Evidence from a Regression Discontinuity Analysis in India 3

9 Tables Table 1: Continuity tests for Baseline Values at the Cutoff VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean (1) Population (2) City Pop (3) Sched Caste Tribe Pop (4) Pct Literate (5) Pct Dist Dark (6) Area Proxy (7) PrivBranches2000 0.839 [35.38] 16.01 [35.38] 16.01 [42.75] -1.344 [13.61] 2.353 [13.61] 2.353 [16.30] -1.436 [8.483] 0.265 [8.483] 0.265 [9.840] 0.0114 [0.0219] 0.0187 [0.0219] 0.0187 [0.0261] -0.00894 [0.0169] -0.0101 [0.0169] -0.0101 [0.0210] -2,485 [2,697] -3,386 [2,697] -3,386 [3,323] 0.192 [3.026] 0.567 [3.026] 0.567 [3.527] 3.500 95 122 176.7 3.500 95 122 28.56 3.500 3.500 95 95 122 122 45.24 0.553 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 3.500 95 122 0.949 3.500 95 122 8150 3.500 95 122 7.198 Note: Estimated using local linear regressions with no controls. Bandwidths are set 3.5 thousand persons per branch and estimated using a triangular kernel. 45 Table 2: Summary Statistics Banking N Banked, Pre‐reform Mean Std. Dev. N Banked, Post‐reform Mean Std. Dev. N Under Banked, Pre‐reform Mean Std. Dev. Under Banked, Post‐Reform N Mean Std. Dev. Branches SBI Nationalised RRB Foreign Old Private New Private 610 610 610 610 610 610 28.618 69.805 21.523 0.121 11.807 2.428 23.095 62.759 21.684 0.624 16.582 4.6 732 732 732 732 732 732 33.238 80.634 23.001 0.243 11.628 7.25 27.971 73.984 22.171 1 15.295 10.687 900 900 900 900 900 900 21.35 45.444 28.221 0.018 4.198 0.794 16.381 44.86 22.147 0.199 9.298 2.372 1080 1080 1080 1080 1080 1080 24.595 51.54 29.207 0.112 4.589 4.049 19.693 50.432 22.946 0.457 10.001 6.154 Public Banks 610 120.154 87.587 732 137.926 105.531 900 95.064 66.491 1080 106.006 76.429 Private Banks 610 14.234 18.58 732 18.878 20.755 900 4.992 10.375 1080 8.638 13.96 Credit Amount SBI 610 5293.635 5980.068 732 11037.746 12248.838 900 3285.651 5986.45 1080 6507.465 8548.992 Nationalised 610 10236.988 13154.392 732 22228.233 33180.444 900 4602.575 5692.052 1080 9362.257 12494.337 RRB 610 870.748 1198.64 732 1738.277 2270.793 900 950.135 1134.78 1080 1869.281 2256.909 Foreign 610 201.344 727.787 732 487.36 1620.559 900 50.173 293.389 1080 191.788 1414.19 Private 610 3813.913 7071.325 732 7637.427 12055.826 900 1354.922 3542.466 1080 2437.963 5464.27 Credit Accounts SBI 610 30945.372 31517.419 732 47639.104 50181.875 900 24107.006 24218.304 1080 38046.444 39105.763 Nationalised 610 60582.561 60584.955 732 89278.02 97041.327 900 37963.999 38526.215 1080 55938.739 58976.202 RRB 610 22255.538 33920.327 732 30088.209 47295.116 900 28251.067 34646.607 1080 36354.233 48093.88 Foreign 610 134.425 772.631 732 319.858 1656.413 900 51.02 564.603 1080 119.098 874.722 Private 610 9792.657 14751.414 732 25507.242 35027.737 900 3214.418 7356.894 1080 9889.303 22363.595 Deposit Amount SBI 607 9599.797 10660.293 732 16412.707 20661.421 892 6104.533 6197.594 1078 10180.87 10886.087 Nationalised 607 20027.738 26126.927 732 33469.464 51159.493 892 9745.183 12975.665 1078 15306.32 20677.413 RRB 607 1340.932 1519.9 732 2212.508 2520.006 892 1807.669 1792.853 1078 2828.679 2818.4 Foreign 607 181.203 1207.168 732 611.752 4849.064 892 20.185 243.413 1078 65.089 603.547 Private 607 4695.24 8722.103 732 8973.14 17799.643 892 1371.376 2938.947 1078 2798.099 5257.67 Deposit Accounts SBI 607 203.438 178.676 732 298.246 276.023 892 147.726 130.511 1078 232.131 228.963 Nationalised 607 502.83 502.301 732 683.751 696.657 892 294.637 342.959 1078 410.485 464.146 RRB 607 76.55 101.221 732 118.76 157.796 892 100.515 109.819 1078 157.422 174.789 Foreign 607 0.98 6.606 732 2.268 14.987 892 0.188 2.342 1078 0.396 2.834 Private 607 91.003 124.465 732 136.977 184.145 892 30.155 63.394 1078 50.568 93.778 Source: RBI Master Office File, BSR 1 and BSR 2 years 2001‐2011. Sample includes years 2001‐2011 for districts falling within 5 thousand persons per branch of the national average. Each year includes 122 banked districts and 180 under banked districts, from a total of 572 districts considered. Amounts are reported in Rupees million adjusted to 2011q4 prices; Accounts are reported in thousands. 46 Table 3: Summary Statistics Continued... Agriculture N Banked, Pre‐reform Mean Std. Dev. N Banked, Post‐reform Mean Std. Dev. Under Banked, Pre‐reform N Mean Std. Dev. Under Banked, Post‐reform N Mean Std. Dev. Cotton Area Output Productivity 403 403 403 32,656 59,959 1.61 53,321 127,462 0.98 Area Output Productivity 560 560 560 11,945 27,988 1.87 20,923 57,175 1.19 349 31,406 349 100,347 349 2.12 56,677 229,598 1.38 619 619 619 31,351 41,581 1.35 64,876 89,199 0.84 471 471 471 37,076 86,119 1.55 75,472 203,562 1.28 470 470 470 28,518 103,819 2.24 968 968 968 16,400 28,449 1.49 32,962 64,162 0.84 761 761 761 16,688 34,070 1.76 36,426 87,053 1.35 Maize 15,124 48,069 2.38 Onion Area 431 1,527 3,714 342 2,036 5,455 743 1,074 2,489 510 1,485 4,019 Output 431 13,885 29,608 342 17,539 36,355 743 14,587 51,185 510 24,189 99,249 Productivity 431 11.71 7.93 342 12.03 8.58 743 11.34 7.48 510 11.38 7.92 Potato Area 351 2,028 4,026 303 2,303 6,024 674 3,014 9,512 587 3,694 12,041 Output 351 28,503 44,128 303 27,843 43,051 674 67,058 248,196 587 71,627 286,377 8.19 Productivity 351 13.75 7.51 303 12.93 7.79 674 12.64 7.55 587 11.76 Rice Area 667 64,626 82,739 544 67,299 85,705 1017 88,839 104,258 784 100,968 120,405 Output 667 173,077 285,059 544 194,407 303,283 1017 160,160 221,919 784 197,829 266,243 Productivity 667 2.30 1.01 544 2.51 1.10 1017 1.61 0.87 784 1.81 0.94 Sesamum Area 573 3,245 6,935 460 2,790 4,742 908 4,826 11,359 749 5,919 15,535 Output 573 1,220 3,198 460 1,119 2,212 908 1,805 5,529 749 2,032 6,103 Productivity 573 0.35 0.23 460 0.38 0.25 908 0.32 0.22 749 0.35 0.24 Sugarcane 419 11,554 22,413 907 8,554 25,972 711 8,866 27,790 Area 523 12,161 23,096 Output 523 955,008 1,797,426 419 902,855 1,738,094 907 590,206 1,786,733 711 588,924 1,878,506 Productivity 523 70.26 35.51 419 67.35 39.47 907 53.13 26.72 711 55.86 30.25 Tobacco Area 166 7,958 16,242 176 8,267 17,829 258 454 1,647 213 620 2,082 Output 166 9,853 22,353 176 10,113 20,766 258 663 2,233 213 1,128 3,622 Productivity 166 1.54 1.53 176 1.53 1.61 258 1.63 1.88 213 1.71 1.57 Wheat 689 52,869 67,471 Area 437 60,088 81,807 349 64,550 81,240 923 49,803 65,451 Output 437 204,344 353,065 349 225,183 353,261 923 126,363 200,516 689 147,671 224,604 Productivity 437 2.21 1.25 349 2.38 1.27 923 1.78 0.97 689 1.93 1.02 Source: Rainfall data from TRMM satellite, crop data from State Agricultural Reports. Sample includes years 2000‐2010 for districts falling within 5 thousand persons per branch of the national average. Observations are crop‐years; the number of districts varies by crop as not every crop is grown in all districts. 