The full text on this page is automatically extracted from the file linked above and may contain errors and inconsistencies.
orKing raper series
R&D and Internal Finance:
A Panel Study of Small Firms
in High-Tech Industries
Charles P. Himmelberg and Bruce C. Petersen
Working Papers Series
Issues in Macroeconomics
Research Department
Federal Reserve Bank of Chicago
December 1991 (WP-91-25)
FEDERAL RESERVE BANK
OF CHICAGO
R&D and Internal Finance: A Panel Study
of Small Firms in High-Tech Industries1
Charles P. Himmelberg
New York University
Bruce C. Petersen
Washington University
Federal Reserve Bank of Chicago
(First submitted: March 23, 1990)
June 12, 1991
1The authors would like to thank Joe Altonji, Charles Calomiris, Steve Fazzari, Zvi Griliches,
Ken K uttner, Dorothy Petersen, Rob Porter, F. M. Scherer, Steve Strongin, two anonym ous referees,
and seminar participants at the Federal Reserve Bank of Chicago, Miami University at Oxford,
Northwestern University, the University of Chicago, the University of W isconsin at Madison and
W ashington University for helpful comments and suggestions. Remaining errors are our own. This
research was begun while both authors were at the Federal Reserve Bank of Chicago.
A b str a c t
Since Schumpeter, economists have argued that internal finance should be an important
determinant of R&D expenditures. Yet almost without exception, previous empirical studies
have not found evidence of such a relation. Using newly available data, we investigate this
puzzle with a panel of 179 small firms in high-tech industries. Under each of the different
estimation strategies we employ, we find an economically large and statistically significant
relationship between R&D expenditures and internal finance. Our results are consistent
with the view that, because of capital market imperfections, the flow of internal finance
is the principal determinant of the rate at which small, high-tech firms acquire technology
through R&D.
Since Schumpeter,1 economists have argued that internal finance is an important de
terminant of R&D expenditures. For example, Ivamien and Schwartz (1978, p.252) state,
“Among the leading characteristics commonly associated with industrial research and de
velopment, one of the most prominent is the virtual necessity for it to be financed internally
from a firm’s current profits and accumulated funds.” Yet almost without exception, pre
vious empirical studies have not found evidence of such a relationship. In this paper, we
investigate this puzzle with new data on small firms in high-tech industries.
The arguments why internal finance, for some firms, may be the principal determinant of
R&D are becoming ever more refined with the development of the economics of information.
Arrow (1962) was among the first to argue that moral hazard problems hinder external
financing of highly risky business activities such as innovation. More recently, Stiglitz and
Weiss (1981) and Myers and Majluf (1984), among others, have developed formal models of
moral hazard and adverse selection in markets for debt and equity which apply particularly
well to high-tech investments .2 These papers provide a formal justification for models of
the firm which assume that the rate at which small, growing firms acquire capital, including
R&D, is determined by access to internal finance (e.g., Kamien and Schwartz (1978) and
Spence (1979)).
Our empirical findings are based on a panel of 179 small firms in high-tech industries.
While the initial size of the firms in our study is under $10 million in capital stock, Acs
and Audretsch (1988) show that firms in this size range account for a major fraction of
new innovation in U.S. manufacturing. Until recently, it would have been nearly impossible
to assemble such a data base .3 Previous R&D studies emphasizing financial considerations
focused on large firms and did not have access to recently developed panel data techniques.
While our major focus is on the effect of internal finance on R&D expenditures, we also
1For Sch um peter’s view s on the potential im portance of internal finance for in n o v a tio n , see Schumpeter
(1942, ch .8). One o f Schu m p eter’s defenses of m onopoly practices was th a t th ey could provide resources for
financing the innovation process. T his remains an provocative though controversial idea.
2 We note th at adverse incentive and selection problem s are com pounded by the absen ce o f collateral value
for investm en ts like R&D. T h e im portance of collateralizable net worth has been em p h asized by Bernanke
and Gertler (1989), C alom iris and Hubbard (1990) and Hubbard and K ashyap (1 990). S ince sm all, high-tech
firms hold m ost of their value in growth opportunities and scientific know ledge, they are likely to have little
or no collateralizable net w orth.
3C om p u stat expanded coverage to such small corporations only recently.
1
consider its effect on physical investment. There are several reasons why it is useful to do
so. First, this approach permits a comparison of our findings to the existing literature on
physical investment under capital market imperfections. Second, it is inappropriate to view
the firm as having access to separate sources of finance for R&D and physical investment.
And finally, as argued by Schumpeter, new knowledge must be embodied in the production
process through investment in new plant and equipment. Hence, the physical investment of
R&D-intensive firms can likely be characterized by a similarly high degree of asymmetric
information.
We find an important role for internal finance in explaining both the R&D and physical
investment expenditures of the firms in our panel. Controlling for unobservable firm effects,
which has not been done in previous R&D studies, we obtain a large and statistically
significant relationship between both forms of investment and internal finance. However,
the conventional within-firm estimates imply an elasticity for R&D that is less than half the
elasticity for physical investment.4 We argue that this is due to high adjustment costs for
R&D. These adjustment costs induce a downward bias in the wi thin-firm estimator if firms
smooth R&D in response to transitory shocks in cash flow. Following a procedure outlined
by Griliches and Hausman (1986), which they apply to a similar problem in the labor
literature, we obtain instrumental variable estimates that imply internal finance elasticities
of 0.670 for R&D and 0.822 for physical investment.
