Recall that the 303-firm Compustat sample supported the following
model for the 1975–1982 period, as well as for the shorter period^8 :
One important factor may be that the 303-firm sample is highly
diversified and includes firms that reported no R&D expenditures. The
NSF/Census sample, which determines the reduced Compustat (RCOMP)
sample, does not contain such (predominately nonmanufacturing) firms.^9
On the positive side, all three models strongly support the importance
of last year’s R&D expenditures as a prediction of this year’s R&D.
When the investment (CE) equation was reestimated, it was perfectly
stable. As can be seen by inspecting Table 7.4, nothing changes. The tax
rate variable is significant in three of the five years, but changes signs (as is
the case with dividends).^10 The investment equation is:
CE = CF, R & D, EF, LIQ
+
f()
+− −
COMP303(75 82− ): R&D = (R& D− 1 NI DIV++−
f t ,, )Financial Decision Estimation Results 191
TABLE 7.6 Two-Stage Least Squares Results. Dependent Variable: New Debt (EF)
Constant CF KD DE CE RD DIV1978: 303 0.002 –0.155 –0.027 0.019 0.322 0.514 0.272
188 Cens 0.023* 0.012 0.023 0.014* 0.082 0.185 –0.751
188 Comp 0.026 0.024 0.003 0.011 0.088 0.023 –0.801
1979: 303 –0.008 –0.085 –0.020 0.013 0.313 0.329 0.381
180 Cens 0.013 0.024 0.005 0.014 0.203 –0.045 –0.725
180 Comp 0.011 0.017 0.004 0.016 0.188 0.097 –0.709
1980: 303 0.011 –0.029 –0.004 –0.007 0.472 0.042 –0.820
177 Cens 0.034 –0.033 0.017 –0.012 0.226 –0.053 –0.505
177 Comp 0.034 –0.031 0.002 –0.012 0.225 –0.061 –0.504
1981: 303 0.023 0.055 0.005 0.003 0.347 –0.072 –1.291
161 Cens –0.005 –0.061 0.008 0.036 0.278 0.157 –0.251
161 Comp –0.005 –0.064 0.009 0.036 0.281 0.183* –0.276
1982: 303 0.009 –0.192 0.032 –0.000 0.302 0.117 0.533
158 Cens 0.038 –0.113 0.002 –0.004 0.348 –0.190 –0.514
158 Comp 0.037 –0.114 –0.001 –0.004 0.348* –0.171 0.488
*Significant at 5% level.
**Significant at 10% level.
Cens—Census R&D database used in analysis.
Comp—Compustat R&D database used in analysis.