firms in each keiretsu. The first portfolio includes only the city banks, trust banks and propertycasualty insurance companies Group A. The second portfo-
lio includes all keiretsu financial firms that had enough valid stock returns over the period of the study Group B.
14
All stock returns were weighted by the combined market value of equity and book value of debt of each firm. Appendix
A outlines the procedure used to generate the weighted-portfolio returns includ- ing an adjustment for cross-holdings.
One difficulty with interpreting the keiretsu portfolio results is that large Japanese life insurance firms are mutuals and thus do not issue traded equity.
This may influence the results of the portfolio regressions because any assetli- ability effects related to the life insurance firms may not be reflected fully in the
portfolio stock returns. Komiya 1990 states that life insurance companies, while holding significant equity and loan portfolios in specific keiretsu firms, have had
few life insurance employees transferred to these firms either as line employees or directors as compared to keiretsu city banks. Thus, they had little influence over
the management of these firms. He also states that the relationship of life in- surance companies to a keiretsu is weak, based partially on their mutual form of
corporate governance.
4. Results
4
.
1
. Description of keiretsu portfolio weights Table 1 provides the names of the financial institutions by keiretsu and type,
and provides firm percentage weightings for each portfolio. The three keiretsu that have the most cohesive group ties Mitsubishi, Sumitomo, and Mitsui have
trust banks whose assets are a larger percentage of total assets for the financial arms of the keiretsu.
15
This indicates one area of greater risk exposure for the more cohesive groups, in that trust banks have recently been exposed to have
some of the more severe asset quality problems and implicit guarantees of sup- port in times of financial distress exist between the keiretsu city banks and
keiretsu trust banks. Note that propertycasualty insurance and trading compa- nies comprise a much smaller percentage of keiretsu financial firm assets. Trading
companies hold more real assets and liabilities with smaller balance sheet valua- tions.
14
There were several small firms that are either joint ventures or with peripheral ties to the keiretsu that had very few, if any, listed stock return data. These firms were not included in the portfolios. Due
to their small size, they should have little influence on the results.
15
The book value of liabilities plus adjusted market value of equity is used as a proxy for total assets.
4
.
2
. Indi6idual keiretsu financial firm regressions Eq. 1 was first estimated for the individual financial firms and then for the two
sets of portfolio returns.
16
Table 2 presents the betas for the individual financial keiretsu firms.
17
All the keiretsu firms have market betas that are significantly different from zero. Individual market betas range from a low of 0.43 to a high of
1.58. Because financial firms, in general, are highly levered, the market equity betas will be levered up to high levels. Still, 76 of the market betas are greater
than one, indicating high systematic market risk for the financial firms in a keiretsu. Twelve out of 14 significant bond return betas are positive indicating a net
asset duration interest rate risk exposure for the financial firms of a keiretsu. Only five out of spread change betas are significant; four with the expected negative sign
because net interest income decreases with spread decreases. Only two of the exchange rate betas are significant indicating the possibility of confounding effects
as changes in the value of foreign asset positions are cancelled by changes in the value of foreign liability positions. Changes in credit quality resulting from changes
in the competitiveness of the Japanese industrial economy may also contribute to create a confounding effect via the exchange rate changes. The ARCH0,
ARCH1, and GARCH1 terms are typically highly significant for the keiretsu firms indicating the influence of past volatility shocks on current stock returns.
4
.
3
. Patterns of betas across financial institution types The city banks three out of six and the trust banks five out of six both exhibit
a substantial number of significant bond interest rate return betas. The interest rate sensitivity of trust banks stock returns may be due, in part, to the large real
estate holdings of trust banks. The trust banks were the most heavily exposed to the property market which exposed them to holding very long-term real estate asset
portfolios due to the very illiquid property market after the 1990 stock market decline, thus causing a large positive duration gap not fully reflected on their
balance sheets. Three out of seven propertycasualty insurance companies also have significantly positive at a 0.10-level or better and economically large bond return
betas.
