram rice in the region satisfies the requirement for
14
use as an instrumental variable
11
. Rice is the most consumed food in India, and its price has a
significant impact on consumption expenditure. The instrument affects the loan amount
through its effect on consumption expenditure only.
4.2 Specification checks
Identification can be achieved by including at least one independent variable that appears
in the selection equation but not in the outcome equation - we need a variable that affects the
selection, but not the outcome (Sartori, 2003). In the specification, there are a number of
additional identifying restrictions that are described below.
Landownership can affect a household’s ability to participate in the credit market as land
can be used as collateral, therefore, it appears in the selection model. However, merely having
land does not affect the amount of loan a borrower can get and other variables such as size of
landholding, quality of land or land titles may be more suitable in the credit amount equation.
To check this, we plugged the landownership dummy in the credit amount equation and found
it does not have any effect on the amount of loan, whereas it positively affects the probability
of participation in credit market. The information regarding the source of loan and purpose of
the loan is only available for people who have taken loan, so it appears in the credit amount
equation. The rest of the variables appears in both selection and loan amount model.
5. Results
5.1 Selection equation
I begin the analysis by estimating a model of the probability of participating in the credit
market using a probit model
12
. The dependent variable is 1 if the client has taken a loan or 0
11
We also test for the relevance of the instrument in the first-stage regression. Staiger and Stock (1997)
proposed a rule of thumb declaring the instruments weak when the first stage F statistic is less than 10. The F-
statistic from the first-stage is sufficiently large in every instance, suggesting that the IV is powerful. Another
approach, by Stock and Yogo (2005) is to reject the null hypothesis of weak instruments when the Cragg and
Donald (1993) F-statistic exceeds a given threshold. In this case, we reject the null hypotheses of the weak
instrument since Cragg-Donald F statistic exceeds the threshold of 16.38 at 10%. By these criteria, we have a
good instrument in the average market price of one kilogram rice in the region.
12
I failed to reject the null hypothesis for Wald test of exogeneity using instrument variable, therefore, a regular
probit regression may be appropriate.
15
otherwise. The estimates of the probit regressions are used to construct the Inverse Mills Ratio
(IMR) for the purpose of correcting the credit amount equation for selection bias as reported
in the later section.
To facilitate the understanding of the effects of coefficients, I present the marginal effects
of the regressors on the probability of participation in the credit market by each caste in Tables
10 and 11 in the Appendix
13
. The results from both time periods show similar results. One
percent increase in consumption increases the probability of taking a loan by 0.1 percent for
GC, 0.14 percent for OBC, 0.17 percent for ST and 0.15 percent for SC in 2005 and 0.11
percent for GC, 0.13 percent for OBC, 0.11 percent for ST and 0.14 percent for SC in 2011-
2012. The results show that gender exerts an influence on taking a loan. Being a female
significantly decreases the probability of taking a loan (except for ST in 2005). The number of
education years completed by the head of the households shows a negative relationship with
the likelihood of participating in the credit market (except for ST in 2011-12). This implies that
highly educated heads are more likely to work in salaried positions and may not require loans.
The land ownership has a positive relationship with the probability of participation in credit
markets. Households living in a strong and permanent dwelling are less likely to participate in
the credit markets (except for SC in 2005 and ST in 2011-12). Households living in rural areas
are more likely to participate in the credit market (except for ST in 2011-12) reflecting the
cyclical nature of an agricultural economy and the relatively long delay between investment
and income. To account for the differences between the sources of income and location, I also
controlled for occupation and states dummies.
5.2 Loan amount equation
I now proceed to estimate regressions for each caste type corrected for selection and
endogeneity (see Tables 14 and 15 in the Appendix). The result shows that a single percentage
increase in consumption increases the loan amount by 0.68 percent for GC, 0.84 percent for
OBC, 1.257 percent for ST and 0.31 percent for SC in 2005 and 0.80 for GC, 0.48 percent for
OBC, 0.52 percent for SC, however, no effect for ST in 2011-12. Education has a positive
relationship with the loan amount. An additional year of education significantly increases the
amount of loan taken by all the castes except for ST in 2005. Age has a quadratic relationship
13
The coefficients from the probit model are shown in Tables 12 and 13 in the Appendix.
16
with the volume of loans implying that lenders are more inclined to give higher loans to older
borrowers (except for SC in 2005). The size of landholding has a negative relationship with
loan amount. This could be due to the following reasons. First, in the absence of land titles,
and poorly administered land records, small and marginal farmers, who account for more than
half of the total land holding, may not be able to use it as collateral (Reserve Bank of India,
2015). Second, the cost of cultivation per unit of land might decrease with an increase in the
size of the land under cultivation (Bhattacharjee & Rajeev, 2014). Third, it’s a non-liquid and
immovable asset so it’s not very suitable for collateral, specially for short term loan for small
and marginal farmers. An alternative such as gold rather than land is preferred by the lenders
(Sarap, 1991). The quality of borrower’s house is a better predictor of loan amount and
significantly increases the loan amount (except for ST in 2005). Having a female head of the
household and living in urban area increases the loan amount for every caste except for ST in
2005. The estimated models have fairly high explanatory power for all four social groups.
5.3 Decomposing the differences in participation in the credit market:
To disentangle the role of observable and unobservable factors on the participation level in the
credit market among various castes, I extend the Oaxaca-Blinder decomposition to nonlinear
methods using Fairlie’s (2005) approach. Table 3 below decompose the probability of
participation in the credit market into explained and unexplained part using the estimates from
the selection equation.
Table 3: Decomposition of the probability of participation in the credit market in 2005 & 2011-
12
2005
2011-12
VARIABLES
GC VS
OBC
GC VS ST
GC VS SC
GC VS
OBC
GC VS ST
GC VS SC
GC
0.331***
0.331***
0.331***
0.458***
0.458***
0.458***
(0.004)
(0.004)
(0.004)
(0.005)
(0.005)
(0.005)
Others
0.469***
0.337***
0.440***
0.600***
0.440***
0.565***
(0.004)
(0.008)
(0.005)
(0.004)
(0.008)
(0.005)
17
Difference
-0.139***
-0.007
-0.109***
-0.142***
0.017*
-0.107***
(0.006)
(0.009)
(0.007)
(0.006)
(0.009)
(0.007)
Explained
-0.132***
-0.040***
-0.066***
-0.109***
-0.031***
-0.052***
(0.004)
(0.008)
(0.005)
(0.004)
(0.008)
(0.004)
Unexplained
-0.007
0.033***
-0.044***
-0.032***
0.048***
-0.055***
(0.006)
(0.011)
(0.007)
(0.006)
(0.011)
(0.007)
Observations
29,714
16,877
21,765
28,444
15,270
20,491
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1.
For the data collected in 2005, I find that all the lower castes have a higher probability
of participating in the credit market compared to the general caste. In 2011-12, I find that the
SC and OBC have a higher probability of participation in the credit market compared to GC
whereas GC has a small advantage over ST. A substantial proportion of the credit gap between
GC and SC and GC and ST remains unexplained. On average, lower castes are more likely to
participate in the credit market compared to general castes after controlling for the selection
variables.
Figure 2: Decomposition of the probability of participation in the credit market. The figure is based on Table 3
comparing caste inequalities between various castes in 2005 and 2011-12.
-0.139
-0.007
-0.109
-0.142
0.017
-0.107
-0.15
-0.1
-0.05
0
0.05
GC vs OBC GC vs ST SC vs GC GC vs OBC GC vs ST SC vs GC
2005 2011-12
Particpation in credit market
Explained Unexplained
18
5.4 Decomposing credit differential with the selection effect
Tables 4 and 5 present the decomposition of the log credit amount differentials between
the general caste and all other castes into explained and unexplained component.
Table 4: Observed credit differentials and selection corrected credit differential in 2005
2005
Observed
credit
differential
1
Adjusted
credit
differential
2
Observed
credit
differential
3
Adjusted
credit
differential
4
Observed
credit
differential
5
Adjusted
credit
differential
6
VARIABLES
GC vs OBC
GC vs OBC
GC vs ST
GC vs ST
GC vs SC
SC vs GC
GC
10.025***
12.368***
10.025***
12.368***
10.025***
12.368***
(0.022)
(0.125)
(0.022)
(0.125)
(0.022)
(0.125)
Others
9.555***
11.278***
8.667***
10.213***
9.140***
10.695***
(0.016)
(0.074)
(0.047)
(0.175)
(0.023)
(0.096)
Difference
0.470***
1.090***
1.358***
2.155***
0.885***
1.673***
(0.027)
(0.146)
(0.052)
(0.215)
(0.032)
(0.158)
Explained
0.470***
0.878***
1.213***
1.405***
0.811***
0.948***
(0.021)
(0.031)
(0.051)
(0.059)
(0.028)
(0.034)
Unexplained
0.000
0.212
0.140***
0.751***
0.073***
0.725***
(0.023)
(0.143)
(0.050)
(0.213)
(0.029)
(0.155)
Observations
12,076
12,077
5,600
5,600
8,105
8,105
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The table shows decomposition of log of
loan into explained and unexplained portion. Columns 1, 3 and 5 show results from equation 2. Columns 2, 4 and
6 show results from equation 3 and are corrected for selection and endogeneity.
Table 5: Observed credit differentials and selection corrected credit differential in 2011-12.
