Documentos de Académico
Documentos de Profesional
Documentos de Cultura
AND
ROHINI PANDE*
1
Bank expansion and economic growth are positively
correlated in cross-country data (e.g., Robert King and Ross
Levine, 1993). The fact that countries with greater growth
potential may attract more banks, however, makes causal
inference difficult.
2
Throughout the paper, locations refer to villages,
towns, and cities as defined by the Indian Census. The
Census defines a location with fewer than 10,000 persons
as rural. The same holds for rural and urban poverty
definitions.
780
VOL. 95 NO. 3
781
I. The Program
Nationalization in 1969 brought the 14 largest commercial banks under the direct control of
the Indian Central Bank. Following this, the
Central Bank launched an ambitious branch expansion program, which sought both to expand
the rural bank branch network and equalize
individual access to banks across Indian states.
This program encouraged branch openings in
rural unbanked locations. Banks were required
to select unbanked locations for branch expansion from a list circulated by the Central Bank.
This list identified all unbanked locations with a
population above a certain number. As the same
population cut-off was applied across India, the
list featured relatively more locations from
states with a lower initial stock of bank
branches per capita. Further, within a state,
more locations were targeted in districts with
fewer bank branches per capita pre-program.3
The list was updated, with a lower population
cutoff, every three years.
The 1949 Banking Regulation Act requires
banks to obtain a license from the Indian Central Bank before opening a new branch. To
ensure that targeted rural unbanked locations
received bank branches, the Central Bank introduced a new branch licensing policy in 1977. It
mandated that a bank can obtain a license to
open a branch in an already banked location
only if it opened branches in four unbanked
locations. This 1:4 licensing policy was aimed
at forcing banks wishing to expand in already
banked locations to open branches in unbanked
locations. The 1:4 licensing policy was discontinued in 1990. Since then, Central Bank policy
has stated that branch expansion should reflect
the need, business potential, and financial viability of the location (Government of India,
1991). Banks cannot, however, close a rural
branch if it is the only one serving a given
location.
To ensure that rural branch expansion
translated into increased credit and savings
opportunities for the rural population, the
Central Bank regulated banks deposit-taking
3
In each Indian district, one commercial bank was selected by the Central Bank to be the lead bank, which was
responsible for coordinating branch expansion activities in
that district.
782
JUNE 2005
7
We are grateful to Gaurav Datt and Martin Ravallion
for providing the state-level poverty figures for 19611994
(see Berk Ozler et al., 1996), and to Gaurav Datt for the
1994 2000 data. The year 1961 is the first, and earliest,
census year preceding bank nationalization for which annual poverty series are available. Figures A2 and A3 in the
Appendix show the state-wise evolution of rural credit and
savings shares, and of rural and urban poverty outcomes,
respectively.
VOL. 95 NO. 3
y it i t B Rit it
783
Bi1961, our measure of initial financial development, denotes the number of bank branches per
capita in state i in 1961. This variable enters the
regression interacted with year dummies, with
t denoting the year-specific coefficients. The
difference between t 1 and t tells us how a
states initial financial development affected rural branch growth between years t and t 1.
Xi1961 denotes a vector of initial state conditions, which includes log real state income per
capita, population density, and the number of
rural locations per capita, all measured in 1961.
These enter the regression with year-specific
coefficients t.
The circles on the solid line in Figure 1 graph
the t coefficients from this regression (the reference year is 1961). Consistent with the idea
that financially more developed states offered
banks greater profit opportunities, we observe
more branch openings in rural unbanked locations in these states between 1961 and 1977.
This is reflected in a positive trend in t coefficients. This trend is reversed in 1977, precisely
when the 1:4 license policy was imposed. Between 1977 and 1990, the t coefficients decrease with timefinancially less developed
states witness higher growth of branch openings
in rural unbanked locations. After 1990, branch
expansion into rural unbanked locations ends.
The shape of this graph is unaltered by the
exclusion of the X61 controls (see Burgess and
Pande, 2003). We also observe identical trend
reversals at the district level in 1977 and 1990.
This indicates that the 1:4 licensing policy
caused banks to target financially less developed districts within a state.8
We summarize these trend reversals by a
linear trend break model:
(3)
B Rit i t t B i1961
t Xi1961 it .
8
For the district-level analysis, see Figure A4 in the
Appendix.
