Does Financial Inclusion Mitigate Social Exclusion? Causal Evidence from India

Full text

Does Financial Inclusion Mitigate Social Exclusion?
Rikhia Bhukta*
, Debayan Pakrashi†
, Sarani Saha‡
, Ashish K. Sedai§
February 14, 2024
Abstract
This paper examines the causal impact of financial inclusion in reducing social exclusion.
Utilizing a quasi-experimental setup involving a 2005 Reserve Bank of India policy incen-
tivizing bank branch openings in underbanked districts, we employ regression discontinuity
design analysis with data from three nationwide surveys and censuses. Our findings reveal
significant consumption increases and poverty reduction among marginalized castes com-
pared to non-marginalized castes, narrowing caste-based welfare inequality. We also observe
enhanced social inclusion for marginalized castes. Three channels—informal finance, busi-
ness finance, and the labor market—contribute to the welfare gains of marginalized castes.
Our results withstand the control for income effect, and various RD design checks. In con-
clusion, this study underscores the role of inclusive formal banking sector in mitigating deep-
rooted social norms.
Keywords: Financial Inclusion, Caste Discrimination, Social Exclusion, Bank Expansion
Policy, Poverty
*Department of Economic Sciences, Indian Institute of Technology, Kanpur. Email: rikhiab20@iitk.ac.in
†Economic Research Unit, Indian Statistical Institute Kolkata and Department of Economic Sciences, Indian
Institute of Technology Kanpur. Email: pakrashide@gmail.com
‡Department of Economic Sciences, Indian Institute of Technology Kanpur. Email: sarani@iitk.ac.in
§Department of Economics, College of Business, University of Texas at Arlington. Email: ashish.sedai@uta.edu.
Research Associate, Centre for Applied Macroeconomic Analysis, Australian National University.

1 Introduction
Many wide-ranging socio-economic impacts of financial inclusion have been well documented
in developing economies (Burgess and Pande,2005;Ashraf et al.,2006;Banerjee et al.,2017;
Kochar et al.,2022;Ngo and Wahhaj,2012;Dupas and Robinson,2013;Jack and Suri,2014;
C´
elerier and Matray,2019;Cramer,2021a;Barboni et al.,2022;Gupta and Sedai,2023). How-
ever, a critical gap remains—a lack of large-scale and comprehensive causal investigation into
how financial inclusion can alleviate deeply entrenched social disparities. These disparities are
fundamentally rooted in discrimination against marginalized groups based on their inherited iden-
tities, and affect consumption, wealth, and overall well-being (Deshpande,2000;Banerjee et al.,
2013;Munshi and Rosenzweig,2006;Arunachalam and Shenoy,2017). Lack of relevant data
and identification issues have constrained adequate understanding of this critical phenomenon.
Given the increasing impetus on financial inclusion of underprivileged and vulnerable groups,
we develop a conceptual model and causally examine whether caste-based discrimination, which
acts as a formidable barrier to financial inclusion and equitable development, can be mitigated
by increasing bank branches, and if so, how? India serves as an ideal context for this analysis,
as it is home to over 80% of the 260 million who endure discrimination rooted in their inherited
caste.1
To conduct a causal longitudinal analysis, we utilize an exogenous nation-wide bank expansion
policy, led by the Reserve Bank of India in 2005. Employing a fuzzy Regression Discontinuity
Design (RDD) framework developed by Young (2017) and Cramer (2021b), we first establish that
the policy significantly improved financial access across all caste categories. Subsequently, we
uncover substantial enhancements in key household welfare indicators for marginalized castes–
increased consumption, asset accumulation, labor force participation, social memberships and
reduced poverty rate, while non-marginalized households only experience higher asset accumu-
lation compared to marginalized households. To disentangle the caste and class effect, we control
1

for permanent income inequality through Machine Learning (ML) techniques to underscore that
the improvements are driven more by addressing social disparities related to caste than by base-
line income inequality.
We adopt a general equilibrium framework that integrates household-level data, administrative
records, and Census data to investigate the multifaceted implications of bank branch expansion
on both households and enterprises by caste. Our analysis extends beyond the direct effect of
increased formal financial inclusion and uncovers indirect effects on informal credit market, la-
bor market and agricultural income and productivity. First, we discern how the policy affects
disparate caste groups: households and firms, access to loans, both in the formal financial sector
and through informal channels. We systematically analyze whether increased bank presence fos-
ters heightened competition, thereby diminishing discriminatory practices in the informal credit,
and how these dynamics create a ripple effect through interest on loans in both the formal and
informal credit market.
We evaluate access to institutional loans by sectors, particularly focusing on agricultural sec-
tor, and how bank expansion affects agricultural technology, labor hours and earnings by caste
groups, aiming to discern whether historically marginalized groups encounter disproportionate
impacts of bank presence, in light of their constrained access to credit in the agricultural sector
(Hoff et al.,2011;Kumar,2013;Kumar and Venkatachalam,2019). Following this, we examine
whether additional bank branches affect non-farm business revenues, household’s overall labor
force participation rates and labor market earnings for marginalized and non-marginalized caste
groups. Additionally, we examine the economic ramifications of bank expansion and its spillover
into the social realm, focusing on household memberships in historically exclusive social groups
and organizations among marginalized and non-marginalized castes.
The RDD is derived from the RBI policy categorizing districts into “banked” and “underbanked”
categories based on district bank-to-population ratio being higher or lower than the national av-
2

erage in 2005. The objective of the policy was to incentivize scheduled commercial banks to
establish branches in underbanked districts. We combine bank branches data from RBI Master
Office File (MOF) with three comprehensive nationwide datasets: the Indian Human Develop-
ment Survey (IHDS) panel, the All India Debt and Investment Survey (AIDIS), and the Eco-
nomic Census (EC) for empirical analysis. This combination provides a unique long-term view
of financial inclusion policies, enabling rigorous pre-policy continuity assessments and various
robustness checks within the RDD framework to establish causal relationships.
Results from the RD estimations demonstrate substantial improvements in welfare measures of
marginalized ’Scheduled’ castes in treated districts following policy implementation. For Sched-
uled Caste households, consumption grew by 16%, and the likelihood of falling below the poverty
line decreased by 59.7%, whereas no discernible impact was observed for non-marginalized
’General’ castes. This indicates a reduction in consumption-based welfare disparities. However,
when considering multidimensional poverty as a broader measure of well-being, we observe a re-
duction across all caste categories, highlighting the overall societal impact of financial inclusion.
Additionally, we find increased membership in social organizations among marginalized castes
following the policy, while no such effects are present for non-marginalized caste groups. Our
analysis delves into both direct and indirect channels driving these welfare improvements, sup-
ported by robustness checks that validate the strength of these results, underscoring the policy’s
positive impact.
As a direct channel of welfare enhancement, we find a significant increase in financial inclu-
sion via bank borrowings and access to insurance, significantly more for marginalized castes
compared to non-marginalized castes after the policy in treatment districts. The likelihood of ob-
taining a bank loan increases by 40% for the most marginalized ‘Scheduled Caste’ households,
whereas the same for non-marginalized ‘General’ caste households is 21%, almost half com-
pared to marginalized households. Apart from loans taken by households, the enterprises owned
by marginalized castes also experience a significantly higher increase of 44% in institutional
3

loans, while the increase for non-marginalized caste is 31%.
We explore three indirect channels that explain why the marginalized castes benefit the most
in terms of welfare gain: informal finance channel, business finance and labor market channel.
In the informal credit market, lenders face competition after the formal sector expansion and are
compelled to reduce discrimination against marginalized castes in the form of lower interest rates,
allowing them to take out more informal loans for consumption purposes. On the other hand, the
non-discriminatory nature of formal loans allows the marginalized castes to take out more agri-
cultural loans, which significantly increases their agricultural productivity and mechanization,
and increases crop income by approximately 87%. We do not find significant changes in these
agricultural outcomes for the non-marginalized groups. We find that non-marginalized groups
acquire institutional loans for business expansion, which creates additional labor demand in the
market and raises the wages of marginalized groups, thereby acting as an additional welfare-
enhancing channel for them.
The rest of the paper is organized in the following sections. Section 2outlines the existing
literature, and section 3lays out a conceptual framework for our study. Section 4describes all
the datasets we use for our study. Section 5outlines the regression discontinuity framework
and provides evidence of its validity. Section 6and 7discuss the main results and mechanisms,
respectively. Section 8furnishes the robustness checks and finally, section 9concludes.
2 Literature Review
Caste has historically been a major source of discrimination in Indian societies (Deshpande,
2011;Hoff et al.,2011;Fehr and Hoff,2011). The caste system originated from the occupation-
based classification or varna system2, which dates back to 1500-500 BCE, with hundreds of jatis
within each varna (Munshi,2019). There are four major caste groups in India, the Scheduled
4

Castes (SCs), Scheduled Tribes (STs), Other Backward Castes (OBCs) and Unreserved Castes,
who are often referred to as General Castes. The SCs were historically the untouchables and
hence encountered the highest level of discrimination (Munshi,2019;Bagde et al.,2016). STs,
the indigenous ethnic groups (tribals), are also economically and socially marginalised (Munshi,
2019). OBCs, described as ’Socially and Educationally Backward Classes’ (SEBC) in Article
340 of the Indian constitution, stand higher than the SCs but lower than the general castes in
the social hierarchy (Deshpande,2011). In this paper, we refer to the non-general castes as
marginalised castes, but the level of marginalisation and discrimination is not the same for all.
The caste identity of an individual has always been a determining factor in education (Munshi
and Rosenzweig,2006;Hanna and Linden,2012;Hoff and Pandey,2014), access to healthcare
services (Luke and Munshi,2007), access to public goods (Anderson,2011) and marital choices
(Munshi and Rosenzweig,2009). The Government of India has undertaken world’s largest af-
firmative action program to eliminate caste-based discrimination and social exclusion, but caste
continues to play a significant role in all facets of Indian society, even among the ostensibly
progressive educated urban population (Banerjee et al.,2013).
In the labour market, the caste identity often restricts occupational mobility (Munshi and Rosen-
zweig,2006) and gives rise to caste-based wage discrimination where marginalised workers
systematically get lower wages compared to their non-marginalised counterparts (Banerjee and
Knight,1985;Ito,2009;Das and Dutta,2007). Furthermore, workers from marginalised castes
often get assigned to less prestigious jobs (Das and Dutta,2007;Deshpande and Sharma,2016).
The implication of caste in the labour market is so intense that workers often decline higher wages
to avoid employments that do not fit with their caste identity (Oh,2023).
Caste-based discrimination acts as a barrier to access to credit in many parts of India, majorly in
informal credit markets where the rate of interest charged can be heavily impacted by the caste-
biasedness of the informal lender (Kumar,2013;Mosse,2018). A vast literature suggests that
5

credit constraint can induce income inequality (Demirg¨
uc¸-Kunt and Levine,2009), hinder agri-
cultural investment and income growth (Kaboski and Townsend,2012) and limit entrepreneurial
opportunities (Banerjee et al.,2017). Credit constraints stemming from caste discrimination and
the resulting interest rate disparity between marginalised and non-marginalised caste borrowers
significantly contribute to the caste disparity in India.
This study adds to the literature that examines the impacts of financial inclusion on economic wel-
fare in developing countries. Leveraging the exogenous branch expansion policy of RBI during
1977 and 1990 in rural unbanked districts,3.Burgess and Pande (2005) establish the effectiveness
of formal banking sector expansion in reducing rural poverty. While their focus was on poverty
and economic development, Burgess and Pande (2005) did not capture the distributional impact
of formal banking expansion on different social groups in India, which we capture in our paper
by leveraging a more recent exogenous bank branch authorization policy (2005) of RBI, which
incentivised the commercial banks (except regional rural banks) to open branches in the under-
banked districts in order to increase their chance of obtaining licenses for favoured locations4.
Existing literature by Young (2017) and Cramer (2021b) have already established that this par-
ticular policy has indeed led to the expansion of bank branches in the treatment (underbanked)
districts. Young (2017) shows that this policy has led to an increase in both agricultural and
manufacturing outputs and, thereby, an increase in the local GDP proxied by night-time lumi-
nosity. Cramer (2021b) further investigated the impact of this policy on health outcomes and
found positive health impacts resulting from higher institutional loan, better offering of health
insurance and better health infrastructure. Gupta and Sedai (2023) found that the impact of the
policy on overall household well-being has been more prominent in urban areas. Our present
study focuses broadly on the distributional impacts of the policy on the financial inclusion and
welfare of different caste groups in India and shows that marginalised caste groups gain signifi-
cantly more compared to their non-marginalised counterparts. In other words, the policy induces
marginalised caste groups to catch up with non-marginalised groups.
6

3 Conceptual Framework
The discussion in section 2outlines the existing literature on caste-based discrimination and so-
cial exclusion in India, which is argued to have resulted in a significant welfare disparity between
marginalised and non-marginalised castes in India (Deshpande,2000;Kijima,2006). For our
study, apart from the social exclusion induced by caste-based discrimination, we focus on two
sectors, informal credit market and labour market, where discrimination against marginalized
castes has led them to significantly fall behind their non-marginalized counterparts in terms of
overall household welfare. In panel A of figure 1, we depict this pre-policy status quo.
Our main findings from this study suggest that the banking expansion policy of 2005 has led to
a significant reduction in caste-based welfare disparity because the benefits of the policy mainly
accrued to the marginalised castes. Why did marginalised castes benefit more from the bank
expansion policy? We seek the answer in a general equilibrium framework, which is depicted in
panel B of figure 1. Our analysis focuses on the informal credit market and the labour market,
and we unravel what happens in those two sectors after the implementation of the policy. As an
immediate impact of the policy, the entry of banks into the market raises competition for informal
lenders. They may also borrow from the formal sector and distribute the funds as informal loans
(Conning and Udry,2007;Sagrario Floro and Ray,1997). Increased competition and higher
availability of funds to lend out induce the informal lenders to mitigate discriminatory practices
towards borrowers from marginalised castes, manifesting in a reduced interest rate.
Establishment of a bank branch enhances the accessibility to business loans and thereby fosters
overall productive activity. This results in higher labour demand, contributing to reduced labour
market discrimination (Becker,2010). With the additional labour demand, business owners find
it rational to hire workers from marginalised castes, whose wages are cheaper compared to non-
marginalised workers (Ito,2009). Furthermore, the diminished prevalence of discrimination in
the informal credit market and labour market has a cascading effect on other spheres of the
7