2011-12
Observed
credit
differential
1
Adjusted
credit
differential
2
Observed
credit
differential
3
Adjusted
credit
differential
4
Observed
credit
differential
5
Adjusted
credit
differential
6
VARIABLES
GC vs OBC
GC vs OBC
GC vs ST
GC vs ST
GC vs SC
SC vs GC
19
GC
10.723***
13.210***
10.723***
13.210***
10.723***
13.210***
(0.020)
(0.108)
(0.020)
(0.108)
(0.020)
(0.108)
Others
10.386***
12.130***
9.593***
11.614***
9.945***
11.801***
(0.014)
(0.059)
(0.042)
(0.191)
(0.019)
(0.084)
Difference
0.337***
1.081***
1.130***
1.597***
0.778***
1.409***
(0.025)
(0.124)
(0.047)
(0.219)
(0.028)
(0.137)
Explained
0.305***
0.726***
1.02***
1.241***
0.649***
0.784***
(0.016)
(0.027)
(0.046)
(0.051)
(0.025)
(0.029)
Unexplained
0.031***
0.355***
0.103***
0.355
0.128***
0.625***
(0.002)
(0.122)
(0.045)
(0.218)
(0.025)
(0.135)
Observations
15,328
15,349
6,887
6,897
10,268
10,283
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The table shows the decomposition of log
of loan into explained and unexplained portion. Columns 1, 3 and 5 show results from equation 2. Columns 2, 4
and 6 show results from equation 3 and are corrected for selection and endogeneity.
Consistent with earlier results, I find that GC is in a more favourable position in the
Indian credit sector. The observed credit differentials show that the GC has 47 percent
advantage over OBC, 135.8 percent over ST, and 88.5 percent over SC in 2005 and 33.7 percent
advantage over OBC, 113 percent over ST, and 77.8 percent over SC in 2011-12 (see observed
credit differentials in Columns 1, 3 and 4 in Tables 4 and 5). Largely these differences can be
explained by the endowments and personal characteristics, and a very small portion remains
unexplained.
20
Figure 3: Credit differential between the general caste and other lower castes for 2005 and 2011-12. The figure is
based on columns 2, 4 and 6 of Tables 4 and 5 comparing caste inequalities between various castes in 2005 and
2011-12.
However, these results should be treated with caution due to self-selection in the credit
market. The adjusted credit differentials in Tables 4 and 5 show that decomposition results are
sensitive to the selection effect. The credit differentials are underestimated without the
correction for selectivity. The adjusted credit differential increases to 109 percent for OBC,
215.5 percent for ST, and 167.3 percent for SC in 2005, and 108 percent for OBC, increases to
159.7 percent for ST, and 140.9 percent for SC in 2011-12. The credit differentials are largely
explained by the differences in endowments and personal characteristics. Out of the total
differences, 19 percent between GC and OBC, 35 percent between GC and ST and 43 percent
between GC and SC remains unexplained in 2005; and 32 percent between GC and OBC, 22
percent between GC and ST and 44 percent between GC and SC remains unexplained in 2011.
The credit differentials have decreased between the general caste and other lower castes
compared to 2005.
1.09
2.155
1.673
1.081
1.597
1.409
0
0.4
0.8
1.2
1.6
2
2.4
GC vs OBC GC vs ST SC vs GC GC vs OBC GC vs ST SC vs GC
2005 2011-12
Adjusted credit differentials between general and other lower castes
Explained Unexplained
21
Even though lower castes have a higher probability of participation in the credit markets
compared to general caste, the credit amount differential between them is very high. Now I
will discuss the factors that will explain these sharp differences. Table 16 in the appendix shows
the variable decomposition of credit differential.
In 2005, the differences in consumption expenditure is responsible for 28 percent of the
explained share of the total difference between GC and OBC. The differences in states
contribute to 38 percent, and the differences in source of the loan, purpose of the loan and
occupations contribute to 16 percent, and the differences in years of education, age and quality
of the house explain the rest. Similarly, 42 percent of the explained portion of the credit
differences between GC and ST in 2005 arise from the differences in consumption expenditure;
source of the loan, purpose of the loan and occupations explain 19 percent; differences in states
contributes to 12 percent and differences in years of education, age and quality of the house
explain the rest. In the case of credit differences between GC and SC in 2005, differences in
consumption explain around 37 percent; differences in the source of the loan, purpose of the
loan and occupations explain 23 percent, and differences in states contribute to 20 percent and
differences in years of education, age and house quality explain the rest.
In 2011, 26 percent of the explained portion of the total credit differences between GC
and OBC is due to the differences in the level of consumption; differences in states contributes
to 12 percent and differences in source of the loan, purpose of the loan and occupations
contribute to 37 percent; and differences in age, house quality and education years explain the
rest. Similarly, 44 percent of the explained portion of the credit differences between GC and
ST in 2005 arise from the differences in consumption expenditure; source of the loan, purpose
of the loan and occupations explain 15 percent; differences in states contributes to 18 percent
and differences in years of education, age and quality of the house explain the rest. In the case
of credit differences between GC and SC in 2011, differences in consumption explain around
37 percent; differences in the source of the loan, purpose of the loan and occupations explain
26 percent, and differences in states contribute to 14 percent and differences in years of
education, age and house quality explain the rest.
Differences in states explain a very significant portion of the credit gap. Further
investigating the variations in states, I find that states with a significant population of OBC and
ST (such as Chhattisgarh, Madhya Pradesh, Orrisa and Karnataka) increase the credit
differences between GC and OBC and GC and ST. I also find that living in states such as
22
Himachal Pradesh, West Bengal, Kerala, Punjab and Maharashtra reduces the credit
differences between caste. Overall, the characteristics disparity between the general castes and
lower castes are largely due to differences in consumption, location (state), years of education,
house quality, source of the loan, purpose of the loan, and occupations.
5.5 Quantile decomposition
In this section, I apply a quantile regression-based decomposition method proposed by Firpo,
Fortin, and Lemieux (2009) to evaluate caste-based differences in the Indian credit sector
14
.
Their methodology relies on an extension of the Oaxaca-Blinder decomposition which
introduces a two-stage procedure; first, carry out the decomposition based on unconditional
quantile regressions (UQR) techniques using a reweighting approach dividing the
distributional changes into structure effect and a composition effect; second, the two
components are further divided into the contribution of each explanatory variable using
recentred influence function (RIF) regressions
15
.
Figures 4 and 5 (based on Tables 23 and 24 in the Appendix) report the quantile
regression decompositions obtained for three quantiles (10th, 50th, and 90th). The quantile
decomposition suggests that credit gaps between GC and OBC are higher at lower (10th) deciles
compared to upper (90th) and middle (50th) deciles for both time period. The share of the
unexplained component of the gap is also higher at the lower end of the credit distribution,
demonstrating the evidence of sticky floor effect. This suggests that borrowers from the OBC
group may be facing greater discrimination at the lower end of credit distribution. However,
this effect reverses in the higher decile where borrowers from lower castes experience negative
discrimination – credit market favours OBC at the higher end of the distribution.
The credit differences between GC and ST are higher at lower and middle deciles in
2005. The unexplained component of the gap is also higher at the lower and middle deciles in
2005 suggesting sticky floor effect. However, this effect reverses in 2011-12, where the credit
differences and the unexplained gap are higher at the higher quantile. The credit differences
between GC and SC and unexplained component is higher at higher quantile suggesting a glass
ceiling effect. In 2011-12, the pattern is not very clear but the credit differences are higher at
14
I used the Stata program rifreg to estimate the unconditional quantiles. The programme can be downloaded
from here: http://faculty.arts.ubc.ca/nfortin/datahead.html.
15
The Firpo et al. method allows us to decompose the caste gap into the contribution of each individual variable.
However due to the space constraint, I haven’t shown this in the paper.
23
the lower deciles. The unexplained component of the gap is also higher at lower deciles
suggesting a sticky floor effect.
In 2011, sticky floor effect for prevails for OBC and SC borrowers suggesting that the
credit market only favours them at the higher end of the distribution. After correcting for
selection, we see the caste differences reduced at upper quantiles implying that lower caste
borrowers at higher quantiles who self-selected in the credit market got a better return for their
characteristics. These borrowers at the upper end of the distribution are more likely to possess
higher levels of entrepreneurial ability, perseverance and drive which improve their
creditworthiness. They are also aware of their rights and might be in a better position to take
action against perceived discrimination. Lenders aware of these possibilities may not be able
to discriminate at the upper end of credit distribution. Moreover, the credit market at the higher
end would be far more structured and rigidly defined, making it harder to discriminate across
caste. Glass ceiling effect for the ST borrowers in 2011 suggest that borrowers from this caste
group may face higher discrimination at the higher end of the distribution. In 2005, we see
sticky floor effect for OBC and ST borrowers and glass ceiling effect for SC borrowers.
It is generally very difficult to disentangle taste-based discrimination and statistical
discrimination. Both kinds of discrimination can easily coexist in the credit market. The
quantile decomposition shows significant variations in the unexplained component across the
distribution. This suggests that a statistical discrimination effect prevails. If the discrimination
was largely due to taste, it would have been constant across the entire credit distribution.
A possible reason for the sticky floor effect for OBC and SC borrowers in 2011 and OBC
and ST borrowers in 2005 could be due to the statistical discrimination practised in the credit
market. In India, division of labour according to the caste system has frequently prevented
individuals from starting businesses (Iyer, Khanna, Varshney, 2013) and hence, the lack of
business experience at the lower end of credit distribution hurts their credit prospects. Even
when Dalits become entrepreneurs, their businesses could suffer due to discrimination present
in most domestic markets and the lack of suitable social and business networks. Because of
this, lower castes are perceived as less creditworthy and riskier to lend to than upper castes. As
a lower caste borrower moves up the economic ladder, lenders are less like to discriminate
against them and; even favour them.
24
Figure 4: Quantile decomposition of log of loan amount for 2011-12. The figure plots the result from quantile
regression decompositions obtained at 10th, 50th, and 90th percentile.
Figure 5: Quantile decomposition of log of loan amount 2005. The figure plots the result from quantile regression
decompositions obtained at 10th, 50th, and 90th percentile.