784
AND
JUNE 2005
Notes: The series rural branches in unbanked locations (with controls) graphs the annual coefficients on initial financial
development (as measured by the number of bank branches per capita in 1961) from a regression of the form described in equation
(2). The series rural branches in unbanked locations (trend break) graphs the annual coefficients implied by the trend break model,
column (1), Table (1). In both cases, the dependent variable is the cumulative number of rural branches opened in unbanked locations.
State and year fixed effects account for permanent differences across states and national
events which may affect branch expansion. [t
1961], [t 1977], and [t 1990] are linear
time trends, which switch on in 1961, 1977, and
1990, respectively. These enter the regression
interacted with our measure of a states initial
financial development, Bi1961. P1977 and P1990
are dummy variables which equal one from
1977 and 1990, respectively.
The main coefficients of interest 1, 2, and
3 measure the average 19611977 trend relationship between a states initial financial development and rural branch expansion, and the
subsequent changes in this trend relationship
(between 1977 and 1990, and between 1990 and
2000). Finally, 4 and 5 measure the intercept
changes in this relationship in 1977 and 1990,
respectively. The set of additional controls,
Xi1961, enters the regression in the same way as
Bi1961. The inclusion of these controls ensures
that any observed trend reversals in Bi1961 do
not proxy for trend breaks in a states economic
and demographic characteristics (as measured
by Xi1961). To account for possible serial correlation in errors, we cluster standard errors by
state (see Marianne Bertrand et al., 2004).
Column 1 of Table 1 reports the results. Between 1961 and 1977, one additional point of
initial financial development increased branch
openings in rural unbanked locations per capita
in a state by 0.07 annually. There was a significant trend reversal in 1977. Between 1977 and
1990, one additional point of initial financial
development reduced annual branch expansion
by 0.18 branches per capita. Finally, after 1990,
a states level of initial financial development
and rural branch expansion were unrelated.
The squares on the dotted line in Figure 1 show
the t coefficients implied by these estimates. The
pattern of coefficients for the unrestricted model
and linear trend break model are extremely similar
and an F test shows that the imposed restrictions
do not cause any significant loss in overall fit.9
9
The value of the F-statistic is 0.04; see William Greene
(1993, p. 208) for the test.
VOL. 95 NO. 3
AS A
FUNCTION
OF INITIAL
Branches
in rural
unbanked
locations
Credit
share
(1)
Number of bank branches per capita
in 1961*(19612000) trend
785
FINANCIAL DEVELOPMENT
Rural bank
Credit share
Savings
share
Branches
in banked
locations
Priority
sector
Cooperative
(2)
(3)
(4)
(5)
(6)
0.07**
(0.03)
0.18
(0.21)
0.03
(0.24)
0.14***
(0.01)
0.08
(0.62)
0.41
(0.34)
0.25***
(0.03)
1.09**
(0.43)
0.82***
(0.25)
0.07***
(0.02)
0.08
(0.86)
0.02
(0.42)
0.17***
(0.04)
0.87***
(0.26)
0.43*
(0.23)
0.10**
(0.04)
0.18
(0.33)
0.03
(1.00)
0.34
(0.25)
0.30
(1.50)
0.17
(0.78)
0.53**
(0.19)
3.37
(2.40)
3.64
(2.22)
0.24
(0.15)
1.95
(1.49)
0.44
(0.53)
0.40***
(0.10)
0.05
(1.86)
3.15
(2.61)
YES
YES
YES
YES
YES
YES
Other controls
YES
YES
YES
YES
YES
YES
Adjusted R-squared
0.96
0.88
0.87
0.98
0.86
0.81
F-test 1
16.87
[0]
12.8
[0]
25.67
[0]
8.97
[0]
0
[0.99]
5.75
[0.03]
F-test 2
0.49
[0.49]
0.1
[0.76]
9
[0]
27.22
[0]
1.79
[0.20]
0.17
[0.69]
636
512
512
636
512
491
Observations
Notes: Standard errors clustered by state are reported in parentheses; p-values are in square brackets. F-test 1 and F-test 2 are
the joint significance test for coefficients in the first two rows and first three rows, respectively. Rural bank credit (saving)
share is the percentage of total bank credit (saving) accounted for by rural branches. Priority credit share is share of bank
lending going to priority sectors. Cooperative credit share is primary agricultural cooperative credit as a percentage of total
cooperative and bank lending. Explanatory variables reported are bank branches in 1961 per 100,000 persons interacted with
(row-wise) (a) a time trend, (b) a post-1976 time trend, (c) a post-1989 time trend. Other controls include state population
density, log state income per capita, and log rural locations per capita, all measured in 1961. They enter the regression in the
same way as branches per capita in 1961. The Data Appendix describes the data sources and the time period for which each
data series is available. * Significant at 10-percent level. ** Significant at 5-percent level. *** Significant at 1-percent level.