302 of 572 districts are eligible for sample. Area is reported in Hectares square, output in tonnes, and productivity is output divided by area. Cotton reported in bales instead of tonnes. Annual Survey of Industries Banked, Pre‐reform N Mean Std. Dev. 42702 3.786 1.403 42824 0.04 0.193 42248 16.002 13.986 Banked, Post‐reform N Mean Std. Dev. 40252 3.954 1.436 40575 0.041 0.203 39268 15.204 13.878 Under Banked, Pre‐reform N Mean Std. Dev. 21133 3.567 1.345 21216 0.021 0.15 20864 14.97 14.197 Under Banked, Post‐reform N Mean Std. Dev. 17976 3.72 1.403 18123 0.025 0.159 17562 14.664 14.332 Log Total Employees Log Number of units Plant Age Log Capital (No Land or Inventory) 42339 14.911 2.876 39707 15.151 3.392 21030 14.576 2.952 17886 14.995 3.135 Log Net Assets 42352 15.679 2.883 39772 15.76 3.294 21040 15.354 2.929 17902 15.602 3.024 Log Working Capital 35823 15.306 3.024 34057 15.259 3.689 18262 15.015 3.105 15818 15.287 3.154 Log Loans 34828 14.869 4.037 32543 14.962 4.199 16258 14.874 4.084 13795 15.062 4.035 Log Total Investment 39950 14.688 3.2 37858 14.943 3.829 20517 14.248 3.298 17468 14.649 3.619 Capital Labor Ratio 42221 6.644 47.52 39543 11.121 237.379 20971 8.133 38.898 17800 10.879 105.471 Log Capital Labor Ratio 42202 0.774 1.535 39535 0.875 1.516 20958 0.89 1.662 17798 1.003 1.645 Source: Annual Survey of Industries, Unit level data 1999‐2010. Sample is restricted to plants reporting being open and reporting a valid urban or rural status. Captital Labor Ratio constructed as average of opening and closing Net Assets divided by the total wage bill plus benefits. States and UTs selected by their share of population being concentrated on one side of the threshold or the other. "Banked States" include Haryana, Uttarakhand, Punjab, Mizoram, Daman and Dimiu, Karnataka, Puducherry, and "Under Banked States" include Rajasthan, Tripura, Jharkhand, Orissa, Dadra and Nagar Haveli. 47 Table 4: Fuzzy RD: Private Bank Branches Licenses VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 (10) 2011 (11) 2012 -0.0469 [0.182] -0.0888 [0.182] -0.0888 [0.215] 0.246 [0.323] 0.318 [0.323] 0.318 [0.379] 0.414 [0.533] 0.869 [0.533] 0.869 [0.643] 0.533 [0.693] 0.936 [0.693] 0.936 [0.831] 0.657 [0.911] 1.067 [0.911] 1.067 [1.075] 2.198* [1.212] 2.628** [1.212] 2.628* [1.423] 3.785** [1.668] 4.153** [1.668] 4.153** [1.939] 5.691** [2.267] 6.265*** [2.267] 6.265** [2.638] 7.313** [2.869] 7.748*** [2.869] 7.748** [3.342] 8.510*** [3.267] 8.910*** [3.267] 8.910** [3.796] 9.665*** [3.545] 10.12*** [3.545] 10.12** [4.123] 3.500 95 122 8.714 3.500 94 122 8.917 3.500 94 122 9.241 3.500 3.500 3.500 3.500 94 94 94 94 122 122 122 122 9.847 10.62 11.92 13.83 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 3.500 94 122 15.31 3.500 94 122 17.13 3.500 94 122 18.47 3.500 94 122 19.99 (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 (10) 2011 (11) 2012 0.155 [0.159] 0.133 [0.159] 0.133 [0.186] 0.320 [0.329] 0.355 [0.329] 0.355 [0.383] 0.538 [0.561] 0.990* [0.561] 0.990 [0.668] 0.579 [0.669] 0.958 [0.669] 0.958 [0.801] 0.629 [0.902] 1.032 [0.902] 1.032 [1.069] 1.128 [1.070] 1.298 [1.070] 1.298 [1.250] 3.017** [1.445] 3.463** [1.445] 3.463** [1.674] 4.512** [1.962] 4.770** [1.962] 4.770** [2.288] 6.630*** [2.541] 6.919*** [2.541] 6.919** [2.961] 8.680*** [3.184] 9.097*** [3.184] 9.097** [3.705] 9.991*** [3.610] 10.41*** [3.610] 10.41** [4.197] 3.500 95 122 8.636 3.500 94 122 8.801 3.500 94 122 9.125 Operating Branches VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean 3.500 3.500 3.500 3.500 3.500 3.500 3.500 3.500 94 94 94 94 94 94 94 94 122 122 122 122 122 122 122 122 9.597 10.34 10.87 12.25 14.42 16.19 17.91 20.00 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 Note: 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 triangular kernel. Under banked status is instrumented for with predicted under banked assignment. Licenses are considered in operation if they are granted for a branch currently operating or pending opening. 48 Table 5: Fuzzy RD: Private Banks Aggregate Credit Private Sector Credit Accounts (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 (10) 2011 -0.264 [0.606] -0.281 [0.606] -0.281 [0.707] 0.119 [0.857] 0.523 [0.857] 0.523 [1.023] 1.147 [1.270] 1.871 [1.270] 1.871 [1.489] 3.678* [1.993] 4.199** [1.993] 4.199* [2.371] 6.725* [4.039] 8.192** [4.039] 8.192* [4.741] 10.42*** [3.698] 12.14*** [3.698] 12.14*** [4.549] 10.01* [5.461] 11.90** [5.461] 11.90* [6.439] 7.781 [4.872] 9.321* [4.872] 9.321 [5.692] 11.74* [7.009] 13.44* [7.009] 13.44 [8.198] 16.96 [10.95] 18.26* [10.95] 18.26 [12.73] 3.500 95 122 5.067 3.500 94 122 5.484 3.500 94 121 6.470 3.500 94 122 16.78 3.500 94 122 17.78 3.500 94 122 22.82 3.500 94 122 25.80 VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean 3.500 3.500 3.500 94 94 94 122 122 122 8.800 12.83 13.77 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 Private Sector Credit Amounts VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 (10) 2011 -550.7 [1,547] -182.5 [1,547] -182.5 [1,760] -867.3 [2,026] -359.5 [2,026] -359.5 [2,326] -159.9 [1,947] 294.2 [1,947] 294.2 [2,221] 994.4 [1,342] 1,813 [1,342] 1,813 [1,598] 1,060 [1,682] 1,893 [1,682] 1,893 [2,072] 1,664 [2,156] 2,523 [2,156] 2,523 [2,623] 3,504 [2,529] 4,854* [2,529] 4,854 [3,314] 1,451 [1,475] 2,703* [1,475] 2,703 [1,879] 1,461 [1,363] 2,314* [1,363] 2,314 [1,679] 2,194 [1,646] 3,044* [1,646] 3,044 [2,054] 3.500 95 122 2641 3.500 94 122 3223 3.500 3.500 3.500 3.500 94 94 94 94 121 122 122 122 2943 3466 3922 4920 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 3.500 94 122 5934 3.500 94 122 5362 3.500 94 122 4932 3.500 94 122 5990 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 triangular kernel. Under banked status is instrumented for with predicted under banked assignment. 49 Table 6: Fuzzy RD: Private Credit to Personal Loans Accounts in Personal Loans (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 (10) 2011 0.147 [0.240] 0.205 [0.240] 0.205 [0.285] 0.460 [0.447] 0.658 [0.447] 0.658 [0.518] 1.109 [0.816] 1.473* [0.816] 1.473 [0.945] 2.898** [1.216] 3.127** [1.216] 3.127** [1.486] 6.498** [2.647] 7.307*** [2.647] 7.307** [3.140] 8.159*** [2.455] 9.538*** [2.455] 9.538*** [3.179] 6.706** [2.732] 7.775*** [2.732] 7.775** [3.359] 6.957** [3.124] 7.839** [3.124] 7.839** [3.707] 8.658*** [3.047] 9.870*** [3.047] 9.870*** [3.614] 10.89* [5.652] 11.98** [5.652] 11.98* [6.562] 3.500 95 122 1.885 3.500 94 122 1.910 3.500 94 121 2.686 3.500 94 122 6.471 3.500 94 122 9.120 3.500 94 122 9.056 3.500 94 122 9.707 VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean 3.500 3.500 3.500 94 94 94 122 122 122 3.681 6.138 6.313 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 Amount to Personal Loans VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 (10) 2011 22.11 [20.50] 33.25 [20.50] 33.25 [23.46] 75.08 [51.87] 103.3** [51.87] 103.3* [60.87] 663.6 [480.0] 759.6 [480.0] 759.6 [497.4] 1,382** [657.6] 1,499** [657.6] 1,499** [763.0] 1,605** [753.1] 1,846** [753.1] 1,846** [901.2] 1,927** [887.3] 2,179** [887.3] 2,179** [1,065] 2,409 [1,533] 3,008** [1,533] 3,008 [2,151] 628.1 [626.7] 687.7 [626.7] 687.7 [735.6] 692.1 [425.4] 807.5* [425.4] 807.5 [493.8] 878.2* [485.3] 925.9* [485.3] 925.9* [556.7] 3.500 95 122 151.8 3.500 94 122 199.6 3.500 94 122 1983 3.500 94 122 1609 3.500 94 122 1200 3.500 94 122 1280 3.500 3.500 3.500 3.500 94 94 94 94 121 122 122 122 644.3 1003 1384 1658 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 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 triangular kernel. Under banked status is instrumented for with predicted under banked assignment. 