Figure 1: Role of Bank Presence in Improving Household Welfare of marginalized Castes
society, ultimately resulting in the broader social integration of marginalized castes. In a nutshell,
reduced interest rates, increased wages and overall social inclusion leads to the extra welfare gain
of marginalised castes, and enables them to catch up with their non-marginalised counterparts in
terms of overall household welfare.
4 Data
For our analysis, we use data from the following four sources: (a) Reserve Bank of India’s (RBI)
Master Office File (MOF), (b) Indian Human Development Survey (IHDS), (c) All India Debt
and Investment Survey (AIDIS) and (d) Economic Census. In figure 2, we provide the timeline
of this study. Below is a detailed discussion of how we use these datasets for our analysis.
8

Figure 2: Timeline of the datasets used
4.1 Master Office File of RBI
The Master Office File (MOF) of RBI contains detailed information about all the bank branches
operating in India. The file is maintained separately for commercial banks and cooperative banks.
Within commercial banks, there are five main classifications: (a) State Bank of India (SBI) and its
associates, (b) Nationalised Banks, (c) Private Sector Banks, (d) Foreign Banks, and (e) Regional
Rural Banks (RRBs). The policy was aimed at all commercial banks excluding RRBs. Hence
our focus is also on the same. From the MOF file, we get the number of bank branches in 581
districts in India, which helps us to construct the running variable for RD design.
4.2 Indian Human Development Survey (IHDS)
IHDS is the largest panel survey in India, conducted in two waves, 2004-05 and 2011-12. Since
wave 2 was conducted six years after the policy introduction, we use it for the main results and
wave 1 data for the pre-policy smoothness check. We first merge the IHDS data with the RBI
MOF data to carry out the RD analysis. A total of 371 districts from 581 districts in MOF data are
merged with IHDS panel. However, this does not threaten our study because the IHDS dataset is
nationally representative. In table A1, we compare the distribution of different caste categories
in IHDS with other nationally representative and contemporary datasets like NFHS-3, NSS and
9

population census, and show that the percentage shares of four caste categories are similar for all
these surveys. Furthermore, in table A8, we furnish evidence that IHDS sub-sample is random
and does not bias the data-generating process.
From IHDS, we use the outcome variables related to financial inclusion and household welfare.
Financial inclusion is measured by five indicators: having a savings/current account with a bank,
taking a loan from bank, having a long-term fixed deposit in the bank, investing in a finan-
cial market security (mutual fund/bonds/shares/unit trust) and purchasing life/health insurance.
Household welfare is measured by another set of five variables: household consumption, house-
hold food consumption, poverty, multidimensional poverty and social inclusion. Consumption
is represented by quintiles of total annual household consumption, whereas food consumption is
measured by quintiles of total monthly food consumption of the household. Poverty is a dummy
variable; takes value 1 if the household falls below the poverty line defined by the Tendulkar cut-
off 5. Multidimensional poverty is another indicator of poverty in a broader sense, which takes
into account three dimensions of overall household welfare: education, health and living stan-
dards6. Following Alkire and Foster (2011), we calculate a multidimensional deprivation score
of the household, ranging from 0 to 1, where a higher score implies that the household is more
multidimensionally poor. Our outcome variable, multidimensional poverty, is a dummy variable
that takes the value 1 if the multidimensional poverty score is above or equal to 0.33 (the standard
cutoff defined by Alkire and Foster (2011)). Lastly, social inclusion is a dummy variable, which
takes the value 1 if the household is a member of more than one social group in the community7.
In table 1, we provide summary statistics of all the outcomes we use from IHDS8.
4.3 All India Debt and Investment Survey (AIDIS)
AIDIS is a nationally representative survey containing detailed information on credit access, asset
holding and indebtedness of Indian households. RBI started this survey under the name of ‘All
10

India Rural Credit Survey’ in 1951-52 with a broad objective of using the collected data for
designing banking policies. In 1971, RBI entrusted the data collection process for this survey to
National Sample Survey Organization (NSSO) and since then, NSSO has been conducting this
survey every ten years. For our current analysis, we choose two rounds of this survey: AIDIS
2003 (NSS 59th round) and AIDIS 2013 (NSS 70th round), and we merge both these datasets
separately with the MOF data. AIDIS 2013 is used for the main results, and AIDIS 2003 for the
pre-policy smoothness check.
AIDIS has a wide set of information on the loans taken by the households, including source of
loan (credit agency), amount of loan, year of taking the loan, type of mortgage, purpose of loan,
interest rate etc. This set of information allows us to study the impact of banking expansion
policy on the credit market more closely and in more detail.
AIDIS also has information on the asset holding of the households. We use the data on agricul-
tural equipment and ownership of livestock to understand a part of the agricultural labour market
channel, which we shall explain in detail while discussing the results. In table 2, we furnish the
summary statistics of the variables we use from AIDIS.
4.4 Economic Census
The economic census of India provides detailed information on all the entrepreneurial units (agri-
cultural and non-agricultural) operating within the geographical boundary of India. It is carried
out by the Ministry of Statistics and Program Implementation (MoSPI), Government of India, and
covers all the districts in all states of India. So far, six economic censuses have been conducted
in 1977,1980,1990, 1998, 2005 and 2013.
The economic census has a set of variables, including the nature of the entrepreneurial unit,
its main source of finance, information about the caste of the owner, the location of the unit
11

(rural/urban) etc. We use the data from the sixth economic census (EC 2013) and merge it
with the RBI MOF data to analyze how the source of finance changes for entrepreneurial units
owned by different caste categories after the policy. The information on the nature of the unit
(agricultural/non-agricultural) further allows us to study the pattern of the change in the source
of finance for these two types of enterprises. Then, we merge the fifth economic census (EC
2003) to check the pre-policy smoothness of the main results. In table 3, we present the summary
statistics of the variables we use from the economic census.
5 Empirical Strategy
5.1 Regression Discontinuity Framework
In 2005, RBI introduced a bank branch authorization policy that incentivizes commercial banks
(excluding Regional Rural Banks) to open branches in ’underbanked’ districts in order to increase
their chance of obtaining branch-opening licenses for favoured locations. The root of the RD
framework lies in the definition of ’underbanked’ districts used by RBI. A district is tagged as
’underbanked’ when the ratio of population to the number of bank branches in the district exceeds
the national ratio. Therefore, the district-level population-to-bank branch ratio acts as the running
variable and the national-level ratio (computed to be 14,780) acts as the cutoff in our regression
discontinuity framework. Panel (a) of figure 3depicts the histogram for the running variable. As
presented in equation 1, a district d is defined as underbanked or treated (Td) if the district-level
ratio is higher than the cutoff, and if the ratio is below the cutoff, the district falls into the set of
’banked’ or ’control’ districts.
T reatedd(Td) = 1P opulation of district d
No of bank branches in district d >National population
T otal no of bank branches (1)
12

(a) Histogram (b) First Stage RD plot
Figure 3: Histogram and First stage RD plot
In 2006, RBI published a list of underbanked districts, but that list does not include the district-
level population-to-branch ratios. So we reconstruct the ratio for each district using district pop-
ulation data from census 2001 and number of bank branches data (during quarter 1 of 2006) from
RBI. However, there are 12 districts for which the predicted underbanked status from our recon-
structed ratio is different from their ’underbanked’ status as per the RBI list. As noted in Cramer
(2021b) and Gupta and Sedai (2023), RBI could have used its own discretion in determining the
underbanked status of these 12 districts, which makes the treatment assignment rule probabilistic
and not deterministic (Hahn et al.,2001). However, this does not pose a threat to our identifica-
tion as we adopt the fuzzy RD design instead of the sharp design (Lee and Lemieux,2010;Dong
and Lewbel,2015). Panel (b) of figure 3, which indicates a jump in the probability of being listed
as underbanked by RBI at the cutoff, provides a graphical justification for our fuzzy design.
Following the fuzzy RD framework, we use the specification in equations 2and 3to estimate the
impact of the bank authorization policy on different caste groups.
Ud=β0+β1Td+β2Rd+β3TdRd+α1Xd+ed(2)
Yh,d =δ0+δ1Ud+δ2Rd+δ3RdTd+α2Xd+vd(3)
13

Here, Udis a dummy variable that takes the value one if district d is listed as underbanked by
RBI, Tdis another dummy variable that takes the value one if the district-level ratio is higher than
the cutoff and Rdis the running variable (the district-level population-to-branches ratio). Yh,d is
the outcome variable of household h in district d and Xdis a vector of controls. Following Abadie
et al. (2023), we cluster the standard errors at the district level as our treatment is homogeneous
across districts. Under the identifying assumption, the coefficient δ1in equation 3can be inter-
preted as the local average treatment effect (LATE) of belonging to a district that has received the
’underbanked’ status.
5.2 Validity of identifying assumption
The main identifying assumption that makes the above RD framework empirically valid is that the
districts above and below the cutoff are similar in all aspects except the status of banked/underbanked,
which can be achieved if the local governments have no power to manipulate the value of the
running variable. One implication of this assumption is the smoothness of the running variable
around the cutoff. Intuitively, given that the district population data comes from 2001 census,
four years before the policy, and the data for the number of branches comes straight from RBI,
manipulation of both these components of the running variable by district administrations seems
logically impossible (Cramer,2021b). The histogram of the running variable around the cutoff
in figure 3tells the same story. Further, to test this ’smoothness around the cutoff’ assumption
formally, we employ the McCrary test, as proposed by McCrary (2008). The result of the test is
depicted in figure A1, which formally establishes the validity of our assumption. Additionally,
to add another layer of robustness to the McCrary test results, we carry out the binomial test
proposed by Cattaneo et al. (2017) in table A2, which further strengthens the validity of the RD
design9.
Another implication of the identifying assumption is that prior to the implementation of the pol-
14

(a) Consumption quintiles (b) Food consumption quintiles
(c) Poverty (d) Social Inclusion
Figure 4: Pre-policy smoothness of household well-being outcomes for SCs
icy, the outcome variables of interest should not be significantly different for the treated and
control districts. We examine this empirically from pre-policy smoothness checks, using IHDS
2004-05, Economic Census 2005 and AIDIS 2003, as discussed in section 4. Results of pre-
policy smoothness checks are furnished in tables A3—A7. As expected, the fuzzy RD coeffi-
cient, δ1, is insignificant for all these results, implying that pre-policy smoothness holds for our
outcome variables of interest. In figure 4, we show the pre-policy smoothness for the main wel-
fare outcomes for SCs. Another potential threat to identification is the existence of analogous
policies during our study period. Here we refer to the previous study by Cramer (2021b) that ex-
tensively furnishes evidence of the absence of similar contemporaneous policies, which provides
strength to our identification.
15

6 Results
6.1 Banks and financial inclusion across caste groups
We start our analysis by examining the causal impact of the banking expansion policy on the
financial inclusion of three caste categories in India: SC, OBC and General. We compare the out-
comes of socially marginalized castes with non-marginalized castes to examine if the marginal-
ized castes catch up with their non-marginalized counterparts in terms of financial inclusion. We
exclude Scheduled Tribes (STs) and Muslims from our analysis. STs were excluded because
historically they do not fall into the so-called ‘Dalit’ caste who face most discrimination (Kumar,
2013) and also because the ST population is concentrated in a few states in India, especially in
North-Eastern states (Kijima,2006), where they form the majority of the population10 and are
not subject to discrimination. Muslims are excluded because of data constraints in IHDS, where
data for castewise categorization is available only for Hindus, not for any other religions.
According to a 2008 report of a government committee, financial inclusion entails extending ac-
cess to several financial services offered by the formal banking sector to underprivileged groups
(Rangarajan et al.,2008). These financial services include access to credit at an affordable rate,
access to bank savings, insurance, remittance facilities and financial consulting/advisory services.
In our analysis, we focus on five dimensions of financial inclusion: having a savings/current ac-
count in the bank, taking loan from the bank, having long-term savings (fixed deposit) in a bank,
buying securities and buying insurance. We use RD estimates to demonstrate that overall finan-
cial inclusion increases across all caste categories. The results are furnished in table 4and the
discontinuities induced by the RD framework are graphically represented in figure 5. In treat-
ment districts, the likelihood of having a bank account increases by 71% for SCs, 27% for OBCs
and 37% for generals. This enormous improvement in bank account opening can be explained
by examining clause 3.b of the relevant RBI policy circular, where it is stated that banks should
16

prioritize basic/no-frills bank accounts. Another important indicator of financial inclusion, bank
loan, increases for SCs and Generals by 40% and 21%, respectively11. Long-term fixed deposits
with banks increase only for OBCs and Generals. The literature has established that marginalized
castes are mostly economically backward compared to other castes (Kijima,2006;Deshpande,
2011;Mosse,2018). They don’t have enough savings for a long-term investment. Hence, we
don’t observe any statistically significant increase in long-term deposits for SCs. Another instru-
ment for long-term investment is purchasing securities, including mutual funds, shares, bonds etc.
The likelihood of investing in these financial instruments increases only for the general category
by 55% in treatment districts.
Insurance markets are closely interlinked with the financial system (Arena,2006). As the finan-
cial system strengthens, insurance companies begin to offer a variety of new insurance products,
particularly tailored to those who previously could not afford coverage. In terms of risk mitiga-
tion, insurance companies are a significant element of the financial system (Holliday,2017). On
the other hand, sometimes banks also offer insurance and sometimes insurance products come
as an add-on with the regular bank account (Anagol et al.,2017). This affordability option
helps marginalized caste groups to purchase life/health insurance products, which is reflected
in a 42.2% increase in the likelihood of being insured for SCs in the treatment districts compared
to SCs in control districts.
To sum up, we observe that after the implementation of the policy, SC households are more
likely to open a bank account, obtain bank loans and purchase insurance. General category
households are also more likely to take bank loans, but unlike SCs, they are likely to make
long-term investments like fixed deposits or buying securities. For OBCs, all the impacts are in
a positive direction but not statistically significant except for fixed deposits. Overall, financial
inclusion increases across all caste categories, albeit not necessarily along the same dimensions.
Insert Table 4about here
17