786
AND
JUNE 2005
Notes: The series rural credit share graphs the annual coefficients on initial financial development (as measured by the
number of bank branches per capita in 1961) from a regression of the form described in equation (2). The dependent variable
is the share of total bank credit disbursed by rural bank branches.
policy driven. In the absence of policy constraints we would expect banks to choose the
locations that offered them the highest expected profits. Between 1961 and 2000, banks
were free to choose where to locate branch
openings in already banked locations. In column 4 of Table 1 we observe that, throughout
our sample period, more of such branch openings occurred in more financially developed
states. This indicates that these states were
more attractive to banks and that regulation
was needed to coerce banks to locate elsewhere. We also observe that the rate of branch
expansion into already banked locations was
lower between 1977 and 1990. This makes
sense, because during this period, branch
openings in bank locations were less profitable, as each such branch opening had to be
accompanied by four branch openings in unbanked locations.
We also check whether bank and state-level
policies, which should be unaffected by the
1:4 licensing policy, exhibit trend reversals in
1977 and 1990. In column 5, we look at the
fraction of bank credit going to priority sectors (small-scale industries, services, and agriculture). Priority sector targets were binding
VOL. 95 NO. 3
AND
787
POVERTY
Notes: The series rural headcount ratio and urban headcount ratio graph the annual coefficients on initial financial
development (as measured by the number of bank branches per capita in 1961) from regressions of the form described in
equation (2). The dependent variables are the rural and urban headcount ratios, respectively.
(4)
ever, the relationship differs by poverty measure. Urban poverty and a states initial financial development are largely uncorrelated. In
contrast, between 1983 and 1990, rural poverty reductions are more pronounced in states
with lower initial financial development. The
graph for rural poverty is thus the inverse of
that for rural branch expansion. To see this
more clearly, we estimate a regression of the
form:
y it i t t B i1961 t X i1961 it
and report the findings in Figure 3. The diamonds on the solid line depict the t coefficients when y it is the rural headcount ratio,
while the squares on the dotted line depict the
t coefficients when y it is the urban headcount
ratio.10 Between 1970 and 1978, and after
1990, both rural and urban poverty declines
were pronounced in more financially developed states. Between 1978 and 1990, how-
10
The rural and urban head-count ratios are defined
as the percentage of rural and urban households with
per capita monthly expenditures below the rural (49 rupees
at 1973June 1974 all-India rural prices) and urban (57 rupees
at 1973June 1974 all-India urban prices) poverty lines.
788
AND
JUNE 2005
Wage
Annual coefficients
rural headcount ratio
Rural
Urban
Aggregate
Agricultural
Factory
(1)
(2)
(3)
(4)
(5)
(6)
0.77***
(0.23)
0.27
(0.24)
0.71***
(0.22)
0.004
(0.006)
0.01
(0.02)
1.15**
(0.42)
0.15
(0.26)
0.99***
(0.33)
0.01
(0.01)
0.01
(0.02)
1.15***
(0.34)
0.31
(0.38)
1.04***
(0.31)
0.05**
(0.02)
0.02
(0.01)
3.77*
(1.94)
2.76
(2.29)
3.53*
(1.71)
0.09*
(0.05)
0.04
(0.05)
1.2
(2.39)
0.5
(0.96)
0.62
(1.82)
0.03
(0.05)
0.01
(0.02)
YES
YES
YES
YES
YES
Other controls
YES
YES
YES
YES
YES
Adjusted R-squared
0.84
0.91
0.88
0.90
0.70
F-test 1
1.5
[0.24]
0.37
[0.55]
1.76
[0.20]
23.95
[0]
0.23
[0.64]
F-test 2
2.97
[0.11]
3.95
[0.07]
4.15
[0.06]
1.88
[0.19]
6.07
[0.03]
627
627
627
545
553
Observations
4.71***
(1.01)
39
Notes: Standard errors clustered by state are in parentheses; p-values are in square brackets. In column (1), the dependent and
explanatory variables are the annual coefficients on the initial financial development variable from running a regression of the
form in equation (4) for the rural headcount ratio, and equation (2) for branches opened in unbanked locations. The column
(1) regression includes the post-1976 and post-1990 dummies as controls. Headcount ratio is the percentage population with
expenditure below the poverty line. Agricultural wage is log real male daily agricultural wage, and factory wage is log real
remunerations per worker in registered manufacturing. The definitions of explanatory variables, other controls, and F-tests for
columns (2) to (6) are in the notes to Table 1. The Data Appendix describes the data sources and the time period for which
each data series is available. * Significant at 10-percent level. ** Significant at 5-percent level. *** Significant at 1-percent
level.
