178 INTERDEPENDENCIES AMONG CORPORATE FINANCIAL POLICIES
TABLE 6.21 (Continued)
2000 2001 2002Variable Coeff T-Stat Coeff T-Stat Coeff T-Stat
Intercept 0.094 7.13 0.097 11.8 –0.027 –1.17
RDL –0.006 –0.55 0.012 0.77 0.057 1.57
Size –0.177 –0.3 0.36 0.74 1.928 1.13
PKL 0.018 0.62 –0.029 –1.81 0.014 0.63
IS –1.232 –13.37 –0.704 –18.74 1.163 19.64
DS 0.34 0.83 –1.31 –3.89 0.638 0.59
FS 1.083 29.4 0.933 25.76 –2.678 –22.22
The Data
The financial variables are divided by total assets to alleviate heteroscedas-
ticity; hence, an “S” denotes standardized variables. We also use annual fi-
nancial data for all U.S. firms during the 1952–2003 period.
Estimated Simultaneous Equations Results
The ordinary least squares (OLS) technique is used to initially estimate equa-
tions (6.2) through (6.5). The simultaneous equation results reported in this
study are produced with the use of two-stage (2SLS) and three-stage (3SLS)
least squares analysis. Although Dhrymes and Kurz (1967) found that the in-
significantly negative association between capital expenditures and dividends
in the two-stage least squares regression estimation became a significantly
negative association in the three-stage least squares estimations, we found no
statistically significant differences in the limited-information (two-stage least
squares) and full-information (three-stage least squares) procedures. The
two-stage least squares regression equation residuals were not highly corre-
lated, providing the statistical basis for the insignificant coefficient differ-
ences in the two- and three-stage least squares estimations. The highest
correlations were found in the annual regression residuals between the new
debt and capital expenditures equations. We found little difference in the
2SLS and 3SLS results; the OLS, 2SLS, and 3SLS squares regression results
are reported in Table 6.1 through Table 6.9. In the Dhrymes and Kurz
equation system, using equations (6.2), (6.3), and (6.4), dividend, capital