18
Three out of seven trading companies also have economically large and
16
Condition number analysis and simple correlation statistics were calculated to test for multi- collinearity of the independent variables. Simple correlation statistics indicated the expected high
correlation between the long-term and short-term interest rate measures. As would be expected, a yendollar exchange rate variable was significantly correlated with the trade-weighted yen exchange rate
variable. No evidence of significantly multicollinearity was found since the values of the condition numbers were all well below 10. Condition number analysis allows for detection of collinearity between
three or more independent variables, a relationship which simple correlation analysis would not detect. The minimum value of the condition index that indicates substantial multicollinearity is given by Belsley
et al. 1980 to be between 10 and 30, based on empirical evidence. Also, inspection of the actual change and innovation data series indicates no exceptionally large values.
17
The data time-series are adjusted so that all regressions have the same number of observations.
18
The three propertycasualty companies are Yasuda MF 8755, Fuyo Group, Tokio MF 8751, Mitsubishi Group and Sumitomo MF 8753, Sumitomo Group.
T .W
. Koch
, A
. Saporoschenko
J .
of Multi
. Fin
. Manag
.
11 2001
165 –
182
175
Table 2 Results of AR1-GARCH1,1 MLE estimation for individual Japanese keiretsu firm stock return series using all innovation variables non-orthogonalized except an actual change spread measure and with
a long-term interest rate return measure
a
A1 ARCH0
ARCH1 GARCH1
R
2
COID LIKHDOBS
INTCPT INMKT
INBD DSPD
INFX Fuyo group
0.007 −
0.145 0.001
0.859 0.212
1.117 0.4453
678.6 346 8317 city
0.006 −
2.366 0.3903
1.000 0.0001
0.0001 0.0145
0.0011 0.0001
0.9644 0.0001
0.148 0.167
2.786E−5 0.126
0.869 −
0.608 0.3324
8404 trust 621.8815 346
0.810 0.364
0.001 0.0478
0.0001 0.7269
0.0001 0.0077
0.4828 0.0638
0.3575 0.0001
0.187 −
1.094 0.001
0.308 0.073
0.874 0.003
1.245 0.5043
625.4161 346 8755 pc
0.4987 1.000
0.0001 0.0987
0.0089 0.0086
0.0001 0.2366
0.2020 0.164
0.557 −
1.414 8002 trading
0.001 0.199
0.4346 641.2616 346
0.002 0.059
1.002 0.0001
1.000 0.3323
0.4059 0.3491
0.0186 0.0001
0.2431 0.0160
Mitsubishi group −
0.182 −
0.235 1.091
0.256 0.000
0.684 0.061
0.5054 711.5689 346
0.003 8315 city
0.301 0.1071
0.0001 0.3526
0.0001 0.0947
0.0047 0.1823
0.7888 0.0001
8402 trust 0.985
1.306 −
0.887 0.001
0.231 0.175
603.9193 346 2.447E−5
0.5183 0.046
0.942 0.0156
0.2954 0.0017
0.0001 0.5773
0.6960 0.2175
0.0010 0.0001
8405 trust −
0.658 1.328
1.398 0.004
0.183 0.181
515.8946 346 0.001
0.3124 0.262
0.514 0.0012
0.0022 0.0250
0.0001 0.1460
0.3543 0.0001
0.0524 0.4806
0.592 −
0.522 1.093
0.090 0.001
8751 pc −
0.012 727.8260 346
1.061E−5 0.5894
0.052 0.937
0.0001 0.0569
0.8349 0.0001
0.0001 0.4420