1.08
(2.33)
5.38**
(2.47)
YES
YES
NO
Post-1989 dummy*(19902000)
trend
Other controls
627
0.92
[0.99]
YES
YES
627
0.82
[0.99]
YES
YES
1.55
(1.76)
1.39
(2.03)
0.46*
(0.23)
4.10**
(1.46)
(5)
IV
460
0.80
YES
YES
2.13
(2.59)
0.43
(0.26)
4.70**
(1.82)
(6)
IV
1961-1989
375
0.81
YES
YES
0.45
(2.90)
0.80*
(0.45)
6.84**
(2.81)
(7)
IV
1977-2000
Rural
375
0.73
[1]
YES
YES
0.79
(2.61)
1.31
(3.32)
0.46
(0.28)
4.21*
(2.26)
(8)
IV
Survey
years
545
0.87
[0.98]
YES
YES
0.11
(0.07)
0.04
(0.06)
0.007
(0.004)
0.08*
(0.04)
(9)
IV
Agricultural
Wage
553
0.70
[0.99]
YES
YES
0.05
(0.05)
0.03
(0.06)
0.01
(0.01)
0.05
(0.08)
(10)
IV
Factory
Notes: Standard errors clustered by state are in parentheses; p-values are in square brackets. The definitions of the dependent and explanatory variables are in the notes to Table
2 and Table 1, respectively. In the IV regressions, the instruments are the number of bank branches per capita in 1961 interacted with (a) a post-1976 time trend and (b) a
post-1989 time trend, respectively. Table 1, column (1), reports the corresponding first-stage regression. In the second row of columns 6 and 7, the number of bank branches
per capita is interacted, respectively, with a (19611989) and a (19772000) trend. The overidentification test we employ is due to John Denis Sargan (1958). The number of
observations times the R-squared from the regression of the stage-two residuals on the instruments is distributed chi-squared (T 1) where T is the number of instruments. The
Data Appendix describes the data sources and the time period for which each data series is available. * Significant at 10-percent level. ** Significant at 5-percent level.
*** Significant at 1-percent level.
627
627
Observations
627
0.76
0.83
0.81
Adjusted R-squared
YES
[0.99]
Overidentification test
YES
0.47
(1.01)
1.42
(2.30)
0.31
(1.23)
Post-1976 dummy*(19772000)
trend
YES
2.06
(1.65)
0.48*
(0.27)
0.43***
(0.17)
0.26*
(0.13)
0.66
(1.07)
2.09**
(0.79)
4.74**
(1.79)
1.16
(1.02)
IV
(4)
(2)
IV
Aggregate
Headcount ratio
Urban
AND
(3)
(1)
OLS
Rural
VOL. 95 NO. 3
BURGESS AND PANDE: DO RURAL BANKS MATTER?
789
790
openings in rural unbanked locations is positive and significant. This is consistent with a
program-based explanation, wherein poorer,
financially less developed states attracted
more rural branches between 1977 and 1990.
The result highlights the pitfalls of using OLS
estimation to assess the impact of rural branch
expansion on poverty. Inclusion of the interaction between a states initial financial development and a time trend, and the vector of
state initial conditions as additional covariates, renders this relationship statistically insignificant (column 2).
Our IV regressions exploit the documented
trend reversals between 1977 and 1990 and
between 1990 and 2000 (relative to the 1961
1977 trend) in the relationship between a
states initial financial development and rural
branch expansion as instruments for branch
openings in rural unbanked locations. The
first stage regression is as in column 1 of
Table 1. The second stage regression takes the
form
y it i t B Rit
1 t 1961 B i1961
2 P 1977 B i1961
3 P 1990 B i1961 u it .
Deviations from the linear state-specific trend,
[t 1961] Bi1961, which we characterize as
[t 1977] Bi1961 and [t 1990] Bi1961, are
our instruments for BRit .
