RDP 2021-01: The Role of Collateral in Borrowing Appendix B: Additional Regressions and Robustness Tests

Nicholas Garvin, David W Hughes and José-Luis Peydró

January 2021

Table B1: Borrowers' Repo Rate Reactions to Stress by Collateral Type
Coefficient estimates, standard errors in parentheses
Market: AGS   Semis   AGS   Semis
(a) (b) (c) (d)
LoanValueslbd 0.138**
(0.053)		   0.020***
(0.003)		   0.142**
(0.054)		   0.020***
(0.003)
LBd *
CPRiskb −0.015
(0.089)		   0.008
(0.006)		   0.020
(0.074)		   0.030
(0.026)
AGSb −0.130
(0.076)		       −0.210*
(0.102)		   −0.075
(0.101)
CPRiskb * AGSb −0.197*
(0.098)		       −0.343*
(0.181)		   −0.081
(0.112)
Semisb     0.007
(0.005)		   0.075
(0.109)		   0.090
(0.109)
CPRiskb * Semisb     0.007
(0.005)		   0.203
(0.193)		   0.092
(0.113)
Fixed effects     Borrower and Lender * Day    
Observations 744   496   615   342
R-squared 0.078   0.043   0.482   0.725
Notes: *, ** and *** denote significance at the 10, 5 and 1 per cent levels, respectively. This table reports estimates from lender-borrower-day level regressions of repo interest rates, separating the repo market by collateral type (using specification (1)). On days with multiple repos in the same lbd unit, the dependent variable is a loan-size weighted average of the rates. Observations for which there are no loans outstanding are excluded. Columns (a) and (c) examine repos collateralised by AGS and columns (b) and (d) examine repos collateralised by semis. The pre-interacted explanatory variables are borrowers' counterparty risk ( CPRiskb ), borrowers' pre-sample AGS holdings ( AGSb ), borrowers' pre-sample semis holdings ( Semisb ) and an indicator variable equal to one after the Lehman Brothers default ( LBd ). CPRiskb , AGSb and Semisb are standardised to zero mean and unit variance. Loans outstanding, in log millions plus one ( LoanValueslbd ) is included in each regression as a control (as suggested by Garvin (2018)). The full variable definitions are in Section 2.4. Standard errors are clustered at the borrower and lender and day levels.
Table B2: 2006 Placebo Tests
Coefficient estimates, standard errors in parentheses
Analysis: Borrower characteristics   Lender characteristics   Borrower substitution
Market: Unsecured   Repo Unsecured   Repo Full sample
Regressand:
 
 		 LoanValueslbd Participationlbd LoanValueslbd Participationlbd LoanValueslbd Participationlbd LoanValueslbd Participationlbd LoanValueslbdm Participationlbdm
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j)
LBd * (1s * explanatory variables)
CPRiski −0.006
(0.105)		 −0.010
(0.020)		   0.105
(0.339)		 0.073
(0.090)		   0.033
(0.074)		 0.013
(0.020)		   0.136
(0.159)		 0.043
(0.042)		   −0.020
(0.175)		 0.051
(0.030)
Collaterali 0.158
(0.321)		 0.015
(0.068)		   −0.122
(0.609)		 0.100
(0.147)		   0.565
(0.365)		 0.085
(0.080)		   0.921
(0.554)		 0.232
(0.152)		   0.045
(0.067)		 0.027*
(0.016)
CPRiski * Collaterali 0.126
(0.301)		 0.006
(0.061)		   0.026
(0.557)		 0.115
(0.139)		   0.830
(0.498)		 0.140
(0.110)		   0.967
(0.682)		 0.211
(0.163)		   0.038
(0.181)		 0.053
(0.043)
Lower-level interactions     .                   yes
Fixed effects Borrower and Lender * Day   Lender and Borrower * Day   Lender * Borrower * Day
Observations 5,073 5,073   2,413 2,413   4,560 4,560   2,565 2,565   16,454 16,454
R-squared 0.208 0.188   0.268 0.241   0.199 0.190   0.241 0.220   0.524 0.510
Notes: *, ** and *** denote significance at the 10, 5 and 1 per cent levels, respectively. This table reports estimates from regressions using the 2006 sample from Monday 11 September to Friday 6 October. The placebo market-stress explanatory variable ( LBd ) is equal to 0 in the first week and 1 afterwards. Columns (a) to (h) report estimates from lender-borrower-day level regressions on sub-samples containing either the unsecured or repo interbank market (i.e. specification (1)), and columns (i) and (j) report lender-borrower-day-market level regressions on the full sample containing both markets (i.e. specification (2)). We regress loans outstanding, in log millions plus one ( LoanValueslbd or LoanValueslbdm ), and an indicator variable for whether LoanValues is positive ( Participationlbd or Participationlbdm ), on interactions between borrower or lender characteristics and market stress. The other pre-interacted explanatory variables are banks' counterparty risk ( CPRiski ), banks' pre-sample high-quality collateral holdings ( Collaterali ) and, in columns (i) and (j), the repo market indicator ( 1s ). CPRiski and Collaterali are standardised to zero mean and unit variance. The full variable definitions are in Section 2.4. Standard errors are clustered at the borrower and lender and day levels.
Table B3: NPL Regressed on Bank Characteristics
Coefficient estimates, standard errors in parentheses
Constant 6.553
(6.444)
Sizeb 0.132
(0.194)
Leverageb −6.585
(6.545)
Foreignb 0.867*
(0.467)
FirstWeekBorrowingb 0.035
(0.070)
Observations 30
R-squared 0.367
Notes: *, ** and *** denote significance at the 10, 5 and 1 per cent levels, respectively. This table reports estimates from regressing borrower NPL (i.e. CPRiskb ) on other borrower characteristics. Sizeb is the log of balance sheet asset value in AUD trillions plus one; Leverageb is balance sheet liabilities over assets; Foreignb is a dummy for whether the bank's ultimate parent is Australian; and FirstWeekBorrowingb is the sum of the bank's interbank borrowing transactions in the week prior to our sample (1 to 5 September 2008), in AUD millions logged plus one. The variables are not standardised. Standard errors are OLS.
Table B4: Robustness to Domicile
Coefficient estimates, standard errors in parentheses
Analysis: Borrower characteristics   Lender characteristics   Borrower substitution
Market: Unsecured   Repo Unsecured   Repo Full sample
Regressand:
 