1.673
0.842 0.743
1.704
1.114
1.93
1.643
1.373 1.384
10 50 90 10 50 90 10 50 90
GC vs OBC GC vs ST GC vs SC
Quantile decomposition of log of loan amount for 2011-12
Explained Unexplained Total Difference
0.947
0.478
0.875
2.374
1.929
1.374 1.381 1.599
1.907
10 50 90 10 50 90 10 50 90
GC vs OBC GC vs ST GC vs SC
Quantile decomposition of log of loan amount for 2005
Explained Unexplained Total Difference
25
5.6 Decomposing credit outcomes differences by lenders:
Since we have data regarding loan applications and approvals from various sources in the IHDS
(II) 2011-12 survey data, we are able to decompose the caste differences in the probability of
loan application and approval from banks, moneylenders, and social networks. The
decomposition analysis reveals that GC has a lower probability of applying but a higher
approval rate on their loan application in lending from banks compared to OBC (Panel A, Table
17 in the Appendix). However, GC has a higher probability of applying and approval rate in
lending from the bank compared to SC and ST (Panel A, Table 17 in the Appendix). Contrary
to that, all lower castes have a higher probability of applying and approval in lending from
money lenders (Panel B, Table 17 in the Appendix). In lending from social networks, OBC and
SC have a higher probability of applying but lower probability of approval compared to GC,
whereas GC has a higher probability and approval compared to ST (Panel C, Table 17 in the
Appendix).
In the following subsection, I will decompose the credit amount differences between castes
in lending from banks, moneylenders and social networks.
Banks: Banks are one of the major sources of credit for Indian borrowers. 27 percent and 32
percent of all the borrowers in the sample in 2005 and 2011, respectively, took their loan from
banks (see Table 13 in the Appendix). With the development of the banking system in India in
the last decade, all the caste groups have increased their lending from banks. Figure 6 (based
on Table 18 in the Appendix) shows that there are large credit differentials between the general
caste and lower castes. However, the credit differentials between GC and OBC have increased,
whereas the credit differentials between GC and SC and GC and ST have decreased over the
period considered. The share of the unexplained portion of the total differences between GC
and lower castes have increased from 2005. A large portion of the differences between GC and
SC remains unexplained suggesting that banks may be discriminating against the SC
borrowers.
26
Figure 6: Adjusted credit differentials between castes in 2005 and 2011-12 in credit taken from banks.
One ST borrower described the process of bank lending as humiliating:
“…they (the bank) keep sending back our documents for the loan. While the loan officer
collects the documents for the Jaats (upper caste) from their houses, we are not even allowed
to sit on the chairs in the bank or offered any help to fill the complicated forms. While they
photocopy the documents of upper castes in the bank, we are asked to get the photocopies from
outside”
One female Dalit interviewee said:
“….even though my documents were complete, I was asked to sweep the bank floor in return
for opening a bank account. This was despite the bank having a cleaning staff. This is
degrading…”
Some of the credit differences between castes in banking lending could also arise from
repayment enforceability of the financial contract (as shown by Rubin and Kuran, 2018).
Successive Indian governments have passed reforms to ensure a substantial flow of credit to
SC/ST for self-employment at concessional interest rates through priority lending and other
special banking schemes. In some cases, loans to SC and ST entrepreneurs are given interest-
0.82
1.59 1.574
0.988
1.193
1.497
0
0.4
0.8
1.2
1.6
GC vs OBC GC vs ST GC vs SC GC vs OBC GC vs ST GC vs SC
2005 2011
Adjusted credit differentials between castes by banks
Explained Unexplained
27
free
16
and waiving the loan all together is in the process
17
. Reserve Bank of India (RBI) updated
its guidelines on credit facilities to Scheduled Castes and Scheduled Tribes in 2016 giving extra
support to these communities in the formal banking sector
18
. However, this may also have
made the lower caste borrower riskier to lend to and less profitable. In this setting where
lending is biased in favour of SC/ST, banks may resort to minimising the risk by giving less
amount of loans to these communities, imposing an intended cost on them. Hence, these large
credit differences may be echoing inherent conflict between allowing the banking system to be
driven by market forces and expecting greater inclusion from the system.
In the absence of credit history or information regarding borrowers’ creditworthiness,
the unexplained gap is more likely to be due to statistical discrimination
19
. In such cases, banks
may be holding lower caste loan applications to higher standards of creditworthiness than upper
castes. For example, lower caste borrowers are more likely to come from poor areas with a
higher risk of default leading a bank loan officer to grade their loan application strictly. When
a substantial part of statistical discrimination is influenced by profit-maximising actors, market
forces are less likely to eliminate it.
Our qualitative enquiries suggest that bank loan officers (largely belonging to general
caste as observed by Fisman, Paravisini, and Rig, 2017) provide more assistance to higher caste
borrowers in loan applications engaging in a subtle form of statistical discrimination, referred
to as the “thick file” phenomenon. This means that the loan application file of a marginal higher
caste borrower us more likely to be thicker with extra documents than those of a marginal lower
caste borrower. The idea here is that upper caste loan officers may have less cultural affinities
with and less knowledge of lower caste applicants. They are more likely to be strict with lower
caste applications, relying on the group characteristics rather than investing resources in
16
Under Chief Minister Scheduled Caste and Scheduled Tribes Entrepreneur Scheme, Bihar government will
provide interest free loans to eligible entrepreneurs from scheduled caste and scheduled tribes category.
17
In the state of Karnataka, the state president has requested to the state government to waive education loans of
the SC/ST students.
18
Under new recommendations, banks are responsible for increasing awareness about new credit facilities among
SC/ST borrowers and helping the borrowers in filling out forms and completing other formalities. Loan proposals
from these communities are encouraged to be considered with utmost sympathy and understanding. To ensure
these policies are followed, a special department has been set up for monitoring the flow of credit to SC/ST
beneficiaries. Under the same guidelines, the Ministry of Rural Development, Government of India has launched
Deendayal Antyodaya Yojana-National Rural Livelihood Mission (DAY-NRLM), which would seek to ensure
adequate coverage of vulnerable sections of the society such that 50% of these beneficiaries are SC/ST. Under
Differential Rate of Interest Scheme, banks will provide finance up to Rs 15,000 at a concessionary rate of interest
of 4 percent per annum to the lower castes for engaging in productive and gainful activities.
19
Bertrand and Mullainathan (2004) show that more information regarding minority applicants’ skills does not
always reduce discrimination in the labour market.
28
gathering more information on the creditworthiness of lower caste borrowers. In such a
situation, extra documentation providing mitigating information could positively affect the
credit outcome of marginal upper caste applications. Although this phenomenon has some
credibility, further investigation is needed to documents its relevance and occurrence.
Money lenders: Although the share of money lenders has reduced significantly over this time,
they still play a major role in financing lower caste borrowers (see Table 13 in the Appendix).
However, Figure 7 (based on Table 19 in the Appendix) shows that the credit differentials
between the general caste and other lower castes have increased since 2005, and a large part of
this gap is unexplained. Since money lenders provide credit to people in regions where formal
finance has not reached; or to borrowers who are not creditworthy for banks and MFIs, the
increase in the credit differentials and unexplained component is worrying.
Figure 7: Adjusted credit differentials between castes in 2005 and 2011-12 in credit taken from money lenders
The qualitative interviews confirm the discriminatory attitude practised by informal money
lenders towards lower castes.
0.66
1.623
1.076
0.805
1.728
1.084
0
0.4
0.8
1.2
1.6
GC vs OBC GC vs ST GC vs SC GC vs OBC GC vs ST GC vs SC
2005 2011
Adjusted credit differentials between castes by money lenders
Explained Unexplained
29
One respondent said:
“… my local money lender always says that people from my caste cannot be trusted, even
though I have never defaulted on a loan in my life. The conditions they set are always
discriminatory. Upper caste lenders think that if they give loans to lower castes, we might
become rich and less dependent on them.”
One Dalit entrepreneur said:
“The bayaj (interest rate) varies depending on your caste. Dalits are also expected to offer
collateral (security) far in excess of the loan amount, and far in excess of other castes”.
These informal money lenders, generally belonging to upper castes, have historically been the
main source of financial credit for lower castes. In this sector of the credit market,
discrimination is frequently overt and extreme. The informal lender I interviewed didn’t
dispute the fact that they discriminate, and based their arguments on old-fashioned prejudice.
One money lender feared loss of face in dealing with lower castes:
“…….if a Dalit dared to default on my loan, people would laugh at me ”
Another questioned the whole idea of Dalit entrepreneurs:
“…if they all have businesses, who will work in our fields or clean our toilets?”
In informal lending, money lenders can force repayments through panchayats (village councils
usually consist of upper caste men whose verdicts are largely partial to the money lenders) or
by keeping the collateral given as a security for the credit. Therefore, the approval rate of lower
caste is higher than general caste in lending from money lenders. Logically, the credit
differentials between general and other lower castes and the unexplained portion of these
differentials should be lower since lower castes are less risky, however, this is not the case
here. Moneylenders are sometimes the last resort of credit for poor and lower caste households.
The qualitative enquiries suggest the informal lenders practice an extreme form of
discrimination against lower caste and are reluctant to fund lower caste entrepreneurs. Hence,
low risk of giving credit to lower caste reduces the credit differences, while discrimination
practised by moneylenders increases it. The results suggest that the latter effect prevails.
30
The qualitative enquiries also suggest that taste-based discrimination is more likely to
be present in lending from money lenders. In the case of better information regarding the
borrower's creditworthiness, as usually in informal lending, the unexplained or discriminatory
component in the total gap is more likely to be due to taste-based discrimination. Berkovec et
al (1994) suggest that taste-based discrimination is likely to be higher when the lenders have
higher market power.