Columns 3 to 5 of Table 3 report IV estimates
for poverty outcomes. A one-point increase in
per capita branch openings in rural unbanked
locations is associated with a 4.74-percent reduction in rural poverty (column 3). Evaluated
at the sample average, our results imply that
rural branch expansion in India can explain a
17-percent reduction in the headcount ratio. In
contrast, rural branch expansion did not affect
urban poverty (column 4).11 Opening a bank
branch in an additional rural location per
11
JUNE 2005
VOL. 95 NO. 3
AND
SAVINGS
AND
791
Rural
(1)
Rural bank credit share
Urban
(2)
1.52**
(0.69)
(3)
0.67
(0.47)
2.22**
(0.78)
Aggregate
(4)
(5)
(6)
1.37**
(0.59)
1.05
(0.67)
2.01***
(0.65)
1.01*
(0.50)
1.51**
(0.54)
0.70**
(0.25)
0.96**
(0.34)
0.96**
(0.41)
1.42***
(0.44)
Post-1976 dummy*(19772000)
trend
2.89
(1.68)
2.05
(2.34)
1.59
(1.98)
1.23
(2.55)
2.6
(1.68)
1.84
(2.52)
Post-1989 dummy*(19902000)
trend
4.4
(2.64)
2.13
(2.65)
2.87
(2.35)
1.88
(1.31)
3.53
(2.35)
1.47
(1.98)
YES
YES
YES
YES
YES
YES
Other controls
YES
YES
YES
YES
YES
YES
Overidentification test
[0.99]
[0.99]
[0.99]
[0.99]
[0.99]
[0.99]
Adjusted R-squared
0.69
0.60
0.90
0.88
0.75
0.67
Observations
503
503
503
503
503
503
Notes: Standard errors clustered by state are in parentheses; p-values are in square brackets. The definitions of the dependent
and explanatory variables are in the notes to Table 2 and Table 1, respectively. The notes to Table 3 describe the instruments
and the overidentification test. Table 1, columns (2) and (3), report the first-stage regressions for rural banks credit and savings
share, respectively. The Data Appendix describes the data sources and the time period for which each data series is available.
* Significant at 10-percent level. ** Significant at 5-percent level. *** Significant at 1-percent level.
792
AND
Urban headcount
ratio
(1)
(2)
(3)
(4)
4.12**
(1.54)
1.75**
(0.70)
10.97
(30.91)
40.84***
(12.39)
3.77**
(1.54)
1.87**
(0.68)
3.31
(28.40)
37.32**
(13.37)
1.05
(1.06)
0.41
(0.29)
23.52
(14.53)
6.31
(12.08)
0.81
(0.91)
0.27
(0.30)
23.74
(14.80)
5.73
(11.89)
YES
YES
[0.99]
0.91
605
0.22
(3.14)
1.62
(3.18)
9.61
(8.36)
1.76
(3.72)
2.34
(4.60)
YES
YES
[0.99]
0.92
603
JUNE 2005
YES
YES
[0.99]
0.80
605
13.07
(8.90)
11.62
(6.90)
6.15
(12.91)
14.81
(9.07)
15.11
(12.91)
YES
YES
[0.99]
0.82
603
Notes: Standard errors clustered by state are in parentheses; p-values are in square brackets.
The definitions of the dependent and bank variables are in the notes to Table 2 and Table
1, respectively. Cumulative land reform is the total number of land reform acts passed by an
Indian state. Health and education spending is the fraction of total state spending on health and
education. Other development spending is the fraction of total state spending on agriculture,
rural development, irrigation, public works, and community development programs. Fraction
Congress, Janata, Hindu, Hard Left, and Regional refer to number of seats held in state
legislatures by parties in these political groupings. The notes to Table 3 describe the
instruments and the overidentification test. The Data Appendix describes the data sources and
the time period for which each data series is available. * Significant at 10-percent level.
** Significant at 5-percent level. *** Significant at 1-percent level.
VOL. 95 NO. 3
12
793
13
Data on resource flows from the Central Bank to
commercial banks are unavailable. The size of the Indian
banking sector, however, is testimony to the state subsidy
being substantial.
14
Sample states are Andhra Pradesh, Assam, Bihar, Gujarat, Haryana (enters in 1965), Jammu and Kashmir, Karnataka, Kerala, Madhya Pradesh, Maharashtra, Orissa,
Punjab, Rajasthan, Tamil Nadu, Uttar Pradesh, and West
Bengal. The total possible number of observations is thus 636.
794
JUNE 2005
defined as towns with fewer than 10,000 persons and villages with between 2,000 and
10,000 persons.
REFERENCES
Adams, Dale W.; Graham, Douglas H. and Von
Pischke, J. D. Undermining rural development
VOL. 95 NO. 3
795