 		 LoanValueslbd Participationlbd LoanValueslbd Participationlbd LoanValueslbd Participationlbd LoanValueslbd Participationlbd LoanValueslbdm Participationlbdm
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j)
LBd * (1s * explanatory variables)
Foreigni 0.141
(0.271)		 0.004
(0.065)		   0.269
(0.249)		 0.103
(0.070)		   −0.091
(0.176)		 −0.022
(0.039)		   −0.206
(0.240)		 −0.075
(0.087)		   0.132
(0.195)		 0.057
(0.052)
Collaterali 0.325**
(0.152)		 0.041
(0.039)		   0.116
(0.116)		 0.030
(0.035)		   −0.294
(0.174)		 −0.055
(0.035)		   −0.341***
(0.088)		 −0.097*
(0.049)		   −0.087
(0.083)		 −0.003
(0.026)
Foreigni * Collaterali −0.249
(0.244)		 −0.027
(0.055)		   0.120
(0.191)		 0.035
(0.053)		   0.014
(0.218)		 0.006
(0.044)		   0.805**
(0.294)		 0.208**
(0.098)		   0.137
(0.110)		 0.028
(0.033)
Lower-level interactions     .           .       yes
Fixed effects Borrower and Lender * Day   Lender and Borrower * Day   Lender * Borrower * Day
Observations 5,800 5,800   3,020 3,020   5,940 5,940   3,100 3,100   15,640 15,640
R-squared 0.225 0.216   0.249 0.220   0.213 0.199   0.248 0.213   0.537 0.509
Notes: *, ** and *** denote significance at the 10, 5 and 1 per cent levels, respectively. This table reports estimates from specifications that replace banks' counterparty risk with banks' domicile. Columns (a) to (h) report estimates from lender-borrower-day level regressions on sub-samples containing either the unsecured or repo interbank market (i.e. specification (1)), and columns (i) and (j) report lender-borrower-day-market level regressions on the full sample containing both markets (i.e. specification (2)). We regress loans outstanding, in log millions plus one ( LoanValueslbd or LoanValueslbdm ), and an indicator variable for whether LoanValues is positive ( Participationlbd or Participationlbdm ), on interactions between borrower or lender characteristics and market stress. The pre-interacted explanatory variables are market stress ( LBd ), equal to 0 in the first week and 1 afterwards, banks' domicile ( Foreigni ), equal to 0 if the bank's parent company is Australian and 1 otherwise, banks' pre-sample high-quality collateral holdings ( Collaterali ) and, in columns (i) and (j), the repo market indicator ( 1s ). Collaterali is standardised to zero mean and unit variance. The full variable definitions are in Section 2.4. Standard errors are clustered at the borrower and lender and day levels.
Table B5: Robustness to Shock Measure
Coefficient estimates, standard errors in parentheses
Analysis: Borrower characteristics   Lender characteristics   Borrower substitution
Market: Unsecured   Repo Unsecured   Repo Full sample
Regressand:
 
