Online Appendix to “Foreign Booms, Domestic Busts: The Global Dimension of Banking Crises” by A. Cesa-Bianchi, and F. Eguren Martin, and G. Thwaites
November 3, 2016
This Online Appendix includes additional details and results to Cesa-Bianchi et al. (2016).
1
2 1,115 34 0.20 0.042 -0.06 0.905 1.11 0.000
Observations Crises Sum of lagged coeffs. of ∆Cred Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆Cred* Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆Cred* x Open Test for sum of lags=0 (p-val) Test for CFE p-value R2 AUROC Standard error
1,115 34 0.25 0.019 -0.16 0.766 1.01 0.001 0.89 0.650 0.10 0.87 0.03
-0.01 (0.026)
-0.01 (0.047) 0.00 (0.053) 0.16** (0.070) 0.08 (0.056) 0.02 (0.107) -0.09 (0.279) 0.34 (0.247) -0.43 (0.267) 0.18 (0.198) -0.16 (0.227) 0.42*** (0.159) 0.41** (0.174) -0.03 (0.141) 0.02 (0.100) 0.19* (0.105) -0.05*** (0.016)
0.27 0.88 0.03
1,115 34 7.55 0.016 -1.61 0.920 59.23 0.004
-3.12*** (1.006)
-3.16 (2.395) -0.10 (2.335) 7.54*** (2.837) 2.39 (2.768) 0.87 (3.693) -5.61 (8.590) 16.25 (10.774) -21.07** (10.666) 15.29 (9.892) -6.47 (10.540) 29.33** (13.665) 30.86*** (11.787) -4.83 (14.041) -3.16 (11.408) 7.03 (15.467) -5.35*** (1.554)
(2) (3) OLS Logit Country None Financial
941 34 9.11 0.008 -52.20 0.010 73.36 0.000 14560 0.000 0.40 0.91 0.02
-17.08*** (1.122)
-1.10 (2.360) 0.03 (1.868) 6.25** (3.150) 3.03 (1.923) 0.90 (1.871) -29.25*** (10.423) 3.79 (13.103) -33.05*** (12.172) 21.24* (12.113) -14.93 (11.688) 36.94*** (12.400) 43.69** (18.094) -4.28 (16.243) -11.64 (12.205) 8.64 (14.410) -4.85*** (1.354)
(4) Logit Country
0.05 0.82 0.04
1,113 34 0.23 0.024 1.14 0.065 0.40 0.299
-0.03 (0.030)
-0.02 (0.047) 0.00 (0.047) 0.16** (0.069) 0.08 (0.059) 0.01 (0.113) 0.49 (0.329) 0.76*** (0.263) -0.51* (0.295) 0.34 (0.242) 0.05 (0.271) 0.10 (0.194) 0.16 (0.186) 0.21 (0.181) -0.14 (0.153) 0.07 (0.167) -0.03 (0.020)
(5) OLS None
1,110 34 0.28 0.013 1.04 0.095 0.44 0.303 0.90 0.638 0.06 0.84 0.04
-0.07** (0.034)
-0.00 (0.046) 0.01 (0.053) 0.17** (0.072) 0.08 (0.057) 0.02 (0.108) 0.49 (0.325) 0.73*** (0.272) -0.54* (0.299) 0.33 (0.243) 0.03 (0.277) 0.08 (0.200) 0.18 (0.193) 0.24 (0.191) -0.14 (0.156) 0.08 (0.180) -0.03 (0.036)
(6) OLS Country
0.22 0.86 0.04
1,113 34 7.92 0.007 16.60 0.462 53.29 0.267
-3.87** (1.799)
-2.08 (2.062) -0.16 (1.775) 6.88*** (2.524) 2.47 (2.231) 0.82 (2.723) 17.85 (12.410) 1.61 (15.674) -24.09* (13.078) 12.35 (12.280) 8.88 (15.321) -2.21 (13.083) 78.41** (36.644) -3.17 (24.754) 2.92 (17.906) -22.66 (26.864) -6.03 (4.136)
Trade
(7) Logit None
936 34 12.40 0.008 -8.45 0.818 90.24 0.268 996.9 0.000 0.29 0.87 0.04
-17.04*** (3.361)
-0.81 (2.771) 0.88 (1.828) 7.64** (3.368) 3.36* (1.718) 1.33 (1.917) 13.67 (18.250) -26.63 (19.088) -23.28 (15.955) 20.83 (19.803) 6.96 (19.108) 0.80 (24.079) 133.87*** (50.238) -9.90 (28.262) -11.74 (31.363) -22.79 (31.873) -8.74 (7.953)
(8) Logit Country
Note. Robust standard errors between parentheses, robust-standard-error-based p-values between brackets. ∆Cred is the growth rate of real lending by domestic banks to domestic households and non-financial corporations, deflated using CPI. For each country and year, ∆Cred* is the average of ∆Cred for the N − 1 remaining countries in the sample, weighted by PPP-adjusted GDP. We consider 5 lags of both ∆Cred and ∆Cred*. FinOpen is the first lag of a proxy for financial openness, namely a country’s gross external liabilities as a share of GDP. T radeOpen is the first lag of a proxy for trade openness, namely the sum of exports and imports normalized by GDP. All specifications include country fixed effects. CFE stands for country fixed effects. AUROC stands for Area Under the Receiving Operating Characteristic curve, a measure of the binary classification ability of the model. Sample covers 38 countries over 1970-2011. The dependent variable is a dummy variable capturing banking crises.