The remainder of the paper is organized as follows. The next section briefly reviews
the theoretical motivation and outlines the empirical predictions. Section 2 explains the
construction of our panel and provides summary statistics. Section 3 reports our empirical
results, and Section 4 concludes.
1
T h e o r e t i c a l a n d E m p i r i c a l Issues
There is an excellent review of the empirical literature on internal finance and R&D in
Kamien and Schwartz (1982, p.98). They conclude that “the empirical evidence that ei
4We adopt the term inology conventionally employed by panel stu d ies by u sin g “b etw een-firm ” to refer
to differences in firm -specific averages across firms, where the averages are co m p u ted over time; the term
“w ithin-firm ” is used to refer to deviations o f variables from these firm -specific m ean s.
2
ther liquidity or profitability are conducive to innovative effort or output appears slim.”
Cross-sectional studies such as Scherer (1965), Mueller (1967) and Elliott (1971) find no
relationship between internal finance and R&D .5 It is important to point out, however,
that previous empirical studies considered only large firms, very often only firms in the
Fortune 500. Because these firms typically generate much more cash flow then they need
for investment purposes, it is unlikely that the existence of financing constraints would have
any effect. This point is made elegantly in a theoretical framework in Kamien and Schwartz
(1978).6
In contrast to the R&D literature, there is a large literature dating back to Meyer
and Kuh (1957) which documents the relationship between internal finance and p h y s ic a l
investment. Most of these studies find an important role for internal finance .7 For example,
a recent study by Fazzari, Hubbard and Petersen (1988) of a panel of U.S. manufacturing
firms finds that a large fraction of the within-firm variation in physical investment can be
explained by variation in cash flow for firms that exhaust their internal finance. There
are also recent studies which document the relationship between physical investment and
internal finance for Japanese and U.K. firms.8
1.1
T h e R o le o f I n t e r n a l F in a n c e
Arguments for why R&D must be funded primarily by internal finance are usually based on
the existence of information asymmetries between firms and suppliers of external finance.
Information asymmetries are easy to motivate, particularly for small, high-tech firms. The
very nature of R&D and innovation-based physical investment precludes outsiders from
making accurate appraisals of value. In addition, even when firms can costlessly transmit
information to outsiders, strategic considerations may induce firms to actively m a in ta in
5An exception is Grabowski (1968, p.296). He exam ines a cross section of large firms in the chem ical, drug
and petroleum industries and finds an econom ically large and statistically significant relationship b etw een
R&D intensity and internal finance only in the drug industry.
6T hev show th at, under quite plausible assum ptions, even if al l firms relied en tirely on internal finance
to fund R&D, only sm all firms would find these constraints to be binding.
7See, for exam ple, Fazzari and A tliey (1987). For a review of this literature, see Fazzari, Hubbard and
Petersen (1988)
8See Hoshi, K ashyap and Scharfstein (1991) for evidence on Japanese firms and D evereux and Schiantarelli
(1989) for evidence on U .K . firms.
3
information asymmetries. Levin, Klevorick, Nelson and Winter (1987) report that firms in
most industries view patents as an ineffective method of appropriating the returns to R&:D,
and instead often prefer to use secrecy.
The effect of information asymmetries on the market for new share issues has been ex
amined by Myers and Majluf (1984) through an extension of Ackerlof’s (1970) well-known
“market-for-lemons” argument.9 Myers and Majluf explain why firms may be forced to sell
stock at a discount (pay a “lemons premium”) if they can sell shares at all. The adverse
selection problems which they describe can be particularly severe for high-tech firms since
the range of actual (but unobservable) values between “good firms” and “lemons” can be
large. 10 Like equity markets, debt markets are also vulnerable to adverse selection prob
lems because of asymmetric information about risk characteristics and default probabilities.
Stiglitz and Weiss (1981) argue that banks may ration credit rather than use interest rates
to clear the market because increases in interest rates may cause low risk borrowers to exit
the application pool. Again, this outcome seems particularly plausible for high-tech firms
where the probability of default can vary widely over a set of observationally equivalent
firms.
In addition to adverse selection, the issuance of new debt is further complicated by
moral hazard problems. Arrow (1962, p.153) argues that this problem is especially relevant
for investment in R&D projects, given that “the output can never be predicted perfectly
from the inputs.” Pursuing this line of reasoning, Stiglitz and Weiss (1981) note that
as interest rates rise, unmonitored borrowers have an incentive to use loans for projects
that are not in the best interest of lenders. In particular, borrowers can invest e x p o s t
in riskier, higher-return projects that increase the probability of bankruptcy, but offer no
offsetting gain to debtholders if success is achieved. This problem is accentuated as firms
become more leveraged. It is for this reason that equity, not debt, is considered the natural
9T h e classic exam p le o f a market w ith asym m etric inform ation and adverse selectio n problem s is A ck
erlof’s (1970) used car m arket. B u t we find this exam ple less convincing than th e n ew -eq u ity m arket for
sm all, R & D -intensive com panies. A p oten tial buyer of a used car can, at relatively low co st, hire a m echanic
to assess the car’s true quality. In contrast, a potential investor m ight have to hire a team o f scien tists to
make an accurate appraisal of the potential value of a firm’s R&D projects.