(a) Bank Account (b) Bank Loan
(c) Insurance
Figure 5: Financial inclusion for SCs
6.2 Banks and household welfare across caste groups
In subsection 6.1, we find that financial inclusion, in a broad sense, increases for all caste cat-
egories. To answer the question ’to which caste category do the majority of benefits accrue?’,
we delve deeper and examine how various household well-being measures are affected across
caste categories following the implementation of the policy. As mentioned before, we look at the
causal impact of the banking expansion policy on five indicators of household welfare: consump-
tion, food consumption, poverty, multidimensional poverty and social inclusion. The results are
furnished in table 5. The graphical representations of the post-policy discontinuities are shown
in figure 6.
18

From panel (a) of table 5, We find that only SC consumption increases by 16% in treatment
districts. Consumption is an important indicator of welfare for SCs because, on average, they
consume less than other caste categories due to lower levels of income and limited access to
resources and opportunities (Deshpande,2000;Kijima,2006;Deshpande,2011). Consequently,
the increase in consumption for SCs, not for other castes, has a broader implication in terms of
convergence of consumption across castes.
Consumption can be subdivided into several categories like food consumption, clothing, house-
hold appliance expenditure etc. The most crucial among these is food consumption, which is
linked to an individual’s overall health/nutritional status and consequentially, to his/her income-
earning potential. As presented in panel (b), food consumption increases by 19% for SCs and
18% for OBCs in treatment districts compared to SCs and OBCs in the control districts.12
Poverty is another indicator of household well-being, especially for SCs. As per the IHDS 2004-
05 data, 27.3% SC households live below the poverty line, whereas the same numbers for OBCs
and generals are comparatively lower, 21% and 9%, respectively. For SC households living
under the poverty line, an additional component of poverty-based social exclusion is added to
the existing caste-based social exclusion. Panel (c) of table 5shows a 59.7% reduction in the
likelihood of falling below the poverty line for SC households, compared to a control mean of
0.174. Poverty reduction for OBCs and generals is, however, not statistically significant. Since
the Tendulkar poverty cutoff is based primarily on consumption, this result is consistent with the
result in panel (a).
Apart from the conventional definition of poverty based on poverty cutoff, we also use a more
comprehensive poverty indicator, multidimensional poverty. As discussed in section 4, it en-
compasses three dimensions of household welfare: education, health and living standards. From
panel (d), we observe that multidimensional poverty declines across all caste categories in the
treatment districts.
19

(a) Consumption quintiles (b) Food consumption quintiles
(c) Poverty (d) Social Inclusion
Figure 6: Household Well-being Outcomes for SCs
20

Our last household welfare indicator, social inclusion, is paramount for our study. Social in-
clusion signifies lesser discrimination against marginalized castes. Not only does it have direct
implications for household welfare (Petrikova,2020), but it also helps in understanding the un-
derlying mechanisms of our findings. We define social inclusion as having a membership in more
than one socioeconomic groups in the area. In panel (e), we show that the likelihood of social
inclusion increases only for the SCs in treatment districts, and it almost doubles (increases by
123%).
Overall, after the policy implementation, SC households in treatment districts experienced the
greatest improvements in household welfare in terms of increased consumption, reduced poverty,
and increased social inclusion. However, multidimensional poverty increases for all caste cat-
egories, indicating that the benefits accrue to other castes as well, albeit to a lesser extent than
SCs. Increased consumption and reduced consumption-based poverty only among SCs in treat-
ment districts imply the convergence of fundamental welfare states across caste groups in India.
Insert Table 5about here
7 Mechanisms
In the literature, it has widely been accepted that financial inclusion improves household welfare.
In section 6, we demonstrate that financial inclusion increases across all caste categories, whereas
household welfare increases predominantly for the SC category. Therefore, there should be some
additional channels alongside the financial inclusion channel that reinforce welfare impacts of the
policy for the SC category. This section attempts to investigate these additional indirect channels.
We are particularly concerned with the indirect effects of the informal credit market and labour
market. Additionally, to examine the labour market channel closely, we also try to comprehend
the business finance channel.
21

7.1 Informal finance channel
In an underbanked area, the credit market is predominated by informal lenders, such as profes-
sional moneylenders, landlords, merchants etc. These informal lenders discriminate extensively
against the marginalized castes (Dreze et al.,1997;Kumar,2013;Kumar and Venkatachalam,
2019). In most cases, the mode of discrimination is the excessive and unrealistic rate of interest
demanded from the SCs, which almost always leads to them falling into a debt trap (Kumar,
2013). According to AIDIS data, the average interest rate in 2003 for SCs was 28%, whereas the
same for OBCs and generals was 22% and 16%.
When a bank branch is established in the area, informal lenders face competition because of
lower interest rates offered by the formal banking sector and their non-discriminatory approach
(Kumar,2013). To counter this competition, informal lenders are compelled to reduce the interest
rate13. There is another reason why informal lenders opt for less discrimination after banks come
in. As noted in Sagrario Floro and Ray (1997); Ghate (1992); Conning and Udry (2007), infor-
mal lenders often acquire formal credits to cater to borrowers’ needs. Therefore, the volume of
available funds in informal credit market rises following the expansion of the formal sector. Due
to the higher availability of funds to lend out, informal lenders consider lowering discrimination
and expanding their customer base from the SC population.
The phenomenon of reduction in discrimination in the form of reduced interest rate is exactly
what is reflected in panel (a) of table 6. The reduction in the annual informal interest rate is
statistically significant for SCs and OBCs. The magnitude and percentage of reduction both are
highest for the SCs. Therefore, the relative informal interest rate faced by SCs compared to
other castes declines after the policy in treatment districts, which results in an increased informal
borrowing for SCs as presented in panel (c). The AIDIS data allows us to examine what SCs
further do with this informal loan. From AIDIS 2013, we find that more than 90% loans taken by
SCs from the informal sector are used for non-productive purposes like household expenditure,
22

medical expenditure, housing, repayment of debt etc. Consequently, we see an increase in con-
sumption for SCs in panel (a) of table 5. On the other hand, OBCs and generals do not encounter
as much discrimination in the informal credit market as SCs do.14 The interest rate charged from
them has always been lower compared to SCs. As a result, their demand for informal loans does
not increase significantly after banks enter the market.
There are multiple theories and concepts that try to study the interaction and coexistence of
informal and formal credit sectors. One such concept is the ’cream-skimming’ theory (Demont
et al.,2010;Mookherjee and Motta,2016), which argues in favor of an increase in informal
interest rate after formal sector comes in, because of the possibility of low-risk borrowers moving
to the formal sector, leaving high-risk borrowers in the informal credit market. At first glance, our
result of reduced informal interest rate does not go hand-in-hand with this theory, but if we look
closely, we can argue that the informal lenders incorporate this higher risk of default indirectly by
taking more mortgages against the loan, not by directly increasing the interest rate. To back this
argument up empirically, in panel (b) of table 6we show that the informal lenders increase the
likelihood of giving mortgaged loans15 to the SC group, alongside reducing the interest rate16.
Thus, we can infer that our findings do not contradict the cream-skimming theory. Summing
up, SCs benefit exclusively from the reduced interest rate in the informal credit market, which
enables them to take up more informal loans, mainly for consumption purposes, and this gets
reflected in higher consumption and reduced poverty for them as shown in table 5.
Insert Table 6about here
7.2 Business finance channel
To comprehend the labour market mechanism, we first need to examine how banks finance enter-
prises owned by various castes. We use the 2013 economic census to study this business finance
23

channel and present, in table 7, how the number of businesses with formal finance as their pri-
mary source of credit has changed as a result of the policy. The economic census data contains
information on the caste of the business owner, which allows us to examine how credit avail-
ability has changed across different caste categories after the policy. The economic census also
categorizes enterprises into two main groups: agricultural and non-agricultural. This enables us
to further analyze which caste group is taking more agricultural loans and which one is taking
more non-agricultural loans.
We can deduce three key findings from table 7. First, overall business loans increase for all
caste categories; second, agricultural loans increase for the SC and OBC owners; and third, non-
agricultural business loans increase for the OBC and general categories. In panel (a) of table
7, we report a significant increase in the number of enterprises with formal finance as the main
source of credit in the treatment districts; 43.69%, 66.27% and 31.37% increase for SC, OBC
and general owned businesses, respectively. This indicates that caste does not play a significant
barrier in accessing formal sector loans.
In panel (b) of table 7, we observe that agricultural loan shows a significant sign of increase for
SC and OBC categories in treatment districts, but not for generals. We argue that this is because
the agricultural loans were collateral-free even before the introduction of the 2005 RBI policy.
In 1998, RBI issued a circular saying agricultural loans up to INR 10,000 should be collateral-
free.17 In 2004, just one year before the 2005 branch authorization policy, this limit was further
increased to INR 50,000,18 and five years after the policy, in 2010, it was again increased to
INR 1,00,000.19 The collateral-free nature of agricultural loan had already made it attractive
for owners from marginalized castes. The RBI policy just removed the barrier to the access of
agricultural loans for them.20
On the other hand, non-agricultural business loans also show improvement for all caste categories
in treatment districts, but the increase is statistically significant only for OBCs and generals.
24

The number of OBC-owned and general-owned non-agricultural enterprises increases by 70.5%
and 52.36%, respectively, in treatment districts. Why do OBCs and generals take more non-
agricultural loans? We argue that, before the policy, these two caste groups were more likely to
own a non-agricultural business compared to SCs. The pre-policy economic census (2005) data
reveals that 37.67% and 39.02% non-agricultural businesses were owned by OBCs and generals,
respectively. In contrast, the same number for SCs was 8.58%. This disparity exists because
setting up a non-agricultural business requires a handsome amount of investment in fixed costs as
well as a wide network in society. SC owners lack both of these most of the time. Predictably, it is
easier for OBCs and generals to obtain a loan to expand their existing businesses than for the SCs
to obtain a larger loan to start a business from scratch. This is precisely why OBCs and generals
opt for more non-agricultural business loans when the availability of formal credit increases as a
result of the policy.
Insert Table 7about here
7.3 Labour market channel
Following our findings that SCs take more agricultural loans and generals take more non-agricultural
business loans, the next question we ask is, what do SCs and generals do with these loans? The
answer will provide insight into the labour market’s response to the policy. Below, we elaborate
on the mechanisms in agricultural sector and non-agricultural sector separately.
7.3.1 Agricultural Sector
Beginning with the agricultural sector, we present our findings in table 8. From AIDIS 2013 data,
we analyze the impact on the value of agricultural machinery, specifically the value of power-
operated agricultural machinery and the number of livestock (columns 1, 2 and 3). In treatment
25

districts, the value of agricultural machinery owned by SCs is 1946.32 rupees (approximately
49 USD) higher than in control districts; if only power-operated machinery is considered, the
difference is INR 15047 (approximately 376 USD21). The number of livestock owned increases
by 23.2% for SCs and 13.8% for OBCs on average, compared to control means of 2.62 and
3.22, respectively. Combining these results, we can infer that as a result of the policy, SCs
mainly use agricultural loans for mechanising their agricultural production and OBCs use them
for expanding their animal stock. Overall, for generals, we do not observe any discernible impact
in the agriculture sector, which is consistent with our findings in panel (b) of table 7.
To further support these findings, we use IHDS data (columns 4 and 5) to show that agricul-
tural labour hours decrease by approximately 7 hours per week (29.25% decrease compared to a
control mean of 24 hours) and that agricultural crop income increases by INR 3625.58 or (approx-
imately 91 USD) (around 87% increase) in SC households in the treatment districts compared to
SC households in the control districts. This indicates an increase in agricultural productivity, ar-
guably caused by the mechanisation of agricultural production by SCs. As anticipated, no change
in productivity is noticed for generals.
Insert Table 8about here
7.3.2 Non-agricultural Sector
Then, we turn to the non-agricultural sector, for which the results are furnished in table 9. The
data used in this table is from IHDS 2011-12. From panel (a), we observe that earnings from
non-agricultural businesses increase for OBCs and generals, by 9.2% and 4.5%, respectively.
This aligns with our previous finding that non-agricultural business loan increases for OBCs and
generals, with the rate of increase for OBCs being higher than that of generals. From this, we can
deduce that OBCs and generals expand their businesses with loans from the formal sector.
26