0.0092 0.4817
0.5052 0.063
− 0.256
1.136 0.293
0.003 8757 pc
0.101 645.5836 346
0.001 0.4806
0.160 0.330
0.0858 0.0495
0.1099 0.2724
0.0001 0.8406
0.8012 0.1302
0.1137 0.318
0.341 1.035
0.439 0.003
8058 trading 0.083
650.3210 346 0.001
0.4588 0.192
0.421 0.0025
0.0005 0.2803
0.0007 0.0001
0.8132 0.2606
0.1407 0.0264
0.004 8359 reg bk
1.817 0.429
− 0.234
− 0.058
0.209 716.3137 346
0.001 0.1921
0.661 0.0001
0.0001 0.0003
1.000 0.0081
0.0001 0.2229
0.6973 0.0001
0.026 8584 cons cr
0.130 2.360
9.035E−5 0.086
0.878 0.4374
570.4457 346 1.253
0.476 0.004
0.0001 0.0001
0.1123 0.0819
0.0115 0.8984
0.1324 0.1261
0.0082 0.059
0.169 3.580E−5
− 0.178
0.061 0.198
0.912 0.000
0.6753 681.4888 346
1.485 8603 broker
0.2808 0.0001
0.8862 0.1817
0.2714 0.4484
0.0054 0.0001
0.8080 Sumitomo group
0.535 0.257
9.318E−5 0.122
0.810 0.681
0.5877 8318 city
676.3000 346 1.295
− 0.078
0.002 0.5933
0.0001 0.0023
0.5800 0.0001
0.0001 0.0423
0.1406 0.0014
0.088 −
1.801 0.000
0.245 0.162
0.457 8403 trust
0.001 0.823
0.5300 623.7500 346
1.339 0.2381
0.0001 0.0480
0.1015 0.0954
0.0001 0.0112
0.5048 0.0086
0.681 1.137
0.003 −
0.117 8753 pc
0.439 0.106
619.3016 346 0.001
0.4598 0.099
0.427 0.1016
0.1089 0.1221
0.2058 0.9317
0.0001 0.0394
0.0455 0.1542
0.396 1.060
0.002 0.471
8053 trading 0.218
0.184 708.5473 346
0.001 0.5631
0.137 0.238
0.0085 0.0398
0.0100 0.0919
0.4790 0.0001
0.0799 0.2144
0.7212 0.002
0.049 1.585
− 0.002
0.297 8601 broker
0.473 0.000
638.3886 346 0.6455
0.695 0.118
0.0001 0.0001
0.4112 0.4348
0.9906 0.2969
0.0861 0.7396
0.0312
T .W
. Koch
, A
. Saporoschenko
J .
of Multi
. Fin
. Manag
.
11 2001
165 –
182
176
Table 2
Continued
COID A1
INTCPT ARCH0
ARCH1 GARCH1
R
2
LIKHDOBS INMKT
INBD DSPD
INFX Sanwa group
8320 city 0.208
0.001 1.143
1.097 −
0.099 0.094
685.3344 346 0.000
0.5063 0.169
0.658 0.0007
0.0070 0.1661
0.0001 0.3806
0.3838 0.5382
0.4285 0.0001
0.001 −
0.125 8407 trust
− 0.220
0.901 0.275
0.086 625.3587 346
1.10108E−5 0.074
0.3431 0.926
0.0001 0.1875
0.8663 0.0001
0.4981 0.0001
0.6620 0.0994
0.003 0.138
1.047 0.439
− 0.016
8754 pc 1.470
0.000 638.0288 346
0.4607 0.629
0.257 0.0001
0.0001 0.0843
0.0385 0.0065
0.9428 0.0457
0.1053 0.0003
1.192 0.281
0.003 8063 trading
0.369 −
0.078 0.031
620.7295 346 0.001
0.4612 0.154
0.0245 0.0001
0.6251 1.000
0.8445 0.1001
0.0001 0.2104
0.7962 0.035
− 0.620
1.139 −
0.014 0.001
8004 trading 0.163
648.3833 346 3.353E−5
0.4218 0.051
0.925 0.0248
0.2283 0.0053
0.0001 0.0001
0.5417 0.9067
0.4305 0.9371
8591 leasing 0.869
0.1999 0.115
0.147 538.2194 346
0.002 −
0.042 0.080
0.743 0.066
0.003 0.6306
0.6983 0.0169
0.9132 0.8891
0.0001 0.0530
0.2248 0.0651
8583 cons cr 0.718
0.4285 0.046
0.141 620.0567 346
0.000 0.245
0.083 1.093
0.843 0.002
0.7899 0.6546
0.0205 0.1780
0.3653 0.0661
0.3071 0.0001
0.0001 Dai-ichi Kangyo group
0.560 0.002
− 0.126
− 1.533
2.693E−5 0.148
0.856 0.4254
669.9640 346 0.820
0.187 8311 city