10A cs and A udretsch (1990, p. 71) report th at only a sm all fraction of new firms receive venture cap
ital financing, su ggestin g th at venture capital is not a q uantitatively im portant m eth o d for overcom ing
inform ation problem s in equity markets.
4
financial instrument for high-tech investment.11 It should also be pointed out that the above
problems are compounded by the lack of collateral value for most R&D investments .12
1.2
R & D a n d P h y s ic a l In v e s tm e n t W i t h F in a n c in g C o n s tr a in ts
We describe the investment problem of a small, high-tech firm by appealing to the model
in Spence (1979). In his model, firm profits in the initial stage of the market are positive
because of low industry capacity relative to demand. Each firm’s rate of growth is con
strained by access to internal finance. The solution in his model is that firms move out
their expansion paths during the growth phase of the industry as rapidly as their internal
finance permits, maintaining equality of marginal products of each type of investment.
In order to introduce R&D investment into the model, the production function is as
sumed to include not only a stock of capital but a stock of technology as well .13 This
follows the productivity literature, where it is common to assume that output is a homothetic function of technology and physical capital, and that the stock of technology is acquired
through R&D expenditures (for a recent review of this literature, see Mairesse and Sassenou
(1991)). The assumption of a homothetic function is supported by the empirical fact that
the R&D-to-sales ratio is approximately constant over firm size in most industries (e.g.,
Griliches (1984)).
For reasons discussed in Section 1 . 1 , we assume that firms face a binding financial
constraint on investment expenditures. For the simplest case in which the firm obtains no
external financing, such a constraint implies that the firm’s total investment expenditures
cannot exceed current cash flow. This stylized view of the financing constraint could be
generalized to include debt as a multiple of internal equity as discussed in Spence (1979).
However, this generalization turns out to not be necessary since the firms in our sample
11 Long and M alitz (1985) provide form al em pirical evidence th a t financial leverage is n egatively correlated
w ith R&D expenditures.
12B ester (1985) em phasizes th at in d eb t m arkets, collateral can be used as b oth a signalling device to
overcom e adverse selection and as an in cen tive device to overcom e moral hazard. H ow ever, these options
are not likely to be available for sm all firms in high-tech industries because there is no collateral value to
failed R&D and innovation-based in vestm en t projects.
13Including technology in the production function clearly accom m odates process R & D . However, m ost
R&D is for new product developm ent. Griliches (1986, p.144) points out th a t the production function
approach also accom m odates new product R&D if ou tp u t is replaced by sales. In our discussion of expansion
paths, this would im ply replacing the isoquant w ith an “isovalue” curve.
5
obtain very little debt finance.
The existence of an internal finance constraint does not change the first order con
ditions determining the r e la tiv e levels of the desired stocks of physical and technological
capital (e.g., Henderson and Quandt (1971)). Rather, the constraint determines the a b
s o lu te
levels of R&D and physical investment. In the absence of adjustment costs, this
allocation is determined by the parameters of the production function. If the production
function is homothetic, then the expansion path is linear, and the optimal allocation of
internal finance between R&D and physical investment is proportional to the (constant)
shares of technology and physical capital in the production function. This fact allows a
loose structural interpretation of the cash flow coefficients in our reduced-form regressions.
We note that if the production function were not homothetic, or if there were adjustment
costs, then expenditure shares would vary over time. In this case, the cash flow coefficients
could be interpreted as a linear approximation of these shares over the time period covered
by our panel.
1.3
A d ju s t m e n t C o s ts
It is important to account for the probable existence of high adjustment costs for R&D
when estimating the effect of internal finance on R&D. The failure to account for adjust
ment costs could bias our results for reasons similar to Griliches and Hausman’s (1986)
explanation of the puzzle that within-firm estimates of labor demand functions often yield
output elasticities of less than one, implying increasing returns to scale. They argue that
because of adjustment costs, labor is hired in anticipation of permanent output, with little
adjustment made in response to transitory movements in output. Since a firm’s R&D in
vestment is predominantly a payment for a flow of services from its stock of highly trained
scientists, engineers and other specialists, the Griliches and Hausman insight and approach
is especially applicable to our problem.
Theoretical explanations and empirical evidence of high adjustment costs for R&D can
be found throughout the economics literature.14 Grabowski (1968) makes a strong case for
14D isciissions of high adjustm ent costs for R&D are found o u tsid e the econom ics literature as w ell. For ex
am ple, the literature on the m anagem ent of technological in n ovation frequently recom m ends th a t tem porary
adjustm ents in R&D expenditures be avoided because o f high ad ju stm en t costs (e.g ., T w iss (1 9 8 6 )).
6
high adjustment costs for R&D and argues that ‘‘research workers, whose salaries constitute
a sizable percentage of total expenditures, are not perfectly elastic in supply and cannot
be alternatively fired and rehired in accordance with temporary changes in business condi
tions.” There are a number of reasons why temporary hiring and firing of research workers
is costly. For one, researchers require a great deal of firm-specific knowledge, and training
new workers is expensive. Perhaps more importantly, fired specialists are able to transmit
valuable knowledge to competitors who hire them. Pakes and Nitzan (1983) describe opti
mal labor contracts designed specifically to retain R&D workers to reduce appropriability
problems. Levin et al. (1987) report that hiring a competitor’s R&D personnel is viewed by
many firms as an effective means of procuring technological capital compared to alternative
channels of information spillover.