0.08 0.85 0.03
0.02 (0.026)
-0.03 (0.049) -0.00 (0.048) 0.15** (0.068) 0.07 (0.058) 0.01 (0.113) -0.06 (0.278) 0.38 (0.244) -0.41 (0.265) 0.17 (0.196) -0.15 (0.225) 0.46*** (0.160) 0.41** (0.176) -0.01 (0.141) 0.05 (0.092) 0.21** (0.100) -0.06*** (0.016)
(1) OLS None
Constant
Open
L5.∆Cred* x Open
L4.∆Cred* x Open
L3.∆Cred* x Open
L2.∆Cred* x Open
L.∆Cred* x Open
L5.∆Cred*
L4.∆Cred*
L3.∆Cred*
L2.∆Cred*
L.∆Cred*
L5.∆Cred
L4.∆Cred
L3.∆Cred
L2.∆Cred
L.∆Cred
Specification Estimation method Fixed effects Openness
Table 1 Banking Crisis Prediction – The Role Of Financial And Trade Openness
3 795 33 11.70 0.000
-0.30 0.627 2367 0.000 0.08 0.72 0.04
0.00 0.910 0.90 0.649 0.03 0.76 0.04
-17.21*** (0.409)
-0.01 (0.150) 0.05 (0.161) -0.56 (0.566) 0.09 (0.239) 0.14 (0.210)
0.82 (1.981) 1.70 (1.741) 6.21** (2.832) 2.46 (1.569) 0.51 (1.712)
951 33 0.29 0.016 1.47 0.015 -0.02 0.388 0.87 0.688 0.06 0.84 0.04
-0.10*** (0.033)
-0.00 (0.051) -0.00 (0.059) 0.17** (0.081) 0.09 (0.062) 0.02 (0.119) 0.59* (0.305) 1.10*** (0.280) -0.40** (0.177) 0.09 (0.186) 0.09 (0.244) -0.01 (0.005) -0.00 (0.008) -0.01 (0.006) -0.00 (0.010) 0.00 (0.008)
(2) (3) Logit OLS Bank inflows to banks
951 33 0.34 0.004
-0.03*** (0.011)
-0.00 (0.006) 0.00 (0.008) -0.01 (0.005) 0.00 (0.009) 0.00 (0.008)
0.04 (0.047) 0.05 (0.057) 0.16** (0.075) 0.09 (0.058) -0.00 (0.109)
(1) OLS
795 33 11.61 0.005 46.09 0.008 -0.83 0.240 1141 0.000 0.20 0.82 0.04
-21.87*** (1.458)
-0.19 (2.465) -0.44 (2.499) 7.61** (3.659) 3.56 (3.167) 1.07 (3.528) 17.42* (9.921) 41.75*** (10.251) -25.23** (12.146) 12.07 (7.656) 0.07 (10.316) -0.20 (0.239) -0.02 (0.212) -0.69 (0.587) -0.05 (0.246) 0.13 (0.267)
(4) Logit
0.04 0.271 0.90 0.646 0.03 0.77 0.04
951 33 0.29 0.018
-0.04*** (0.013)
0.02 (0.015) 0.01 (0.013) -0.00 (0.008) 0.01 (0.009) 0.01 (0.010)
0.02 (0.049) 0.04 (0.058) 0.15** (0.076) 0.08 (0.060) -0.00 (0.112)
(5) OLS
1.19 0.395 1884 0.000 0.09 0.74 0.05
795 33 10.68 0.002
-17.40*** (0.683)
0.48 (0.388) 0.27 (0.337) -0.37 (0.461) 0.44 (0.332) 0.37 (0.334)
0.58 (2.171) 1.26 (1.894) 6.17** (2.902) 2.43 (1.747) 0.24 (1.850)
951 33 0.25 0.045 1.40 0.017 0.03 0.448 0.86 0.701 0.06 0.83 0.04
-0.10*** (0.033)
-0.01 (0.052) -0.01 (0.060) 0.17** (0.082) 0.08 (0.064) 0.02 (0.120) 0.55* (0.300) 1.08*** (0.278) -0.41** (0.176) 0.09 (0.189) 0.09 (0.240) 0.02 (0.013) 0.00 (0.013) -0.01 (0.008) 0.01 (0.009) 0.01 (0.010)
(6) (7) Logit OLS Bank inflows to non-banks
795 33 10.42 0.010 44.95 0.008 0.89 0.490 893 0.000 0.21 0.82 0.04
-22.16*** (1.614)
-0.74 (2.459) -1.18 (2.679) 7.78** (3.658) 3.85 (3.241) 0.71 (3.411) 17.49* (10.200) 41.43*** (10.086) -26.69** (11.840) 12.21* (7.309) 0.51 (9.801) 0.65* (0.365) 0.04 (0.345) -0.38 (0.519) 0.42 (0.287) 0.16 (0.263)
(8) Logit
1.12 0.002 0.82 0.776 0.07 0.83 0.03
793 29 0.25 0.070
-0.07*** (0.019)
0.25** (0.117) 0.25** (0.104) 0.11** (0.051) 0.26* (0.146) 0.25 (0.176)
-0.00 (0.055) 0.06 (0.065) 0.17** (0.078) 0.07 (0.057) -0.05 (0.116)
(9) OLS
36.43 0.000 3009 0.000 0.19 0.83 0.03
622 29 12.01 0.031
-20.87*** (0.987)
4.04 (2.907) 9.00** (4.367) 14.95*** (5.793) 3.84 (6.101) 4.60 (5.210)
0.37 (3.242) 2.50 (2.659) 7.65** (3.737) 1.74 (1.272) -0.26 (1.336)
793 29 0.20 0.172 1.49 0.031 0.94 0.004 0.80 0.799 0.10 0.87 0.03
-0.14*** (0.040)
-0.04 (0.058) 0.00 (0.068) 0.19** (0.087) 0.08 (0.061) -0.03 (0.126) 0.57* (0.319) 1.30*** (0.326) -0.52** (0.217) 0.09 (0.213) 0.05 (0.252) 0.23** (0.113) 0.17* (0.093) 0.08* (0.047) 0.30** (0.144) 0.16 (0.166)
(10) (11) Logit OLS Portfolio inflows
622 29 12.22 0.056 45.96 0.037 28.39 0.003 2868 0.000 0.32 0.89 0.04
-24.82*** (2.181)
-2.95 (3.459) -0.48 (3.510) 11.49** (4.921) 3.88 (3.127) 0.27 (1.977) 23.02* (13.544) 50.43*** (14.504) -35.30** (17.253) 9.01 (10.301) -1.21 (14.692) 3.56 (2.405) 4.09 (3.835) 14.16** (6.343) 7.37 (6.741) -0.79 (6.129)
(12) Logit
Note. Robust standard errors between parentheses, robust-standard-error-based p-values between brackets. ∆Cred is the growth rate of real lending by domestic banks to domestic households and non-financial corporations, deflated using CPI. For each country and year, ∆Cred* is the average of ∆Cred for the N − 1 remaining countries in the sample, weighted by PPP-adjusted GDP. XB-Cred are three different type of inflows, depending on the specification (see heading). We consider 5 lags of ∆Cred, ∆Cred* and XB-Cred. All specifications include country fixed effects. CFE stands for country fixed effects. AUROC stands for Area Under the Receiving Operating Characteristic curve, a measure of the binary classification ability of the model. Sample covers 38 countries over 1978-2011 (1980-2011 for portfolio flows). The dependent variable is a dummy variable capturing banking crises.