 		 LoanValueslbd Participationlbd LoanValueslbd Participationlbd LoanValueslbd Participationlbd LoanValueslbd Participationlbd LoanValueslbdm Participationlbdm
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j)
TEDd * (1s * explanatory variables)
CPRiski −0.082*
(0.039)		 −0.015
(0.009)		   −0.052
(0.057)		 0.003
(0.015)		   0.026
(0.054)		 0.005
(0.011)		   0.030
(0.046)		 0.001
(0.016)		   0.040**
(0.018)		 0.015**
(0.006)
Collaterali 0.020
(0.037)		 0.003
(0.008)		   0.136**
(0.052)		 0.049***
(0.010)		   −0.051
(0.081)		 −0.010
(0.018)		   0.145
(0.095)		 0.027
(0.029)		   0.029
(0.024)		 0.016**
(0.006)
CPRiski * Collaterali −0.045
(0.049)		 −0.003
(0.012)		   0.136**
(0.061)		 0.050***
(0.009)		   0.166*
(0.086)		 0.030
(0.018)		   0.223*
(0.128)		 0.046
(0.034)		   0.050***
(0.015)		 0.016**
(0.006)
Lower-level interactions     .           .       yes
Fixed effects Borrower and Lender * Day   Lender and Borrower * Day   Lender * Borrower * Day
Observations 5,780 5,780   3,020 3,020   5,360 5,360   3,060 3,060   15,600 15,600
R-squared 0.226 0.218   0.253 0.225   0.233 0.211   0.244 0.208   0.533 0.505
Notes: *, ** and *** denote significance at the 10, 5 and 1 per cent levels, respectively. This table reports estimates from specifications that replace the binary market stress measure with the TED spread. Columns (a) to (h) report estimates from lender-borrower-day level regressions on sub-samples containing either the unsecured or repo interbank market (i.e. specification (1)), and columns (i) and (j) report lender-borrower-day-market level regressions on the full sample containing both markets (i.e. specification (2)). We regress loans outstanding, in log millions plus one ( LoanValueslbd or LoanValueslbdm ), and an indicator variable for whether LoanValues is positive ( Participationlbd or Participationlbdm ), on interactions between borrower or lender characteristics and market stress. The pre-interacted explanatory variables are the TED spread ( TEDd ), in percentage points and lagged one day, banks' counterparty risk ( CPRiski ), banks' pre-sample high-quality collateral holdings ( Collaterali ) and, in columns (i) and (j), the repo market indicator ( 1s ). TEDd , CPRiski and Collaterali are standardised to zero mean and unit variance. The full variable definitions are in Section 2.4. Standard errors are clustered at the borrower and lender and day levels.
Table B6: Robustness to Standard Error Method
Coefficient estimates, standard errors in parentheses
Analysis: Borrower characteristics   Lender characteristics   Borrower substitution
Market: Unsecured   Repo Unsecured   Repo Full sample
Regressand:
 
 		 LoanValueslbd Participationlbd LoanValueslbd Participationlbd LoanValueslbd Participationlbd LoanValueslbd Participationlbd LoanValueslbdm Participationlbdm
(a) (b) (c) (d) (e) (f) (g) (h) (i) (j)
LBd * (1s * explanatory variables)
CPRiski −0.247**
(0.101)		 −0.049**
(0.021)		   −0.019
(0.127)		 0.023
(0.033)		   −0.023
(0.100)		 −0.006
(0.021)		   0.106
(0.118)		 0.017
(0.034)		   0.166**
(0.066)		 0.047**
(0.019)
Collaterali 0.019
(0.098)		 0.001
(0.022)		   0.267**
(0.119)		 0.084***
(0.032)		   −0.189
(0.136)		 −0.040
(0.030)		   0.409**
(0.192)		 0.082
(0.055)		   0.061
(0.089)		 0.029
(0.019)
CPRiski * Collaterali −0.151
(0.131)		 −0.021
(0.028)		   0.358**
(0.156)		 0.117***
(0.038)		   0.162
(0.147)		 0.022
(0.032)		   0.613***
(0.230)		 0.136**
(0.058)		   0.151**
(0.069)		 0.043**
(0.016)
Lower-level interactions     .           .       yes
Fixed effects Borrower and Lender * Day   Lender and Borrower * Day   Lender * Borrower * Day
Observations 5,780 5,780   3,020 3,020   5,360 5,360   3,060 3,060   15,600 15,600
R-squared 0.227 0.219   0.251 0.222   0.230 0.218   0.246 0.210   0.533 0.505
Notes: *, ** and *** denote significance at the 10, 5 and 1 per cent levels, respectively. This table reports estimates from specifications that use different standard error clustering. Columns (a) to (h) cluster at the borrower * lender, borrower * day and lender * day levels, and columns (i) and (j) cluster at the borrower and lender levels. Columns (a) to (h) report estimates from lender-borrower-day level regressions on sub-samples containing either the unsecured or repo interbank market (i.e. specification (1)), and columns (i) and (j) report lender-borrower-day-market level regressions on the full sample containing both markets (i.e. specification (2)). We regress loans outstanding, in log millions plus one ( LoanValueslbd or LoanValueslbdm ), and an indicator variable for whether LoanValues is positive ( Participationlbd or Participationlbdm ), on interactions between borrower or lender characteristics and market stress. The pre-interacted explanatory variables are market stress ( LBd ), equal to 0 in the first week and 1 afterwards, banks' counterparty risk ( CPRiski ), banks' pre-sample high-quality collateral holdings ( Collaterali ) and, in columns (i) and (j), the repo market indicator ( 1s ). CPRiski and Collaterali are standardised to zero mean and unit variance. The full variable definitions are in Section 2.4.