Empirical evidence on adjustment costs is reported by Bernstein (1986) and Bernstein
and Nadiri (1988, 1989). Bernstein and Nadiri (1989) estimate returns for R&D and physical
investment as well as the marginal adjustment costs for these inputs for firms in four twodigit industries. The estimated marginal adjustment costs were higher for R&D in all four
industries. In particular, the marginal adjustment cost for R&D was two to several times
higher for two-digit SIC codes 28 and 35, two of the four industries in our panel.
The existence of high adjustment costs for R&D implies the following modification to
our description of financially constrained firms. In order to minimize both the current and
future adjustment costs, firms set the level of R&D expenditures in accordance with the
“permanent” level of internal finance. When the firm believes that a change in the flow of
internal funds is “transitory,” it attempts to maintain the planned level of R&D expendi
tures by adjusting physical investment, or, if available, working capital. The econometric
specification in Section 3.3 accommodates this description of firm behavior by postulating
that current cash flow can be decomposed into a permanent component and a transitory
component. Since high adjustment costs imply that R&D is relatively unresponsive to tran
sitory movements, the full impact of the financing constraint is revealed by the relationship
between R&D and permanent cash flow.
7
1 .4
E m p ir ic a l P r e d ic t io n s
The existence of financial constraints and high costs of adjustment for R&D yields the fol
lowing empirical predictions which we investigate in section 3. First, the division of internal
finance among competing investments is determined by the parameters of the production
function. If R&D and physical investment were the only components of total investment,
then the cash flow coefficients would sum to one. In reality, of course, while R&D and
physical investment are the principal components, there are other uses (sources) of funds
such as working capital. Hence, the coefficients should sum to a number that is large but
less than one. Second, if adjustment costs are important, R&D may not respond equally to
transitory and permanent shocks to cash flow. Since the conventional within-firm estimator
does not distinguish between transitory and permanent movements in cash flow, it may be
a downward biased estimate of the effect of permanent movements. For this reason, we
emphasize an instrumental variables procedure which is designed to control for both this
bias and the existence of individual firm effects.
2
2.1
T h e D a t a a n d S u m m a r y Statistics
C o n s tr u c tio n o f th e P a n e l
The firm data for this study are taken from the May 1989 release of Standard and Poor’s
Compustat file. Compustat follows virtually every company listed on the American and the
New York Stock Exchanges and the Over-the-Counter Markets. Out of the initial universe
of 3035 manufacturing firms, we construct our panel using five selection criteria. A firm
is included in our panel if (i) its primary location is domestic and it is not a subsidiary,
(ii) the replacement value of its capital stock in 1983 is between Si and $ 1 0 million, (iii)
there are no missing values for essential variables from 1983 to 1987, (iv) there are no ma
jor mergers, acquisitions, or divestitures.15, and (v) its industry is one of four identified
as high tech: chemicals and drugs, machinery, electrical equipment and communications,
15Further details are described in an appendix available from the authors. T h is criterion excludes any
firm for which the discrepancy betw een in vestm en t expenditures and the reported change in the gross book
value of the capital stock net of retirem ents is greater than 15 percent.
8
and instruments .16 These four industries have the highest R&D-to-sales ratios in manu
facturing, and collectively account for approximately one half of all R&D expenditures in
manufacturing (see Scherer (1980, p.410)).
We examine the five-year time period 1983-1987 because using a longer period results
in a sharp reduction in the number of small firms listed continuously in Compustat. The
ten million dollar size cutoff is chosen to focus on small firms.17 Domestic, non-subsidiary
firms with historical data back to 1982 comprise approximately 61 percent of the Compustat
universe of manufacturing firms. Of these, 56 percent are in high-tech industries, and of
these, 3 7 percent have capital stocks of under $ 1 0 million at the start of our sample period.
Deleting firms with major mergers and acquisitions leaves us with a final sample of 179
firms.18
Table 1 documents the beginning and ending average sizes of the firms in our panel.
The average beginning value of the capital stock is $4.35 million, and the average number
of employees is 237. On average, these firms accumulated capital at a very high average real
rate of growth - over 12 percent annually over the five year period. The resulting average
ending capital stock was 11.93 million and the average number of employees was 407. The
standard deviation of the distribution of capital stocks grew from 2.5 in 1983 to 23.6 in
1987, reflecting a wide range of growth rates across firms. Very high growth rates are not
uncommon in high-tech industries. For example, in the computer industry, a number of
startups reached Fortune 500 size in just a few decades .19
The last row of Table 1 reports the ratio of R&D to R&D plus physical investment for
our sample of high-tech firms. On average, these firms allocate as much funding to R&D
as to physical investment. This ratio varies little across our four high-tech industries: the
16Griliches and Mairesse (1984) identified the tw o-digit SIC codes 28, 35, 36 and 38 as the science-based
industries. A sim ilar set of industries was identified by B ern stein and Nadiri (1988).