Observations Crises Sum of lagged coeffs. of ∆Cred Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆Cred* Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆XB-Cred Test for sum of lags=0 (p-val) Test for CFE p-value R2 AUROC Standard error
Constant
L5.∆XB-Cred
L4.∆XB-Cred
L3.∆XB-Cred
L2.∆XB-Cred
L.∆XB-Cred
L5.∆Cred*
L4.∆Cred*
L3.∆Cred*
L2.∆Cred*
L.∆Cred*
L5.∆Cred
L4.∆Cred
L3.∆Cred
L2.∆Cred
L.∆Cred
Specification Estimation method
Table 2 Banking Crisis Prediction – Cross-border Capital Inflows
4 1,118 34 0.27 0.014 1.42 0.006 0.91 0.625 0.06 0.84 0.04
-0.09*** (0.027)
-0.00 (0.045) 0.01 (0.053) 0.17** (0.072) 0.08 (0.056) 0.02 (0.108) 0.56** (0.243) 0.87*** (0.214) -0.34* (0.180) 0.24 (0.144) 0.10 (0.178)
(1) OLS
941 34 10.66 0.008 49.71 0.001 1405 0.000 0.21 0.83 0.04
-21.11*** (1.299)
-0.85 (2.208) -0.40 (2.334) 7.71** (3.559) 3.09 (2.891) 1.11 (3.469) 19.43** (9.081) 43.01*** (10.547) -28.54*** (9.084) 14.68** (6.669) 1.13 (6.490)
(2) Logit
1,118 34 0.25 0.022 2.26 0.000 0.86 0.711 0.08 0.88 0.03
0.05** (0.023)
-0.02 (0.048) -0.01 (0.055) 0.17** (0.073) 0.08 (0.058) 0.04 (0.110) 0.93*** (0.289) 0.77*** (0.203) -0.12 (0.189) 0.22 (0.145) 0.46** (0.196) -0.08*** (0.015)
(3) OLS
941 34 11.70 0.006 75.69 0.000 1118 0.000 0.32 0.89 0.03
-13.69*** (1.409)
-1.09 (2.237) -1.00 (2.587) 8.54** (4.084) 3.97 (3.363) 1.29 (3.456) 25.16*** (9.590) 44.78*** (10.369) -28.03** (11.709) 17.58* (10.648) 16.20 (10.218) -4.62*** (0.852)
(4) Logit
1,118 34 0.27 0.014 1.32 0.016 0.91 0.625 0.06 0.84 0.04
-0.08** (0.032)
-0.14 (0.190)
-0.00 (0.045) 0.01 (0.053) 0.17** (0.072) 0.08 (0.056) 0.02 (0.108) 0.52** (0.243) 0.86*** (0.217) -0.36** (0.182) 0.23 (0.146) 0.07 (0.187)
(5) OLS
941 34 10.88 0.012 42.43 0.008 1464 0.000 0.22 0.83 0.04
-21.09*** (1.390)
-14.50 (9.270)
-0.78 (2.355) -0.32 (2.376) 7.95** (3.724) 2.93 (2.835) 1.10 (3.260) 14.77* (8.824) 45.95*** (11.095) -32.58*** (9.188) 15.45** (6.691) -1.16 (6.889)
(6) Logit
761 31 0.29 0.043 1.65 0.028 0.80 0.795 0.08 0.84 0.04
-0.12*** (0.041)
0.01 (0.040)
-0.01 (0.055) -0.02 (0.073) 0.21** (0.097) 0.10 (0.073) 0.01 (0.138) 0.75** (0.313) 1.40*** (0.362) -0.59** (0.262) 0.07 (0.274) 0.01 (0.329)
(7) OLS
630 31 12.01 0.028 34.60 0.072 967 0.000 0.22 0.84 0.04
-21.73*** (1.489)
0.59 (0.924)
-0.60 (2.286) -0.95 (2.781) 9.57** (3.988) 3.45 (3.317) 0.55 (2.659) 19.11** (9.700) 50.81*** (14.551) -31.19** (14.405) 4.76 (12.663) -8.90 (14.410)
(8) Logit
1,118 34 0.26 0.019 1.53 0.007 0.90 0.648 0.06 0.83 0.04
-0.09*** (0.025)
-1.03 (0.695)
-0.00 (0.045) 0.00 (0.054) 0.16** (0.073) 0.08 (0.057) 0.02 (0.110) 0.66** (0.279) 0.89*** (0.221) -0.33* (0.182) 0.21 (0.141) 0.10 (0.180)
(9) OLS
941 34 10.64 0.008 48.39 0.001 1330 0.000 0.21 0.83 0.04
-20.77*** (1.298)
-19.29 (38.025)
-0.77 (2.157) -0.56 (2.465) 7.68** (3.583) 3.23 (3.055) 1.06 (3.718) 20.19** (9.028) 42.43*** (10.066) -27.96*** (8.884) 13.43* (6.926) 0.31 (6.449)
(10) Logit
925 32 0.20 0.082 1.17 0.046 0.81 0.791 0.09 0.91 0.03
0.07* (0.039)
-0.01*** (0.001)
-0.04 (0.050) -0.02 (0.062) 0.18** (0.082) 0.08 (0.064) 0.01 (0.115) 0.59* (0.307) 0.82*** (0.245) -0.31* (0.184) 0.04 (0.179) 0.02 (0.230)
(11) OLS
772 32 11.22 0.043 20.10 0.173 541 0.000 0.37 0.90 0.03
-11.48*** (1.771)
-0.40*** (0.092)
-2.45 (2.617) 0.07 (3.317) 10.69** (4.527) 2.58 (3.081) 0.34 (2.017) 2.13 (14.942) 46.16*** (17.388) -52.94** (23.897) 36.48** (17.021) -11.72 (15.086)
(12) Logit
733 31 0.28 0.044 3.28 0.001 0.77 0.841 0.11 0.86 0.03
0.58*** (0.151) -0.21*** (0.058)
-0.02 (0.058) -0.03 (0.075) 0.23** (0.099) 0.11 (0.076) -0.00 (0.140) 1.13*** (0.406) 1.36*** (0.353) 0.00 (0.274) 0.66** (0.318) 0.14 (0.342)
(13) OLS
606 31 11.40 0.020 47.17 0.007 659 0.000 0.29 0.86 0.03
20.58*** (5.849) -21.83*** (1.367)
-1.91 (2.690) -0.25 (2.663) 10.30** (4.255) 3.12 (2.920) 0.13 (2.156) 5.21 (13.376) 59.35*** (15.390) -7.48 (20.660) 11.49 (18.245) -21.40* (12.508)
(14) Logit