0.0184 0.0001
0.0074 0.0001
0.2430 0.0369
0.4242 0.1371
0.0001 0.426
0.180 0.001
0.192 0.051
− 0.193
0.070 0.458
0.2752 686.5660 346
8765 pc 0.002
0.4380 0.5004
0.8582 0.0001
0.2025 0.3336
0.0001 0.1595
0.0001 0.319
− 0.815
1.132 0.258
0.000 8001 trading
0.107 677.9889 346
0.001 0.5400
0.246 0.236
0.0001 0.0035
0.1267 0.2163
0.0001 0.1405
0.3791 0.2307
0.7934 0.408
8585 leasing 0.062
− 0.117
0.001 0.110
0.489 0.3777
645.9076 346 0.883
0.001 −
1.625 0.6800
0.1189 0.3729
0.0001 0.3737
0.5926 0.1391
0.0500 0.0507
8607 broker 1.165
− 0.001
− 0.155
0.3898 0.062
655.5574 346 8.946E−6
0.115 0.823
0.047 0.894
0.3448 0.1498
0.2527 0.0971
0.7841 0.0001
0.6525 0.0001
0.0001 Mitsui group
− 0.423
1.066 0.001
0.891 8314 city
0.055 0.101
662.9165 346 0.000
0.3788 0.437
0.558 0.0001
0.0001 0.2239
0.0001 0.1906
0.0001 0.7361
0.0562 0.4138
0.452 0.083
1.009 0.172
0.003 8401 trust
0.149 598.0407 346
0.001 0.3783
0.346 0.380
0.0001 0.0010
0.0502 0.0030
0.0001 0.9479
0.0447 0.1444
0.3976 0.324
− 0.710
0.554 0.445
1.226 0.106
0.004 8752 pc
0.001 636.1751 346
0.142 0.5069
0.3068 0.1065
0.0535 0.1040
0.1156 0.0001
0.5556 0.0377
0.0152 0.264
0.590 0.388
0.264 1.110
0.180 0.002
8031 trading 0.001
671.0938 346 0.291
0.5001 0.1265
0.0001 0.0040
0.0001 0.0588
0.2758 0.1309
0.0202 0.6157
a
OBS refers to the number of observations according to the SAS GARCH estimation procedure. P-values are reported below the parameter estimates. R
2
is total R-squared, i.e., 1 − sum of squares for the original response variables corrected for the meanerror sum of squares. Company names are provided in Table 1.
significantly positive bond return betas indicating their status as financial intermediaries.
19
4
.
4
. Keiretsu financial firm portfolio regressions The results of the portfolio regressions using keiretsu financial firm stock
returns are presented in Table 3. For ease of comparison, the regression results for the appropriate keiretsu city bank are also included in Table 3. Few substan-
tial differences are noted between the keiretsu city bank sensitivities and the sensitivities for the relevant keiretsu portfolios. The Fuyo and Mitsubishi portfo-
lio stock returns exhibit significant and positive sensitivity to the long-term inter- est rate measure while the relevant city bank stock returns do not exhibit
significant sensitivity to this variable. As evidenced in Table 2, the increased interest rate sensitivity is probably due to the interest rate sensitivity of the
relevant keiretsu propertycasualty insurance companies. The Fuyo keiretsu prop- ertycasualty insurance firm, Yasuda MF, has a significant bond return beta of
0.874, while the Fuyo city bank has an insignificant bond return beta of 0.212. The Mitsubishi keiretsu propertycasualty firm, Tokio MF, has a significant
bond return beta of 0.592, while Mitsubishi city bank has an insignificant interest rate beta of 0.301.