17Size cutoffs are always arbitrary, but ten million is a convenient cutoff b ecau se it is a focal point and
because it closely corresponds to the 500-em ployee cutoff used in other stu d ies o f sm all firm behavior. Cutoffs
ranging from five to twenty million yield the sam e pattern of findings reported in the n ex t section.
18We exam ined the distribution of the d a ta sam ple selected by the above criteria and identified eight
d istin ct outliers, all of which had initial capital stocks o f under $2 m illion. See the d a ta appendix available
from the authors for further details.
19T he Wall Street Journal, Sept. 8, 1989, reports: “In the 32 years since a $70,000 capital investm ent
launched D igital Equipm ent Corp, hundreds of electronics pioneers have sta rted com puter-hardware com
panies. More than a dozen of these startups have turned into Fortune 500 com p an ies.”
9
Table 1: B asic S tatistics
Variable
M ean
Standard D ev ia tio n
C apital Stock, 1983
C apital Stock, 1987
Num ber of Em ployees, 1983
Num ber of Em ployees, 1987
Sales, 1983
Sales, 1987
R & D /(R & D + Investm ent)
4.35
11.93
237
407
16.48
38.91
0.528
2.50
23.61
225
1285
16.56
89.22
0.239
Note: A ll financial figures reported in m illions of 1982 dollars.
ratio is 0.481 for pharmaceuticals, 0.530 for non-electrical equipment, 0.516 for electrical
equipment, and 0.541 for scientific instruments. These ratios are approximately twice as
high as the ratios for the balance of the manufacturing industries in Compustat.
We followed the standard practice in the investment literature of dividing each variable
by the beginning-of-period replacement value of property, plant and equipment. This trans
formation from levels to ratios makes it possible to compare investment and R&D ratios
over time and across firms. In a panel with firms that are growing over time as well as
starting at different sizes, such a transformation yields trend-stationary series and controls
for heterogeneity as well.20
2.2
S u m m a r y S ta tis tic s
Our key summary statistics appear in Table 2 , which is divided into a section reporting
investment and a section reporting sources of finance. These variables are scaled by the
firm’s capital stock. The first two columns of the table report the mean and the value at
the 75th percentile for each variable. The last two columns of the table decompose the
20Since firms record accum ulated cap ital stocks at book value, the replacem ent value of capital is con
stru cted using a perpetual inventory m ethod (for further d etails, see Salinger and Sum m ers (1983)). T he
physical in vestm en t, R&D, and the changc-in-sales variables correspond to the u sual accounting definitions.
Since firms treat R&D as an expense, we add R&D back into the usual accou n tin g definition of cash flow.
W e n ote th a t if R&D is su b ject to classical m easurem ent error, this construction w ould bias the least squares
regressions of R&D on cash flow reported below; the instru m en tal variable e stim a tes reported Section 3.3
elim inate this bias. A d a ta appendix detailing the con stru ction o f the rem aining variables (including T o b in ’s
q) is available from the authors on request.
10
Table 2: Sum m ary S tatistics
V ariance
B etw een -F irm
W ithin-Firm
Variable
Mean
75th percentile
R&D
P h y sica l InvestmentT o ta l In vestm en t
(R & D -f P h ysical Investm ent)
0.240
0.257
0.497
0.344
0.280
0.624
0.054
0.078
0.195
0.016
0.128
0.194
Cash Flow
N et Long Term D ebt Financing
N et Short Term D ebt Financing
N et N ew Share Issues
All observations
E xcluding upper 5% tail
0.444
0.035
0.023
0.610
0.023
0.021
0.265
0.027
0.007
0.157
0.124
0.124
0.344
0.082
0.045
0.030
0.512
0.029
1.912
0.075
D ividends
0.008
0.000
0.000
0.000
N ote: A ll variables first scaled by capital stock.
variance of each variable into its between-firm and its within-firm component. We adopt
the terminology conventionally employed by panel data studies by using “between-firm” to
refer to differences in firm-specific averages across firms, where the averages are computed
over time; the term “within-firm” is then used to refer to deviations of variables from these
firm-specific means.
Reading across the first two rows of Table 2, the means of the investment and the R&D
ratios are almost the same; that is, on average, firms allocate as much resources to R&D
as to physical investment. In addition, the relative shares are invariant across firm size,
suggesting an approximately linear expansion path. As already noted, this is a very high
level of R&D spending relative to physical investment compared to firms not in high-tech
industries. In addition, the absolute size of the R&D and physical investment ratios is large,
which is consistent with the high average growth rates observed in Table 1 .
The variance decompositions of investment and R&D are given in the next two columns.
The between-firm variances of these two ratios are of the same magnitude, with the vari
ance of the investment ratio being somewhat greater than the variance of the R&D ratio.
However, the within-firm variances are very different. The within-firm variance of physical
11
investment is nearly te n tim e s greater than the within-firm variance of R&D. Moreover,
for physical investment, the wi thin-firm variance amounts to approximately two thirds of
its total variance. In contrast, the within-firm variance of R&D is only 20 percent of its
total variance. Hall and Hayashi (1989), among others, document a similar pattern in the
variance decompositions of physical investment and R&D. This “sm ooth” behavior of R&D
expenditures is consistent with the hypothesis that adjustment costs are high for R&D.