Note. Robust standard errors between parentheses, robust-standard-error-based p-values between brackets. ∆Cred is the growth rate of real lending by domestic banks to domestic households and non-financial corporations, deflated using CPI. For each country and year, ∆Cred* is the average of ∆Cred for the N − 1 remaining countries in the sample, weighted by PPP-adjusted GDP. We consider 5 lags of both ∆Cred and ∆Cred*. Spread is the spread between US Baa-rated corporate bond yields and US Treasury bills. r ST is log(1 + F F R)/log(1 + Π), where FFR is the effective Fed funds rate and Π is ex-post realised CPI inflation. TED is the spread between 3-month interbank rate in the US and 3-month US Treasury bills. Slope is log(1 + i10y )/log(1 + F F R), where i10y is the yield of a 10-y maturity Treasury bond in the US, and FFR is defined above. VIX is the CBOE Volatility Index. LEV is (the growth rate in) the leverage of main US broker-dealer banks, taken from Bruno and Shin (2015). CFE stands for country fixed effects. AUROC stands for Area Under the Receiving Operating Characteristic curve, a measure of the binary classification ability of the model. Sample covers 38 countries over 1970-2011. The dependent variable is a dummy variable capturing banking crises.
Observations Crises Sum of lagged coeffs. of ∆Cred Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆Cred* Test for sum of lags=0 (p-val) Test for CFE p-value R2 AUROC Standard error
Constant
LEV
VIX
Slope
TED
r ST
Spread
L5.∆Cred*
L4.∆Cred*
L3.∆Cred*
L2.∆Cred*
L.∆Cred*
L5.∆Cred
L4.∆Cred
L3.∆Cred
L2.∆Cred
L.∆Cred
Specification Estimation method
Table 3 Banking Crisis Prediction – The Role Of Other Global Variables
5 1,272 53 11.53 0.000 -5.18 0.213
0.06 0.69 0.04
1.00 0.445 0.03 0.73 0.04 0.02 0.69 0.04
-3.69*** (0.258)
0.00 (1.981) 7.35*** (2.391) 0.95 (2.898) 0.51 (1.300) 2.72* (1.560) 0.51 (3.826) -2.47 (4.567) 7.63** (3.474) -5.95* (3.568) -4.90* (2.784)
(3) Logit None
1,272 53 0.48 0.003 -0.20 0.193
0.01 (0.015)
-0.02 (0.107) 0.31** (0.122) 0.05 (0.118) 0.01 (0.063) 0.13 (0.081) 0.03 (0.156) -0.15 (0.203) 0.29* (0.170) -0.17 (0.115) -0.21* (0.118)
(2) OLS Country
1,272 53 0.48 0.001 -0.20 0.178
0.03*** (0.007)
-0.02 (0.106) 0.30** (0.121) 0.05 (0.118) 0.02 (0.062) 0.13 (0.080) 0.03 (0.157) -0.14 (0.201) 0.29* (0.171) -0.17 (0.115) -0.21* (0.116)
(1) OLS None
7.78 0.858 0.08 0.73 0.04
1,272 53 11.32 0.001 -5.86 0.189
-4.31*** (0.744)
-0.07 (1.971) 7.58*** (2.736) 0.79 (3.090) 0.53 (1.398) 2.49 (1.629) 0.25 (3.830) -2.86 (4.531) 8.09** (3.611) -6.45* (3.717) -4.88* (2.866)
(4) Logit Country
0.06 0.81 0.03
1,272 53 0.56 0.000 0.76 0.002 -0.11 0.766
-0.01 (0.106) 0.30** (0.118) 0.08 (0.119) 0.02 (0.063) 0.16* (0.081) 0.21 (0.227) -0.10 (0.265) 0.73*** (0.258) -0.05 (0.148) -0.02 (0.150) -0.05 (0.251) 0.14 (0.280) -0.53** (0.268) 0.24 (0.167) 0.09 (0.181) 0.01 (0.033) -0.08 (0.080) 0.13 (0.081) -0.06* (0.032) -0.10*** (0.039) -0.00 (0.009)
(5) OLS None
1,272 53 0.54 0.001 0.78 0.003 -0.15 0.750 1.07 0.385 0.06 0.83 0.03
-0.01 (0.107) 0.30** (0.119) 0.08 (0.118) 0.02 (0.064) 0.15* (0.081) 0.21 (0.226) -0.10 (0.267) 0.71*** (0.256) -0.05 (0.147) -0.01 (0.151) -0.07 (0.269) 0.12 (0.291) -0.51* (0.274) 0.23 (0.170) 0.08 (0.187) 0.01 (0.035) -0.08 (0.080) 0.13 (0.081) -0.06* (0.032) -0.10** (0.039) -0.03 (0.017)
(6) OLS Country
Note. Standard errors between parentheses (based on robust standard errors for Logit specifications only as in ST), and p-values between brackets. ∆Cred is the growth rate in lending by domestic banks to domestic households and non-financial corporations. For each country and year, ∆Cred* is the average of ∆Cred for the N − 1 remaining countries in the sample, weighted by PPP-adjusted GDP. We consider 5 lags of both ∆Cred and ∆Cred*. N on − mobile K is a dummy variable that takes the value of one between 1945 and 1971, and zero otherwise (note that WWI and WWII years are dropped as they are clear outliers among many dimensions.) All specifications include country fixed effects. CFE stands for country fixed effects. Sample covers 14 advanced countries over 1870-2008. The dependent variable is a dummy variable capturing banking crises.