The existence of discrimination against marginalized castes in the form of wage differential is an
established phenomenon in the literature (Madheswaran and Attewell,2007;Thorat and Attewell,
2007;Ito,2009). The extent of discrimination was the highest for SCs, whereas OBC’s position
was somewhere in between SCs and generals. Our data reflect the same story. In IHDS 2004-
05 data, the average hourly wages for the SC and OBC categories were INR 17.44 (0.43 USD)
and INR 19.62 (0.49 USD), while the same for the general category was INR 34.94 (0.87 USD).
The business expansion in the non-agricultural sector resulting from increased access to formal
credit creates an additional labour demand in the market. Therefore, the business owners find
it more rational to meet that excess demand by raising the wages of cheaper SC labourers. We
furnish evidence for this in panel (b) of table 9by showing an increase in SC wages by 11.9% in
treatment districts. In a broader sense, this implicates a reduction in labour market discrimination
against marginalized castes, which contributes to the reduction in caste-based welfare disparity.
On the other hand, from the labour supply side, we see a 14.2% increase in the number of wage
or salary jobs in SC households. We argue that the disguised agricultural labourers in those
households move to wage or salary jobs after the increase in agricultural productivity, as seen in
columns (4) and (5) of table 8. This is another additional welfare-improving channel for the SCs.
Insert Table 9about here
8 Robustness checks
8.1 Class vs Caste Effect
One inevitable question that arises from the results in table 5is how do we ensure our results are
attributable to caste effect and not class effect? The RD coefficients in our results compare the
outcome difference between a particular caste group in the treatment districts with the same caste
27

group in the control districts. This eliminates the class effect while comparing the same caste
groups.
To further eliminate the class effect within a particular caste group, we control for baseline perma-
nent income and re-estimate the models in table 5, which is presented in table A9. We construct
the permanent income using the adaptive LASSO model22. Our results remain intact even after
controlling for permanent income, which ensures the origin of the impact is caste discrimination.
8.2 Quadratic Estimations
All the results presented in sections 6and 7are linear RD estimates. According to Gelman and
Imbens (2019), quadratic approximation is the highest order polynomial that researchers should
use because using higher degree polynomials results in noisy estimates, sensitivity to the order
of the polynomial, and inadequate coverage of confidence intervals resulting in poor inference.
We present the quadratic polynomial estimates in tables A10,A11,A12,A13,A14 and A15.
These estimates are similar to our main results (linear estimates) in tables 4,5,6,7,8and 9,
respectively. We find that 88% of the main results and 83% of the mechanism results remain
statistically significant for second-order polynomial estimates. Thus, we can safely assert that
the results are robust to quadratic estimation as well.
8.3 Placebo cutoffs
Checking for the smoothness around placebo cutoffs is considered to be another robustness check
in RD literature. Intuitively, since the likelihood of obtaining the treatment changes discontinu-
ously only at the true cutoff value, the outcomes should also change discontinuously only at that
cutoff value (Cattaneo and Titiunik,2022). The underlying assumption for this placebo cutoff
28

test is the absence of any similar policy during the same period. Following Cramer (2021b), we
can say that this assumption is satisfied for our RBI branch authorization policy (2005).
To carry out the placebo cutoff test, we consider six placebo cutoffs, three on each side of the
true cutoff (which is normalized to zero), ±750, 1,500 and 2,250. Results of the placebo cutoff
tests are shown in table A16 for the main results, and in tables A17 and A18 for the mechanism
results23. We find that 96% of the results pass through the placebo cutoff test.
8.4 Bandwidth Selector
Next, we show that our results are robust to the choice of bandwidth selection method. In the
main results, we use the common Mean-square-error (MSE)-optimal method following (Calonico
et al.,2019), which selects equal bandwidth for both sides of the cutoff. As an alternative, we
use a two-sided MSE optimal bandwidth selector that separately chooses optimal bandwidth for
each side of the cutoff. We also use Coverage-error-rate (CER)-optimal bandwidth selector (both
common and two-sided) following Calonico et al. (2020). The main difference between MSE and
CER methods is that the former aims to minimize the mean square error of the point estimator,
whereas the latter aims to minimize the coverage error of the interval estimator (Calonico et al.,
2020).
We present the results using different bandwidth selection methods in tables A19,A20 and A21.
72% of the main results and 75% of the mechanism results remain statistically significant, which
suggests that our results are robust to different bandwidth selection methods.
29

8.5 Bandwidth Multipliers
Another method of checking whether the coefficients remain statistically significant for different
bandwidth choices is to check for bandwidth multipliers. We consider multipliers in the range
of 0.50 to 1.50, with gaps of 0.25. That is, if the MSE-optimal bandwidth in the main result
is x, we additionally examine bandwidths of 0.50x, 0.75x, 1.24x, and 1.50x. The results are
furnished in tables A22,A23 and A24. 80% of the main results and 72% of the mechanism
results remain statistically significant. Hence, our results are robust to the selection of different
bandwidth multipliers.
8.6 Donut Hole Test
The broad objective of the donut hole sensitivity test is to check if our results are drastically
determined by the observations closest to the cutoff (Cattaneo et al.,2023). To carry out this
test, we create the so-called ’donut’ (Dowd,2021) by omitting the closest 1% observations from
both sides of the cutoff and re-estimate the RD treatment effects. The results remain intact for
the donut, as furnished in table A25, which signifies that our results pass through the donut-hole
sensitivity test.
9 Conclusion
In a general equilibrium framework, our study examines the causal effect of a bank branch autho-
rization policy on caste-based welfare disparities in India. Our findings demonstrate that commer-
cial bank branch expansion policies in underbanked districts enhanced access to formal financial
services across all caste categories, thereby promoting inclusivity, diversity and unbiasedness.
Marginalized caste categories, who historically faced the highest degree of social exclusion and
30

discrimination in Indian societies, benefit the most from expansion of banks in underbanked
areas, in terms of increased consumption and reduced poverty. We examine all direct and indi-
rect channels that facilitated the welfare gain of marginalized castes, including informal finance,
business finance and labour market channels. Our results are based on a regression discontinuity
framework and are, therefore, causal in nature. The results are also robust to standard RD checks,
including bandwidth multiplier test, bandwidth selector test, placebo cutoff test, quadratic esti-
mates and the donut hole test. The robustness of our findings underlines the substantial impor-
tance of this study for policymakers, particularly in relation to policies formulated by the central
bank. We highlight the importance of strengthening the formal banking sector in order to reduce
sticky social norms of social discrimination.
Notes
1Link to the United Nations Article on the number of people enduring caste based discrimination.
2Brahmins (priests), Kshatriyas (warriors), Vaishyas (merchants), and Shudras (menial workers) were the four
major varnas in India. One other group (the so-called Dalits) fell outside the caste system, and they were considered
“untouchables” (Bidner and Eswaran,2015).
3This policy is often termed as the Social Banking Policy of India. For more details, see Burgess et al. (2005);
Burgess and Pande (2005).
4For details of the policy framework, please refer to section 5.1.
5Tendulkar Committee was formed in 2009 with Suresh Tendulkar as the chairperson. This is a consumption-
based poverty line, which has been used by the Planning Commission of India (Thorat et al.,2017). Instead of using
the Universal Reference period, the Committee used the Mixed Reference period and recommended new poverty
lines for urban and rural areas.
6For the education dimension, we use two indicators: school enrolment of children, years of education of adult
members. Indicators for the health dimension are infant mortality and malnutrition of the members. Indicators
31

for living standards are clean cooking fuel, electricity, safe drinking water, proper sanitation, pucca flooring and
possession of household durables like TV, motorcycle, refrigerator etc.
7From IHDS, we get information on whether the household is a member of different social groups, like religious
groups, caste associations, cooperatives, unions, ROSCAs, self-help groups, Panchayet, and NGOs. We contend that
membership in these groups can be an indicator of social inclusion.
8All these outcome variables are drawn from IHDS household file, except multidimensional poverty, which is
drawn from the individual data. This is because we need education and health data of each household member to
calculate the multidimensional poverty.
9The binomial test verifies if the number of observations in the control and treatment groups around the cutoff is
significantly different from the expected number in a random sample of Bernoulli trials with a specific probability.
Unlike the McCrary test, this test does not depend on asymptotic approximations (Cattaneo and Titiunik,2022).
10For example, the ST population share in Arunachal Pradesh is 88.15%, in Nagaland 80.59%, in Mizoram
99.35%, in Meghalaya 82.09% etc (Calculated from IHDS).
11The number of observations in the bank loan variable is smaller compared to the other financial inclusion
indicators. This is because we restrict our sample to all the households who took any loan in last five years and
our objective is to understand if the loan-taking households in treatment districts opt more for formal bank loans
compared to control districts.
12We estimate the impacts on other types of consumption as well, including household appliances, clothing and
utensils. All of these significantly increase only for the SCs.
13Reduced interest rate for SCs implicates lower discrimination against them in the credit market, which ripples
through other spheres of the society. The evidence of an increase in social inclusion for SCs supports this hypothesis.
14In some parts of India, OBCs face discrimination to some extent, but not as prominent as SCs.
15Mortgage is a dummy variable; takes the value 1 if mortgage is taken with the loan and 0 if the loan is mortgage-
free.
16This does not impact the borrowers much because, with lower interest rate, they are now more able to pay back
the loan. But increased mortgaged loans help the informal lenders to minimize their risk of lending.
32

17Circular number: RPCD. No. PLFS. BC. 123/05.05.18/1997-98 dated May 20, 1998.
18Circular number: RPCD. Plan. BC. No. 87 /04.09.01/2003-04 dated May 18, 2004.
19Circular number: RPCD.PLFS. BC. No. 85/05.04.02/2009-10 dated June 18, 2010.
20One might argue that loans for micro, small and medium enterprises (MSME) are also collateral-free in India.
However, MSME loans up to INR 5 lakhs are made collateral-free in 2009, four years after the policy. Naturally,
awareness for this policy is expected to be less compared to the 1998 collateral-free agricultural loan policy. Also,
MSME loans are generally larger than agricultural loans, which makes them non-feasible for most of the marginal-
ized caste business owners.
21INR to USD conversion is done using 2012 exchange rate, which was around 40 INR/USD.
22LASSO (Least Absolute Shrinkage and Selection Operator) is a regression-based method to select a set of
variables for prediction from a large pool of variables. LASSO is also a regularization method that penalizes for
over-fitting. Adaptive LASSO is an improvement over the LASSO in the sense that it has the oracle properties.
23For space constraint, we have shown the results only for the SC category, but similar results hold for OBC and
generals as well.
10 Tables
33

Table 1: Summary statistics from IHDS data
IHDS 1 IHDS 2
SC OBC Gen SC OBC Gen
(a) Financial inclusion
Bank account - - - 0.51 (0.49) 0.55 (0.49) 0.68 (0.46)
Observations - - - 8560 13573 8604
Bank loan 0.09 (0.28) 0.13 (0.34) 0.14 (0.34) 0.16 (0.36) 0.26 (0.44) 0.25 (0.43)
Observations 8533 13908 8428 8577 13614 8627
Fixed deposit - - - 0.06 (0.25) 0.08 (0.27) 0.17 (0.38)
Observations - - - 8562 13574 8599
Securities - - - 0.01 (0.09) 0.01 (0.10) 0.03 (0.16)
Observations - - - 8562 13575 8602
Insurance 0.15 (0.36) 0.21 (0.41) 0.35 (0.47) 0.30 (0.45) 0.38 (0.48) 0.47 (0.49)
Observations 8518 13891 8396 8569 13585 8617
(b) Household welfare
Consumption 2.89 (1.39) 3.13 (1.39) 3.75 (1.28) 2.75 (1.34) 3.05 (1.38) 3.61 (1.30)
Observations 8526 13896 8413 8580 13610 8624
Food consumption 3.50 (1.25) 3.61 (1.24) 4.01 (1.09) 2.02 (1.08) 2.06 (1.09) 2.53 (1.20)
Observations 8526 13896 8413 7601 12081 7749
Poverty 0.27 (0.44) 0.21 (0.41) 0.09 (0.29) 0.22 (0.41) 0.15 (0.36) 0.07 (0.25)
Observations 8526 13896 8413 8580 13610 8624
Multidimensional poverty 0.90 (0.28) 0.86 (0.33) 0.73 (0.44) 0.86 (0.33) 0.82 (0.38) 0.67 (0.46)
Observations 30908 51599 31086 31276 50481 31990
Social inclusion 0.12 (0.33) 0.16 (0.37) 0.12 (0.32) 0.14 (0.34) 0.17 (0.38) 0.18 (0.38)
Observations 8487 13823 8380 8538 13530 8593
(c) Agricultural Sector
Agriculture hours 24.17 (13.82) 20.75 (12.49) 22.83 (12.66) 20.08 (12.90) 17.50 (12.47) 20.01 (13.04)
Observations 4803 5708 1457 5196 5995 1343
Agriculture income 7723.22 (35896) 22991.65 (90388) 32591.49 (113804) 9635.52 (131654) 27986.5 (133602) 38560.28 (160651)
Observations 31462 52191 31579 31655 51113 32333
(d) Non-agri Sector
Business Revenue 10.89 (1.35) 11.11 (1.32) 11.85 (1.26) 11.06 (1.27) 11.37 (1.31) 11.90 (1.31)
Observations 1106 2806 1916 1089 2808 1938
Hourly Wage 17.43 (17.61) 19.62 (23.24) 34.94 (42.38) 23.77 (23.13) 24.82 (28.28) 39.24 (43.71)
Observations 9412 11619 5449 12317 14093 6878
Number of jobs 1.11 (0.36) 1.10 (0.36) 1.02 (0.17) 1.33 (0.67) 1.27 (0.66) 1.29 (0.41)
Observations 4657 5696 2924 6081 7077 3468
Source: Authors’ calculation. Standard deviations in parenthesis. Missing values indicate that there is no
data for that particular variable at that particular time point.
34