The second part of the table presents statistics on the financial behavior of the firms
in our panel. Internal finance is obviously an important source of funds. The mean value
of the cash flow ratio is only slightly less than the sum of the means of the investment
and R&D ratios. Debt usage, on the other hand, is small; the sum of the mean values
of both short term and long term new debt is barely ten percent of the mean of internal
finance. This comparison is even more obvious when these ratios are compared at the 75th
percentile. A striking result in the table is the fact that essentially none of the firms in
our sample pay dividends, which is again consistent with the assumption that they face
financing constraints.
The mean ratio of new share issues to capital is much larger than the mean ratio for
debt. However, this number is misleading because of the effect of a few large outliers, as
shown by the very low value of new share issues at the 75th percentile (the median is zero),
and the very high total variance of this ratio. The high value of the mean results from a
few very small firms (between $ 1 and $2 million in capital stock) making proportionately
very large new share issues in the first year of our sample .21 If the upper five percent tail
of the distribution is excluded, the ratio declines to 0.082, as reported in the next row of
the table .22
To summarize, the typical firm in our sample has the following profile. It pays no divi
dends, rarely issues new equity, and makes only modest use of debt finance. This financing
pattern is consistent with the discussion on the role of internal finance in Section 1 . 1 . In
21T w o exam p les include Priam C orporation, which in 1983 had a cap ital sto ck o f 2.4 m illion and a new
share issue o f 67.6 m illion, and Lym phom ed, which had a cap ital stock o f 1.4 m illion, and a new share issue
of 13.8 m illion.
22A nother w ay to m easure the im portance of internal finance relative to extern al finance is sim ply to add
up the dollar values for a l l firms w ithout scaling; over the tim e period 1984-1987, in tern a l finance am ounted
to 3.3 billion, net new d eb t contributed 745 million and net new share issues am ou n ted to 536 million, where
all figures are com p u ted in constant 1982 dollars.
12
addition, internal finance is approximately equal to the sum of physical investment and
R&D. All of this evidence supports our working hypothesis that the typical firm in our
panel faces binding internal finance constraints.
3
E c o n o m e t r i c Specification a n d Results
In this section, we first present the within-firm results and emphasize the potential impor
tance of accounting for unobservable firm effects. We next present the between-firm results
for reasons which we motivate below. Finally, we consider an econometric specification
which explicitly allows for a differential response of R&D to the permanent and transitory
components of cash flow.
3.1
W i t h i n - F i r m R e s u lts
An important econometric issue not addressed in previous empirical studies of R&D and
internal finance is the existence of individual firm effects. Controlling for unobservable firm
effects is important for our study since the firm effect is likely to be positively correlated
with both internal finance and R&D. The most obvious source of the correlation is that
firms differ with respect to managerial abilities and that superior managers b o th generate
higher cash flows and seek to expand their firms faster than inferior managers .23 Failing to
account for this firm effect can be viewed as a specification error which is likely to bias the
estimate of the effect of internal finance on R&D.
We treat the firm-specific components of the respective error terms as fixed effects. Thus
our baseline specification for the within-firm estimation is
R D { t — Po
where
ol{ is
+ P cj C F n + a { + v t + e,-<,
(1)
the individual firm effect and vt is the year effect. We use the standard method
of sweeping out the fixed effects by transforming variables to deviations from their firmspecific means .24 The error term e a accommodates measurement error in the dependent
23See M undlak (1978) and Hsiao (198G).
24See Hsiao (1986) for a detailed discussion of this approach.
13
variable and the effect of unobserved explanatory variables assumed to be uncorrelated
with internal finance and the firm and year effects. We do not view reverse causation as
a problem in Equation 1 since there is a sizable gestation lag as well as an application lag
between the outlay of an R&D dollar and the beginning of the associated revenue stream .25
The physical investment equation is estimated using the same specification.
Our within-firm results are reported in Table 3. The top half of the table reports R&D
regressions (A 1 through A4) while the bottom half contains physical investment regressions
(B 1 through B4). Heteroskedasticity consistent standard errors are reported in parentheses
(see White (1980)). Our baseline specifications A 1 and B 1 are based on the hypothesis
that the firm’s investment rate is determined by internal finance. The remainder of Table 3
examines the effect of including demand variables such as Tobin’s q and the change in sales
to control for the possible expectations role of cash flow. These variables have been used
in past studies to investigate the physical investment behavior of mature firms. We include
them here to check the robustness of our interpretation of the baseline specification.
For our baseline specification the cash flow coefficient is 0.197 for R&D and 0.482 for
physical investment; these coefficients are precisely estimated. The implied elasticities (eval
uated at the means in Table 2 ) are 0.355 and 0.833, respectively. The R 2 statistics for the
R&D and physical investment regressions are 0.43 and 0.30, respectively. Thus, a fairly
large percentage of the within-firm variation, particularly for R&D, is well explained by
within-firm variation in internal finance alone. As already noted, we suspect that the lower
estimated elasticity for cash flow for R&D may reflect the fact that firms “smooth” R&D
expenditures because it is expensive to respond to t r a n s i t o r y movements in cash flow. The
extent to which the within-firm cash coefficient for R&D coefficient reflects this bias is the
focus of subsections 3.2 and 3.3.