Observations Crises Sum of lagged coeffs. of ∆Cred Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆Cred* Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆Cred* x Non-mobile K Test for sum of lags=0 (p-val) Test for CFE p-value R2 AUROC Standard error
Constant
L5.Non-mobile K
L4.Non-mobile K
L3.Non-mobile K
L2.Non-mobile K
L.Non-mobile K
L5.∆Cred* x Non-mobile K
L4.∆Cred* x Non-mobile K
L3.∆Cred* x Non-mobile K
L2.∆Cred* x Non-mobile K
L.∆Cred* x Non-mobile K
L5.∆Cred*
L4.∆Cred*
L3.∆Cred*
L2.∆Cred*
L.∆Cred*
L5.∆Cred
L4.∆Cred
L3.∆Cred
L2.∆Cred
L.∆Cred
Specification Estimation method Fixed effects
Table 4 Banking Crisis Prediction - Domestic And Foreign Credit In Schularick And Taylor (2012) Database
6 690 22 16.59 0.018 114.70 0.000
645.3 0.000 0.45 0.68 0.07
0.96 0.519 0.10 0.75 0.06
-28.13*** (2.966)
-8.89* (5.147) 5.06 (5.400) 13.08 (12.722) -2.66 (7.709) 10.01* (5.168) 48.29*** (16.796) 86.81*** (27.332) -55.83** (22.204) 28.35** (13.718) 7.03 (13.625)
(2) OLS Country
804 22 0.22 0.092 2.45 0.000
-0.14*** (0.033)
-0.16* (0.080) 0.04 (0.092) 0.30* (0.151) -0.05 (0.098) 0.09 (0.067) 0.98*** (0.278) 1.09*** (0.260) -0.19 (0.193) 0.31** (0.128) 0.25 (0.197)
(1) OLS None
AEs
801 22 0.21 0.095 0.97 0.131 0.83 0.007 0.83 0.698 0.14 0.77 0.06
-0.06** (0.033)
-0.17** (0.085) 0.02 (0.096) 0.30* (0.156) -0.06 (0.096) 0.12* (0.072) 0.31 (0.304) 0.60* (0.307) -0.21 (0.284) 0.28 (0.182) -0.01 (0.259) 0.38** (0.160) 0.35** (0.177) -0.08 (0.139) 0.01 (0.094) 0.17 (0.106) -0.04*** (0.016)
(3) Logit None
690 22 30.34 0.085 -49.97 0.226 145.60 0.012 656.9 0.000 0.60 0.75 0.06
-17.81*** (2.162)
-14.28*** (5.457) 6.16 (5.747) 19.29 (17.577) -1.75 (9.369) 20.92*** (6.302) -39.35** (18.295) 35.04* (21.102) -45.10* (26.363) 30.50* (17.238) -31.06 (19.183) 89.19*** (34.530) 36.90 (24.489) -4.27 (33.767) -14.35 (20.474) 38.17 (31.569) -11.51** (4.550)
(4) Logit Country
0.96 0.489 0.06 0.52 0.05
314 12 0.29 0.065 -1.36 0.119
0.12 (0.085)
0.08 (0.060) -0.01 (0.069) 0.13 (0.081) 0.11 (0.066) -0.02 (0.135) -0.46 (0.480) 0.29 (0.387) -0.81* (0.462) 0.05 (0.420) -0.44 (0.408)
(5) OLS None
5.51 0.788 0.13 0.52 0.05
251 12 8.54 0.007 -37.61 0.117
-1.70 (1.991)
2.19 (2.311) -0.15 (1.546) 3.58** (1.774) 2.72* (1.637) 0.20 (1.596) -13.41 (13.100) 9.45 (13.377) -24.59** (12.372) 5.24 (11.771) -14.30 (11.146)
(6) OLS Country
EMs
314 12 0.29 0.074 -4.20 0.023 3.41 0.254 0.97 0.480 0.10 0.73 0.05
0.21* (0.119)
0.09 (0.082) 0.01 (0.073) 0.13* (0.081) 0.10 (0.067) -0.04 (0.133) -2.01* (1.099) -0.81 (0.959) -0.96 (0.968) 0.04 (0.782) -0.46 (0.719) 2.04 (1.525) 1.65 (1.395) 0.08 (1.223) -0.14 (0.994) -0.22 (1.076) -0.09 (0.170)
(7) Logit None
251 12 12.22 0.005 -154.90 0.000 112.60 0.115 17.98 0.035 0.26 0.68 0.05
1.99 (2.612)
5.07 (3.596) -0.41 (1.815) 4.81*** (1.657) 2.80 (1.987) -0.05 (1.523) -69.01*** (23.573) -24.49 (16.844) -54.38*** (20.216) 22.59 (22.189) -29.60 (22.881) 70.11** (31.468) 39.07* (20.282) 36.20* (21.660) -28.65 (27.763) -4.18 (33.289) -2.44 (3.619)
(8) Logit Country
Note. Robust standard errors between parentheses, robust-standard-error-based p-values between brackets. ∆Cred is the growth rate of real lending by domestic banks to domestic households and non-financial corporations, deflated using CPI. For each country and year, ∆Cred* is the average of ∆Cred for the N − 1 remaining countries in the sample, weighted by PPP-adjusted GDP. We consider 5 lags of both ∆Cred and ∆Cred*. FinOpen is the first lag of a proxy for financial openness, namely a country’s gross external liabilities as a share of GDP. All specifications include country fixed effects. CFE stands for country fixed effects. AUROC stands for Area Under the Receiving Operating Characteristic curve, a measure of the binary classification ability of the model. Sample covers 38 countries (25 advanced and 13 emerging, as classified by the IMF — see the Appendix for the definitions) over 1970-2011. The dependent variable is a dummy variable capturing banking crises.
