Evidence from a Regression Discontinuity Analysis in India 2

8 Figures Figure 1: Banking Sector Structure in India Reserve Bank of India (RBI) Public Sector Banks State Bank of India (SBI) Foreign Banks Nationalised Banks Private Sector Banks Old Private Banks New Private Banks Regional Rural Banks (RRBs) Figure 2: Policy Time Line for Bank Branching and Related Reforms Vyas committee commissioned Branching Policy Reform Expectation of Competition Interim report released (Refined) 2003 2004 2005 Branching Policy Reform Priority sector reform ABEP Grace Period Expires 2006 2007 34 Metro markets made ineligible 2008 2009 2010 Figure 3: Bank-year Specific Shares of New Licenses Issued to the Bank in Districts on the 2006 List of Under Banked Districts Share of New Licenses in Under Banked Districts .2 .4 .6 0 0 Share of New Licenses in Under Banked Districts .2 .4 .6 .8 Nationalised Banks 2006−2011 .8 Nationalised Banks 1994−2004 2.5 3 3.5 Log New Branch Licenses 4 4.5 2 3 4 5 Log New Branch Licenses Share of New Licenses in Under Banked Districts .2 .4 .6 0 0 Share of New Licenses in Under Banked Districts .2 .4 .6 .8 Private Banks 2006−2011 .8 Private Banks 1994−2004 6 2.5 3 3.5 Log New Branch Licenses 4 4.5 2 3 4 5 Log New Branch Licenses 6 Note: Each observation is the percentage of new licenses for general and specialized branches issued during a calendar year for each bank-year pair in the respective Nationalised and Private bank groups. The value is plotted against the log of total new licenses issued to the bank in that year. Figure 4: RD Visual First Stage: Under Banked Status by District Population Per Branch .8 .6 .4 .2 0 District Under Banked Status = 1 1 District Under Banked Status from RBI List By Population Per Branch −12 −10 −8 −6 −4 −2 0 2 4 6 8 10 12 14 16 18 20 District Population Per Branch (1,000s) Centered on Cutoff Note: The dots report the under banked status of a district, taking a value equal to one if the district appeared on the list of under banked districts in the 2006 RBI MC on Branching Authorisation Policy, and zero otherwise. The forcing variable, district population per branch centered on the national average, is on the x-axis scaled to thousands of persons per branch. Values to the right of the cutoff are predicted to have under banked status. 368 districts of 572 have under banked status, with 6 incorrect predictions based on the rule. 35 Figure 5: Maps of Under Served Areas by Formal Banking Population per Branch High Low Districts in Local Linear Regressions Districts at Threshold Under Banked Banked Excluded Under Banked Banked Excluded Upper: District population per branch. Lower Left: Including only districts used in local linear regressions. Lower Right: Including only those districts within 1,000 persons per branch of the cutoff. 0 .02 .04 .06 .08 Figure 6: Visual McCrary Test -20 -10 0 10 20 District Pop Per Branch - Cutoff (1000s) 30 Note: Districts censored at 22 above the cutoff due to sparsity of districts. The graph plots a density of districts along the forcing variable, district population per branch, centered on the cutoff. The discontinuity estimate in the log difference in height is 6.6 with a standard error of 22. I fail to reject the null hypothesis of continuity at the cutoff, suggesting a lack of manipulation. 36 Figure 7: Continuity Around the Threshold Baseline District Characteristics Scheduled Caste and Tribe Population 0 0 −100 −12 −10 −8 −6 −4 −2 0 2 4 6 8 10 12 14 16 18 20 −12 −10 −8 −4 −2 0 2 4 6 8 10 12 14 16 18 20 16 18 20 Percent of Workers in Agriculture .6 .8 .8 Percent of District Literate −6 .2 0 .2 .4 .4 .6 District Characteristics 50 100 100 200 150 300 Population from Cities Over 100k −12 −10 −8 −6 −4 −2 0 2 4 6 8 10 12 14 16 18 20 −12 −10 −8 −6 −4 −2 0 2 4 6 8 10 12 14 District Population Per Branch (1,000s) Centered on Cutoff Note: The figure presents baseline district characteristics taken from the 2001 Population Census of India, with dots reporting local averages for districts falling within non-overlapping 200 persons per branch bins. The horizontal axis is the forcing variable of district population per branch centered on the cutoff. Districts predicted to have under banked status fall to the right of the cutoff. The estimated y-value from a local linear regression of bandwidth 3.5 thousand persons per branch is shown at each x-value, allowing for different slopes on either side of the cutoff, with 5% confidence intervals. 