Specifications A 2 and B 2 add the change in sales to the baseline specification. This
variable is presumably better than cash flow as a proxy for changes in product demand.
Despite the high degree of correlation between cash flow and the change in sales, this regres25Pakes and Schankerm an (1984, pp. 82-84) review the literature and report e stim a ted gestation lags
of at least one year; R avcnscraft and Scherer (1982) find even longer g estation lags, reporting m ean lags
betw een four and six years. See also the discussion in Griliches (1979, p. 101). R everse ca u sa tio n problem s
are further m itigated by the use of instrum ental variables in Section 3.3.
14
T able 3: W ithin-Firm OLS Regressions
E quation
Independent Variables
Cash Flow
A Sales
T o b in ’s q
df
K2
710
0.43
709
0.46
648
0.46
647
0.46
710
0.30
709
0.31
648
0.30
647
0.31
R&D R egressions
A1
A2
A3
A4
0.197
(0.020)
0.173
(0.022)
0.174
(0.019)
0.160
(0.023)
0.012
(0.009)
0.008
(0.008)
0.002
(0.0007)
0.002
(0.0006)
In vestm en t R egressions
Bl
B2
B3
B4
0.482
(0.053)
0.393
(0.053)
0.490
(0.056)
0.402
(0.055)
0.046
(0.021)
0.049
(0.021)
-0.0005
(0.0015)
-0.0014
(0.0014)
Note: T o b in ’s q is available for only 161 firms.
Note: E stim a ted w ith year and firm dum m ies (not reported).
Note: H eteroskedasticily consistent standard errors reported in p aren th eses
15
sor is not significant in the ll&D regression.26 However, the change in sales is significant in
the physical investment regression, lowering the cash flow coefficient by approximately 20
percent.
In specifications A3 and B3, we include Tobin’s 5 , which is tax adjusted following
Salinger and Summers (1983).27 Our results show that Tobin’s q enters significantly for
our R&D regression but not the physical investment regression .28 The cash flow coefficient
for R&D is reduced by just 10 percent. We considered alternative ways of entering Tobin’s
q
in the regression. For example, we included leads of Tobin’s q to capture the potential
information role that current cash flow might be playing. This had no additional effect
on the results reported in Table 3. Finally, specifications A4 and B4 include both Tobin’s
q
and change-in-sales; in both cases, the reduction in the magnitude and the statistical
significance of cash flow coefficient is still small.
The results in specifications A2-A4 and B2-B4 indicate that the explanatory power of
cash flow is robust to the inclusion of demand variables. This evidence is suggestive, but it
is still possible that sales and Tobin’s q do not completely control for the expectations role
played by cash flow. For this reason, we re-emphasize the summary statistics in Table 2
which reveal that the typical firm in the sample re-invests 10 0 percent of its earnings, yet
obtains little external finance. When combined with this auxiliary evidence, the above
regressions strongly suggest that cash flow is important as a source of finance rather than
as a proxy for firms’ investment opportunities.
3.2
B e tw e e n - F ir m R e s u lts
We have two reasons for presenting between-firm results, in spite of the potential importance
of controlling for unobservable firm effects. One reason is that over th r e e f o u r th s of the
26T he correlation betw een cash flow and the change in sales is 0.68.
27T he d ata required to con stru ct T o b in ’s q are available for only 161 firms.
28Hayashi and Inoue (1989) have pointed ou t th at when there are m ultiple quasi-fixed factors, it is neces
sary to assum e the existen ce o f a cap ital aggregator and then to redefine T o b in ’s q in term s o f th is aggregate.
In their fram ework, T o b in ’s q is a w eighted average of shadow prices of the com p on en ts o f the aggregate
m easure of quasi-fixed factors. Since we wish to consider the determ inants of these factor dem ands sepa
rately, neither the traditional nor this augm ented version of T o b in ’s q is strictly applicable. N evertheless,
we construct a traditional m easure of T o b in ’s q and include it in som e specifications for sake o f com parison
to previous studies.
16
variance of the R&D ratio is in the cross-sectional dimension .29 More importantly, the
between-firm results are of interest because the transitory component of cash flow tends
to average out over time. Hence, these estimates provide evidence on the extent to which
the within-flrm estimates are biased downward due to the unresponsiveness of R&D to the
transitory component of cash flow.
The standard approach to obtaining the between-firm result is to regress the firm-specific
means of the dependent variable on the firm-specific means of the independent variables.
For our panel, this amounts to regressing the 1983-1987 firm average of R&D on the 19831987 firm average of cash flow. To permit direct comparisons with our within-firm results,
Table 4 reports the similar specifications.
The results for our baseline regressions (C l and D l) appear in Table 4 below. The
between-firm estimate of the cash flow coefficient are 0.328 for R&D and 0.306 for physical
investment. These results are consistent with our summary statistics that show that our
firms, on average, allocate roughly equal amounts to R&D and physical investment. The
increase in the R&D coefficient and the nearly offsetting decline in the physical investment
coefficient is consistent with the view that firms smooth R&D expenditures to transitory
shocks in cash flow at the expense of physical investment. The remaining specifications in
Table 4 consider the robustness of the baseline specification. The inclusion of observable
firm characteristics such as average sales and average q may also help control for fixed
firm effects. As is true with the within-firm specifications, the coefficient on Tobin’s q
is significant, and has some effect on the cash flow coefficient for R&D, but the effect of
cash flow is still large and statistically significant. These results are robust to a number of
alternative specifications which we do not report.30
29Griliches and M airesse (1984, p .345), facing a similar situ ation , also present b o th the within-firm and
the betw een-firm results, n otin g th at to not do so can lead to discarding m ost of th e variance in the sam ple.