50 Table 7: RD from Reduced Form: Credit from Public Sector Banks Public Sector Credit Accounts (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 (10) 2011 -3.827 [3.191] -3.866 [3.191] -3.866 [3.814] -0.621 [4.109] -0.196 [4.109] -0.196 [5.045] 0.269 [5.598] 1.956 [5.598] 1.956 [6.903] -1.419 [7.397] 1.412 [7.397] 1.412 [9.092] -1.086 [10.56] 1.662 [10.56] 1.662 [12.92] 4.170 [10.96] 8.924 [10.96] 8.924 [13.43] -0.156 [11.23] 2.854 [11.23] 2.854 [13.69] 2.630 [12.24] 7.809 [12.24] 7.809 [14.88] 4.258 [15.05] 8.348 [15.05] 8.348 [17.98] -1.188 [15.09] 2.285 [15.09] 2.285 [18.26] Bandwidth N Banked N UBanked DepMean 3.500 94 122 99.43 3.500 94 122 102.5 3.500 3.500 3.500 3.500 94 94 94 94 122 122 122 122 105.6 120.6 132.4 141.5 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 3.500 93 122 151.7 3.500 94 122 154.2 3.500 94 122 167.2 3.500 94 122 177.1 VARIABLES (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 (10) 2011 585.4 [528.0] 531.9 [528.0] 531.9 [599.5] 791.0 [774.0] 1,086 [774.0] 1,086 [1,058] 795.6 [982.6] 1,073 [982.6] 1,073 [1,343] 252.2 [2,482] 639.5 [2,482] 639.5 [2,974] -669.7 [3,326] -421.5 [3,326] -421.5 [3,959] 27.51 [3,029] 865.7 [3,029] 865.7 [3,546] -849.1 [4,649] 322.0 [4,649] 322.0 [5,400] 2,754 [3,457] 4,478 [3,457] 4,478 [4,124] 2,948 [3,487] 4,171 [3,487] 4,171 [4,329] 3,923 [3,949] 5,329 [3,949] 5,329 [4,995] 3.500 94 122 10544 3.500 94 122 11953 VARIABLES Conventional Bias-Corrected Robust Public Sector Credit Amounts Conventional Bias-Corrected Robust Bandwidth N Banked N UBanked DepMean 3.500 3.500 3.500 3.500 3.500 3.500 3.500 3.500 94 94 94 94 93 94 94 94 122 122 122 122 122 122 122 122 13493 17693 21386 23326 27547 29581 31372 34125 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 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 triangular kernel. Public sector banks include State Bank of India and Associated Banks, Nationalised Banks, IDBI and Regional Rural Banks. 51 Table 8: Fuzzy RD: Percentage Change in Private Credit Amount to Rural and Semi-Urban Areas Direct to Agriculture VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 (10) 2011 0.0681 [0.108] 0.120 [0.108] 0.120 [0.131] 0.141 [0.166] 0.266 [0.166] 0.266 [0.198] 0.216 [0.209] 0.301 [0.209] 0.301 [0.243] 0.699** [0.343] 0.917*** [0.343] 0.917** [0.409] 1.033** [0.418] 1.284*** [0.418] 1.284** [0.507] 0.734 [0.534] 0.899* [0.534] 0.899 [0.668] 0.945 [0.666] 1.236* [0.666] 1.236 [0.812] 1.795*** [0.679] 2.104*** [0.679] 2.104** [0.829] 1.407** [0.686] 1.719** [0.686] 1.719** [0.842] 1.731*** [0.659] 1.918*** [0.659] 1.918** [0.805] 3.500 95 122 -0.0700 3.500 94 122 0.0481 3.500 94 121 0.164 3.500 3.500 3.500 94 94 94 122 122 122 0.433 0.550 0.964 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 3.500 94 122 1.478 3.500 94 122 1.419 3.500 94 122 1.953 3.500 94 122 2.376 Indirect to Agriculture VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 (10) 2011 -0.00565 [0.257] 0.0720 [0.257] 0.0720 [0.290] 0.333 [0.350] 0.449 [0.350] 0.449 [0.424] 0.534 [0.423] 0.670 [0.423] 0.670 [0.511] 0.519 [0.572] 0.703 [0.572] 0.703 [0.685] 0.614 [0.505] 0.866* [0.505] 0.866 [0.606] -0.0643 [0.617] 0.0355 [0.617] 0.0355 [0.742] 0.292 [0.689] 0.357 [0.689] 0.357 [0.844] 1.637** [0.705] 1.997*** [0.705] 1.997** [0.846] 1.713*** [0.556] 2.075*** [0.556] 2.075*** [0.666] 2.207*** [0.595] 2.507*** [0.595] 2.507*** [0.711] 3.500 95 122 0.317 3.500 94 122 0.199 3.500 94 121 0.188 3.500 94 122 1.039 3.500 94 122 1.313 3.500 94 122 1.133 3.500 3.500 3.500 3.500 94 94 94 94 122 122 122 122 0.237 0.257 0.453 0.963 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 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 triangular kernel. 52 Table 9: Reduced Form RD: Rainfall Averaged Percentage Deviation from the Mean VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean (1) 2000 (2) 2001 (3) 2002 (4) 2003 (5) 2004 (6) 2005 (7) 2006 (8) 2007 (9) 2008 (10) 2009 (11) 2010 (12) 2011 -0.0405 [0.0796] -0.0498 [0.0796] -0.0498 [0.0973] -0.0212 [0.0595] -0.0128 [0.0595] -0.0128 [0.0738] -0.0237 [0.0448] -0.0278 [0.0448] -0.0278 [0.0532] -0.0396 [0.0444] -0.0354 [0.0444] -0.0354 [0.0531] -0.00818 [0.0720] -0.0147 [0.0720] -0.0147 [0.0876] 0.00883 [0.0514] 0.00278 [0.0514] 0.00278 [0.0624] -0.0638 [0.0707] -0.0874 [0.0707] -0.0874 [0.0849] -0.0373 [0.0761] -0.0580 [0.0761] -0.0580 [0.0923] 0.0247 [0.0928] 0.0359 [0.0928] 0.0359 [0.115] 0.0520 [0.0707] 0.0720 [0.0707] 0.0720 [0.0872] 0.0633 [0.0694] 0.0644 [0.0694] 0.0644 [0.0864] 0.00992 [0.0760] 0.0131 [0.0760] 0.0131 [0.0922] 3.500 95 120 -0.0500 3.500 95 120 -0.154 3.500 95 120 -0.0411 3.500 94 120 -0.208 3.500 3.500 3.500 94 94 94 119 120 120 -0.0298 0.0111 0.0970 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 3.500 94 120 0.0506 3.500 94 120 0.110 3.500 94 120 0.0545 3.500 94 120 -0.127 3.500 94 120 0.174 Note: Estimated using local linear regressions with no controls. Bandwidths are set 3.5 thousand persons per branch and estimated using a triangular kernel. 