37 Figure 8: Histogram of Branch Licenses Showing ABEPs for a Large Private Sector Bank 0 50 Licenses 100 150 200 250 Branch License Dates Private Sector Bank Example 03 20 ov n 18 04 20 jul 08 5 06 00 20 p2 jul se 01 08 07 20 jul 02 08 20 jul 01 09 09 10 20 20 20 jul ec jul 01 01d 01 20 jul 01 11 Branch License Date Note: Branch license dates are from the MOF. Bin widths are set to 4 days. Though annual branch expansion plans (ABEP) may not be observed directly, the large spikes in branch licenses set approximately a year apart after 2005 are consistent with licenses issued through ABEPs. The dates of Master Circular releases are shown, with vertical red lines at the 2005 policy reform and the subsequent reform in December 2009. Branches acquired through mergers and acquisitions are excluded. 38 Figure 9: Visual RD: Operating Private Bank Branches Operating Private Sector Bank Branches at the District Level Comparing Pre and Post Reform 30 20 0 10 20 0 10 Branches 30 40 2012 40 2000 −5 −4 −3 −2 −1 0 1 2 3 4 5 −5 −4 −3 −2 −1 0 1 2 3 4 5 District Population Per Branch (1,000s) Centered on Cutoff Note: Each plot presents the number of operating private sector bank branches within a district, in respective years, with dots reporting local averages for districts falling within non-overlapping 200 persons per branch bins. The horizontal axis is the forcing variable of district population per branch centered on cutoff and scaled to thousands of persons per district. The estimated y-value from local linear regressions, with a 3.5 thousand persons per district bandwidth and triangular kernel, at each x-value along with 5% confidence intervals is shown, allowing for different slopes on either side of the cutoff. The year 2000 in the left plot shows a pre-reform example of branches around the cutoff. The figure on the right shows the cumulative effect of the policy on operating branches since its implementation in 2005. Local averages greater than 40 are not shown in the plots, but were included in local linear regressions. Local averages greater than 40 did not occur close to the cutoff. 39 Figure 10: Discontinuity from Reduced Form: Operating Private Bank Branches 20 15 10 5 5 10 15 20 25 Operating Branches 25 Licenses 2002 2004 2006 2008 2010 2012 2002 2004 2006 Year Banked at Cutoff 2008 2010 Year Under−Banked At Cutoff Banked at Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. Under−Banked At Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. Note: Estimated using local linear regressions with controls for district population and its square, and the prerandomization 2001 value of the dependent variable. Bandwidths are set 3.5 thousand persons per branch and estimated using a uniform kernel. The figure plots the estimated intercepts at the cutoff from the estimation of the RD equation repeated annually. The red dashed line provides the estimated intercept from approaching the threshold along the under banked side. The solid blue line reports the corresponding intercept approaching from the banked side. The distance between the two, reported for each year, shows the estimated discontinuity at the threshold. A vertical red line between the two points indicates a positive discontinuity with under banked districts exhibiting a higher value at the threshold than banked districts, with significance at least at the 10% level. A vertical dashed green line indicates a negative discontinuity estimated at least at the 10% level. The thin vertical red line at 2006 represents the first estimation made after the reform implementation. 40 2012 Figure 11: Discontinuity from Reduced Form: Private Banks Aggregate Credit Amount 0 1000 2000 10 3000 4000 20 5000 30 6000 Accounts 2002 2004 2006 2008 2010 2002 2004 2006 Year Banked at Cutoff 2008 2010 Year Under−Banked At Cutoff Banked at Cutoff Under−Banked At Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. Local Polynomial Degree: 1. Bandwidth: 3.5. Note: Accounts reported in thousands. Amounts reported in millions of rupees. Estimated using local linear regressions with controls for district population and its square, and the pre-randomization 2001 value of the dependent variable. Bandwidths are set 3.5 thousand persons per branch and estimated using a uniform kernel. See notes from Table 10 for graph description. 