Target Setting and Allocative Inefficiency in Chinese Lending

Figures and Tables 0 ln(change rate in loan value) .0005 .001 .0015 Figure 1: Log Changes in Loan Value (City-level data) -15 -10 -5 0 5 Day (relative to the month end) 10 15 -.0005 0 Coefficients .0005 .001 .0015 (a) Raw plots -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) (b) Coefficients plots Note: dates more than +/-13 days away from the month end are excluded from the raw data graph; days outside of the +/-13 day window are the omitted category in the coefficient plots, which are conditional on year-month, day-of-week, and city fixed effects. Standard errors are clustered at the city level. -.1 0 Coefficients .1 .2 .3 Figure 2: Value of New Lending Issued, Coefficients Plots (Branch-level data) -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) Note: dates more than +/-13 days away from the month end are the omitted category; coefficient plots are conditional on year-month, day-of-week, and city fixed effects; standard errors are clustered at the branch level. -.1 -.005 0 0 Coefficients .005 .01 Coefficients .1 .2 .015 .3 .02 .4 Figure 3: Extensive and Intensive Margins of New Loan Issued -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) (b) Intensive: log(N ewLoans|N ewLoans > 0) -.1 -.05 0 0 Coefficients Coefficients .05 .1 .1 .15 .2 .2 (a) Extensive: P rob(Anylending) -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) (c) Intensive: log(AverageSize) -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) (d) Intensive: log(Contracts)) Note: dates more than +/-13 days away from the month end are the omitted category; coefficient plots are conditional on year-month, day-of-week, and city fixed effects; standard errors are clustered at the branch level. Coefficients (Share of Bad Loan (Loan value)) -.02 -.01 0 .01 .02 .03 Figure 4: Share of Bad Loans (weighted by loan amount), Coefficients Plots -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) Note: dates more than +/-13 days away from the month end are the excluded category; coefficient plots are conditional on year-month, day-of-week, and city fixed effects; standard errors are clustered at the branch level. 0 .005 Coefficients (Log(1+loan size)) .1 .2 .3 Coefficients (Any lending) .01 .015 .02 .025 .4 .03 Figure 5: Firms with good vs bad previous history, Coefficients Plots -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) Good History -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) Bad History Good History (b) Extensive Margin 14 .1 Coefficients (Log of loan size) 14.2 14.4 14.6 Coefficients (Share of bad loan) .15 .2 .25 14.8 (a) Overall Effects Bad History -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) Good History Bad History (c) Intensive Margin -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) Good History Firms Bad History Firms (d) Quality Note: good-history firms are defined as firms with unblemished repayment records up to that points; bad-history firms are defined as those with at least one previous loan that had been classified as bad up to that points. Dates more than +/-13 days away from the month end are the excluded category; coefficient plots are conditional on year-month, day-of-week, and city fixed effects; standard errors are clustered at the branch level. 0 0 Coefficients (Any lending) .01 .02 Coefficients (Log(1+loan size)) .1 .2 .3 .4 .03 Figure 6: SOE vs Non-SOE Firms, Coefficients Plots -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) SOEs -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) Non-SOEs SOEs (b) Extensive Margin 0 14 Coefficients (Log of loan size) 14.5 15 Coefficients (Share of bad loan) .05 .1 .15 .2 15.5 (a) Overall Effects Non-SOEs -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) SOEs Non-SOEs (c) Intensive Margin -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) SOEs Non-SOEs (d) Quality Note: SOE firms are defined as firms owned or controlled by the state. Dates more than +/-13 days away from the month end are the excluded category; coefficient plots are conditional on year-month, day-of-week, and city fixed effects; standard errors are clustered at the branch level. .01 .2 Coefficients (any lending) .02 .03 .04 .05 Coefficients (log(1+loan size)) .4 .6 .8 .06 1 Figure 7: High Growth vs Low Growth Cities, Coefficients Plots -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) Top quartile GDP growth -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) Bottom quartile GDP growth Top quartile GDP growth (b) Extensive Margin 14.6 .02 Coefficients (loan size) 14.8 15 15.2 Coefficients (Share of bad loan) .04 .06 .08 .1 15.4 (a) Overall Effects Bottom quartile GDP growth -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) Top quartile GDP growth Bottom quartile GDP growth (c) Intensive Margin -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) Top quartile GDP growth Bottom quartile GDP growth (d) Quality Note: high-growth cities are defined as cities with GDP growth rates above the sample’s 75 percentile value in a given year; low-growth cities are defined as those with GDP growth rates below the sample’s 25 percentile value in a given year; the set of high- and low-growth cities vary across years. Dates more than +/-13 days away from the month end are the excluded category; coefficient plots are conditional on year-month, day-of-week, and city fixed effects; standard errors are clustered at the branch level. Table 1: Summary Statistics Mean Median Std Obs 43.95 0 195.00 1.13 499,346 494,241 Panel A: City-level data Cumulative Loan Value (million RMB) 106.19 Change Rate in Lending (%) 0.056 Panel B: Loan-level data aggregated to the branch level (balanced) Loan Value (million RMB) 0.13 0 6.39 10,137,096 Prob(Any Lending) (%) 1.82 0 13.37 10,137,096 Panel C: Loan-level data aggregated to the branch level Loan Value (million RMB) 7.23 Average Loan Size (million RMB per contract) 5.60 Number of Contracts 1.50 Share of Bad Loans (%) 12.90 (unbalanced sub-sample) 2.00 46.9 183,744 1.50 41.1 183,744 1.00 2.55 183,744 0 32.48 183,744 Notes. The branch-level summary statistics in Panel A are based on city-level daily aggregates of loans outstanding to a partially privatized bank during the years 2006 – 2010. The loan-level data in Panels B and C are based on the daily new loans issued to a randomly selected subset of borrowers from a state bank during the years 1997 – 2010, aggregated to the subbranch level. In Panel B, branch-day observations in which there was no lending to these borrowers are coded as zeros. In Panel C branch-day observations with no lending are excluded. See the text for further details on the sample construction and detailed variable definitions. Table 2: Loan Quantity and Quality Responses to Month End Incentives Dependent Variables: Quantity Response Quality Response ∆log(Loans) log(1 + N ewLoans) AnyLoan log(N ewLoans) log(Loans/Contracts) log(Contracts) Share of Bad Loans (1) (2) (3) (4) (5) (6) (7) Last 5 days 0.121 (0.0694) 0.173 (0.00320) 0.012 (0.000224) 0.123 (0.00863) 0.038 (0.00800) 0.085 (0.00314) 0.016∗∗∗ (0.00207) First 5 days 0.000 (0.0000618) 0.005 494241 -0.084∗∗∗ (0.00186) 0.026 10131984 -0.006∗∗∗ (0.000130) 0.025 10131984 -0.042∗∗∗ (0.0106) 0.490 183744 -0.026∗∗ (0.00989) 0.554 183744 -0.016∗∗∗ (0.00329) 0.117 183744 -0.005 (0.00243) 0.313 158762 A Yes Yes Yes B Yes Yes Yes B Yes Yes Yes C Yes Yes Yes C Yes Yes Yes C Yes Yes Yes C Yes Yes Yes R2 Observations Sample City FEs Year-month FEs Day-of-week FEs ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Notes. Column (1) uses branch level data (the sample from Table 1(A)), and the coefficients and standard errors have been multiplied by 1000 for ease of exposition; columns (2) – (3) use the balanced panel of loans aggregated to the branch level, in which days without any recorded loans are coded as zero (the sample from Table 1(B)); columns (4) – (7) use the sub-sample of the balanced panel that includes only branch-day observations when at least one loan was issued (the sample from Table 1(C)). The reference groups are all other days not in the first or last 5 days. OLS estimates, robust standard errors in parentheses are clustered at the branch level. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. A Appendix: Modeling Delayed Lending The model in Section 2 assumes that a branch manager cannot defer a loan application until later, effectively removing the option value of holding a potential loan in reserve in case later loan applications are of low quality. In this appendix, we present a model which highlights that, as long as there is some probability that a deferred loan is no longer available at the point at which the quality of later loan applicants is realized, we still obtain the model’s main intuition that loan quantity increases over the month and loan quality decreases. We again assume a two-period model where the branch manager receives a single loan application of size 1 in each period. We depart from the baseline model (in which lending decisions have to be made at the time of loan application), we allow the branch manager (if he wishes) to defer a decision on the first-period application until the next period. If the manager choose to postpone the decision, she risks losing the application with probability β. This may reflect, for example, the risk that a borrower switches to another lender, or the termination of the project due to the failure to obtain funds in time. However, with probability (1 − β), the manager will have two applicants in hand to choose from in the second period. The decision rule for the second period depends on the first-period lending decision, and also whether the first-period loan is still available. As in our baseline model, if a loan has already been made in the first period, the manager makes a second loan if and only if p > R1 . If no loan was made during the first period, the manager offers a loan to the higher quality applicant if the first-period applicant is still available, and to any second-period applicant with p > 0 if the first-period loan application is unavailable. In the first period, the branch manager offers a loan if and only if πloan > πnoloan : Z 1 Z 1 Z 1 (Rq − 1)dq > β · (Rq − 1)dq + (1 − β)[(Rp − 1)p + (Rq − 1)dq] Rp − 1 + 1 R 0 p This simplifies to (1 − β)(Rp)2 − 2R(RP ) + 2R − 1 6 0 . We first notice that the benchmark model we presented in Section 2 is a special case of 1 this expression, in which β = 1, leading to the same cutoff of p∗ = R1 (1 − 2R ) as before. With the option of delay in lending, the manager will set a higher cutoff p∗∗ in the first period. To see this, we notice from the first-period decision rule that 0 6 (1 − β)(Rp∗∗ )2 = 2R(Rp∗∗ ) − 2R + 1 which simplifies to 1 1 (1 − ) = p∗ R 2R Now we can compare the probability of no lending in the first period (p∗∗ ) and in the second period ((1 − p∗∗ ) R1 ): p∗∗ > (1 − p∗∗ ) 1 R < p∗ 6 p∗∗ . Again, the probability of a loan (as well as the probability of default) is strictly higher in expectation in the second period. B Appendix Figures and Tables -.1 0 Coefficients .1 .2 .3 Figure B1: Value of New Lending Issued with Branch Fixed Effects -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) Note: dates more than +/-13 days away from the month end are the omitted category; coefficient plots are conditional on year-month, day-of-week, and branch fixed effects; standard errors are clustered at the branch level. 12 log(number of loan value) 14 16 18 Figure B2: Value of New Lending Issued (Branch-level data) -15 -10 -5 0 5 Day (relative to the month end) 10 15 Note: dates more than +/-13 days away from the month end are excluded from the raw data graph. 17 20 Number of branches that lend 40 60 log(Value of new loans per day) 17.5 18 18.5 19 80 Figure B3: Extensive and Intensive Margins of New Loans Issued, Raw plots -15 -10 -5 0 5 Day (relative to the month end) 10 15 -10 -5 0 5 Day (relative to the month end) 10 15 (b) Intensive: log(N ewLoan|N ewLoan > 0) 14.4 2.8 log(Number of new loan contracts per day) 3 3.2 3.4 3.6 3.8 log(Number of new loan contracts per day) 14.6 14.8 15 15.2 15.4 (a) Extensive: P rob(anylending) -15 -15 -10 -5 0 5 Day (relative to the month end) 10 (c) Intensive: log(AverageSize) 15 -15 -10 -5 0 5 Day (relative to the month end) 10 15 (d) Intensive: Log(Contracts) Note: dates more than +/-13 days away from the month end are excluded from the raw data graph. Coefficients (Share of Bad Loan (Contract number)) -.02 -.01 0 .01 .02 .03 Figure B4: Alternative Presentations of Loan Quality Response -13-12-11-10 -9 -8 -7 -6 -5 -4 -3 -2 -1 0 1 2 3 4 5 6 7 8 9 10 11 12 13 Day (relative to the month end) .13 Share of Bad Loan (Loan value) .14 .15 .16 .17 (a) Share of Bad Loans (unweighted), Coefficients Plots -15 -10 -5 0 5 Day (relative to the month end) 10 15 (b) Share of Bad Loans (weighted by loan amount),Raw Plots Note: dates more than +/-13 days away from the month end are excluded from the raw data graph; days outside of the +/-13 day window are the omitted category in the coefficient plots, which are conditional on year-month, day-of-week, and city fixed effects. Standard errors are clustered at the branch level. Table B1: Loan Quantity