53 Table 10: RD Results: Individual Crops Cotton Yield (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 -0.0596 [0.205] -0.0693 [0.205] -0.0693 [0.258] 0.192 [0.358] 0.193 [0.358] 0.193 [0.444] 0.314 [0.521] 0.479 [0.521] 0.479 [0.665] 0.900 [0.681] 1.165* [0.681] 1.165 [0.829] 0.444 [0.563] 0.524 [0.563] 0.524 [0.730] 0.387 [0.632] 0.596 [0.632] 0.596 [0.767] 0.144 [0.649] 0.368 [0.649] 0.368 [0.812] -0.124 [0.676] -0.128 [0.676] -0.128 [0.849] -0.380 [0.595] -0.369 [0.595] -0.369 [0.734] 3.500 55 83 1.291 3.500 54 83 1.327 3.500 3.500 3.500 3.500 53 37 55 46 83 52 72 54 1.739 1.910 1.771 2.007 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 3.500 51 68 1.861 3.500 54 69 1.757 3.500 53 63 1.805 VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean Cotton Output (Bales) (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 15,129 [12,529] 15,837 [12,529] 15,837 [17,640] 710.2 [13,759] -6,044 [13,759] -6,044 [17,434] 40,656 [34,687] 53,965 [34,687] 53,965 [49,786] 62,240 [41,323] 82,530** [41,323] 82,530 [56,638] 101,178* [54,006] 113,763** [54,006] 113,763 [71,802] 176,076** [86,326] 216,181** [86,326] 216,181** [107,247] 148,302* [81,145] 173,677** [81,145] 173,677* [103,780] 98,366 [65,372] 110,198* [65,372] 110,198 [85,300] 79,421 [58,139] 87,096 [58,139] 87,096 [75,475] 3.500 55 83 44580 3.500 54 83 38189 3.500 51 68 111801 3.500 54 69 108616 3.500 53 63 115830 VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean 3.500 3.500 3.500 3.500 53 37 55 46 83 52 72 54 64526 115579 97665 132027 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 Wheat Yield VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 0.125 [0.143] 0.149 [0.143] 0.149 [0.161] 0.124 [0.167] 0.116 [0.167] 0.116 [0.193] 0.0764 [0.166] 0.126 [0.166] 0.126 [0.204] 0.173 [0.158] 0.221 [0.158] 0.221 [0.188] 0.479 [0.426] 0.643 [0.426] 0.643 [0.541] 0.431 [0.291] 0.546* [0.291] 0.546 [0.355] 0.459* [0.240] 0.629*** [0.240] 0.629** [0.290] 0.402 [0.260] 0.542** [0.260] 0.542* [0.322] 0.313 [0.216] 0.369* [0.216] 0.369 [0.263] 3.500 58 93 2.001 3.500 57 88 1.944 3.500 55 80 2.172 3.500 53 87 2.212 3.500 49 69 2.207 3.500 3.500 3.500 3.500 58 59 53 49 88 92 88 64 2.070 1.921 1.969 2.211 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 Wheat Output (Tonnes) VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 13,076 [11,915] 14,898 [11,915] 14,898 [14,158] 2,897 [10,099] -1,283 [10,099] -1,283 [12,315] 4,849 [18,090] 5,066 [18,090] 5,066 [21,197] 6,177 [20,159] 4,761 [20,159] 4,761 [22,151] 63,023 [56,729] 79,457 [56,729] 79,457 [73,073] 15,449 [30,231] 12,576 [30,231] 12,576 [36,954] 49,503 [39,057] 52,895 [39,057] 52,895 [46,181] 26,082 [36,240] 27,974 [36,240] 27,974 [46,379] -10,164 [27,953] -17,642 [27,953] -17,642 [35,750] 3.500 58 93 161445 3.500 57 88 146242 3.500 3.500 3.500 3.500 58 59 53 49 88 92 88 64 162996 151854 158668 196304 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 3.500 55 80 191220 3.500 53 87 204942 3.500 49 69 194492 Note: Cotton output measured in bales rather than tonnes. 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 triangular kernel. 54 Table 11: RD Results: Crop Yield Index Reduced Form Separately for Each Year VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean (1) 2003 (2) 2004 (3) 2005 (4) 2006 (5) 2007 (6) 2008 0.0113 [0.101] -0.00324 [0.101] -0.00324 [0.120] -0.114 [0.146] -0.123 [0.146] -0.123 [0.173] 0.120 [0.132] 0.117 [0.132] 0.117 [0.155] 0.380 [0.268] 0.492* [0.268] 0.492 [0.358] 0.187 [0.157] 0.226 [0.157] 0.226 [0.198] 0.284* [0.170] 0.337** [0.170] 0.337 [0.213] 3.500 3.500 3.500 3.500 47 44 38 47 74 72 67 74 -0.115 -0.0303 0.0340 0.0133 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 3.500 44 60 -0.0710 3.500 45 72 -0.0449 Fuzzy RD Instrumenting for Private Bank Credit Accounts, Pre-reform and Post-Reform VARIABLES Conventional Bias-corrected Robust (1) preref (2) postref -0.0786 [0.214] -0.109 [0.214] -0.109 [0.247] 0.0290* [0.0154] 0.0362** [0.0154] 0.0362* [0.0193] Bandwidth 3.500 3.500 N Banked 91 174 N UBanked 146 273 DepMean -0.0731 -0.0168 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 Note: Index of crop yield using weighted averages of the crops rice, wheat, jowar and groundnut. Weighted by crop revenue share. Estimated using local linear regressions with controls for district average rainfall percentage deviation from the mean, district population and its mean and the pre-randomization 2002 value of the dependent variable. Bandwidths are set at 3.5 thousand persons per branch and estimated using a triangular kernel. Pre-reform years are considered 2003-2004 and post-reform is 2005-2008. 55 Table 12: Fuzzy RD: Percentage Change in Private Credit Amount to Manufacturing and Processing from 2001 Level (1) 2002 (2) 2003 (3) 2004 (4) 2005 (5) 2006 (6) 2007 (7) 2008 (8) 2009 (9) 2010 (10) 2011 -0.141 [0.528] -0.151 [0.528] -0.151 [0.640] 0.0501 [0.614] 0.139 [0.614] 0.139 [0.752] 0.0285 [0.533] 0.0544 [0.533] 0.0544 [0.657] -0.367 [0.488] -0.328 [0.488] -0.328 [0.614] -0.756 [0.624] -0.860 [0.624] -0.860 [0.766] -0.114 [0.590] -0.232 [0.590] -0.232 [0.706] 0.543 [0.631] 0.703 [0.631] 0.703 [0.758] 0.985 [0.610] 1.395** [0.610] 1.395* [0.740] 1.402** [0.655] 1.824*** [0.655] 1.824** [0.794] 0.913 [0.686] 1.214* [0.686] 1.214 [0.825] 3.500 95 122 0.934 3.500 94 122 1.098 3.500 3.500 3.500 3.500 3.500 94 94 94 94 94 121 122 122 122 122 0.678 0.763 0.553 0.694 1.117 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 3.500 94 122 1.231 3.500 94 122 1.287 3.500 94 122 1.410 VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean 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 triangular kernel. Table 13: Diff n Diff: States Selected around Under Banked Threshold, 1999-2010 (1) Ln Net Assets (2) Ln Working Capital (3) Ln Loans (4) Ln Tot Investment (5) Cap Labor Ratio TreatXPost2006 0.171 [0.142] 0.264* [0.136] 0.235* [0.116] 0.197* [0.106] 3.426* [1.724] Observations R-squared State FEs State Trend 118,236 0.270 Yes Yes 101,566 95,269 113,296 0.195 0.082 0.200 Yes Yes Yes Yes Yes Yes Robust standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 Standard Errors Clustered at State level 118,128 0.012 Yes Yes VARIABLES Note: Banked States include Haryana, Uttarakhand, Punjab, Mizoram, Daman and Dimiu, Karnataka and Puducherry. Under Banked States include Rajasthan, Tripura, Jharkhand, Orissa and Dadra and Nagar Haveli. All regressions control for post 2006 and treated state individual effects, logged number of units in firm and the logged number of employees in the enterprise, plant age and its square, a year trend, state specific year trends and state fixed effects. 56 Table 14: Difference in Log Mean District Light from 2004 Discontinuity from Reduced Form Estimated Annually VARIABLES Conventional Bias-corrected Robust Bandwidth N Banked N UBanked DepMean (1) 2005 (2) 2006 (3) 2007 (4) 2008 (5) 2009 (6) 2010 (7) 2011 -0.0258 [0.0193] -0.0297 [0.0193] -0.0297 [0.0210] 0.00485 [0.0272] 0.00720 [0.0272] 0.00720 [0.0300] 0.0942*** [0.0297] 0.108*** [0.0297] 0.108*** [0.0319] 0.0916*** [0.0322] 0.104*** [0.0322] 0.104*** [0.0353] 0.0346 [0.0707] 0.0426 [0.0707] 0.0426 [0.0773] 0.0972 [0.0605] 0.119** [0.0605] 0.119* [0.0663] 0.0810* [0.0492] 0.105** [0.0492] 0.105** [0.0531] 3.500 94 122 0.355 3.500 94 122 0.219 3.500 94 122 -0.139 3.500 3.500 3.500 3.500 94 94 94 94 122 122 122 122 -0.0808 0.114 0.0722 0.0259 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 Fuzzy RD Instrumenting for Private Bank Branches, Pre-reform and Post-Reform VARIABLES Conventional Bias-corrected Robust (1) preref (2) postref -0.0156 [0.100] -0.0111 [0.100] -0.0111 [0.117] 0.00482 [0.00413] 0.0119*** [0.00413] 0.0119** [0.00491] Bandwidth 3.500 3.500 94 658 N Banked N UBanked 122 854 DepMean -0.139 0.143 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 Note: Reduced form 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 kernel. The fuzzy regression discontinuity is estimated using local linear regressions. The number of operating private bank branches is instrumented with predicted under banked assignment. Controls include district population and its square. Pre-reform year is 2005 using 2004 as the base year for the approximate percentage change. Post-reform years are 2006-2012. 