Using different classifications of keiretsu member cohesion such as attendance at multiple presidents’ councils, cross-holdings of equity, and exchange of corpo-
rate directors, the Dai-ichi Kangyo, Fuyo, and Sanwa keiretsu are cited at the low end of group cohesion by Sheard 1994, Berglof and Perotti 1994 and
Dodwell Marketing Consultants, Industrial Groupings in Japan 1992. The Mit- subishi, Sumitomo, and to a lesser extent, Mitsui keiretsu are thus cited on the
high-end of group cohesion. Interestingly, none of the firms in the Sanwa keiretsu exhibit significant bond return interest rate sensitivity Table 2. Only
the Dai-ichi Kangyo keiretsu firms exhibit a similar pattern though the Dai-ichi Kangyo city bank has a significant bond beta. Also, examination of the Group B
regressions indicates that the Sumitomo, Mitsui, and Mitsubishi keiretsu all have market betas larger than the Dai-ichi Kangyo, Sanwa and Fuyo keiretsu. A
question for further study is if risk-taking by individual financial firms increases with increased keiretsu cohesion.
The volatility persistence of the portfolio regressions is similar to that of the relevant city bank regressions. The adjusted R
2
s typically increase for each keiretsu group as the regressions proceed from the regression for the individual
keiretsu city bank to the portfolio regressions with the least number of firms to the regressions with the most firms in a portfolio. This is probably a result of an
increase in fit, as a portfolio becomes closer to the market portfolio.
19
The three trading companies are Marubeni Corp. 8002, Fuyo Group, Sumitomo Corp. 8053, Sumitomo Group, and Mitsui and Co. 8031, Mitsui Group.
T .W
. Koch
, A
. Saporoschenko
J .
of Multi
. Fin
. Manag
.
11 2001
165 –
182
178 Table 3
Results of AR1-GARCH1,1 MLE estimation for Japanese keiretsu financial firm stock return portfolios using all innovations except for the spread variable
a
INFX INTCPT
A1 ARCH0
ARCH1 GARCH1
Likelihood INMKT
INBD DSPD
GROUP Total R
2
Fuyo group 0.46022
− 2.29739
0.99155 0.04015
0.006067 GRP1A
− 0.17863
742.461 0.000357
1.04630 1.000
0.0001 0.5006
0.0002 0.0001
0.0001 0.0001
0.7275 0.0001
0.0036 0.49190
− 2.55196
0.98227 0.0549
0.006402 GRP1B
− 0.1785
758.213 0.000322
1.06548 1.0000
0.0001 0.5318
0.0001 0.0001
0.0001 0.0001
0.0003 0.0001
0.6169 8317
0.21204 0.006042
− 2.36565
1.11701 0.00689
− 0.1448
678.597 0.000607
0.85917 1.000
0.0001 0.4254
0.0145 0.0001
0.3903 0.0001
0.9644 0.0001
0.0011 Mitsubishi group
0.002056 0.42296
0.25995 −
0.49819 0.000066
− 0.14213
0.01535 748.764
1.11473 GRP2A
0.70156 0.0001
0.0001 0.1669
0.5917 0.8272
0.6197 0.0118
0.1998 0.0001
0.0210 0.24499
− 0.39919
0.000032 −
0.08764 −
0.04678 0.35267