Observations Crises Sum of lagged coeffs. of ∆Cred Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆Cred* Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆Cred* x FinOpen Test for sum of lags=0 (p-val) Test for CFE p-value R2 AUROC Standard error
Constant
FinOpen
L5.∆Cred* x FinOpen
L4.∆Cred* x FinOpen
L3.∆Cred* x FinOpen
L2.∆Cred* x FinOpen
L.∆Cred* x FinOpen
L5.∆Cred*
L4.∆Cred*
L3.∆Cred*
L2.∆Cred*
L.∆Cred*
L5.∆Cred
L4.∆Cred
L3.∆Cred
L2.∆Cred
L.∆Cred
Specification Estimation method Fixed effects
Table 5 Banking Crisis Prediction - Advanced And Emerging Market Economies
7 -0.04* (0.022) 1,118 0.19 0.044 0.91 0.042
Constant
Observations Sum of lagged coeffs. of ∆Cred Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆Cred* Test for sum of lags=0 (p-val) Test for CFE p-value R2 AUROC Standard error
1,118 0.25 0.020 0.87 0.054 0.91 0.633 0.08 0.87 0.03
-0.07*** (0.025)
0.33*** (0.094)
-0.03 (0.045) 0.01 (0.052) 0.16** (0.072) 0.08 (0.058) 0.03 (0.110) 0.27 (0.221) 0.66*** (0.189) -0.28 (0.183) 0.03 (0.156) 0.19 (0.182)
(2) OLS Country
0.19 0.84 0.04
1,118 6.81 0.035 31.82 0.084
-6.50*** (1.074)
4.93** (2.136)
-2.23 (1.675) -0.94 (2.384) 7.26** (2.841) 2.24 (3.243) 0.47 (4.138) 8.76 (10.183) 36.03*** (10.085) -24.67*** (8.676) 5.91 (8.612) 5.80 (7.718)
(3) Logit None
941 10.30 0.006 30.65 0.062 1197 0.000 0.23 0.84 0.04
-20.38*** (1.242)
5.42*** (1.839)
-1.73 (2.097) -0.18 (2.434) 7.94** (3.693) 3.15 (2.795) 1.12 (3.038) 8.60 (9.940) 35.75*** (10.071) -25.10*** (9.446) 5.89 (8.181) 5.50 (7.655)
(4) Logit Country
0.09 0.86 0.03
1,115 0.18 0.053 -0.37 0.491
0.03 (0.026)
-0.04 (0.048) -0.00 (0.047) 0.14** (0.068) 0.07 (0.059) 0.02 (0.114) -0.18 (0.281) 0.23 (0.243) -0.34 (0.268) -0.00 (0.212) -0.07 (0.224) 0.39** (0.155) 0.40** (0.174) -0.02 (0.138) 0.05 (0.088) 0.21** (0.094) -0.06*** (0.015) 0.25*** (0.084)
(5) OLS None
1,115 0.24 0.024 -0.49 0.362 0.88 0.672 0.11 0.88 0.03
-0.01 (0.026)
-0.03 (0.046) 0.00 (0.053) 0.16** (0.071) 0.07 (0.057) 0.03 (0.108) -0.22 (0.281) 0.18 (0.245) -0.37 (0.269) 0.00 (0.213) -0.08 (0.227) 0.35** (0.155) 0.41** (0.171) -0.04 (0.137) 0.02 (0.095) 0.19* (0.098) -0.05*** (0.015) 0.26*** (0.084)
(6) OLS Country
0.28 0.89 0.03
1,115 7.22 0.016 -9.66 0.597
-2.89*** (1.019)
-3.47 (2.197) -0.02 (2.375) 7.54*** (2.827) 2.30 (2.679) 0.86 (3.546) -8.21 (8.923) 14.12 (11.274) -22.50** (11.180) 12.72 (10.944) -5.79 (11.438) 25.93* (15.104) 28.85** (11.385) -0.19 (15.722) -6.42 (11.925) 8.67 (16.086) -5.09*** (1.665) 2.89 (2.478)
(7) Logit None
941 9.08 0.008 -52.55 0.019 476 0.000 0.40 0.92 0.02
-17.06*** (1.522)
-1.12 (2.257) 0.03 (1.878) 6.25** (3.153) 3.02 (1.950) 0.89 (1.878) -29.21*** (10.423) 3.58 (14.204) -33.11*** (12.227) 21.06 (12.901) -14.87 (11.791) 36.32** (17.753) 43.68** (18.090) -4.05 (17.356) -11.71 (12.400) 8.55 (15.139) -4.81*** (1.644) 0.25 (3.425)
(8) Logit Country
Note. Robust standard errors between parentheses, robust-standard-error-based p-values between brackets. ∆Cred is the growth rate of real lending by domestic banks to domestic households and non-financial corporations, deflated using CPI. For each country and year, ∆Cred* is the average of ∆Cred for the N − 1 remaining countries in the sample, weighted by PPP-adjusted GDP. We consider 5 lags of both ∆Cred and ∆Cred*. FinOpen is the first lag of a proxy for financial openness, namely a country’s gross external liabilities as a share of GDP. Crisis∗ is the weighted average of banking crises taking place at time t in the N − 1 remaining countries in the sample, weighted by PPP-adjusted GDP. All specifications include country fixed effects. CFE stands for country fixed effects. AUROC stands for Area Under the Receiving Operating Characteristic curve, a measure of the binary classification ability of the model. Sample covers 38 countries over 1970-2011. The dependent variable is a dummy variable capturing banking crises.