57 Table 15: NREGA Discontinuity in District Phase Assignment VARIABLES Conventional Bias-Corrected Robust (1) Phase 1 (2) Phase 2 (3) Phase 3 -0.0648 [0.119] -0.121 [0.119] -0.121 [0.139] 0.0145 [0.0909] 0.0710 [0.0909] 0.0710 [0.109] 0.0503 [0.135] 0.0497 [0.135] 0.0497 [0.160] Bandwidth 3.500 3.500 3.500 N Banked 93 93 93 N UBanked 121 121 121 DepMean 0.285 0.201 0.514 Standard errors in brackets *** p<0.01, ** p<0.05, * p<0.1 Note: Reduced form 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 kernel. NREGA was rolled out in 3 phases between 2006 and 2009 based on some measure of expected program need by district. 58 10 10.1 Data Appendix Districts The majority of analysis in this paper is conducted at the administrative district level in India. Districts constitute the administrative level directly below the state government (and union territory). Data sets at the district level rarely provide numerical identifiers. When available, these identifiers typically do not easily map to other data sets. Further, the anglicized spelling of district names is often inconsistent across and even within data sets. Renaming and redistricting also occur relatively frequently in India. As such, each data set required the assignment of a numerical identifier before conducting analysis. To ensure consistent measures in the data across time, I adjust all data to their 2001 district boundaries from the Population Census. I first assign each district its 2001 state and district numerical codes from the 2001 Population Census, or an auxiliary district code if the district was formed post 2001. Then using the atlas provided in the 2011 Population Census, I map new districts back to their source districts in 2001. Although super-districts, created when newly formed districts drew land from more than one source district, are identified, they are dropped from the analysis.20 District websites, newspapers and other internet based resources were used to help map alternative spellings to numerical codes. 10.2 Banking Branches and Licenses Data on the number of operating branches and licenses are from the Master Office File (MOF) accessed from the RBI website in spring 2012. Opening and closing dates (when applicable) are provided for each bank, as well as information regarding branch location and type of business conducted at the branch (e.g. General Banking, Specialized Banking, ATM). “Brick and mortar” branches are used in the analysis, meaning branches 20 New districts since 2001 that claimed territory from more than one source district are dropped along with the source districts due to issues with the aggregation. In addition, Thane and Pune districts in Maharashtra are dropped. These districts are located close to Mumbai but are not technically classified as belonging to the greater Mumbai area. They constitute outliers as they achieve rapid growth more likely to be attributable to their proximity to Mumbai. Thane is on the under banked list while Pune is not, though the RBI amended the policy to 2008 to make centers within 100km of Mumbai ineligible for under banked status, effectively removing Thane’s status. Varanasi district in Uttar Pradesh is also dropped after 2002 due to the 2002 merger of the private sector Banaras State Bank with Bank of Baroda which is a nationalised bank. Banaras State Bank primarily operated in Uttar Pradesh with the bulk of its branches in districts designated as under banked. However, 20 branches operated in Varanasi which happens to be located right at the cutoff on the banked side. The vast majority of branches affected by the merger belonged to districts designated as under banked. However, the reclassification of 20 branches to public sector bank status just on the banked side of the cutoff results in a sudden drop in the banked intercept in 2003 for private banks. Since most of these branches continued to operate under the public sector, the drop-off creates an exaggerated representation of the policy effect, which does not accurately represent the change to the banking environment. While these branches could be “added back” using the detailed data from the MOF, the same cannot be done for the aggregated data on credit. 59 classified at being general banking or specialized banking. Not Administratively Independent Offices such as extension counters and ATMs are excluded from analysis. The number of operating branches for each year is calculated as the number of branches with an opening date prior to January 1st of that year and a closing date afterward or missing. Operating branches by subsets of bank group classification are calculated similarly. Licenses are considered to be operating if issued before January 1st of a given year with a branch close date afterward or missing. Thus, licenses can be in “operation” even if branch opening occurs at a later date. After the December 2009 reform granting blanket permissions to open in low population centers, the incidence of unreported license dates for branches in such centers increased. The assumption is made that these constitute branches exercising the blanket permission, such that the effective license date is taken to be the date of branch opening.21 Constructing the forcing variable In constructing the forcing variable and national average I follow the APPBO procedure 22 described for identifying deficit districts during the policies of the 1980s and also that for identifying under banked states in the RBI Report of the Group to Review Branch Authorisation Policy (RBI, 2009). I take the number of operating branches on September 7th, 2005, the day prior to the 2005 Master Circular issue date that implemented the branching policy reform. Following the rule that Under Banked Status = 1(district population per branch ¿ national average) yields nearly an exact match to the official 2006 list of under banked districts in the 2006 master circular.23 Out of 572 districts only 6 fail to match their official status. Due to redistricting and the level of aggregation of credit and deposits data, I aggregate all districts bifurcating since 2001 back to their 2001 boundaries. In cases that new districts form from two or more source districts, these are aggregated into a single super district, resulting in 572 districts. Of these, I denote 202 districts as banked (with 204 on the official list) and 370 under banked (368 officially). After dropping super districts from the sample, 4 misassigned districts remain. Replicating the analysis taking the number of operating branches on January 1st, 2006 yields similar results. 