GRP2B 0.002137
0.73996 807.897
1.10159 0.5221
0.5370 0.0001
0.0265 0.6563
0.3506 0.0001
0.0196 0.1050
0.0001 0.68419
− 0.23532
0.30086 −
0.18199 1.09053
0.06062 0.002556
8315 0.000096
711.569 0.25646
0.0001 0.0001
0.3788 0.0047
0.1071 0.1823
0.0947 0.0001
0.3526 0.7888
Sumitomo group 1.32292
0.31115 0.003121
GRP3A 0.05665
0.61686 0.16722
705.289 0.000850
0.18278 1.000
0.0001 0.6345
0.0206 0.0325
0.7441 0.6832
0.0001 0.0495
0.0085 1.31148
− 0.00309
0.002908 GRP3B
0.08112 0.55857
0.15088 731.923
0.000782 0.09602
1.000 0.0001
0.6707 0.0219
0.1551 0.5379
0.9976 0.0121
0.0582 0.0001
0.6815 0.53495
1.29458 −
0.07798 0.002084
8318 0.2568
676.3 0.000093
0.81002 0.12183
0.0001 0.0423
0.5054 0.0001
0.0014 0.0001
0.5933 0.0023
0.1406 0.5800
Sanwa group GRP4A
0.74336 0.69329
− 0.07703
0.07817 727.353
0.000141 0.12994
1.10151 0.16715
0.001797 0.5725
0.5719 0.2542
0.5555 0.5768
0.0001 0.2217
0.0088 0.0001
0.0002 789.531
0.59058 0.71697
− 0.03464
0.1191 0.0603
1.0861 0.000089
0.002114 GRP4B
0.15698 0.7528
0.0001 0.6340
0.0014 0.5426
0.0170 0.0001
0.3611 0.0939
0.6093 0.20752
1.14283 1.09724
− 0.09888
0.001440 8320
0.09395 685.334
0.000218 0.65766
0.16903 0.0001
0.0070 0.4453
0.1661 0.0007
0.3806 0.5382
0.0001 0.4285
0.3838
T .W
. Koch
, A
. Saporoschenko
J .
of Multi
. Fin
. Manag
.
11 2001
165 –
182
179 Table 3 Continued
INFX INTCPT
A1 ARCH0
ARCH1 GARCH1
Likelihood INMKT
INBD DSPD
GROUP Total R
2
DKB Daichi Kangyo
group 0.55433
− 1.51669
0.81758 −
0.12471 0.001868
GRP5A 0.18521
672.481 0.000026
0.85602 0.14784
0.0001 0.0391
0.4264 0.0080
0.0001 0.1441
0.0001 0.4244
0.2393 0.0188
0.51698 −
1.50679 0.8671
− 0.05199
0.002030 GRP5B
0.13883 735.57
0.000021 0.83948
0.16332 0.0001
0.0714 0.5189
0.0541 0.0001
0.0001 0.0933
0.0076 0.1385
0.6927 669.964
− 1.53258
0.85563 −
0.12591 0.56016
0.1865 0.81991
0.000026 0.001868
8311 0.14763
0.4242 0.0001
0.5877 0.0001
0.0184 0.0074
0.2430 0.0369
0.0001 0.1371
Mitsui group 1.09967
0.63425 0.001627
GRP6A 0.11382
− 0.04837
0.01276 713.312
0.000142 0.47492
0.54081 0.0001
0.0002 0.5189
0.8615 0.0001
0.2125 0.4052
0.0001 0.2580
0.8151 0.01678
0.72535 1.11997
0.1091 0.001625
GRP6B 0.00398
736.206 0.000143
0.45104 0.54928
0.0001 0.0001
0.4551 0.9580
0.0001 0.1565
0.0001 0.4042
0.2283 0.9289
0.55821 0.89054
− 0.42262
0.0552 1.06634
0.10078 0.001464
8314 0.000169
662.917 0.43699
0.0001 0.0001
0.5063 0.0001
0.0562 0.7361
0.0001 0.4138
0.2239 0.1906
a
Variables are unorthogonalized and the long-term interest rate measure is used. P-values are reported below the parameter estimates. The results for the appropriate keiretsu city bank are also reported for comparison. The group designations are defined in Table 1.
5. Conclusions and issues for further study