Social networks such as friends and relatives: Credit in the informal sector is highly
segmented, and is based around people of the same caste, religion and kinship (Gupta and Mitra
2002). Hence, poor and lower castes are significantly disadvantaged due to a lack of networks,
income, land, and education in obtaining loans from friends and relatives. The proportion of
those taking loans from friends and relatives have marginally changed over the years. Figure 8
(based on Table 20 in the Appendix) shows that the credit differentials in lending from social
networks such as friends and family have decreased significantly between GC and other lower
castes.
Figure 8: Adjusted credit differentials between castes by credit taken from friends and relatives
However, these changes in credit differentials above are not easy to explain without significant
additional research. The individual results from each year are perhaps easier to interpret: if
1.034
2.397
1.571
0.812
1.064
0.985
-0.4
0
0.4
0.8
1.2
1.6
2
2.4
GC vs OBC GC vs ST GC vs SC GC vs OBC GC vs ST GC vs SC
2005 2011
Adjusted credit differentials between castes by friends and relatives
Explained Unexplained
31
lower caste borrowers seek credit from lower caste friends, while general caste borrowers seek
credit from general caste friends, then it is logical that there would be a wide differential in the
availability and ease of credit – in this scenario, the general caste ‘lenders’ simply have more
credit available to give. It is, though, difficult to explain why this differential has changed so
dramatically over the time period unless the overall level of wealth within the pool of lower
caste lenders has increased at a greater rate than that of their general caste equivalents – and
there is little evidence to suggest that this is the case. This could also be due to the lower caste
abandoning formal channels of finance because of the poor treatment and discouragement,
thereby increasing their reliance on their own caste.
In terms of this study, however, the reasons for the change are not directly relevant. What
is relevant is that this is the only category of lending in which both borrower and lender are
likely to share the same caste, and it is also the only category which shows improvement in the
credit differentials between the general caste and all three lower castes. Clearly, this represents
a positive development, but it also indicates that caste-based discrimination may be a
significant factor in driving the credit differentials in other types of lending. In other words,
caste-based differences may be decreasing, but only if lower caste borrowers are borrowing
from lower caste lenders.
5.7 Decomposing credit differences by residence:
Figure 9 (based on Tables 21 and 22 in the Appendix) presents the caste differences in the
credit amount by place of residence – urban and rural areas. There are stark differences between
credit differentials in rural and urban areas. Compared to 2005, the credit differences between
GC and other lower castes have increased in the urban area and decreased in the rural area. The
credit differences between GC and OBC and GC and ST are larger in the rural compared to the
urban area in 2005 while the credit differences between GC and other lower castes are higher
in urban area in 2011.
32
Figure 9: The figure compares credit inequalities between various caste in urban and rural areas in 2011-12 and
2005.
While caste may be losing its relevance in traditional custom in urban areas, caste
differences and prejudices are being reinforced by high gaps in credit amount. The credit gap
has increased over this time suggesting that situation of lower caste has actually worsened in
urban areas. A greater proportion of the lower castes live in rural areas, and the credit
differentials between upper castes and lower castes, although decreased over the years, are
overwhelmingly high. Successive Indian governments have failed to improve the village
banking infrastructure in India. Even though 70 percent of India’s population lives in rural
areas, they only have 37 percent of the total number of bank branches of the country (Reserve
Bank of India, 2015). Thus, a significant proportion of rural households, especially lower
castes, are still outside the formal fold of the banking system.
6. Discussion and conclusion
The study of discrimination in economics is motivated both by the moral case for equality
and the consequent loss of efficiency in the market. Advances in research methods and designs
have produced a significant interest in the field which has generated new insights into the nature
0.569
1.157
1.919
1.157
2.213
2.004
1.158
1.934
1.544
0.901 1.068 1.081
-0.4
0.1
0.6
1.1
1.6
2.1
GC vs
OBC GC vs
ST GC vs
SC GC vs
OBC GC vs
ST GC vs
SC GC vs
OBC GC vs
ST GC vs
SC GC vs
OBC GC vs
ST GC vs
SC
2005 2011-12 2005 2011-12
Urban Rural
Adjusted credit differentials between general and other lower castes by
residence
Explained Unexplained
33
of discrimination. Guryan and Charles (2013) argue that a deeper understanding of the sources
and causes of discrimination is needed in order to formulate policies to reduce its incidence
and effects; however, in order to do this, it is first necessary to identify the nature and scale of
discrimination clearly. That is the focus of our study.
The empirical evidence in this paper suggests that caste is still a worryingly potent
determinant of lending outcomes in India. There are substantial credit differences between the
general caste and other lower castes and these differences have decreased over the years. A
significant part of the credit differences between general caste and other lower castes (specially
SC) remains unexplained. Hence, it can be argued that the disparities between the loans granted
to general castes and other lower castes in India are not only because lower castes possess less
human and physical capital than general castes, but also because these groups may be facing
extensive and persistent discrimination in the credit sector. I also find that the loan application
and approval rates are higher for general caste in the formal sector whereas lower castes are
generally more likely to participate and have a higher approval on their loans in informal sector.
However, the differences in the amount of credit granted is still a cause of apprehension, and
in some cases, the situation appears to be growing worse, not better.
Using a quantile regression-based decomposition method I analysed the caste gap across
the entire credit distribution. I found the evidence of sticky floor effect in lending to OBC (for
both time period), ST (in 2005) and SC (in 2011-12), whereas glass ceiling effect prevails in
lending to SC (in 2005) and ST (in 2011-12). It’s also important to note that there are large
credit differentials between the general caste and lower castes in almost every instance in
question: this includes rural and urban areas, credit taken from banks, money lenders and social
network. In many instances, the credit differentials have actually increased over the time period
considered in this research.
In attempting to explain the results, I recognise that the unexplained portion may include
unmeasurable or unobservable characteristics, for instance, drive, determination or other
attitudes which are likely to affect the credit outcomes and thus, it does not necessarily mean
explicit discrimination against lower castes. It is worth noting that the analysis seeks to measure
the effects of a social variable named ‘caste’ which is itself composed of a number of ill-defined
and unquantifiable elements. For instance, lower castes may possess higher levels of
unmeasured characteristics like perseverance and determination which improve their
34
creditworthiness but display more traits such as humility and lowered expectations which limit
their credit requests. Hence, the issue of the unexplained components including the effects of
unobservable or unmeasurable characteristics is a standard limitation in the decomposition
analysis.
Altogether, the evidence is consistent that lower caste individuals are disadvantaged in
the credit sector. Recognising this, the Indian government has launched various programmes
to improve the provision of financial services to the lower castes. However, the government
can only play a direct role in the formal sector. Since banks and other government programmes
have become the major source of finance for borrowers in India, a broader intervention from
the government is much needed. Furthermore, the differences in credit amount sanctioned and
loan approval rates between the general caste and other lower castes in lending from banks are
high. Schemes to promote the economic empowerment of lower castes through finance have
been implemented on a large scale since the 1990s, but if we take anything from the results in
this research, they have not been very effective.
A large endowments difference between social groups indicates that there is a need to
promote educational and training opportunities for the lower castes. The government should
also ensure that the disadvantaged sections of society get full participation in schooling,
employment, health programmes to reduce pre-market discrimination. Policy-makers need to
adopt a broader range of strategies to tackle the deep-seated and multi-faceted challenge of
systemic discrimination. Initiatives need to include the improvement of financial literacy across
lower castes, encouragement of positive discrimination, improving the functioning and
competitiveness of the financial sector, active monitoring of caste bias, and more focused social
research into the causes and nature of caste discrimination.
7. References
Agrawal, T. (2014). Gender and caste-based wage discrimination in India: some recent
evidence. Journal for Labour Market Research, 47(4), 329-340.
Ahuja, A., & Ostermann, S. L. (2016). Crossing caste boundaries in the modern Indian
marriage market. Studies in Comparative International Development, 51(3), 365-387.
Aristei, D., & Gallo, M. (2016). Does gender matter for firms' access to credit? Evidence from
international data. Finance Research Letters, 18(C), 67-75.
35
Arrow, K. (1973). The theory of discrimination. Discrimination in labor markets, 3(10), 3-33.
Balasubramaniam, D., Chatterjee, S., & Mustard, D. B. (2014). Got water? Social divisions
and access to public goods in rural India. Economica, 81(321), 140-160.
Banerjee, A., & Somanathan, R. (2007). The political economy of public goods: Some
evidence from India. Journal of Development Economics, 82(2), 287-314.
Banerjee, A. V., Breza, E., Duflo, E., & Kinnan, C. (2017). Do credit constraints limit
entrepreneurship? Heterogeneity in the returns to microfinance. Heterogeneity in the Returns
to Microfinance. Buffett Institute Global Poverty Research Lab Working Paper, (17-104).
Banerjee, B., & Knight, J. B. (1985). Caste discrimination in the Indian urban labour
market. Journal of Development Economics, 17(3), 277-307.
Becker, G. (1971). The Economics of Discrimination.
Berkovec, James A & Canner, Glenn B. & Gabriel, Stuart A. & Hannan, Timothy H., 1994.
"Race, Redlining, and Residential Mortgage Loan Performance," The Journal of Real
Estate Finance and Economics, Springer, vol. 9(3), pages 263-294.
Bhattacharjee, M., & Rajeev, M. (2014). Accessibility to Credit and its Determinants: A State-
level Analysis of Cultivator Households in India. Margin: The Journal of Applied
Economic Research, 8(3), 285–300. https://doi.org/10.1177/0973801014531137
Blinder, A. S. (1973). Wage discrimination: reduced form and structural estimates. Journal of
Human resources, 436-455.
Borooah, V. K. (2005). Caste, inequality, and poverty in India. Review of Development
Economics, 9(3), 399-414.
Borooah, V. K., & Iyer, S. (2005). Vidya, Veda, and Varna: The influence of religion and caste
on education in rural India. The Journal of Development Studies, 41(8), 1369-1404.