0.06 0.83 0.04
0.33*** (0.094)
-0.04 (0.047) -0.00 (0.047) 0.15** (0.069) 0.07 (0.060) 0.02 (0.115) 0.28 (0.224) 0.67*** (0.186) -0.27 (0.182) 0.03 (0.157) 0.20 (0.181)
(1) OLS None
Crisis*
FinOpen
L5.∆Cred* x FinOpen
L4.∆Cred* x FinOpen
L3.∆Cred* x FinOpen
L2.∆Cred* x FinOpen
L.∆Cred* x FinOpen
L5.∆Cred*
L4.∆Cred*
L3.∆Cred*
L2.∆Cred*
L.∆Cred*
L5.∆Cred
L4.∆Cred
L3.∆Cred
L2.∆Cred
L.∆Cred
Specification Estimation method Fixed effects
Table 6 Banking crisis prediction – Controlling for foreign crises
8 0.05 0.79 0.04
1,118 0.18 0.067 1.60 0.002
-0.07*** (0.027)
-0.03 (0.048) -0.00 (0.049) 0.13* (0.071) 0.06 (0.060) 0.02 (0.118) 0.73*** (0.245) 0.54*** (0.204) 0.10 (0.180) 0.25 (0.184) -0.01 (0.205)
(1) OLS None
1,118 0.23 0.036 1.58 0.002 0.87 0.688 0.06 0.84 0.04
-0.11*** (0.030)
-0.02 (0.048) 0.00 (0.056) 0.14* (0.074) 0.07 (0.057) 0.03 (0.113) 0.73*** (0.243) 0.53** (0.207) 0.09 (0.179) 0.24 (0.185) -0.01 (0.203)
0.17 0.81 0.04
1,118 6.35 0.063 50.49 0.003
-7.56*** (1.199)
-2.24 (1.968) -0.83 (2.648) 6.62** (3.092) 2.46 (3.428) 0.33 (4.950) 27.39*** (9.280) 17.81*** (6.717) -8.90 (7.716) 15.70 (10.362) -1.51 (8.677)
(2) (3) OLS Logit Country None Equal
941 9.55 0.011 50.53 0.002 1161 0.000 0.21 0.83 0.04
-21.38*** (1.439)
-1.53 (2.204) 0.13 (2.668) 6.97* (3.703) 2.90 (3.012) 1.08 (4.152) 27.78*** (9.744) 17.18** (6.812) -8.31 (7.905) 16.00 (10.116) -2.12 (8.296)
(4) Logit Country
0.03 0.80 0.04
913 0.23 0.080 0.55 0.094
-0.01 (0.012)
-0.07 (0.057) 0.03 (0.070) 0.17** (0.081) -0.00 (0.050) 0.09 (0.066) 0.42** (0.187) 0.18 (0.168) -0.02 (0.217) 0.21 (0.210) -0.25 (0.172)
(5) OLS None
913 0.33 0.027 0.68 0.042 0.89 0.634 0.04 0.84 0.04
-0.06*** (0.017)
-0.04 (0.056) 0.04 (0.073) 0.20** (0.088) 0.01 (0.050) 0.11 (0.074) 0.48** (0.197) 0.17 (0.170) 0.00 (0.214) 0.21 (0.209) -0.19 (0.186)
0.11 0.80 0.05
913 7.75 0.033 18.03 0.039
-5.27*** (0.527)
-2.63* (1.582) 0.46 (2.727) 7.83*** (2.891) -1.25 (2.410) 3.34 (2.419) 15.24** (6.641) 7.66 (6.760) -0.09 (11.575) 9.77 (8.927) -14.55* (8.468)
(6) (7) OLS Logit Country None Banking exposures
799 13.74 0.002 28.41 0.007 880 0.000 0.16 0.83 0.04
-21.72*** (1.313)
-1.95 (2.133) 0.88 (3.388) 11.01** (4.341) -1.98 (3.201) 5.79** (2.921) 19.01*** (6.604) 7.13 (6.335) 2.32 (10.951) 11.13 (7.717) -11.18 (9.461)
(8) Logit Country
0.02 0.77 0.04
1,065 0.21 0.048 0.48 0.117
-0.00 (0.012)
-0.02 (0.046) 0.01 (0.048) 0.12* (0.069) 0.09 (0.058) 0.00 (0.112) 0.28* (0.168) 0.35* (0.195) 0.13 (0.222) -0.13 (0.210) -0.15 (0.225)
(9) OLS None
1,065 0.30 0.012 0.65 0.070 0.90 0.635 0.04 0.83 0.03
-0.06*** (0.019)
0.00 (0.044) 0.02 (0.053) 0.14* (0.073) 0.10* (0.056) 0.03 (0.108) 0.38** (0.181) 0.35* (0.201) 0.14 (0.228) -0.13 (0.210) -0.08 (0.239)
0.10 0.77 0.04
1,065 6.85 0.029 16.69 0.053
-5.05*** (0.478)
-1.08 (1.588) -0.39 (1.891) 5.52** (2.582) 2.84 (2.679) -0.05 (3.014) 9.03 (6.308) 15.01* (8.288) 7.09 (11.930) -7.61 (10.043) -6.83 (8.270)
(10) (11) OLS Logit Country None Exports
915 10.60 0.008 22.51 0.020 37620 0.000 0.14 0.81 0.04
-20.62*** (0.584)
-0.46 (2.030) 0.20 (2.202) 6.53* (3.560) 3.44 (2.770) 0.89 (3.040) 13.19* (7.985) 13.09 (9.750) 8.06 (12.693) -7.21 (11.716) -4.62 (10.419)
(12) Logit Country
Note. Robust standard errors between parentheses, robust-standard-error-based p-values between brackets. ∆Cred is the growth rate of real lending by domestic banks to domestic households and non-financial corporations, deflated using CPI. For each country and year, ∆Cred* is the average of ∆Cred for the N − 1 remaining countries in the sample, alternatively weighted by equal weights, weights based on a country’s banking sector bilateral exposures to other countries in the sample, and on a country’s exports to other countries in the sample. We consider 5 lags of both ∆Cred and ∆Cred*. All specifications include country fixed effects. CFE stands for country fixed effects. AUROC stands for Area Under the Receiving Operating Characteristic curve, a measure of the binary classification ability of the model. Sample covers 38 countries over 1970-2011. The dependent variable is a dummy variable capturing banking crises.