21 A similar pattern for license dates from branches in urban centers in the Northeast region that had a special exception for blanket permissions for urban centers, and only in that region, provides additional support for this assumption. 22 The Average Population Per Bank Office was constructed using the district population from the most recent population census, in this case that from 2001, and dividing that by the number of bank offices in that district. I restrict the set of offices to those conducting general and specialized bank business which may depart from the actual algorithm used by the RBI. The national average to which the value is compared is the total population of India divided by the number of bank offices. 23 A list of under banked districts was issued with the 2005 master circular. A slightly revised list was reissued with the 2006 master circular and remained unchanged through 2009, after which the districts of some states were dropped. The national average computed using September 7th, 2005 as the policy date was 14,915 persons per branch in India. 60 Credit The Basic Statistical Returns 1 (BSR1) provides information on credit accounts, credit limits and credit outstanding by scheduled commercial banks including RRBs (last accessed spring 2014). The data are reported annually by banks with values as of March 31st for that year. Credit captured by BSR1 relates to gross bank credit such as term loans, cash credit, overdrafts, etc. Detailed descriptions are provided by the RBI. The financial year 200X-200Y is reported as 200Y in the paper and is reported with consistent notation across analyzed data. Values are delineated by bank group and population group at the district level (e.g. number of credit accounts with Nationalised Banks, by semi-urban areas in Rangareddy). Locations, such as semi-urban Rangareddy, represent the area of credit utilization for loans exceeding 2 lakh Rs. for which detailed account information is collected. Loans of lesser amounts are reported with less information, and the RBI assumes they are utilized in the same area as which the loan was sanctioned. Credit amounts are further delineated by utilization purpose, coined “occupation,” and include : agriculture, industry, professional and other services, personal, trade, transport operators, finance and all other. These are broken down further for agriculture into “direct” and “indirect,” for industry by “construction” “mining” “manufacturing and processing” and “electricity, gas and water” and trade by “retail” and “wholesale.” Personal loans are also presented disaggregated, but the delineation between subgroups appears to be inconsistent through time so are always treated as aggregated personal loans in the analysis. A reclassification of loans to make occupations consistent with a 2004 update of industrial codes occurred in 2008. The reclassification should not have affected aggregate measures of account and amounts, though caution should be taken when attempting to draw comparisons at the occupation level before and after 2008.24 The BSR2 provides analogous information for deposits and is structured similarly (last accessed spring 2014). Values are reported for the number of deposit accounts and deposit amounts. The BSR7 provides quarterly data on credit, deposits and reporting branches. Analysis on BSR7 is not included in this paper. All credit and deposit limits and amounts are adjusted using the Consumer Price Index for Industrial Workers provided by India’s Labour Bureau. I adjust all values to 2011, fourth quarter prices. Amounts are reported in Rupees. Population Groups The RBI follows a specific assignment procedure for population groups. Based on the Population Census, locations with populations less than ten thousand are desig24 Two districts exhibit measures of credit accounts and amounts that appear to reflect coding errors in the data. Mallapuram, Kerala is dropped in 2004 due to an unexplainable jump in the magnitude of credit unmatched in the district in any other years. Ghaziabad, Gujarat in 2008 displays even more erratic values for certain credit measures. These values are set to missing as the remaining appear unaffected. In both instances, private sector banks with a presence in the concerned district were acquired by the public sector. The reclassification of the bank to the public sector may have created underlying issues in the data reported in those places for those years. 61 nated rural, 10,000 - 100,000 semi-urban, 100,000 - 1 million urban and greater than 1 million metropolitan. Prior to 2005 locations were assigned status based on their 1991 Population Census values. The switch to the 2001 Population Census for reports in 2006 and later make strict comparisons between the sets of years complicated at the disaggregated population group level. The problem appears to be greater for the metropolitan and urban population groups, as fewer centers exist in these categories. The scope for problems appears smaller for rural and semi-urban classifications due to the high volume of centers in these categories. Still, the caveat should be kept in mind for analysis at the disaggregated level. 10.3 Agriculture Crop output and area The data on crop output and area are reported in the Annual Crop Yields at District Level from the Crop Production Statistics. The production output in tonnes and area cultivated in square hectares are reported by crop at the district level either annually or by season, depending on the crop and state. Reported crops vary across districts, and the detail of information on variety and growing season also varies across states and years. I develop the data from a file made available from the Government of India for years 19981999 to 2010-2011 (years reported July-June). Extensive cleaning of district and crop names, as well as accounting for redistricting, is required to analyze the data as a panel. I match each district reported to their 2001 Population Census identification number or to a 2011 ID number constructed for this analysis when dealing with new districts since 2001. Analysis is restricted to years 2001-2010 which exhibit lower frequencies of missing data. Missings values after 2010 are reported to be due to unfiled state