Chowdhury, M.J.A., Ghosh, D. and Wright, R.E., (2005). The impact of micro-credit on
poverty: evidence from Bangladesh. Progress in Development studies, 5(4), pp.298-309.
Cotton, J. (1988). On the Decomposition of Wage Differentials. The Review of Economics and
Statistics, 70(2), 236-243.
Das, M. B., & Dutta, P. V. (2007). Does caste matter for wages in the Indian labor market. Draft
Paper, World Bank, Washington, DC.
Demirgüç-Kunt, A., & Levine, R. (2009). Finance and inequality: Theory and evidence.
Annual Review of Financial Economics.
36
Desai, S., and Kulkarni, V. (2008). Changing educational inequalities in India in the context of
affirmative action. Demography, 45(2), 245-270.
Deshpande, A. (2011). The Grammar of Caste: Economic Discrimination in Contemporary
India. OUP Catalogue.
Deshpande, A. (2000). Does caste still define disparity? A look at inequality in Kerala,
India. American Economic Review, 90(2), 322-325.
Deshpande, A., & Sharma, S. (2014). Is Self-Employment the Answer to Caste Discrimination?
Decomposing the Earnings Gap in Indian Household Nonfarm Businesses.
Doan, T., Gibson, J. and Holmes, M., (2014). Impact of household credit on education and
healthcare spending by the poor in peri-urban areas, Vietnam. Journal of Southeast Asian
Economies (JSEAE), 31(1), pp.87-103.
Firpo, S., Fortin, N. M., & Lemieux, T. (2009). Unconditional quantile
regressions. Econometrica, 77(3), 953-973.
Fisman, R., Paravisini, D., & Vig, V. (2017). Cultural proximity and loan outcomes. American
Economic Review, 107(2), 457-92.
Gang, I. N., Sen, K., & Yun, M. Caste, Ethnicity and Poverty in Rural India.
Government of India. (2011). Census of India. Ministry of Home Affairs. New Delhi.
Government of India (2013). Report No. 68/1.0, Key Indicators of Household Consumer
Expenditure in India, NSS 68th Round (July 2011–June 2012).
Gupta, I., and Mitra, A. (2002). Rural migrants and labour segmentation: micro-level evidence
from Delhi slums. Economic and Political Weekly, , 163-168.
Guryan, J., and Charles, K. K. (2013). Taste‐based or Statistical Discrimination: The
Economics of Discrimination Returns to its Roots. The Economic Journal, 123(572),
F432.
Ladd, H. F. (1998). Evidence on discrimination in mortgage lending. Journal of Economic
Perspectives, 12(2), 41-62.
Heckman, J. J. (1979). Sample Selection Bias as a Specification Error. Econometrica, 47(1),
153-161.
Hnatkovska, V., Lahiri, A., & Paul, S. (2012). Castes and labor mobility. American Economic
Journal: Applied Economics, 4(2), 274-307.
37
Hoff, K., & Pandey, P. (2004). Belief systems and durable inequalities: An experimental
investigation of Indian caste. The World Bank.
Iyer, L., Khanna, T., & Varshney, A. (2013). Caste and entrepreneurship in India. Economic
and Political Weekly, 52-60.
Jann, B. (2008). The Blinder-Oaxaca decomposition for linear regression models. The Stata
Journal, 8(4), 453-479.
Jodhka, S. S. (2010). Dalits in Business: Self-Employed Scheduled Castes in North-West
India. Economic & Political Weekly, 45(11), 41.
Kaboski, J. P., & Townsend, R. M. (2012). The impact of credit on village
economies. American Economic Journal: Applied Economics, 4(2), 98-133.
Kijima, Y. (2006). Caste and tribe inequality: evidence from India, 1983–1999. Economic
Development and Cultural Change, 54(2), 369-404.
Kochar, A. (1997). An empirical investigation of rationing constraints in rural credit markets
in India. Journal of Development Economics, 53(2), 339-371.
Kumar, S. M. (2016). Why does caste still influence access to agricultural credit?
Kuran, T., & Rubin, J. (2017). The financial power of the powerless: socio‐economic status
and interest rates under partial rule of law. The Economic Journal, 128(609), 758-796.
Munshi, K., & Rosenzweig, M. (2006). Traditional Institutions Meet the Modern World: Caste,
Gender, and Schooling Choice in a Globalizing Economy. American Economic
Review, 96(4), 1225-1252.
Narula, S. (1999). Broken People: Caste Violence Against India's" untouchables". Human
Rights Watch.
Neumark, D. (1988). Employers' Discriminatory Behavior and the Estimation of Wage
Discrimination. Journal of Human Resources, 279-295.
National Campaign on Dalit Human Rights. (2009). National Dalit Watch Annual Report 2009
Oaxaca, R. (1973). Male-Female Wage Differentials in Urban Labor Markets. International
Economic Review, 14(3), 693-709.
Phelps, E. S. (1972). The Statistical Theory of Racism and Sexism. American Economic
Review, 62(4), 659-661.
38
Reserve Bank of India. (2016) Master Circular Priority Sector Lending-Credit facilities to
Scheduled Castes (SCs) & Scheduled Tribes (STs).
Sarap, K. (1991). Collateral and other forms of guarantee in rural credit markets: evidence
from eastern India. Indian Economic Review, 167-188.
Sargan, J. D. (1958). The estimation of economic relationships using instrumental
variables. Econometrica: Journal of the Econometric Society, , 393-415.
Sartori, A. E. (2003). An Estimator for Some Binary-Outcome Selection Models Without
Exclusion Restrictions. Political Analysis, 11(2), 111-138.
Saxena, V., & Bhattacharya, P. C. (2018). Inequalities in LPG and electricity consumption in
India: The role of caste, tribe, and religion. Energy for Sustainable Development, 42, 44-
53.
Sen, B. (2014). Using the Oaxaca–Blinder decomposition as an empirical tool to analyze racial
disparities in obesity. Obesity, 22(7), 1750-1755.
Sharma, S. (2015). Caste-based crimes and economic status: Evidence from India. Journal of
Comparative Economics, 43(1), 204-226.
Thorat, S. (2005). Caste, Social Exclusion and Poverty Linkages–Concept, Measurement and
Empirical Evidence. Concept Paper for PACS, New Delhi, October,
Thorat, S., & Attewell, P. (2007). The legacy of social exclusion: A correspondence study of
job discrimination in India. Economic and Political Weekly, 4141-4145
Thorat, S. (2009). Dalits in India: search for a common destiny. SAGE Publications Ltd.
Thorat, S., & Neuman, K. S. (2012). Blocked by caste: economic discrimination in modern
India. Oxford University Press.
8. Appendix: Tables
Table 6: Primary income generating occupational activities of various caste by percentage
2011-12
2005
39
Occupational activities
GC
OBC
ST
SC
GC
OBC
ST
SC
Cultivation
25.89
26.52
36.48
13.63
23.24
26.97
35
12.84
Allied agriculture
0.77
1.27
0.85
0.63
1.02
0.98
0.44
0.76
Agricultural wage labour
4.06
8.99
15.48
18.08
5.76
12.61
22
24.76
Non-agricultural wage labour
12.51
23.38
22.87
33.82
10.81
17.7
17
28.45
Artisan/Independent
1.58
1.91
0.8
1.35
5.26
8.01
2
4.9
Petty shop/Small business
13.72
12.87
4.39
6.87
5.68
5.03
2.44
2.54
Organized Trade/Business
2.4
1.32
0.41
0.4
9.02
5.58
2.06
2.92
Salaried employment
26.67
15.81
14.69
18.45
28.56
16.25
15
17.38
Other Professions
1.04
0.45
0.22
0.39
1.41
0.98
0.44
0.64
Pension/Rent/Dividend
8.23
4.28
2.72
3.7
6.05
3.25
2.27
2.51
Others
3.14
3.2
1.1
2.67
3.19
2.63
1.37
2.32
Table 7: Purpose of the loan in 2011-2012 and 2005
Purpose
GC
OBC
ST
SC
Purpose
GC
OBC
ST
SC
2011-12
2005
House
15.56
14.24
12.96
16.46
House
19
15.21
14.16
14.16
Land*
1.97
1.53
1.45
1.05
Land*
1.21
0.94
0.52
0.52
Marriage
13.92
17.07
18.68
19.76
Marriage
13.05
15.58
12.78
12.78
Agriculture*
18.38
18.21
23.14
9.8
Agri/business*
35.13
32.88
33.33
33.33
Business*
10.25
7.77
4.47
5.35
Consumption
8.38
12.29
18.05
18.05
Consumption
13.16
13.8
16.6
15.65
Car/appliance
2.76
1.1
0.78
0.78
Car/Jeep
3.34
1.22
1.01
0.95
Education*
2.88
2.44
1.81
1.81
Two-wheeler
1.23
1.02
0.44
0.97
Medical
10.89
13.83
12.09
12.09
Truck/Bus*
0.56
0.3
0.38
0.1
Other
6.69
5.73
6.48
6.48
Educational*
5.61
4.91
3.77
4.69
Medical Exp
12.21
15.71
13.77
19.76
Others
3.82
4.22
3.33
5.45
* Loans for productive purposes
Table 8: Source of the loan in 2011-12 and 2005
Source
GC
OBC
ST
SC
GC
OBC
ST
SC
40
2011-12
2005
Employer
2.36
2.06
2.45
3.64
2.00
1.65
1.81
1.91
Money Lender
10.22
20.19
20.57
24.81
19.47
33.58
32.9
42.17
Friend
10.91
10.3
11.19
11.71
8.86
9.13
13.21
10.32
Relative
19.85
21.45
23.77
21.23
18.39
19.19
19.78
17.41
Bank*
43.81
32.31
26.42
22.9
37.36
26.28
21.07
19.62
NGO*
0.91
0.78
1.45
1.39
0.13
0.17
0.26
0.14
Credit Group*
3.28
2.29
2.39
3.12
2.61
2.06
3.97
1.83
Govt. Program*
1.02
0.48
0.38
0.64
1.44
0.8
1.64
1.12
Self-help group*
3.45
6.04
6.92
7.83
9.76
7.15
5.35
5.49
Kisan Credit*
2.06
2.53
2.58
0.85
Prov Funds*
0.3
0.19
0.13
0.02
Suppliers*
0.26
0.22
0.31
0.4
Others
1.58
1.16
1.45
1.47
* Loan from formal sources.