Table 2: Summary Statistics of variables from AIDIS
AIDIS 2003 AIDIS 2013
SC OBC Gen SC OBC Gen
(a) Informal finance
Informal rate of interest 23.79 (25.77) 22.10 (21.91) 16.36 (19.73) 34.72 (16.97) 32.51 (15.77) 30.25 (16.83)
Observations 6302 14287 6189 7401 16,856 6426
No mortgage informal loan 0.95 (0.20) 0.95 (0.20) 0.96 (0.19) 0.87 (0.32) 0.88 (0.31) 0.92 (0.26)
Observations 9927 22018 13368 12127 27828 14037
Informal loan 0.55 (0.49) 0.56 (0.49) 0.46 (0.49) 0.31 (0.46) 0.29 (0.45) 0.21 (0.41)
Observations 17789 39319 28812 38436 95887 64034
(b) Agricultural Sector
Value of
agricultural machinery 1420.73 (11850) 3330.55 (21041) 5240.86 (27735) 2931.55 (11499) 6117.15 (21043) 8572.37 (64259)
Observations 26,632 63174 50881 17033 28850 18120
Value of agricultural
machinery (power operated) 8430.45 (34180) 15738.65 (47996) 20015.51 (56202) 14831.85 (29938) 19694.27 (39383) 26779.65 (126547)
Observations 2980 10770 11003 1154 6721 4461
Number of livestocks 2.47 (1.70) 3.18 (2.51) 3.30 (2.40) 2.39 (1.55) 3.05 (2.38) 3.12 (2.45)
Observations 7532 17438 13459 4912 13481 8133
Source: Authors’ calculation. Standard deviations in parenthesis.
Table 3: Summary statistics of variables from Economic Census
EC 2005 EC 2013
SC OBC Gen SC OBC Gen
All Enterprises 151.54 (265.66) 808.38 (1597.79) 1106.76 (2217.45) 147.01 (260.86) 672.21 (1310.49) 806.20 (1604.54)
Observations 581 581 581 581 581 581
Agricultural Enterprises 25.28 (99.52) 103.86 (394.35) 115.34 (764.75) 4.11 (25.63) 29.86 (193.70) 24.71 (119.95)
Observations 581 581 581 581 581 581
Non-agricultural Enterprises 126.25 (200.19) 704.51 (1286.61) 991.42 (1733.73) 142.91 (256.48) 642.35 (1285.99) 781.49 (1542.98)
Observations 581 581 581 581 581 581
Source: Authors’ calculation. Standard deviations in parenthesis.
35

Table 4: Banks and financial inclusion of SC, OBC and General caste
(1) (2) (3)
SC OBC Gen
(a) Bank account (0/1)
Treatment 0.400***
(0.157)
0.149**
(0.084)
0.263***
(0.087)
Control mean 0.56 0.55 0.72
Robust p value 0.008 0.035 0.004
Bandwidth 3600 5060 5342
Effective obs 3920 8510 4673
Observations 8,451 13,291 8,425
(b) Bank loan (0/1)
Treatment 0.143*
(0.074)
0.095
(0.077)
0.120*
(0.072)
Control mean 0.35 0.50 0.58
Robust p value 0.066 0.242 0.095
Bandwidth 5664 4107 6605
Effective obs 2991 4297 2767
Observations 4,813 8,088 3,997
(c) Fixed deposit (0/1)
Treatment 0.036
(0.024)
0.115***
(0.041)
0.082**
(0.043)
Control mean 0.09 0.11 0.22
Robust p value 0.113 0.007 0.035
Bandwidth 5666 3490 7255
Effective obs 5107 6362 6649
Observations 8,453 13,292 8,420
(d) Securities (0/1)
Treatment -0.002
(0.002)
0.009
(0.007)
0.022*
(0.010)
Control mean 0.01 0.02 0.04
Robust p value 0.309 0.330 0.085
Bandwidth 2253 3834 6749
Effective obs 2664 6862 5949
Observations 8,453 13,293 8,423
(e) Insurance (0/1)
Treatment 0.135**
(0.054)
0.045
(0.066)
-0.004
(0.048)
Control mean 0.32 0.43 0.53
Robust p value 0.024 0.613 0.745
Bandwidth 4418 5940 6639
Effective obs 4467 6989 5894
Observations 8,452 13,293 8,437
Note: This table presents the effectiveness of the RBI bank expansion policy of
2005 in enhancing financial inclusion across three caste categories. We look into
the impacts on the likelihood of having a savings/current account with the bank
(panel a), taking a loan from the bank (panel b), having a fixed deposit account
(panel c), purchasing securities such as mutual fund, shares, unit trust and bonds
(panel d) and having life/health insurance (panel e). Robust standard errors in
parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at
district level. Data used: IHDS 2011-12. District population and number of
bank branches in 1996 are controlled for. Source: Authors’ calculation.
36

Table 5: Banks and household welfare of SC, OBC and General caste
(1) (2) (3)
SC OBC Gen
(a) Consumption quintiles
Treatment 0.488**
(0.300)
0.121
(0.246)
-0.005
(0.223)
Control mean 3.08 3.37 3.83
Robust p value 0.050 0.550 0.864
Bandwidth 3176 3851 5931
Effective obs 3652 7053 5285
Observations 8,580 13,610 8,624
(b) Food consumption quintiles
Treatment 0.416**
(0.179)
0.398**
(0.202)
0.110
(0.215)
Control mean 2.19 2.23 2.68
Robust p value 0.014 0.040 0.457
Bandwidth 3792 3997 6103
Effective obs 3610 6375 4973
Observations 7,601 12,081 7,749
(c) Poverty (0/1)
Treatment -0.104***
(0.049)
-0.033
(0.049)
-0.011
(0.024)
Control mean 0.174 0.098 0.047
Robust p value 0.009 0.393 0.260
Bandwidth 4635 4997 5160
Effective obs 4718 8605 4751
Observations 8,580 13,610 8,624
(d) Multidimensional poverty (0/1)
Treatment -0.061**
(0.033)
-0.090*
(0.062)
-0.127**
(0.067)
Control mean 0.802 0.727 0.609
Robust p value 0.041 0.068 0.034
Bandwidth 5483 3561 7071
Effective obs 18363 23980 24424
Observations 31,090 50,181 31,333
(e) Social inclusion (0/1)
Treatment 0.172*
(0.084)
-0.018
(0.096)
0.083
(0.078)
Control mean 0.14 0.24 0.21
Robust p value 0.058 0.628 0.290
Bandwidth 5843 3676 6938
Effective obs 5198 6747 6420
Observations 8,538 13,530 8,593
Note: This table presents the impact of the RBI bank expansion policy of 2005 on five household
welfare outcomes for three caste categories. Poverty in panel (c) represents the likelihood of falling
below the Tendulkar poverty line. Multidimensional poverty in panel (d) represents the likelihood
of being multidimensionally poor, which is calculated using three dimensions: health, education
and asset/infrastructure. Social inclusion in panel (e) is a dummy that takes value 1 if the household
is a member of one or more socioeconomic groups in the community. Robust standard errors in
parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data
used: IHDS 2011-12 household file for panels (a), (b), (c), (e) and individual file for panel (d).
District population and number of bank branches in 1996 are controlled for. Source: Authors’
calculation. 37

Table 6: Banks and informal finance channel
(1) (2) (3)
SC OBC Gen
(a) Annual interest rate on informal loan (%)
Treatment -8.113**
(3.329)
-5.025*
(2.574)
-3.377
(2.586)
Control mean 32.89 30.17 27.79
Robust p value 0.016 0.092 0.189
Bandwidth 3016 4312 4670
Effective obs 3613 9958 3604
Observations 7,401 16,856 6,426
(b) No mortgage informal loan (0/1)
Treatment -0.152*
(0.097)
-0.131**
(0.080)
0.004
(0.021)
Control mean 0.87 0.87 0.93
Robust p value 0.075 0.050 0.949
Bandwidth 3521 3819 5234
Effective obs 5554 13374 7827
Observations 12,127 27,828 14,037
(c) Informal loan (0/1)
Treatment 0.090**
(0.040)
-0.022
(0.029)
-0.006
(0.035)
Control mean 0.32 0.28 0.21
Robust p value 0.015 0.587 0.814
Bandwidth 4395 5307 5732
Effective obs 19693 60570 37941
Observations 38,436 95,887 64,034
Note: This table presents the informal finance channel to explain why SCs benefit the most from the RBI bank expansion
policy of 2005. We look into the impacts on annual interest rate paid for informal loans (panel a), the likelihood of having an
informal loan without any mortgage and lastly, the likelihood of taking an informal loan. Robust standard errors in parentheses
(*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: AIDIS 2013. District population
and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
38

Table 7: Banks and the number of enterprises with formal finance as main source of credit
(1) (2) (3)
SC OBC Gen
(a) All enterprises
Treatment 80.80*
(50.85)
647.41**
(298.73)
376.27**
(237.17)
Control mean 184.92 976.91 1199.38
Robust p value 0.081 0.031 0.048
Bandwidth 4481 4486 4805
Effective obs 284 284 296
Observations 581 581 581
(b) Agricultural enterprises
Treatment 7.64*
(5.39)
31.31*
(21.33)
7.24
(22.03)
Control mean 5.64 39.98 35.93
Robust p value 0.083 0.093 0.305
Bandwidth 4160 4361 4237
Effective obs 260 277 268
Observations 581 581 581
(c) Non-agricultural enterprises
Treatment 83.51
(65.86)
661.1**
(291.9)
609.28*
(375.38)
Control mean 179.28 936.94 1163.45
Robust p value 0.159 0.024 0.073
Bandwidth 4608 4405 4960
Effective obs 280 280 307
Observations 581 581 581
Note: This table presents the business finance channel to examine how the number of enterprises in the district
with formal finance as main source of credit has changed following the RBI bank expansion policy of 2005. Panel
(a) considers all types of enterprises, panel (b) considers only agricultural enterprises and panel (c) considers only
non-agricultural business enterprises. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).
Standard errors clustered at district level. Data used: Economic Census 2013. District population, number of bank
branches in 1996 and pre-policy values of the outcome variables are controlled for. Source: Authors’ calculation.
39

Table 8: Banks and agricultural sector
(1) (2) (3) (4) (5)
Value of agricultural
machinery
Value of agricultural
machinery: power-operated
Number of
livestock
Labour hours:
agriculture (weekly)
Income:
agriculture
(a) SC
Treatment 1,946.328**
(838.012)
15047***
(7,288.128)
0.608*
(0.352)
-7.109**
(3.241)
3,625.582**
(1,969.989)
Control mean 2118.502 8778.52 2.62 24.30 4166.69
Robust p value 0.024 0.004 0.060 0.042 0.047
Bandwidth 3926 3394 2709 4336 3666
Effective obs 3767 401 1232 3160 4033
Observations 9,632 1,154 4,912 5,196 8,583
(b) OBC
Treatment -729.233
(1,315.220)
2,760.711
(2,190.110)
0.446**
(0.248)
-4.671
(3.398)
-1,281.121
(4,680.583)
Control mean 3882.38 12336.34 3.22 22.87 9225.86
Robust p value 0.668 0.123 0.044 0.401 0.804
Bandwidth 4791 4621 8414 3712 3997
Effective obs 14929 3305 10703 3157 7275
Observations 28,850 6,721 13,481 5,995 13,619
(c) Gen
Treatment 2,104.938
(1,488.745)
693.683
(2,151.968)
0.230
(0.417)
-4.973
(4.246)
-4,361.575
(6,035.694)
Control mean 5330.11 14374.01 3.45 22.59 13092.28
Robust p value 0.102 0.597 0.513 0.549 0.556
Bandwidth 4408 7004 4289 4110 5739
Effective obs 7936 3231 3572 761 5136
Observations 18,120 4,461 8,133 1,343 8,630
Note: This table presents the impact of the RBI bank expansion policy of 2005 on agricultural sector outcomes for three caste categories to
explain the labour market channel. All monetary values are in Indian rupees. Robust standard errors in parentheses (*** p<0.01, ** p<0.05,
* p<0.1). Standard errors clustered at district level. Data used: IHDS 2011-12 and AIDIS 2013. District population and number of bank
branches in 1996 are controlled for. Source: Authors’ calculation.
40

Table 9: Banks and Non-agricultural business sector
(1) (2) (3)
SC OBC Gen
(a) Log nonfarm business revenue
Treatment 0.481
(0.332)
1.06***
(0.327)
0.551**
(0.306)
Control mean 11.20 11.48 12.04
Robust p value 0.103 0.001 0.039
Bandwidth 4879 3486 5070
Effective obs 581 1301 1026
Observations 1089 2808 1938
(b) Hourly wage/salary (Rs)
Treatment 2.84*
(1.72)
2.58
(2.24)
2.59
(2.28)
Control mean 23.75 24.53 27.61
Robust p value 0.079 0.363 0.270
Bandwidth 5567 3700 5067
Effective obs 7285 6839 3083
Observations 11,464 13,076 5,591
(c) Number of wage/salary jobs in the household
Treatment 0.172*
(0.105)
0.139
(0.120)
0.035
(0.074)
Control mean 1.21 1.12 1.08
Robust p value 0.081 0.191 0.566
Bandwidth 4488 4791 5946
Effective obs 3348 4419 2157
Observations 6,081 7,077 3,468
Note: This table presents the impact of the RBI bank expansion policy of 2005 on non-agricultural business sector
outcomes for three caste categories to explain the labour market channel. Robust standard errors in parentheses (***
p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: IHDS 2011-12, household
file [for panels (a) and (c)] and individual file [for panel (b)]. District population and number of bank branches in
1996 are controlled for. For hourly wage, we winsorize the outcome at 5th and 95th percentile to remove outliers
and additionally control for pre-policy values of the outcome, education, union membership and region dummies to
separate out the wage increment induced by reduced labour market discrimination. Source: Authors’ calculation.
41