and Quality Responses Relative to the Middle 5 Days of the Month Dependent Variables: Quantity Response Quality Response ∆log(Loans) log(1 + N ewLoans) AnyLoan log(N ewLoans) log(Loans/Contracts) log(Contracts) Share of Bad Loans (1) (2) (3) (4) (5) (6) (7) Last 5 days 0.179 (0.0785) 0.188 (0.00355) 0.013 (0.000247) 0.170 (0.0111) 0.075 (0.0105) 0.095 (0.00373) 0.014∗∗∗ (0.00276) First 5 days 0.000∗ (0.0000751) 0.006 243175 -0.068∗∗∗ (0.00216) 0.027 4992658 -0.005∗∗∗ (0.000151) 0.027 4992658 0.005 (0.0129) 0.488 95236 0.008 (0.0121) 0.557 95236 -0.003 (0.00412) 0.144 95236 -0.007∗ (0.00307) 0.315 82087 A Yes Yes Yes B Yes Yes Yes B Yes Yes Yes C Yes Yes Yes C Yes Yes Yes C Yes Yes Yes C Yes Yes Yes R2 Observations Sample City FEs Year-month FEs Day-of-week FEs ∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Notes. Column (1) uses city level data (the sample from Table 1(A)), and the coefficients and standard errors have been multiplied by 1000 for ease of exposition; columns (2) – (3) use the balanced panel of loans aggregated to the branch level, in which days without any recorded loans are coded as zero (the sample from Table 1(B)); columns (4) – (7) use the sub-sample of the balanced panel that includes only branch-day observations when at least one loan was issued (the sample from Table 1(C)). The reference groups are days (the 13th – 17th ) in the middle of the month. OLS estimates, robust standard errors in parentheses are clustered at the branch level. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Table B2: Loan Quantity and Quality Responses Controlling for Holidays Dependent Variables: Quantity Response Quality Response ∆log(Loans) log(1 + N ewLoans) AnyLoan log(N ewLoans) log(Loans/Contracts) log(Contracts) Share of Bad Loans (1) (2) (3) (4) (5) (6) (7) Last 5 days 0.188 (0.0782) 0.189 (0.00355) 0.013 (0.000247) 0.168 (0.0111) 0.074 (0.0105) 0.094 (0.00374) 0.014∗∗∗ (0.00277) First 5 days 0.000∗∗ (0.0000781) 0.006 243175 -0.055∗∗∗ (0.00207) 0.028 4992658 -0.004∗∗∗ (0.000144) 0.027 4992658 -0.001 (0.0129) 0.488 95236 0.004 (0.0121) 0.557 95236 -0.005 (0.00415) 0.144 95236 -0.007∗ (0.00308) 0.315 82087 A Yes Yes Yes Yes B Yes Yes Yes Yes B Yes Yes Yes Yes C Yes Yes Yes Yes C Yes Yes Yes Yes C Yes Yes Yes Yes C Yes Yes Yes Yes R2 Observations Sample Holiday Indicator City FEs Year-month FEs Day-of-week FEs ∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Notes. Column (1) uses branch level data (the sample from Table 1(A)), and the coefficients and standard errors have been multiplied by 1000 for ease of exposition; columns (2) – (3) use the balanced panel of loans aggregated to the branch level, in which days without any recorded loans are coded as zero (the sample from Table 1(B)); columns (4) – (7) use the sub-sample of the balanced panel that includes only branch-day observations when at least one loan was issued (the sample from Table 1(C)). The reference groups are days (the 13th – 17th ) in the middle of the month. OLS estimates, robust standard errors in parentheses are clustered at the branch level. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01. Table B3: Loan Quantity and Quality Responses with Year-month-city FEs Dependent Variables: Quantity Response Quality Response log(1 + N ewLoans) AnyLoan log(N ewLoans) log(Loans/Contracts) log(Contracts) Share of Bad Loans (1) (2) (3) (4) (5) (6) Last 5 days 0.188 (0.00311) 0.013 (0.000213) 0.141 (0.0149) 0.053 (0.0136) 0.087 (0.00539) 0.009∗ (0.00367) First 5 days -0.068∗∗∗ (0.00222) 0.060 4992658 -0.005∗∗∗ (0.000154) 0.059 4992658 0.023 (0.0176) 0.673 94290 0.023 (0.0163) 0.717 94290 -0.000 (0.00580) 0.409 94290 -0.007 (0.00426) 0.551 81334 B Yes Yes B Yes Yes C Yes Yes C Yes Yes C Yes Yes C Yes Yes R2 Observations Sample Y ear × M onth × City(branch) FEs Day-of-week FEs ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ ∗∗∗ Notes. Columns (1) – (2) use the balanced panel of loans aggregated to the branch level, in which days without any recorded loans are coded as zero (the sample from Table 1(B)); columns (3) – (6) use the sub-sample of the balanced panel that includes only branch-day observations when at least one loan was issued (the sample from Table 1(C)). The reference groups are days (the 13th – 17th ) in the middle of the month. OLS estimates, robust standard errors in parentheses are clustered at the branch level. ∗ p < 0.10, ∗∗ p < 0.05, ∗∗∗ p < 0.01.