Table 9: Application for loans from various sources
All
GC
OBC
SC
ST
Banks
Didn't apply
75
75
72
81
82
Rejected
3
3
3
3
4
Approved
22
22
25
16
15
Money Lenders
Didn't apply
81.5
89
77.5
78
84
Rejected
3.5
3.5
4
3
3
Approved
15
8
18.5
19
13
Relative and friends
Didn't apply
70.5
76.5
66.5
70
71.5
Rejected
3.5
3.5
4
3.5
4
Approved
26
20
29.5
26.5
24.5
Note: Numbers in percentages
Table 10: The marginal effect of participation in the credit market by various castes in 2005.
(1)
(2)
(3)
(4)
VARIABLES
GC
OBC
ST
SC
CONSUMPTION
0.105***
0.142***
0.170***
0.155***
(0.008)
(0.008)
(0.018)
(0.011)
41
AGE
0.005***
0.007***
0.006
0.001
(0.002)
(0.002)
(0.004)
(0.003)
AGESQ
-0.000***
-0.000***
-0.000**
-0.000
(0.000)
(0.000)
(0.000)
(0.000)
EDUCATION
-0.007***
-0.008***
-0.008***
-0.007***
(0.001)
(0.001)
(0.003)
(0.002)
LAND
0.065***
0.080***
0.070***
0.104***
(0.014)
(0.013)
(0.025)
(0.016)
SEX
-0.054***
-0.060***
0.010
-0.042**
(0.015)
(0.015)
(0.032)
(0.020)
URBAN RURAL
-0.076***
-0.083***
-0.122***
-0.035**
(0.013)
(0.012)
(0.034)
(0.017)
HOUSE QUALITY
-0.037***
-0.044***
0.053**
-0.009
(0.012)
(0.011)
(0.026)
(0.015)
STATE DUMMIES
YES
YES
YES
YES
OCCUPATIONAL
DUMMIES
YES
YES
YES
YES
Observations
13,332
16,213
3,270
8,304
Delta standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is 1 if the
client has taken a loan or 0 otherwise. All predictors at their mean value. The independent variables are:
log of consumption, age, sex of the head of the household, size of the household, number of education
years completed by the head, log of amount of land, dummy whether the household has a ration card,
dummy for quality of the house (good/bad), dummy whether household is in urban area, and various
dummies for occupation and the state where the household is located.
Table 11: The marginal effect of participation in the credit market by various castes in 2011-12
(1)
(2)
(3)
(4)
VARIABLES
GC
OBC
ST
SC
CONSUMPTION
0.119***
0.132***
0.113***
0.141***
(0.009)
(0.007)
(0.018)
(0.010)
AGE
0.009***
0.013***
0.021***
0.013***
(0.002)
(0.002)
(0.005)
(0.003)
AGESQ
-0.000***
-0.000***
-0.000***
-0.000***
(0.000)
(0.000)
(0.000)
(0.000)
EDUCATION
-0.009***
-0.009***
0.004
-0.007***
(0.001)
(0.001)
(0.003)
(0.001)
42
LAND
0.078***
0.078***
0.092***
0.062***
(0.015)
(0.012)
(0.025)
(0.016)
SEX
-0.048***
-0.057***
-0.109***
-0.033*
(0.015)
(0.012)
(0.029)
(0.017)
URBAN RURAL
-0.083***
-0.093***
-0.024
-0.058***
(0.015)
(0.011)
(0.035)
(0.016)
HOUSE QUALITY
-0.034**
-0.031***
0.003
-0.035**
(0.014)
(0.011)
(0.026)
(0.014)
STATE DUMMIES
YES
YES
YES
YES
OCCUPATIONAL
DUMMIES
YES
YES
YES
YES
Observations
11,680
16,763
3,590
8,807
Delta standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Same as Table 3
Table 12: Probit model for 2005
(1)
(2)
(3)
(4)
VARIABLES
GC
OBC
ST
SC
CONSUMPTION
0.304***
0.358***
0.482***
0.397***
(0.022)
(0.019)
(0.050)
(0.029)
AGE
0.016***
0.018***
0.017
0.002
(0.006)
(0.005)
(0.012)
(0.007)
AGE SQ
-0.000***
-0.000***
-0.000**
-0.000
(0.000)
(0.000)
(0.000)
(0.000)
EDUCATION
-0.021***
-0.021***
-0.023***
-0.017***
(0.003)
(0.003)
(0.008)
(0.004)
LAND OWN
0.187***
0.201***
0.200***
0.266***
(0.040)
(0.032)
(0.071)
(0.042)
SEX HEAD
-0.156***
-0.151***
0.027
-0.108**
(0.044)
(0.038)
(0.092)
(0.052)
URBAN
-0.218***
-0.209***
-0.348***
-0.091**
(0.037)
(0.030)
(0.096)
(0.043)
HOUSE QUALITY
-0.107***
-0.112***
0.151**
-0.023
(0.034)
(0.028)
(0.073)
(0.038)
STATE DUMMIES
YES
YES
YES
YES
OCCUPATIONAL
YES
YES
YES
YES
43
DUMMIES
Constant
-3.673***
-3.047***
-4.932***
-3.314***
(0.517)
(0.276)
(0.629)
(0.501)
Observations
13,332
16,213
3,270
8,304
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is 1 if the client
has taken a loan or 0 otherwise. The independent variables are: log of consumption, age, sex of the head of the
household, size of the household, number of education years completed by the head, log of amount of land,
dummy whether the household has a ration card, dummy for quality of the house (good/bad), dummy whether
household is in urban area, and various dummies for occupation and the state where the household is located.
Table 13: Probit Model for 2011-12
(1)
(2)
(3)
(4)
VARIABLES
GC
OBC
ST
SC
CONSUMPTION
0.301***
0.345***
0.290***
0.360***
(0.022)
(0.018)
(0.046)
(0.027)
AGE
0.024***
0.034***
0.053***
0.032***
(0.006)
(0.005)
(0.012)
(0.007)
AGE SQ
-0.000***
-0.000***
-0.001***
-0.000***
(0.000)
(0.000)
(0.000)
(0.000)
EDUCATION
-0.022***
-0.023***
0.010
-0.017***
(0.003)
(0.003)
(0.007)
(0.004)
LAND OWN
0.198***
0.203***
0.235***
0.157***
(0.038)
(0.030)
(0.064)
(0.040)
SEX HEAD
-0.122***
-0.148***
-0.279***
-0.084*
(0.039)
(0.033)
(0.074)
(0.043)
URBAN
-0.208***
-0.242***
-0.062
-0.147***
(0.037)
(0.029)
(0.090)
(0.040)
HOUSE QUALITY
-0.087**
-0.080***
0.007
-0.090**
(0.036)
(0.028)
(0.066)
(0.035)
44
STATE DUMMIES
YES
YES
YES
YES
OCCUPATIONAL
DUMMIES
YES
YES
YES
YES
Constant
-3.872***
-3.393***
-3.306***
-4.127***
(0.862)
(0.267)
(0.605)
(0.503)
Observations
11,680
16,763
3,590
8,807
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Same as Table 15.
Table 14: Selection corrected loan amount equation estimates for 2005
(1)
(2)
(3)
(4)
VARIABLES
GC
OBC
ST
SC
CONSUMPTION
0.684***
0.848***
1.257***
0.318*
(0.152)
(0.129)
(0.331)
(0.177)
AGE
-0.022**
-0.038***
-0.014
0.026**
(0.009)
(0.007)
(0.018)
(0.011)
AGESQ
0.000***
0.000***
0.000
-0.000
(0.000)
(0.000)
(0.000)
(0.000)
EDUCATION
0.047***
0.041***
0.026**
0.037***
(0.006)
(0.004)
(0.011)
(0.006)
LAND UNIT
-0.084***
-0.063***
-0.082*
-0.031
(0.022)
(0.017)
(0.042)
(0.024)
SEX
0.322***
0.486***
0.277**
0.183***
(0.067)
(0.050)
(0.138)
(0.071)
URBAN
0.454***
0.291***
0.056
0.338***
(0.060)
(0.047)
(0.189)
(0.057)
HOUSE QUALITY
0.198***
0.188***
-0.042
0.176***
(0.058)
(0.042)
(0.116)
(0.058)
MILLS
-2.589***
-2.414***
-1.837***
-2.095***
(0.136)
(0.101)
(0.206)
(0.125)
Constant
6.948***
2.913*
-5.794
9.958***
45
(1.687)
(1.491)
(3.792)
(1.949)
LOAN PURPOSE
YES
YES
YES
YES
LOAN SOURCE
YES
YES
YES
YES
OCCUPATIONAL
DUMMIES
YES
YES
YES
YES
STATE DUMMIES
YES
YES
YES
YES
Observations
4,444
7,633
1,156
3,661
R-squared
0.511
0.460
0.628
0.496
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The dependent variable is the
log of amount of loan. The independent variables are age, age square, number of education years
completed, unit of land owned, predicted values of the first stage regression replacing the original value
of log of consumption, and various dummies for loan source, its purpose and the state where the
household is located.