reports. Districts never reporting positive statistics for a crop over the sample period are dropped from analysis for that individual crop. In years a district reports a missing value for a crop that is reported in that district in other years, the value is interpreted as null and replaced with a zero value. Crop prices The data on crop prices are from the Farm Harvest Prices of Principle Crops. States are responsible for reporting crop prices for a set of prominent crops each year. The prices are supposed to be collected during the peak harvest times of each crop and account for variations in quality. States vary in their reporting of crop prices by season and detail on variety. Further, states vary in reporting price for some crops by product (e.g. some report prices for sugarcane while others only report prices for raw sugar, cotton lint or whole cotton, etc.) Technical conversion factors for raw crops to agricultural outputs provided by the Statistic Division of the FAO are used where applicable to match prices to corresponding crop outputs. Prices are reported in Rupees per Quintal (an Indian quintal is 100 kg) and must be converted to Rupees per tonne for consistent units with the output data. I have developed the data from pdf reports available in separate sets by state for 2001-2002 to 2003-2004, 2004- 62 2005, 2005-2006, and 2006-2007 to 2007-2008. Efforts to process the remaining years of the data are under way. Extensive cleaning of district names, accounting for redistricting, and assignment to identification numbers was similarly required. Crop yield index Annual crop yield is calculated as crop output in tonnes per hectare cultivated for that crop. To create the index of crop yields as in Jayachandran (2006), I match the crop prices data to the crop output and area data. Four of the top five revenue producing crops for India identified in Jayachandran (2006) are used in the index, rice, wheat, jowar and groundnut. Sugar is excluded due to concerns regarding the accuracy of conversions of sugarcane to raw sugar production in order to match the two data sets, and whether the reported prices for sugar capture actual prices faced by farmers after accounting for delay of payments bargaining. Crop yields are normalized to have mean values equal to one in each year for comparability across crops. Weighted averages of the log values of the four crop yields are taken for each district year, using the crop revenue share of the total crop revenue of the district from those four crops as weights. When matching the price and production data sets, season and variety matches are made when the detail of data from both sets allow. Otherwise, the mean of price data by district and crop are calculated (if price is broken out by variety or season) and matched to the production data for that crop-year. To increase the number of matches, when prices are missing for a crop at the district level, the weighted state average prices provided in the reports are used. Missing crop prices at the district level generally correspond to relatively low levels of output in the production data. The index is currently constructed for 2002-2008, with efforts to process the remaining years of data under way. 10.4 Industry Annual Survey of Industries The Annual Survey of Industries (ASI) is a detailed survey of registered manufacturing firms in India conducted by the Central Statistical Organisation. The ASI is used extensively in economic research (Hsieh and Klenow, 2009; Bollard et al., 2013) to name just a few). I use fiscal years 2001-2010 in my analysis. In these years, all firms with greater than 100 workers were enumerated, as were all firms operating in the five less developed states/UTs (Manipur, Meghalaya, Nagaland, Tripura and Andaman & Nicobar Islands). The remainder of registered firms (those with greater than 10 workers, assuming compliance) were surveyed from samples representative at the State by NIC-2004 4 digit industry code. In addition to the values reported directly in the ASI, I construct the capital labor ratio as the average of the opening and closing values of assets net of depreciation divided by the sum of the firm’s wage bill plus benefits, as in Hsieh and Klenow (2009). Due to the joint census-sampling methodology, I conduct my analysis at the state level in order 63 to apply proper weighting for a representative sample of all registered firms. A thorough discussion of the ASI data can be found in Bollard et al. (2013). 10.5 Remote Sensing DMSP-OLS Nightlights The Defense Meteorological Satellite Program (DMSP) maintains data sets with of night lights data, constituting a yearly average of the amount of light emitted into space at night for a roughly 1km square grid. Using satellite images, algorithms to control for reflection, cloud cover and other confounding factors assign a digital number between 0 and 63 for each cell that may be downloaded as a finely pixelated map of the Earth. Using the boundary outline of India’s administrative districts in 2001, I construct the district average of the digital numbers in each district. I then calculate the percentage change of this average as the log of the district mean value minus the log district mean from 2004. Analyzing changes in growth across districts, as opposed to levels is important due to measurement error introduced through machine learning and the algorithms applied to eliminate glare light bleed. I have processed data from satellites F16 and F18, that cover calendar years 2004-2012. Efforts are under way to process the data from F15 that would extend the data set back to year 2000. A thorough discussion of the nightlights data is included in Henderson et al (2012). TRMM Rainfall Data Rainfall strongly affects agricultural productivity. To the extent that rainfall varies annually across districts, conditioning on it will improve my precision for estimates related to agriculture. I use the publicly available data collected by the Tropical Rainfall Measuring Mission (TRMM) satellite jointly maintained by the National Aeronautics and Space Administration (NASA) and the Japan Aerospace and Exploration Agency (JAXA). Fetzer (2014) gives a detailed description of these data and their verification processes. These data are collected from a satellite orbiting approximately 250 miles above the Earth’s surface that completes an orbit several times a day and is able to detect rainfall falling as lightly as 0.7 millimeters per hour. Daily rainfall measures are available from 1998-2012 on a 0.25 by 0.25 degree grid, making it the finest available spatial resolution for India to the best of my knowledge. These data are likely favorable to those generated using ground rainfall gauges as the latter require