2002 2004 2006 2008 −1000 0 0 5 1000 10 2000 15 3000 Figure 12: Discontinuity from Reduced Form: Private Credit to Personal Loans 2010 2002 2004 2006 Year Banked at Cutoff 2008 2010 Year Under−Banked At Cutoff Banked at Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. Under−Banked At Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. Note: Accounts reported in thousands. Amounts reported in millions of rupees. Estimated using local linear regressions with controls for district population and its square, and the pre-randomization 2001 value of the dependent variable. Bandwidths are set 3.5 thousand persons per branch and estimated using a uniform kernel. See notes from Table 10 for graph description. 41 Figure 13: Discontinuity from Reduced Form: Credit from Public Sector Banks Amount 2002 2004 2006 2008 2010 10000 100 15000 120 20000 140 25000 160 30000 35000 180 Accounts 2002 2004 2006 Year Banked at Cutoff 2008 2010 Year Under−Banked At Cutoff Banked at Cutoff Under−Banked At Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. Local Polynomial Degree: 1. Bandwidth: 3.5. Note: Accounts reported in thousands. Amounts reported in millions of rupees. Estimated using local linear regressions with controls for district population and its square, and the pre-randomization 2001 value of the dependent variable. Bandwidths are set 3.5 thousand persons per branch and estimated using a uniform kernel. See notes from Table 10 for graph description. Public sector banks include State Bank of India and Associated Banks, Nationalised Banks, IDBI and Regional Rural Banks. 0 0 .5 1 1 2 1.5 2 3 Figure 14: Discontinuity from Reduced Form: Percentage Change in Private Credit Amount to Agriculture in Rural and Semi-Urban Areas 2002 2004 2006 2008 2010 2002 2004 2006 Year Banked at Cutoff 2008 Year Under−Banked At Cutoff Banked at Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. Under−Banked At Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. (Left) Direct Agriculture, (Right) Indirect Agriculture. Note: Percentage change is approximated using difference in logs relative the value reported in 2001. Estimated using local linear regressions with controls for district population and its square. Bandwidths are set 3.5 thousand persons per branch and estimated using a uniform kernel. 42 2010 −.2 −.1 0 .1 .2 Figure 15: Discontinuity from Reduced Form: Rainfall 2000 2002 2004 2006 Year Banked at Cutoff 2008 2010 Under−Banked At Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. Note: District Average Percentage Deviation from Mean. Estimated using local linear regressions. Bandwidths are set 3.5 thousand persons per branch and estimated using a uniform kernel. See notes from Table 10 for graph description. Figure 16: Discontinuity from Reduced Form: Individual Crops Cotton Statistics Output 2002 2004 2006 Year Banked at Cutoff 2008 2010 −50000 1.2 1.4 0 1.6 50000 1.8 2 100000 2.2 150000 Yield 2002 Under−Banked At Cutoff 2004 2006 Year 2008 Banked at Cutoff 2010 Under−Banked At Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. Local Polynomial Degree: 1. Bandwidth: 3.5. Wheat Statistics Output 1.6 120000 1.8 140000 2 160000 180000 2.2 200000 2.4 220000 Yield 2002 2004 Banked at Cutoff 2006 Year 2008 2010 Under−Banked At Cutoff 2002 2004 Banked at Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. 2006 Year 2008 2010 Under−Banked At Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. Note: Yield [Tonnes/Hectare] (Left), Output [Tonnes] (Right). Cotton output measured in bales rather than tonnes. 43 0 .5 1 1.5 2 Figure 17: Discontinuity From Reduced Form: Percentage Change in Private Credit Amount to Manufacturing and Processing from 2001 Level 2002 2004 2006 2008 2010 Year Banked at Cutoff Under−Banked At Cutoff Local Polynomial Degree: 1. Bandwidth: 3.5. Note: Percentage change is approximated using difference in logs relative the value reported in 2001. Estimated using local linear regressions with controls for district population and its square. Bandwidths are set 3.5 thousand persons per branch and estimated using a uniform kernel. −.1 0 .1 .2 .3 .4 Figure 18: Discontinuity from Reduced Form: Difference in Log Mean District Light from 2004 Level 2005 2007 2009 2011 Year Banked at Cutoff Under−Banked At Cutoff Local Polynomial Degree: 2. Bandwidth: 3.5. Note: Estimated using local quadratic regressions with controls for district population and its square. Bandwidths are set to 3.5 thousand persons per branch and estimated using a triangular uniform. 44