Table 15: Selection corrected loan amount equation estimates for 2011-12
(1)
(2)
(3)
(4)
VARIABLES
GC
OBC
ST
SC
CONSUMPTION
0.809***
0.484***
-0.225
0.528***
(0.160)
(0.102)
(0.500)
(0.148)
AGE
-0.047***
-0.049***
-0.074***
-0.044***
(0.008)
(0.007)
(0.025)
(0.009)
AGESQ
0.001***
0.001***
0.001***
0.001***
(0.000)
(0.000)
(0.000)
(0.000)
EDUCATION
0.050***
0.044***
0.013
0.028***
(0.005)
(0.003)
(0.013)
(0.005)
LAND UNIT
-0.085***
-0.034***
-0.059**
-0.055***
(0.017)
(0.011)
(0.026)
(0.017)
SEX
0.381***
0.372***
0.453***
0.295***
(0.053)
(0.038)
(0.120)
(0.056)
URBAN
0.483***
0.468***
0.519***
0.223***
(0.055)
(0.036)
(0.157)
(0.052)
HOUSE QUALITY
0.206***
0.267***
0.342***
0.257***
46
(0.053)
(0.037)
(0.115)
(0.043)
MILLS
-3.273***
-3.121***
-3.050***
-3.012***
(0.140)
(0.103)
(0.287)
(0.134)
Constant
3.688**
8.339***
17.829***
7.108***
(1.870)
(1.319)
(5.562)
(1.634)
LOAN SOURCE
DUMMIES
YES
YES
YES
YES
LOAN PURPOSE
DUMMIES
YES
YES
YES
YES
OCCUPATIONAL
DUMMIES
YES
YES
YES
YES
STATE DUMMIES
YES
YES
YES
YES
Observations
5,316
10,012
1,571
4,952
R-squared
0.447
0.47
0.630
0.473
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. Same as Table 6
Table 16: Decomposition of the log of credit amount differential for a selection corrected equation in
2005 and 2011-12
(1)
(2)
(3)
(4)
(5)
(6)
VARIABLES
GC VS OBC
GC VS ST
GC VS SC
GC vs OBC
GC vs ST
GC vs SC
2005
2011-12
CONSUMPTION
0.249***
0.592***
0.353***
0.190***
0.554***
0.292***
(0.027)
(0.077)
(0.047)
(0.019)
(0.070)
(0.036)
AGE
-0.038***
-0.037
0.002
-0.091***
-0.119***
-0.117***
(0.010)
(0.025)
(0.022)
(0.014)
(0.027)
(0.024)
AGE SQ
0.048***
0.074***
0.029
0.112***
0.140***
0.154***
(0.012)
(0.026)
(0.021)
(0.016)
(0.028)
(0.024)
EDUCATION
0.081***
0.162***
0.109***
0.074***
0.104***
0.103***
(0.007)
(0.019)
(0.012)
(0.006)
(0.015)
(0.010)
LAND UNIT
-0.008***
0.004
-0.023***
-0.007***
0.003
-0.034***
(0.002)
(0.003)
(0.007)
(0.002)
(0.003)
(0.007)
URBANRURAL
0.027***
0.094***
0.037***
0.025***
0.073***
0.027***
(0.004)
(0.014)
(0.006)
(0.004)
(0.011)
(0.004)
47
SEX
-0.001
0.001
-0.002
0.001
0.003
-0.002
(0.002)
(0.002)
(0.002)
(0.002)
(0.003)
(0.002)
HOUSE QUALITY
0.022***
0.059***
0.028***
0.017***
0.060***
0.040***
(0.004)
(0.018)
(0.009)
(0.003)
(0.016)
(0.006)
PURPOSE OF LOAN
0.028***
0.050***
0.041***
0.015***
0.024***
-0.003
(0.005)
(0.009)
(0.008)
(0.005)
(0.009)
(0.007)
SOURCE OF LOAN
0.061***
0.113***
0.102***
0.051***
0.115***
0.143***
(0.007)
(0.013)
(0.011)
(0.007)
(0.014)
(0.010)
OCCUPATION
0.065***
0.116***
0.079***
0.063***
0.050***
0.067***
(0.007)
(0.016)
(0.014)
(0.006)
(0.014)
(0.012)
STATES
0.342***
0.178***
0.194***
0.275***
0.233***
0.113***
(0.023)
(0.035)
(0.019)
(0.017)
(0.030)
(0.015)
Observations
12,077
5,600
8,105
15349
6897
10,283
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The table decomposes the
explained component from the equation 3 to identify the contribution of each specific characteristic in
generating credit differences.
Table 3: Decomposition of loan application and loan approval rates.
(1)
(2)
(3)
(4)
(5)
(6)
VARIABLES
GC vs
OBC
GC vs ST
GC vs SC
GC vs
OBC
GC vs ST
GC vs SC
Loan application rate
Loan approval rate
Panel A: Banks
GC
0.257***
0.257***
0.257***
0.963***
0.963***
0.963***
(0.004)
(0.004)
(0.004)
(0.004)
(0.004)
(0.004)
Others
0.274***
0.187***
0.192***
0.937***
0.909***
0.898***
(0.003)
(0.007)
(0.004)
(0.004)
(0.012)
(0.008)
Difference
-0.017***
0.070***
0.065***
0.026***
0.054***
0.065***
(0.005)
(0.008)
(0.006)
(0.005)
(0.012)
(0.008)
Explained
-0.015***
0.044***
0.069***
0.020***
0.030***
0.050***
(0.004)
(0.007)
(0.004)
(0.003)
(0.008)
(0.006)
Unexplained
-0.002
0.026**
-0.003
0.006
0.024*
0.015
(0.006)
(0.010)
(0.007)
(0.005)
(0.013)
(0.009)
Observations
27,988
15,050
20,240
7,256
3,431
4,421
48
Panel B: Money
Lenders
GC
0.112***
0.112***
0.112***
0.806***
0.806***
0.806***
(0.003)
(0.003)
(0.003)
(0.012)
(0.012)
(0.012)
Others
0.223***
0.164***
0.221***
0.878***
0.894***
0.900***
(0.003)
(0.006)
(0.004)
(0.006)
(0.013)
(0.007)
Difference
-0.111***
-0.052***
-0.109***
-0.072***
-0.088***
-0.095***
(0.004)
(0.007)
(0.005)
(0.013)
(0.018)
(0.014)
Explained
-0.088***
-0.036***
-0.063***
-0.048***
-0.107***
-0.066***
(0.003)
(0.005)
(0.004)
(0.008)
(0.016)
(0.010)
Unexplained
-0.023***
-0.016*
-0.047***
-0.024*
0.019
-0.029*
(0.005)
(0.009)
(0.006)
(0.013)
(0.020)
(0.015)
Observations
27,988
15,050
20,240
4,655
1,670
2,991
Panel B: Social
Networks
GC
0.233***
0.233***
0.233***
0.926***
0.926***
0.926***
(0.004)
(0.004)
(0.004)
(0.005)
(0.005)
(0.005)
Others
0.336***
0.287***
0.302***
0.922***
0.940***
0.925***
(0.004)
(0.008)
(0.005)
(0.004)
(0.008)
(0.005)
Difference
-0.102***
-0.053***
-0.068***
0.004
-0.014
0.001
(0.005)
(0.009)
(0.006)
(0.006)
(0.009)
(0.007)
Explained
-0.073***
-0.061***
-0.056***
0.002
-0.022***
-0.014***
(0.003)
(0.007)
(0.004)
(0.004)
(0.008)
(0.005)
Unexplained
-0.029***
0.008
-0.012*
0.002
0.008
0.015**
(0.006)
(0.011)
(0.007)
(0.007)
(0.011)
(0.007)
Observations
27,988
15,050
20,240
7,883
3,457
5,070
Robust standard errors in parentheses.*** p<0.01, ** p<0.05, * p<0.1. The dependent variable is 1 if applied for
a loan at a bank, from money lender or in social network, 0 otherwise (Columns 1-3). The independent variable
is 1 if approved by bank, money lender or in social network (Columns 4-6). The dependent variables are same
as used in selection model.
Table 4: Adjusted differences between castes by banks
2005
2011
VARIABLES
GC vs OBC
GC vs ST
GC vs SC
GC vs OBC
GC vs ST
GC vs SC
GC
12.918***
12.918***
12.918***
13.674***
13.674***
13.674***
(0.205)
(0.205)
(0.205)
(0.160)
(0.160)
(0.160)
Others
12.097***
11.328***
11.344***
12.686***
12.481***
12.177***
(0.133)
(0.341)
(0.219)
(0.096)
(0.424)
(0.165)
Difference
0.820***
1.590***
1.574***
0.988***
1.193***
1.497***
(0.244)
(0.398)
(0.300)
(0.187)
(0.453)
(0.230)
Explained
0.794***
1.401***
0.802***
0.710***
0.960***
0.702***
(0.050)
(0.119)
(0.062)
(0.038)
(0.084)
(0.048)
Unexplained
0.027
0.189
0.772***
0.278
0.232
0.795***
49
(0.239)
(0.391)
(0.296)
(0.183)
(0.449)
(0.228)
Observations
3,664
1,900
2,377
5,541
2,743
3,460
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The table shows the
decomposition results corrected for selection and endogeneity for sample of borrowers from banks.
Table 5: Credit differences between castes by money lenders
2005
2011
VARIABLES
GC vs OBC
GC vs ST
GC vs SC
GC vs OBC
GC vs ST
GC vs SC
GC
11.553***
11.553***
11.553***
12.752***
12.752***
12.752***
(0.254)
(0.254)
(0.254)
(0.317)
(0.317)
(0.317)
Others
10.893***
9.930***
10.478***
11.947***
11.024***
11.667***
(0.126)
(0.303)
(0.140)
(0.128)
(0.373)
(0.166)
Difference
0.660**
1.623***
1.076***
0.805**
1.728***
1.084***
(0.284)
(0.395)
(0.290)
(0.342)
(0.490)
(0.358)
Explained
0.465***
0.778***
0.413***
0.457***
1.068***
0.487***
(0.052)
(0.083)
(0.050)
(0.058)
(0.106)
(0.062)
Unexplained
0.195
0.845**
0.663**
0.348
0.659
0.597*
(0.282)
(0.394)
(0.287)
(0.340)
(0.487)
(0.354)
Observations
3,430
1,246
2,409
2,579
871
1,775
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The table shows the
decomposition results corrected for selection and endogeneity for sample of borrowers from money
lenders.