References
Abadie, A., Athey, S., Imbens, G. W., and Wooldridge, J. M. (2023). When should you adjust
standard errors for clustering? The Quarterly Journal of Economics, 138(1):1–35.
Alkire, S. and Foster, J. (2011). Counting and multidimensional poverty measurement. Journal
of Public Economics, 95(7-8):476–487.
Anagol, S., Cole, S., and Sarkar, S. (2017). Understanding the advice of commissions-motivated
agents: Evidence from the Indian life insurance market. Review of Economics and Statistics,
99(1):1–15.
Anderson, S. (2011). Caste as an impediment to trade. American Economic Journal: Applied
Economics, 3(1):239–263.
Arena, M. (2006). Does Insurance Market Activity Promote Economic Growth?: A
Cross-country Study for Industrialized and Developing Countries, volume 4098. World Bank
Publications.
Arunachalam, R. and Shenoy, A. (2017). Poverty traps, convergence, and the dynamics of house-
hold income. Journal of Development Economics, 126:215–230.
Ashraf, N., Karlan, D., and Yin, W. (2006). Tying odysseus to the mast: Evidence from a commit-
ment savings product in the philippines. The Quarterly Journal of Economics, 121(2):635–672.
Bagde, S., Epple, D., and Taylor, L. (2016). Does affirmative action work? caste, gender, college
quality, and academic success in india. American Economic Review, 106(6):1495–1521.
Banerjee, A., Duflo, E., Ghatak, M., and Lafortune, J. (2013). Marry for what? caste and mate
selection in modern india. American Economic Journal: Microeconomics, 5(2):33–72.
Banerjee, A. V., Breza, E., Duflo, E., and Kinnan, C. (2017). Do credit constraints limit en-
trepreneurship? heterogeneity in the returns to microfinance. Heterogeneity in the Returns to
Microfinance (September 1, 2017). Global Poverty Research Lab Working Paper, (17-104).
Banerjee, B. and Knight, J. B. (1985). Caste discrimination in the indian urban labour market.
Journal of Development Economics, 17(3):277–307.
Barboni, G., Field, E., and Pande, R. (2022). Rural banks can reduce poverty: Experimental
evidence from 870 indian villages. Working Paper.
Becker, G. S. (2010). The economics of discrimination. University of Chicago press.
Bidner, C. and Eswaran, M. (2015). A gender-based theory of the origin of the caste system of
india. Journal of Development Economics, 114:142–158.
Burgess, R. and Pande, R. (2005). Do rural banks matter? evidence from the indian social
banking experiment. American Economic Review, 95(3):780–795.
42

Burgess, R., Pande, R., and Wong, G. (2005). Banking for the poor: Evidence from india. Journal
of the European Economic Association, 3(2-3):268–278.
Calonico, S., Cattaneo, M. D., and Farrell, M. H. (2020). Optimal bandwidth choice for ro-
bust bias-corrected inference in regression discontinuity designs. The Econometrics Journal,
23(2):192–210.
Calonico, S., Cattaneo, M. D., Farrell, M. H., and Titiunik, R. (2019). Regression discontinuity
designs using covariates. Review of Economics and Statistics, 101(3):442–451.
Cattaneo, M. D., Keele, L., and Titiunik, R. (2023). A guide to regression discontinuity designs
in medical applications. arXiv preprint arXiv:2302.07413.
Cattaneo, M. D. and Titiunik, R. (2022). Regression discontinuity designs. Annual Review of
Economics, 14:821–851.
Cattaneo, M. D., Titiunik, R., and Vazquez-Bare, G. (2017). Comparing inference approaches
for rd designs: A reexamination of the effect of head start on child mortality. Journal of Policy
Analysis and Management, 36(3):643–681.
C´
elerier, C. and Matray, A. (2019). Bank-branch supply, financial inclusion, and wealth accumu-
lation. Review of Financial Studies, 32(12):4767–4809.
Conning, J. and Udry, C. (2007). Rural financial markets in developing countries. Handbook of
Agricultural Economics, 3:2857–2908.
Cramer, K. F. (2021a). Bank presence and health. Working Paper.
Cramer, K. F. (2021b). Bank presence and health. Available at SSRN 3917526.
Das, M. B. and Dutta, P. V. (2007). Does caste matter for wages in the indian labor market.
Washington, DC, USA: The World Bank.
Demirg¨
uc¸-Kunt, A. and Levine, R. (2009). Finance and inequality: Theory and evidence. Annu.
Rev. Financ. Econ., 1(1):287–318.
Demont, T. et al. (2010). The impact of microfinance on the informal credit market: An adverse
selection model. Technical report, Citeseer.
Deshpande, A. (2000). Does caste still define disparity? a look at inequality in kerala, india.
American Economic Review, 90(2):322–325.
Deshpande, A. (2011). The grammar of caste: Economic discrimination in contemporary India.
Oxford University Press.
Deshpande, A. and Sharma, S. (2016). Disadvantage and discrimination in self-employment:
caste gaps in earnings in indian small businesses. Small Business Economics, 46:325–346.
Dong, Y. and Lewbel, A. (2015). Identifying the effect of changing the policy threshold in
regression discontinuity models. Review of Economics and Statistics, 97(5):1081–1092.
43

Dowd, C. (2021). Donuts and distant cates: Derivative bounds for rd extrapolation. Available at
SSRN 3641913.
Dreze, J., Lanjouw, P. F., and Sharma, N. (1997). Credit in rural india: a case study. LSE
STICERD Research Paper No. DEDPS06.
Dupas, P. and Robinson, J. (2013). Savings constraints and microenterprise development: Ev-
idence from a field experiment in kenya. American Economic Journal: Applied Economics,
5(1):163–92.
Fehr, E. and Hoff, K. (2011). Introduction: Tastes, castes and culture: The influence of society
on preferences. The Economic Journal, 121(556):F396–F412.
Gelman, A. and Imbens, G. (2019). Why high-order polynomials should not be used in regression
discontinuity designs. Journal of Business & Economic Statistics, 37(3):447–456.
Ghate, P. B. (1992). Interaction between the formal and informal financial sectors: The Asian
experience. World Development, 20(6):859–872.
Gupta, N. and Sedai, A. (2023). Disentangling the effects of financial inclusion on household
well-being. Available at SSRN 4393250.
Hahn, J., Todd, P., and Van der Klaauw, W. (2001). Identification and estimation of treatment
effects with a regression-discontinuity design. Econometrica, 69(1):201–209.
Hanna, R. N. and Linden, L. L. (2012). Discrimination in grading. American Economic Journal:
Economic Policy, 4(4):146–68.
Hoff, K., Kshetramade, M., and Fehr, E. (2011). Caste and punishment: The legacy of caste
culture in norm enforcement. The economic journal, 121(556):F449–F475.
Hoff, K. and Pandey, P. (2014). Making up people—the effect of identity on performance in a
modernizing society. Journal of Development Economics, 106:118–131.
Holliday, S. C. (2017). Insurance: supporting development goals and creating markets. Technical
report, The World Bank.
Ito, T. (2009). Caste discrimination and transaction costs in the labor market: Evidence from
rural North India. Journal of Development Economics, 88(2):292–300.
Jack, W. and Suri, T. (2014). Risk sharing and transactions costs: Evidence from kenya’s mobile
money revolution. American Economic Review, 104(1):183–223.
Kaboski, J. P. and 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.
44

Kochar, A., Nagabhushana, C., Sarkar, R., Shah, R., and Singh, G. (2022). Financial access and
women’s role in household decisions: Empirical evidence from India’s national rural liveli-
hoods project. Journal of Development Economics, 155:102821.
Kumar, S. M. (2013). Does access to formal agricultural credit depend on caste? World
Development, 43:315–328.
Kumar, S. M. and Venkatachalam, R. (2019). Caste and credit: a woeful tale? The Journal of
Development Studies, 55(8):1816–1833.
Lee, D. S. and Lemieux, T. (2010). Regression discontinuity designs in economics. Journal of
Economic Literature, 48(2):281–355.
Luke, N. and Munshi, K. (2007). Social affiliation and the demand for health services: Caste and
child health in south india. Journal of development economics, 83(2):256–279.
Madheswaran, S. and Attewell, P. (2007). Caste discrimination in the indian urban labour market:
Evidence from the national sample survey. Economic and political Weekly, pages 4146–4153.
McCrary, J. (2008). Manipulation of the running variable in the regression discontinuity design:
A density test. Journal of Econometrics, 142(2):698–714.
Mookherjee, D. and Motta, A. (2016). A theory of interactions between mfis and informal
lenders. Journal of Development Economics, 121:191–200.
Mosse, D. (2018). Caste and development: Contemporary perspectives on a structure of discrim-
ination and advantage. World development, 110:422–436.
Munshi, K. (2019). Caste and the indian economy. Journal of Economic Literature, 57(4):781–
834.
Munshi, K. and 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.
Munshi, K. and Rosenzweig, M. (2009). Why is mobility in india so low? social insurance,
inequality, and growth. Technical report, National Bureau of Economic Research.
Ngo, T. M.-P. and Wahhaj, Z. (2012). Microfinance and gender empowerment. Journal of
Development Economics, 99(1):1–12.
Oh, S. (2023). Does identity affect labor supply? American Economic Review, 113(8):2055–83.
Petrikova, I. (2020). Perpetuating poverty through exclusion from social programmes: Lessons
from andhra pradesh. Oxford Development Studies, 48(1):33–55.
Rangarajan, C. et al. (2008). Report of the committee on financial inclusion. Ministry of Finance,
Government of India.
45

Sagrario Floro, M. and Ray, D. (1997). Vertical links between formal and informal financial
institutions. Review of Development Economics, 1(1):34–56.
Thorat, A., Vanneman, R., Desai, S., and Dubey, A. (2017). Escaping and falling into poverty in
india today. World development, 93:413–426.
Thorat, S. and Attewell, P. (2007). The legacy of social exclusion: A correspondence study of
job discrimination in India. Economic and political weekly, pages 4141–4145.
Young, N. (2017). Banking and growth: Evidence from a regression discontinuity analysis.
EBRD Working Paper.
46

Online Appendix
Table A1: Comparison of IHDS with other nationally representative datasets
Caste
Survey Survey year Other backward classes Scheduled Castes Scheduled Tribes Other
IHDS 2004-2005 41.79 21.14 7.06 30.01
NFHS -III 2005-2006 39.6 19.2 8.4 31.9
NSS 2004-2005 40.96 19.59 8.64 30.81
CENSUS 2001 NA 16.2 8.2 NA
Source: Indian Human Development Survey: Technical Paper No. 1 (Table 2)
Accessible from https://www.icpsr.umich.edu/web/pages/DSDR/idhs-II-data-guide.html
1

Figure A1: McCrary Test
Note: The null hypothesis of this test (McCrary,2008) is that the running variable’s density function is
continuous around the cutoff. The McCrary estimate is -0.1996, and the associated p-value is 0.8418.
Hence, we fail to reject the null hypothesis and conclude that there is no evidence of manipulation
around the cutoff. Data used: RBI master office file data. Source: Authors’ calculation.
2

Table A2: Binomial test
Window length/2
(h)
Observations
below cutoff
Observations
above cutoff p-value
600.000 11 20 0.1496
1200.000 35 42 0.4944
1800.000 54 62 0.5159
2400.000 71 90 0.1558
3000.000 88 111 0.1186
Note: Cutoff (c) is normalized to 0. Window (W)= [c-h, c+h]. The p-values associated with this
binomial test are calculated using an exact binomial distribution with probability=0.5.
3

Table A3: Pre-policy smoothness of household welfare outcomes
(1) (2) (3)
SC OBC Gen
(a) Consumption quintiles
Treatment 0.037 0.163 -0.389
(0.328) (0.297) (0.238)
Observations 8,526 13,896 8,413
(b) Food consumption quintiles
Treatment -0.071 0.068 -0.189
(0.314) (0.300) (0.161)
Observations 8,526 13,896 8,413
(c) Poverty
Treatment 0.080 0.075 0.046
(0.073) (0.064) (0.034)
Observations 8,526 13,896 8,413
(d) Multidimensional poverty
Treatment -0.043 -0.058 -0.060
(0.028) (0.037) (0.059)
Observations 30,639 51,329 30,534
(e) Social inclusion
Treatment 0.039 0.044 -0.097
(0.086) (0.099) (0.073)
Observations 8,487 13,823 8,380
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch
Authorization Policy in 2005. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).
Standard errors clustered at district level. Data used: IHDS 2004-05. District population and number of
bank branches in 1996 are controlled for. Source: Authors’ calculation.
4

Table A4: Pre-policy smoothness: Informal finance channel
(1) (2) (3)
SC OBC Gen
(a) Informal interest rate
Treatment 4.913 3.918 1.766
(5.402) (4.653) (3.329)
Observations 6,302 14,287 6,189
(b) No mortgage loan
Treatment -0.009 -0.009 0.000
(0.011) (0.013) (0.014)
Observations 9,927 22,018 13,368
(c) Informal loan
Treatment 0.081 0.041 -0.035
(0.056) (0.073) (0.053)
Observations 17,789 39,319 28,812
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch
Authorization Policy in 2005. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).
Standard errors clustered at district level. Data used: AIDIS 2003. District population and number of bank
branches in 1996 are controlled for. Source: Authors’ calculation.
5

Table A5: Pre-policy smoothness: Business finance channel
(1) (2) (3)
SC OBC Gen
(a) All enterprises
Treatment -1.099 -218.088 660.540
(115.199) (474.035) (1,162.673)
Observations 581 581 581
(b) Agricultural enterprises
Treatment 28.878 33.292 418.807
(61.384) (139.516) (477.259)
Observations 581 581 581
(c) Non-agricultural enterprises
Treatment -44.227 -248.278 137.121
(64.704) (386.039) (670.851)
Observations 581 581 581
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: Economic Census
2005. District population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
6

