the dividend variable becomes negative and significant, while cash flow
drops out. Finally, when the NSF/Census R&D data are substituted, the
debt/equity ratio enters as a significant positive variable. This is counter-
intuitive, since the cost of debt should rise as the debt/equity ratio rises,
thus discouraging new debt. This latter result was found by Guerard and
McCabe (1992).
Relation of Current Results to Prior Research
Three new models of statistically significant relationships between financial
decisions and the stock market value of a firm have been estimated. They
have then been compared with a fourth reference model developed in an
earlier study.
1.The compression of the time frame from 1975–1982 to 1978–1982
did not change the structure of the R&D equation or the capital expendi-
ture equation associated with the 303-firm Compustat sample. This means
that the interdependence of the R&D, net income, dividends, and capital
expenditures decisions were preserved, implying support for the imperfect
markets hypothesis.^11
2.The reduced Compustat data set includes only the firms that are
present in both the Compustat (1978–1982) and the NSF/Census
(1978–1982) samples (see Guerard, Bean, and Andrews 1987). An im-
portant result of this analysis is that the R&D decision is no longer de-
pendent on the net income and dividend decisions, but is driven only by
previous R&D. This finding tends to weaken support for the imperfect
markets hypothesis. The capital expenditures equation continues to be
stable, including a positive association between R&D and capital expen-
ditures, which is consistent with expectations and prior results. The fact
that the RCOMP (1978–1982) sample contains only R&D-performing
firms, and thus includes a limited range of observed levels of R&D ex-
penditures (lacking a zero point, for example), may explain why R&D
expenditures are no longer interdependent with other financial decisions.
The dividend equation is consistent with the COMP303 (1978–1982)
version. New debt continues to be strongly affected by capital expendi-
tures, but the relationship to dividends changes signs again, thus indicat-
ing instability based on sample composition as well as time frame (the
preceding item 1).
3.The substitution of the NSF/Census R&D expenditure data for the
Compustat R&D data in the reduced sample suggests only minor modifi-