Table 6: Credit differences in social networks
2005
2011
VARIABLES
GC vs OBC
GC vs ST
GC vs SC
GC vs OBC
GC vs ST
GC vs SC
GC
11.933***
11.933***
11.933***
12.699***
12.699***
12.699***
(0.236)
(0.236)
(0.236)
(0.207)
(0.207)
(0.207)
Others
10.899***
9.536***
10.362***
11.886***
11.635***
11.714***
(0.148)
(0.274)
(0.175)
(0.115)
(0.348)
(0.157)
Difference
1.034***
2.397***
1.571***
0.812***
1.064***
0.985***
(0.279)
(0.362)
(0.294)
(0.237)
(0.405)
(0.260)
Explained
0.696***
1.312***
0.766***
0.489***
1.302***
0.443***
(0.050)
(0.101)
(0.054)
(0.043)
(0.085)
(0.043)
Unexplained
0.338
1.085***
0.805***
0.323
-0.238
0.542**
(0.277)
(0.360)
(0.292)
(0.235)
(0.403)
(0.258)
50
Observations
3,374
1,594
2,228
4,836
2,186
3,272
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The table shows the
decomposition results corrected for selection and endogeneity for sample of borrowers from social
networks.
Table 7: Adjusted credit differential between castes in urban areas.
2005
2011
VARIABLES
GC vs OBC
GC vs ST
GC vs SC
GC vs OBC
GC vs ST
GC vs SC
GC
13.225***
13.225***
13.225***
14.493***
14.493***
14.493***
(0.250)
(0.250)
(0.250)
(0.230)
(0.230)
(0.230)
Others
12.656***
12.068***
11.306***
13.336***
12.280***
12.490***
(0.187)
(0.651)
(0.205)
(0.156)
(0.596)
(0.166)
Difference
0.569*
1.157*
1.919***
1.157***
2.213***
2.004***
(0.313)
(0.698)
(0.323)
(0.277)
(0.639)
(0.283)
Explained
0.949***
0.204
0.861***
0.712***
0.338**
0.646***
(0.061)
(0.179)
(0.063)
(0.054)
(0.136)
(0.052)
Unexplained
-0.380
0.953
1.059***
0.445
1.875***
1.358***
(0.306)
(0.680)
(0.315)
(0.274)
(0.626)
(0.277)
Observations
3,815
1,754
2,534
4,877
2,082
3,213
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The table shows the
decomposition results corrected for selection and endogeneity for sample of borrowers from urban
area only.
Table 8: Adjusted credit differential between castes in rural areas.
2005
2011
VARIABLES
GC vs OBC
GC vs ST
GC vs SC
GC vs OBC
GC vs ST
GC vs SC
GC
11.970***
11.970***
11.970***
12.549***
12.549***
12.549***
(0.136)
(0.136)
(0.136)
(0.117)
(0.117)
(0.117)
Others
10.812***
10.036***
10.426***
11.647***
11.481***
11.468***
(0.076)
(0.192)
(0.105)
(0.058)
(0.191)
(0.093)
Difference
1.158***
1.934***
1.544***
0.901***
1.068***
1.081***
(0.156)
(0.235)
(0.172)
(0.131)
(0.224)
(0.149)
Explained
0.697***
1.170***
0.787***
0.639***
1.083***
0.701***
51
(0.033)
(0.062)
(0.039)
(0.028)
(0.054)
(0.034)
Unexplained
0.461***
0.765***
0.757***
0.262**
-0.015
0.380**
(0.153)
(0.234)
(0.170)
(0.129)
(0.225)
(0.150)
Observations
8,262
3,845
5,571
10,472
4,805
7,068
Robust standard errors in parentheses. *** p<0.01, ** p<0.05, * p<0.1. The table shows the
decomposition results corrected for selection and endogeneity for sample of rural area only.
52
Table 23: Quantile decomposition of log of loan amount for 2011-12
VARIABLE
GC VS OBC
GC VS ST
GC VS SC
Percentiles
10th
25th
50th
75th
90th
10th
25th
50th
75th
90th
10th
25th
50th
75th
90th
GC
11.467***
12.471***
12.961***
14.552***
15.385***
11.467***
12.471***
12.961***
14.552***
15.385***
11.467***
12.471***
12.961***
14.552***
15.385***
(0.019)
(0.021)
(0.018)
(0.023)
(0.021)
(0.213)
(0.173)
(0.135)
(0.205)
(0.236)
(0.213)
(0.173)
(0.135)
(0.205)
(0.236)
Others
9.794***
10.995***
12.119***
13.285***
14.642***
9.762***
11.711***
11.847***
12.220***
13.455***
9.824***
10.430***
11.589***
12.649***
14.001***
(0.007)
(0.011)
(0.012)
(0.014)
(0.016)
(0.428)
(0.342)
(0.267)
(0.320)
(0.425)
(0.111)
(0.089)
(0.101)
(0.118)
(0.198)
Difference
1.673***
1.476***
0.842***
1.267***
0.743***
1.704***
0.760**
1.114***
2.332***
1.930***
1.643***
2.041***
1.373***
1.904***
1.384***
(0.020)
(0.024)
(0.021)
(0.027)
(0.027)
(0.478)
(0.383)
(0.299)
(0.380)
(0.486)
(0.240)
(0.195)
(0.168)
(0.236)
(0.308)
Explained
0.384***
0.581***
0.659***
0.937***
0.997***
0.925***
1.109***
1.193***
1.532***
1.369***
0.547***
0.679***
0.698***
0.891***
1.003***
(0.016)
(0.021)
(0.020)
(0.026)
(0.026)
(0.081)
(0.070)
(0.056)
(0.074)
(0.096)
(0.037)
(0.034)
(0.031)
(0.039)
(0.049)
Unexplained
1.289***
0.895***
0.183***
0.330***
-0.255***
0.779
-0.350
-0.079
0.801**
0.561
1.096***
1.361***
0.674***
1.012***
0.381
(0.007)
(0.005)
(0.004)
(0.005)
(0.006)
(0.484)
(0.385)
(0.299)
(0.379)
(0.492)
(0.237)
(0.194)
(0.168)
(0.232)
(0.307)
Observation
15,361
15,361
15,361
15,361
15,361
6,904
6,904
6,904
6,904
6,904
10,295
10,295
10,295
10,295
10,295
Robust standard errors in parentheses. p<0.01, ** p<0.05, * p<0.1. The table shows the result from quantile regression decompositions of log of loan amount obtained at 10%,
25%, 50%, 75%, and 90%
53
Table 24: Quantile decomposition of log of loan amount for 2005
VARIABLE
GC VS OBC
GC VS ST
GC VS SC
Percentiles
10th
25th
50th
75th
90th
10th
25th
50th
75th
90th
10th
25th
50th
75th
90th
GC
10.264***
10.93***
12.086***
13.425***
14.628***
10.264***
10.92***
12.086***
13.425***
14.628***
10.264***
10.93***
12.086***
13.425***
14.62***
(0.019)
(0.017)
(0.020)
(0.024)
(0.029)
(0.202)
(0.135)
(0.150)
(0.188)
(0.355)
(0.202)
(0.135)
(0.150)
(0.188)
(0.355)
Others
9.318***
9.682***
11.607***
12.730***
13.752***
7.891***
9.162***
10.157***
11.502***
13.254***
8.884***
9.681***
10.487***
11.559***
12.720***
(0.015)
(0.009)
(0.017)
(0.017)
(0.019)
(0.367)
(0.276)
(0.212)
(0.340)
(0.450)
(0.154)
(0.121)
(0.109)
(0.135)
(0.206)
Difference
0.947***
1.250***
0.478***
0.695***
0.875***
2.374***
1.770***
1.929***
1.923***
1.374**
1.381***
1.251***
1.599***
1.866***
1.907***
(0.024)
(0.019)
(0.026)
(0.029)
(0.035)
(0.419)
(0.307)
(0.259)
(0.389)
(0.573)
(0.254)
(0.182)
(0.185)
(0.231)
(0.410)
Explained
0.698***
0.512***
0.920***
1.032***
1.226***
1.211***
1.146***
1.262***
1.589***
1.774***
0.724***
0.680***
0.813***
0.957***
1.162***
(0.023)
(0.016)
(0.027)
(0.029)
(0.033)
(0.091)
(0.065)
(0.059)
(0.077)
(0.124)
(0.041)
(0.032)
(0.032)
(0.039)
(0.060)
Unexplained
0.249***
0.739***
-0.442***
-0.337***
-0.351***
1.162***
0.623**
0.666**
0.334
-0.401
0.657***
0.571***
0.786***
0.909***
0.745*
(0.005)
(0.005)
(0.004)
(0.004)
(0.007)
(0.422)
(0.308)
(0.259)
(0.393)
(0.582)
(0.254)
(0.182)
(0.184)
(0.229)
(0.404)
Observation
12,081
12,081
12,081
12,081
12,081
5,602
5,602
5,602
5,602
5,602
8,108
8,108
8,108
8,108
8,108
Robust standard errors in parentheses. p<0.01, ** p<0.05, * p<0.1. The table shows the result from quantile regression decompositions of log of loan amount obtained at 10%, 25%, 50%, 75%,
and 90%.