Table A6: Pre-policy smoothness: Agricultural sector
(1) (2) (3) (4) (5)
Value of agricultural
machinery
Value of agricultural
machinery: power-operated
Number of
livestock
Labour hours:
agriculture
Income:
agriculture
(a) SC
Treatment 60.997 1,087.810 0.282 0.128 1,012.208
(150.143) (920.832) (0.194) (3.571) (2,232.781)
Observations 26,632 2,980 7,532 4,803 8,533
(b) OBC
Treatment 106.004 149.144 0.120 0.002 -6,480.221
(314.894) (1,077.033) (0.428) (3.767) (4,716.493)
Observations 63,174 10,770 17,438 5,708 13,908
(c) Gen
Treatment 426.281 305.365 0.221 -0.897 -5,039.682
(445.848) (1,170.995) (0.289) (4.780) (5,061.666)
Observations 50,881 11,003 13,459 1,457 8,428
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: AIDIS 2003. District
population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
7

Table A7: Pre-policy smoothness: Non-agricultural business sector
(1) (2) (3)
SC OBC Gen
(a) Log nonfarm business revenue
Treatment 0.558 0.298 0.036
(0.545) (0.292) (0.312)
Observations 565 1,484 1,046
(b) Hourly wage/salary (Rs)
Treatment 1.428 4.019 3.653
(2.479) (2.922) (4.087)
Observations 8,451 10,267 4,517
(c) Number of wage/salary jobs in the household
Treatment 0.034 -0.130 -0.024
(0.064) (0.094) (0.023)
Observations 4,657 5,696 2,924
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Autho-
rization Policy in 2005. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors
clustered at district level. Data used: IHDS 2004-05. District population and number of bank branches in 1996 are
controlled for. Source: Authors’ calculation.
8

Table A8: Validity of RD Design For Survey Subsamples
(1) (2) (3) (4)
Full Sample SC Subsample OBC Subsample General Subsample
Treatment -0.0002 -0.0414 0.0279 0.0854
(0.1483) (0.1646) (0.1507) (0.1579)
Robust p-value 0.965 0.827 0.826 0.572
Bandwidth 4398 4065 4364 4337
Effective Observations 280 254 277 276
Observations 581 581 581 581
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India branch authorization policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Since all the coefficients are
insignificant here, we can conclude that in IHDS, full sample along with SC, OBC and General subsamples separately satisfy the randomization
prerequisite for carrying out RD analysis. Source: Authors’ calculation.
9

Table A9: Disentangling Class and Caste
(1) (2) (3)
SC OBC General
(a) Consumption Quintiles
Treatment 0.369* 0.118 -0.135
(0.230) (0.215) (0.146)
Observations 7965 12448 7428
(b) Food Consumption Quintiles
Treatment 0.303** 0.366** -0.033
(0.158) (0.183) (0.200)
Observations 7066 11078 6732
(c) Poverty
Treatment -0.065* -0.027 -0.016
(0.050) (0.039) (0.028)
Observations 7965 12448 7428
(d) Multidimensional Poverty
Treatment -0.058** -0.079** -0.102*
(0.026) (0.044) (0.063)
Observations 23898 38045 22278
(e) Social Inclusion
Treatment 0.164* -0.042 0.033
(0.087) (0.100) (0.073)
Observations 7926 12374 7401
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India branch authorization policy
in 2005. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level.
Data used: IHDS 2011-12. In addition to district population and number of bank branches in 1996, permanent income of the
household, predicted using the adaptive LASSO model, is controlled for. This helps us to disentangle caste effect from class
effect. Source: Authors’ calculation.
10

Table A10: Polynomial 2: Banks and financial inclusion
(1) (2) (3)
SC OBC Gen
(a) Bank account
Treatment 0.399*** 0.105* 0.301***
(0.159) (0.081) (0.102)
Observations 8,451 13,291 8,425
(b) Bank loan
Treatment 0.172** 0.094 0.144
(0.080) (0.078) (0.096)
Observations 4,813 8,088 3,997
(c) Fixed deposit
Treatment 0.043* 0.088*** 0.139***
(0.027) (0.032) (0.058)
Observations 8,453 13,292 8,420
(d) Securities
Treatment -0.003 0.014 0.006**
(0.004) (0.006) (0.015)
Observations 8,453 13,293 8,423
(e) Insurance
Treatment 0.144** 0.058 -0.007
(0.062) (0.053) (0.060)
Observations 8,452 13,293 8,437
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch
Authorization Policy in 2005. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).
Standard errors clustered at district level. Data used: IHDS 2011-12. District population and number of
bank branches in 1996 are controlled for. Source: Authors’ calculation.
11

Table A11: Polynomial 2: Banks and household welfare outcomes
(1) (2) (3)
SC OBC Gen
(a) Consumption quintiles
Treatment 0.607** 0.093 -0.092
(0.325) (0.255) (0.269)
Observations 8,580 13,610 8,624
(b) Food consumption quintiles
Treatment 0.434** 0.403** 0.194
(0.202) (0.210) (0.263)
Observations 7,601 12,081 7,749
(c) Poverty
Treatment -0.118** -0.012 -0.030
(0.057) (0.040) (0.035)
Observations 8,580 13,610 8,624
(d) Multidimensional poverty
Treatment -0.023 -0.073 -0.150**
(0.036) (0.061) (0.078)
Observations 31,090 50,181 31,333
(e) Social inclusion
Treatment 0.140* -0.020 0.125
(0.088) (0.085) (0.110)
Observations 8,538 13,530 8,593
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch
Authorization Policy in 2005. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).
Standard errors clustered at district level. Data used: IHDS 2011-12. District population and number of
bank branches in 1996 are controlled for. Source: Authors’ calculation.
12

Table A12: Polynomial 2: Informal finance channel
(1) (2) (3)
SC OBC Gen
(a) Informal interest rate
Treatment -8.113*** -5.025* -3.377*
(3.329) (2.574) (2.586)
Observations 7,401 16,856 6,426
(b) No mortgage loan
Treatment -0.152 -0.131* 0.004
(0.097) (0.080) (0.021)
Observations 12,127 27,828 14,037
(c) Informal loan
Treatment 0.090* -0.022 -0.006
(0.040) (0.029) (0.035)
Observations 38,436 95,887 64,034
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch
Authorization Policy in 2005. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).
Standard errors clustered at district level. Data used: AIDIS 2013. District population and number of bank
branches in 1996 are controlled for. Source: Authors’ calculation.
13

Table A13: Polynomial 2: Business finance channel
(1) (2) (3)
SC OBC Gen
(a) All enterprises
Treatment 6.552 752.849* 422.479
(42.749) (388.800) (274.271)
Observations 581 581 581
(b) Agricultural enterprises
Treatment 10.917 39.953 19.697
(8.389) (61.826) (28.549)
Observations 581 581 581
(c) Non-agricultural enterprises
Treatment 89.499 739.586* 739.006
(79.884) (385.309) (470.991)
Observations 581 581 581
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: Economic census
2013. District population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
14

Table A14: Polynomial 2: Agricultural sector
(1) (2) (3) (4) (5)
Value of agricultural
machinery
Value of agricultural
machinery: power-operated
Number of
livestocks
Labour hours:
Agriculture
Income:
Agriculture
(a) SC
Treatment 2,013.037*** 16,206.719** 0.239 -7.480** 3,093.469**
(841.527) (7,288.128) (0.316) (3.614) (1,910.163)
Observations 9,632 1,154 4,912 5,196 8,583
(b) OBC
Treatment -979.696 2,494.162* 0.669** -4.256 -2,542.783
(1,244.245) (1,986.704) (0.334) (3.356) (4,176.103)
Observations 28,850 6,721 13,481 5,995 13,619
(c) Gen
Treatment 2,900.308** 1,429.467 0.345 -3.060 -4,367.131
(1,684.957) (2,778.128) (0.486) (4.683) (7,417.267)
Observations 18,120 4,461 8,133 1,343 8,630
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: AIDIS 2013 and IHDS
2011-12. District population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
15

Table A15: Polynomial 2: Non-agricultural business sector
(1) (2) (3)
SC OBC Gen
(a) Log nonfarm business revenue
Treatment 0.558 0.298 0.036
(0.545) (0.292) (0.312)
Observations 565 1,484 1,046
(b) Hourly wage/salary (Rs)
Treatment 3.518 2.068 4.035
(2.276) (2.159) (2.684)
Observations 11,464 13,076 5,591
(c) Number of wage/salary jobs in the household
Treatment 0.203** 0.176 0.070
(0.124) (0.127) (0.090)
Observations 6,081 7,077 3,468
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: IHDS 2011-12. District
population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
16

Table A16: Placebo cutoff test: Financial inclusion and household welfare outcomes for SCs
(1) (2) (3) (4) (5) (6) (7)
c=-2250 c=-1500 c=-750 c=0 c=750 c=1500 c=2250
Financial inclusion (SC)
(a) Bank account
Treatment 0.02 0.77 0.01 0.40*** -0.35 1.34* 1.47
(1.18) (0.44) (0.35) (0.16) (0.45) (0.45) (0.88)
Observations 8,451 8,451 8,451 8,451 8,451 8,451 8,451
(b) Bank loan
Treatment 0.26 0.72 0.07 0.14* 1.26 0.56 0.71
(0.43) (0.49) (0.17) (0.07) (0.96) (0.39) (0.49)
Observations 4,813 4,813 4,813 4,813 4,813 4,813 4,813
(c) Insurance
Treatment 0.96* 0.31 0.22 0.13** 0.21 0.36 -0.05
(1.00) (0.21) (0.18) (0.05) (0.39) (0.26) (0.38)
Observations 8,452 8,452 8,452 8,452 8,452 8,452 8,452
Welfare outcomes (SC)
(d) Consumption quintiles
Treatment -3.66 0.15 0.12 0.49** 1.99 0.51 2.75
(5.00) (1.14) (0.91) (0.30) (4.72) (1.39) (2.07)
Observations 8,580 8,580 8,580 8,580 8,580 8,580 8,580
(e) Food consumption quintiles
Treatment -4.20 -0.50 3.00 0.42** 1.04 0.80 2.21*
(4.65) (0.83) (9.86) (0.18) (2.56) (0.86) (1.13)
Observations 7,601 7,601 7,601 7,601 7,601 7,601 7,601
(f) Poverty
Treatment 0.14 0.25 -1.22 -0.11*** -0.24 -0.66 -0.84
(0.41) (0.26) (27.36) (0.05) (0.50) (0.39) (0.43)
Observations 8,580 8,580 8,580 8,580 8,580 8,580 8,580
(g) Multidimensional poverty
Treatment -0.74 0.26 -0.17 -0.06** -0.11 -0.30 -0.36
(0.91) (0.23) (0.22) (0.03) (0.29) (0.23) (0.26)
Observations 31,090 31,090 31,090 31,090 31,090 31,090 31,090
(h) Social inclusion
Treatment 0.26 -0.16 0.12 0.17* -0.81 0.25 0.46
(0.57) (0.39) (0.24) (0.08) (1.04) (0.49) (0.49)
Observations 8,538 8,538 8,538 8,538 8,538 8,538 8,538
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: IHDS 2011-12. District
population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
17

Table A17: Placebo cutoff test: Informal and business finance channel for SCs
(1) (2) (3) (4) (5) (6) (7)
c=-2250 c=-1500 c=-750 c=0 c=750 c=1500 c=2250
Informal finance (SC)
(a) Interest rate: SC
Treatment 599.23 -123.54 6.10 -8.11** 90.61 8.15 -14.87
(50,636.60) (182.32) (17.33) (3.33) (439.70) (23.93) (17.44)
Observations 7,401 7,401 7,401 7,401 7,401 7,401 7,401
(b) No mortgage: SC
Treatment 0.43 0.51 -0.82 -0.15* -0.05 -0.73 -0.11
(0.57) (0.64) (0.84) (0.10) (0.13) (0.40) (0.27)
Observations 12,127 12,127 12,127 12,127 12,127 12,127 12,127
(c) Informal loan: SC
Treatment -0.18 -0.03 0.27 0.09** -2.89 0.72 0.23
(0.29) (0.16) (0.27) (0.04) (10.84) (0.28) (0.15)
Observations 38,436 38,436 38,436 38,436 38,436 38,436 38,436
Business finance (SC)
(d) All Enterprises: SC
Treatment -293.22 153.36 -10,585.61 80.80* 883.37 163.02 177.66
(368.85) (229.41) (135,554.07) (50.86) (1,704.50) (207.78) (122.53)
Observations 581 581 581 581 581 581 581
(e) Agri Enterprises: SC
Treatment -58.48 12.23 5.82 7.65* -25.34 28.91 1.49
(64.25) (47.51) (16.05) (5.39) (25.55) (39.03) (6.10)
Observations 581 581 581 581 581 581 581
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: AIDIS 2013 and EC
2013. District population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
18