local monitoring and maintenance, the quality of which may vary systematically with the prosperity of districts. Further, the spatial diffusion of gauges is not uniform across India, requiring different levels of interpolation between rain gauges that can introduce measurement error that may be difficult to account for and change in less transparent ways as the number and location of gauges vary across time. 64 11 Theoretical Framework Appendix This section sketches out the theoretical framework for anticipated competition in the second period leading to increased levels of credit at the time of policy announcement in the first. Details of intermediate steps are omitted in the interest of space. The participation constraint for borrower i in each period is given by,  E[π1 ] = P (RA )[RA (1 − θ ) − (1 + r1 )] − s > µ P eriod 1 s i s s i (3) E[π2 ] = {P (RA )[RA (1 − θ ) − (1 + r2 )] − s1(F irst T ime Borrower) > µ P eriod 2 s i s s i where θi ∼ unif orm[0, 1] is a privately known cost to the borrower that is constant across periods (as is being a safe type), rt denotes the interest rate in period 1 or 2, and 1(F irst T ime Borrower) is the indicator function for the borrower’s first period of borrowing from the specific bank. Consider the borrower participation constraint from period t. The indifferent borrower with type θi facing interest rate rt will satisfy Ps (RsA )RsA (1 − θi ) = Ps (RsA )(1 + rt )] + s1(F irst T ime Borrower) + µ (4) Rearranging terms, the indifferent borrower may be expressed as a function of the interest rate rt , θ̂i (rt ) = 1 − Ps (RsA )(1 + rt ) + s1(F irst T ime Borrower) + µ Ps (RsA )RsA (5) such that all borrowers with θi < θ̂i (rt ) will demand a loan with interest rate rt . For a market of size M , total demand for loans at interest rate rt will be M β θ̂i (rt ). Assume banks are profit maximizers, face an exogenous marginal cost of funds plus administrative costs of lending equal to (1 + ρ), and cannot discriminate in the interest rate offered to repeat and first time borrowers. The bank’s participation constraint from each period is, E[πtB ] = Ps (RsA )(1 + rt )θ(rt ) > (1 + ρ)θ(rt ) f or t = 1, 2 (6) In deciding the interest rate for each period, the incumbent bank will anticipate its outcome in the second period if facing entry and take that into consideration in setting its first period interest rate. Specifically, if a new bank enters the market in the second period, the incumbent will expect to compete in interest rates such that the entrant must offer his zeroprofit condition interest rate and the incumbent will offer the interest rate making his first period borrowers that do not pay the screening fee if they stay indifferent between borrowing from him and the incumbent. 65 Sketch of Proof: If the incumbent offers an interest rate higher than that making first period borrowers indifferent between borrowing from the incumbent while avoiding switching costs and borrowing from the entrant while incurring the switching costs, then the incumbent loses the entire market to the entrant. If the incumbent offered an interest rate lower than that value, then he loses profits from the locked in first period borrowers but gains no new borrowers since new borrowers must pay the screening fee regardless and the entrant’s interest rate is strictly lower. If instead the entrant offered a price above the zero profit condition interest rate, then the incumbent would increase his rate to earn higher profits off of his first period borrowers. This, however, creates incentive for the entrant to lower his interest rate a small amount and capture the entire market. If the entrant instead lowers his interest rate he will serve the entire market at a loss. Taking the second period equilibrium into consideration, the incumbent knows his second period interest rate when facing entry will be 1 + r2I = 1+ρ+s Ps (RsA ) by equating demand for the zero profit interest rate and demand for an interest rate when the switching cost need not be incurred. The incumbent will then maximize first period interest taking the second period predetermined interest rate into consideration as the first period interest rate will determine the demand faced in both periods. Thus, the incumbent’s maximization problem is maxPs (RsA )(1 + r1I )θ(r1I ) + δPs (RsA )(1 + r2I )θ(r2I ) − [(1 + ρ)θ(r1I ) + δ(1 + ρ)θ(r2I )] r1I , r2I (7) Substituting in the value for r2I and setting demand equal in both periods reduces the problem to maxPs (RsA )(1 + r1I )θ(r1I ) + δPs (RsA )( r1I 1+ρ+s )θ(r1I ) − (1 + δ)(1 + ρ)θ(r1I ) Ps (RsA ) (8) Taking the first order condition with respect to r1I , setting it equal to zero and solving for the optimal first period interest rate for the incumbent yields, 1 + r1I∗Entry = 1 {Ps (RsA )RsA − (1 + δ)s − µ + (1 + ρ)} 2Ps (RsA ) (9) Intuitively, the incumbent increases the interest rate with the expected payoff of the project to capture additional surplus as well as the cost of lending the funds and lowers the interest rate with the borrower’s reservation utility. The incumbent lowers the interest rate as the switching cost increases, as this relaxes the constraint on the interest rate he offers in the second period, allowing for higher profits from each continuing first period borrower. To determine the effect of anticipated competition on first period lending, consider an incumbent that does not expect entry in the second period. He will find it optimal to set 66 interest rates so as to maximize total profit from both periods, increasing the interest rate in the second period to extract the additional surplus from the repeat borrowers no longer paying the screening cost. Since no other changes occur to the environment, the incumbent will maximize profits by serving the same set of borrowers in both periods, setting the second period interest rate so as to make the marginal borrower indifferent between accepting the loan and not. The maximization for the incumbent not expecting entry may be expressed as, maxPs (RsA )(1 + r1I )θ(r1I ) + δPs (RsA )(1 + r1I + s)θ(r1I ) − (1 + δ)(1 + ρ)θ(r1I ) r1I (10) Taking the first order condition with respect to r1I , setting it equal to zero and solving for the optimal first period interest rate for the incumbent yields, 1 + r1I∗N oEntry = 1 δP (RsA ) A A {P (R )R − (1 + )s − µ + (1 + ρ)} s s s 2Ps (RsA ) (1 + δ) (11) Finally, since the interest rate determines the first period quantity of credit, anticipated competition will lead to an expansion of credit if 1+r1I∗Entry < 1+r1I∗N oEntry . This inequality reduces to the simple expression, Ps (RsA ) 1+δ < 1 that must always be true. Hence, introducing the potential of future competition leads to an expansion of credit at the time announcement. 67