Observations Sum of lagged coeffs. of ∆Cred Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆Cred* Test for sum of lags=0 (p-val) Test for CFE p-value R2 AUROC Standard error
Constant
L5.∆Cred*
L4.∆Cred*
L3.∆Cred*
L2.∆Cred*
L.∆Cred*
L5.∆Cred
L4.∆Cred
L3.∆Cred
L2.∆Cred
L.∆Cred
Specification Estimation method Fixed effects Weighting method
Table 7 Banking crisis prediction – Alternative weighting schemes
9 1,194 4.35 0.092 36.10 0.001
0.13 0.77 0.04
0.95 0.563 0.05 0.82 0.04
0.03 0.76 0.05
-6.16*** (0.765)
-2.19 (1.481) 3.69 (3.385) 2.86 (4.222) 22.31*** (7.654) 26.50*** (7.382) -12.71* (6.789)
(3) Logit None
1,194 0.14 0.122 1.07 0.005
-0.06*** (0.020)
-0.04 (0.063) 0.10 (0.082) 0.09 (0.089) 0.59*** (0.208) 0.61*** (0.181) -0.13 (0.168)
(2) OLS Country
1,194 0.12 0.170 1.13 0.003
-0.03** (0.017)
-0.05 (0.063) 0.10 (0.086) 0.07 (0.092) 0.62*** (0.210) 0.61*** (0.183) -0.10 (0.167)
(1) OLS None
2406 0.000 0.17 0.81 0.04
1,006 4.75 0.151 34.99 0.001
-20.08*** (0.856)
-1.40 (1.437) 3.02 (2.559) 3.13 (3.606) 21.26*** (7.762) 28.36*** (7.186) -14.64** (7.180)
(4) Logit Country
0.07 0.81 0.04
1,189 0.09 0.295 -0.10 0.799 0.90 0.001
0.03 (0.019)
-0.07 (0.064) 0.09 (0.087) 0.07 (0.092) 0.06 (0.242) 0.17 (0.198) -0.33 (0.226) 0.40** (0.160) 0.36** (0.143) 0.14 (0.094) -0.05*** (0.013)
(5) OLS None
1,189 0.12 0.187 -0.17 0.668 0.81 0.002 0.96 0.538 0.09 0.87 0.03
-0.01 (0.019)
-0.06 (0.064) 0.10 (0.083) 0.08 (0.089) 0.01 (0.237) 0.13 (0.193) -0.31 (0.226) 0.36** (0.158) 0.39*** (0.138) 0.06 (0.094) -0.04*** (0.012)
(6) OLS Country
0.22 0.85 0.03
1,189 3.86 0.143 -4.43 0.712 48.90 0.000
-2.79*** (0.838)
-2.35 (1.636) 3.79 (3.130) 2.42 (3.708) 5.22 (7.453) 1.39 (10.198) -11.04 (9.004) 18.47** (8.488) 32.05** (13.782) -1.63 (8.509) -4.44*** (1.160)
(7) Logit None
1,001 2.78 0.300 -43.18 0.001 65.47 0.000 1356 0.000 0.36 0.89 0.03
-15.59*** (0.786)
-0.91 (1.208) 2.28 (2.160) 1.41 (2.310) -16.77* (10.036) -8.90 (12.453) -17.51* (10.327) 26.83** (11.562) 48.44** (20.769) -9.80 (12.635) -4.19*** (1.095)
(8) Logit Country
Note. Robust standard errors between parentheses, robust-standard-error-based p-values between brackets. ∆Cred is the growth rate of real lending by domestic banks to domestic households and non-financial corporations, deflated using CPI. For each country and year, ∆Cred* is the average of ∆Cred for the N − 1 remaining countries in the sample, weighted by PPP-adjusted GDP. We consider 3 lags of both ∆Cred and ∆Cred* (instead of the 5 lags considered in the main specification). FinOpen is the lag of a proxy for financial openness, namely a country’s gross external liabilities as a share of GDP. All specifications include country fixed effects. CFE stands for country fixed effects. AUROC stands for Area Under the Receiving Operating Characteristic curve, a measure of the binary classification ability of the model. Sample covers 38 countries over 1970-2011. The dependent variable is a dummy variable capturing banking crises.
Observations Sum of lagged coeffs. of ∆Cred Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆Cred* Test for sum of lags=0 (p-val) Sum of lagged coeffs. of ∆Cred* x FinOpen Test for sum of lags=0 (p-val) Test for CFE p-value R2 AUROC Standard error
Constant
FinOpen
L3.∆Cred* x FinOpen
L2.∆Cred* x FinOpen
L.∆Cred* x FinOpen
L3.∆Cred*
L2.∆Cred*
L.∆Cred*
L3.∆Cred
L2.∆Cred
L.∆Cred
Specification Estimation method Fixed effects
Table 8 Banking crisis prediction – Alternative lag structure
References Bruno, V. and H. S. Shin (2015): “Cross-Border Banking and Global Liquidity,” Review of Economic Studies, 82, 535–564. Cesa-Bianchi, A., F. Eguren Martin, and G. Thwaites (2016): “Foreign Booms, Domestic Busts: The Global Dimension of Banking Crises,” Unpublished manuscript.
10