Table A18: Placebo cutoff test: Agricultural and non-agricultural business sector for SCs
(1) (2) (3) (4) (5) (6) (7)
VARIABLES c=-2250 c=-1500 c=-750 c=0 c=750 c=1500 c=2250
Agricultural sector (SC)
(a) Agri machinery value
Treatment 14,590.50 -3,262.34 854.53 1,946.33** 1,273.64 -2,822.09 4,709.80
(10,037.06) (5,890.82) (2,159.05) (838.01) (1,267.63) (5,363.33) (3,264.33)
Observations 9,632 9,632 9,632 9,632 9,632 9,632 9,632
(b) Agri (power) machinery value
Treatment -24,287.61 -147,386.87 -1,565.18 15,047.18*** -7,542.18 -38,122.90 12,303.80
(379,336.97) (727,351.34) (17,366.47) (5,332.61) (21,367.12) (33,533.90) (17,515.23)
Observations 1,154 1,154 1,154 1,154 1,154 1,154 1,154
(c) Number of livestocks
Treatment 1.03 2.98 1.51 0.61* -3.90 1.96 -0.07
(3.14) (1.79) (2.42) (0.35) (5.22) (1.11) (1.16)
Observations 4,912 4,912 4,912 4,912 4,912 4,912 4,912
(d) Agri labour hours
Treatment -18.16 -21.00 -1.21 -7.11** -18.50 -36.61* 12.36
(18.37) (20.79) (8.23) (3.24) (53.80) (17.47) (38.02)
Observations 5,196 5,196 5,196 5,196 5,196 5,196 5,196
(e) Agri income
Treatment -58,543.24 8,655.79 -67,158.16 3,625.58** -4,958.76 -20,014.77 18,065.87
(52,030.10) (10,642.79) (164,374.99) (1,969.99) (11,690.45) (11,764.45) (19,667.69)
Observations 8,583 8,583 8,583 8,583 8,583 8,583 8,583
Non-agricultural business sector (SC)
(f) Nonfarm wage
Treatment 2.58 28.88 -5.44 2.84* -1.58 6.35 32.27*
(15.50) (26.23) (4.19) (1.72) (9.83) (8.50) (15.11)
Observations 11,464 11,464 11,464 11,464 11,464 11,464 11,464
(g) Number of jobs
Treatment 0.32 0.58 1.87 0.17* -0.50 -0.42 -0.38
(1.14) (0.47) (3.61) (0.11) (1.00) (0.72) (1.10)
Observations 6,081 6,081 6,081 6,081 6,081 6,081 6,081
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: AIDIS 2013 and IHDS
2011-12. District population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
19

Table A19: Bandwidth selector test: Financial inclusion and household welfare outcomes for
SCs
MSE optimal CER optimal
(1) (2) (3) (4)
Common Two-sided Common Two-sided
Financial inclusion (SC)
(a) Bank account
Treatment 0.40*** 0.43* 0.41** 0.49
(0.16) (0.24) (0.19) (0.46)
Observations 8,451 8,451 8,451 8,451
(b) Bank loan
Treatment 0.14* 0.21** 0.18** 0.22
(0.07) (0.10) (0.09) (0.14)
Observations 4,813 4,813 4,813 4,813
(c) Insurance
Treatment 0.13** 0.21** 0.16** 0.24
(0.05) (0.10) (0.07) (0.19)
Observations 8,452 8,452 8,452 8,452
Welfare outcomes (SC)
(d) Consumption quintiles
Treatment 0.49** 0.47 0.55* 0.96
(0.30) (0.36) (0.36) (0.74)
Observations 8,580 8,580 8,580 8,580
(e) Food consumption quintiles
Treatment 0.42** 0.52** 0.47** 0.75**
(0.18) (0.20) (0.20) (0.30)
Observations 7,601 7,601 7,601 7,601
(f) Poverty
Treatment -0.11*** -0.12** -0.13** -0.18**
(0.05) (0.06) (0.06) (0.08)
Observations 8,580 8,580 8,580 8,580
(g) Multidimensional poverty
Treatment -0.06** -0.08 -0.07* -0.09
(0.03) (0.04) (0.04) (0.06)
Observations 31,090 31,090 31,090 31,090
(h) Social inclusion
Treatment 0.17* 0.21 0.18* 0.32
(0.08) (0.12) (0.10) (0.22)
Observations 8,538 8,538 8,538 8,538
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch
Authorization Policy in 2005. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1).
Standard errors clustered at district level. Data used: IHDS 2011-12. District population and number of
bank branches in 1996 are controlled for. Source: Authors’ calculation.
20

Table A20: Bandwidth selector test: Informal and business finance channel for SCs
MSE optimal CER optimal
(1) (2) (3) (4)
Common Two-sided Common Two-sided
Informal finance (SC)
(a) Interest rate: SC
Treatment -8.11** -7.28 -6.74* -9.69*
(3.33) (4.71) (4.16) (4.69)
Observations 7,401 7,401 7,401 7,401
(b) No mortgage: SC
Treatment -0.15* -0.12* -0.19** -0.20**
(0.10) (0.07) (0.10) (0.10)
Observations 12,127 12,127 12,127 12,127
(c) Informal loan: SC
Treatment 0.09** 0.05** 0.10** 0.09**
(0.04) (0.04) (0.04) (0.04)
Observations 38,436 38,436 38,436 38,436
Business finance (SC)
(d) All Enterprises: SC
Treatment 80.80* 29.09 92.65 85.23*
(50.86) (47.83) (61.27) (54.19)
Observations 581 581 581 581
(e) Agri Enterprises: SC
Treatment 7.65* 5.91 11.77* 6.27*
(5.39) (3.65) (7.20) (3.69)
Observations 581 581 581 581
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: AIDIS 2013 and EC
2013. District population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
21

Table A21: Bandwidth selector test: Agricultural and non-agricultural business sector for SCs
MSE optimal CER optimal
(1) (2) (3) (4)
Common Two-sided Common Two-sided
Agricultural Sector (SC)
(a) Agri machinery value
Treatment 1,946.33** 1,819.46*** 2,047.61** 1,793.12***
(838.01) (558.82) (1,039.91) (592.51)
Observations 9,632 9,632 9,632 9,632
(b) Agri (power) machinery value
Treatment 15,047.18*** 10,020.05** 18,141.54*** 9,935.63***
(5,332.61) (3,543.99) (6,739.51) (3,427.54)
Observations 1,154 1,154 1,154 1,154
(c) Number of livestocks
Treatment 0.61* 0.61** 0.44 0.62*
(0.35) (0.30) (0.38) (0.37)
Observations 4,912 4,912 4,912 4,912
(d) Agri labour hours
Treatment -7.11** -6.60 -7.95** -7.84*
(3.24) (3.62) (3.81) (3.91)
Observations 5,196 5,196 5,196 5,196
(e) Agri income
Treatment 3,625.58** 8,663.61 5,099.88** 41,356.79
(1,969.99) (9,732.22) (2,288.57) (347,737.36)
Observations 8,583 8,583 8,583 8,583
Non agricultural business sector (SC)
(f) Nonfarm wage
Treatment 2.84* 2.94 3.76* 2.62
(1.72) (1.94) (2.12) (3.00)
Observations 11,464 11,464 11,464 11,464
(g) Number of jobs
Treatment 0.17* 0.50 0.23* 0.72
(0.11) (0.38) (0.12) (0.86)
Observations 6,081 6,081 6,081 6,081
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: AIDIS 2013 and IHDS
2011-12. District population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
22

Table A22: Bandwidth multiplier test: Financial inclusion and household welfare of SCs
(1) (2) (3) (4) (5)
0.50x 0.75x 1.00x 1.25x 1.50x
Financial inclusion (SC)
(a) Bank account
Treatment 0.37 0.41 0.40*** 0.33** 0.26***
(0.35) (0.19) (0.16) (0.13) (0.11)
Observations 8,451 8,451 8,451 8,451 8,451
(b) Bank loan
Treatment 0.19 0.18* 0.14* 0.13** 0.12**
(0.11) (0.09) (0.07) (0.06) (0.06)
Observations 4,813 4,813 4,813 4,813 4,813
(c) Insurance
Treatment 0.21* 0.16*** 0.13** 0.13*** 0.13**
(0.10) (0.07) (0.05) (0.05) (0.04)
Observations 8,452 8,452 8,452 8,452 8,452
Welfare outcomes (SC)
(d) Consumption quintiles
Treatment 1.11*** 0.55** 0.49** 0.47 0.37*
(0.87) (0.36) (0.30) (0.26) (0.23)
Observations 8,580 8,580 8,580 8,580 8,580
(e) Food consumption quintiles
Treatment 0.65** 0.47*** 0.42** 0.38** 0.30***
(0.28) (0.20) (0.18) (0.16) (0.15)
Observations 7,601 7,601 7,601 7,601 7,601
(f) Poverty
Treatment -0.15*** -0.13** -0.11*** -0.07*** -0.03**
(0.07) (0.06) (0.05) (0.04) (0.04)
Observations 8,580 8,580 8,580 8,580 8,580
(g) Multidimensional poverty
Treatment -0.05 -0.07 -0.06** -0.03** -0.02*
(0.05) (0.04) (0.03) (0.03) (0.03)
Observations 31,090 31,090 31,090 31,090 31,090
(h) Social inclusion
Treatment 0.13 0.18 0.17* 0.14* 0.12*
(0.12) (0.10) (0.08) (0.07) (0.06)
Observations 8,538 8,538 8,538 8,538 8,538
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in
2005. Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used:
IHDS 2011-12. District population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
23

Table A23: Bandwidth multiplier test: Informal and business finance channel for SCs
(1) (2) (3) (4) (5)
0.50x 0.75x 1.00x 1.25x 1.50x
Informal finance (SC)
(a) Interest rate: SC
Treatment -6.47** -6.87 -8.11** -8.44 -7.72**
(5.38) (4.11) (3.33) (2.94) (2.70)
Observations 7,401 7,401 7,401 7,401 7,401
(b) No mortgage: SC
Treatment -0.28 -0.19 -0.15* -0.13** -0.10**
(0.16) (0.10) (0.10) (0.10) (0.09)
Observations 12,127 12,127 12,127 12,127 12,127
(c) Informal loan: SC
Treatment 0.11 0.10 0.09** 0.07** 0.06**
(0.05) (0.04) (0.04) (0.04) (0.03)
Observations 38,436 38,436 38,436 38,436 38,436
Business finance (SC)
(d) All Enterprises: SC
Treatment 147.35 91.44** 80.80* 54.99* 40.72*
(76.73) (60.62) (50.86) (43.61) (38.91)
Observations 581 581 581 581 581
(d) Agri Enterprises: SC
Treatment 16.15 11.40 7.65* 8.79 7.00
(11.39) (7.01) (5.39) (6.03) (6.08)
Observations 581 581 581 581 581
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: AIDIS 2013 and EC
2013. District population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
24

Table A24: Bandwidth multiplier test: Agricultural and non-agricultural business sector for SCs
(1) (2) (3) (4) (5)
0.50x 0.75x 1.00x 1.25x 1.50x
Agricultural Sector (SC)
(a) Agri machinery value
Treatment 3,053.95* 2,053.58* 1,946.33** 1,942.25* 1,573.30**
(1,745.13) (1,029.76) (838.01) (700.04) (603.11)
Observations 9,632 9,632 9,632 9,632 9,632
(b) Agri (power) machinery value
Treatment 18,228.79 18,202.46** 15,047.18*** 11,610.53*** 9,521.20***
(6,742.08) (6,816.27) (5,332.61) (4,393.23) (3,787.66)
Observations 1,154 1,154 1,154 1,154 1,154
(c) Number of livestocks
Treatment 0.26 0.45 0.61* 0.45** 0.24**
(0.42) (0.38) (0.35) (0.31) (0.27)
Observations 4,912 4,912 4,912 4,912 4,912
(d) Agri labour hours
Treatment -6.33 -7.96 -7.11** -6.26* -6.05*
(4.46) (3.83) (3.24) (2.80) (2.56)
Observations 5,196 5,196 5,196 5,196 5,196
(e) Agri income
Treatment 7,301.54* 5,087.81** 3,625.58** 2,979.76*** 2,500.19**
(3,844.46) (2,290.45) (1,969.99) (1,781.08) (1,642.89)
Observations 8,583 8,583 8,583 8,583 8,583
Non agricultural business sector (SC)
(f) Nonfarm wage
Treatment 3.62 3.75 2.84* 2.11* 1.61*
(2.57) (2.12) (1.72) (1.45) (1.28)
Observations 11,464 11,464 11,464 11,464 11,464
(g) No of jobs
Treatment 0.25* 0.23** 0.17* 0.14** 0.09*
(0.16) (0.12) (0.11) (0.09) (0.09)
Observations 6,081 6,081 6,081 6,081 6,081
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005. Robust
standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: AIDIS 2013 and IHDS
2011-12. District population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
25

Table A25: Donut Hole Test
Financial Inclusion Household Well-being
SC OBC Gen SC OBC Gen
(a) Bank account (f) Consumption
Treatment 0.360***
(0.110)
0.177**
(0.096)
0.293***
(0.086) Treatment 0.345*
(0.244)
0.088
(0.234)
0.056
(0.220)
Observations 8,356 13,144 8,364 Observations 8,485 13,462 8,561
(b) Bank loan (g) Food consumption
Treatment 0.150*
(0.073)
0.093
(0.076)
0.126*
(0.066) Treatment 0.457***
(0.172)
0.434**
(0.208)
0.293
(0.232)
Observations 4,752 8,002 3,952 Observations 7,513 11,956 7,691
(c) Fixed Deposit (h) Poverty
Treatment 0.036
(0.027)
0.136***
(0.042)
0.168***
(0.058) Treatment -0.102**
(0.050)
-0.043
(0.046)
-0.036
(0.028)
Observations 8,358 13,145 8,359 Observations 8,485 13,462 8,561
(d) Securities (i) MPI
Treatment -0.002
(0.002)
0.011
(0.007)
0.029***
(0.011) Treatment -0.052*
(0.032)
-0.078
(0.063)
-0.169**
(0.081)
Observations 8,358 13,146 8,362 Observations 30,713 49,648 31,074
(e) Insurance (j) Social inclusion
Treatment 0.143**
(0.054)
0.082
(0.065)
0.014
(0.047) Treatment 0.162*
(0.094)
-0.016
(0.103)
0.124
(0.085)
Observations 8,357 13,147 8,376 Observations 8,444 13,382 8,530
Note: Treatment is district-level expansion of bank branches following the Reserve Bank of India, Branch Authorization Policy in 2005.
Robust standard errors in parentheses (*** p<0.01, ** p<0.05, * p<0.1). Standard errors clustered at district level. Data used: IHDS 2012.
District population and number of bank branches in 1996 are controlled for. Source: Authors’ calculation.
26