30Our m ain concern was the possibility of reverse causation from R&D to cash flow caused by regressing
five-year averages on five-year averages. To address this issue, we re-estim ated the betw een-firm regression
on the last three years of the panel, using the first two years of cash flow as in stru m en ts. For our baseline
specification, the cash flow coefficient for ll& D rose by 25 percent, while the coefficient for ph ysical investm ent
fell by 13 percent. In both cases, the coefficients were highly significant.
17
Table 4: B etw een-Firm OLS R egressions
Equation
Independent Variables
Cash Flow
Sales
T ob in ’s q
df
K2
174
0.53
173
0.52
155
0.54
154
0.54
174
0.31
173
0.30
155
0.31
154
0.31
R&D Reg]cessions
Cl
C2
C3
C4
0.328
(0.024)
0.301
(0.034)
0.253
(0.038)
0.232
(0.041)
0.009
(0.010)
0.007
(0.010)
0.004
(0.0016)
0.004
(0.0016)
In vestm en t R egressions
Dl
D2
D3
D4
0.306
(0.048)
0.278
(0.067)
0.282
(0.060)
0.260
(0.076)
0.010
(0.014)
0.008
(0.014)
0.001
(0.0019)
0.001
(0.0019)
Note: T o b in ’s q is available for only 161 firms.
Note: E stim ated w ith industry dum m ies (n o t reported).
Note: H eteroskedasticity consistent standard errors reported in p a ren th eses.
18
3 .3
I n s t r u m e n t a l- V a r ia b le
R e s u lts
We now consider an econometric specification which explicitly recognizes and controls for
the downward bias induced by high adjustment costs for R&D. In particular, assume that
observed cash flow, C F u , consists of a permanent component, C F * t , plus a transitory com
ponent, w n • If R&D expenditures respond primarily to the permanent component of cash,
then the specification in Equation 1 becomes
R D a = (3o
+ P cjC F ? t + a,* + v t + et
*t,
(2 )
where C F { t = C F * t + W{t . We can rewrite this model in terms of observable cash flows as
R D n = Po
+ P c f C F a + a i + v t + e it — P c f w i t •
(3)
Since the w a component of the composite error term in Equation 3 is negatively cor
related with observed cash flow, the within-firm and first-differenced estimates of (3cf are
downward biased .31 Griliches and Hausman (1986) describe a Hausman test for the exis
tence of this bias that compares the wi thin-firm and first-differenced estimates of Equation 3.
They show that under most conditions, the existence of the transitory component will bias
the first-differenced estimate more than the within-firm estimate. In Row 1 of Table 5,
Columns 1 and 2 report the within-firm and first-differenced estimates of Equation 3 for
the full, five-year panel. For R&D, the point estimates of the cash flow coefficient are 0.197
(as reported in Table 3) and 0.133, respectively. Given the precision of these estimates,
it is obvious that the first-differenced estimate is significantly lower. In addition, we also
computed the “long-differenced” estimates and found that the cash flow coefficient was
monotonically increasing in the length of the difference operator.32 As shown by Griliches
and Hausman (1986), these findings provide additional evidence that R&D is unrespon
sive to the transitory component in cash flow, and hence that both the within-firm and
first-differenced estimates understate the effect of cash flow on R&D.33
31 N ote th at this specification is formally identical to the classical errors-in-variables problem .
32For R&D, the second, third and fourth differenced coefficient estim ates were 0.197, 0.211 and 0.233,
respectively.
33T h e first-differenced estim ate for physical investm ent is also lower, although the proportional decline is
19
T able 5: Instrum ental Variable Regressions
Specification
5-year sam ple_____
__________________ 3-year sam p le
1. W ithin
2. F.D .
3. F.D .
4. F .D . (G M M )
5. F .D . (G M M )
(O LS)
(OLS)________ (O LS)
i.n .i.d . Error
M A (1) Error
R&D R egression
1- 0 c ,
2. d . f .
3. Instrum ents
4.
5. Prob. V alue
0.197
(0.020)
0.133
(0.021)
0.085
(0.024)
710
n.a.
n.a.
n.a.
711
n.a.
n.a.
n.a.
355
n.a.
n.a.
n.a.
355
355
CF.2
CF-3
0.78(2)
0.677
0 .1 4 (2 )
0.932
0.482
(0.053)
0.395
(0.053)
0.250
(0.065)
0.476
(0.091)
0.461
(0.208)
710
n.a.
n.a.
n.a.
711
n.a.
n.a.
n.a.
355
n.a.
n.a.
n.a.
0.362
(0.049)
0.344
(0.086)
Investm en t R egression
6. P c
7. d . f .
8. Instrum ents
9. H ( d . f . )
10. Prob. Value
355
355
CF-2
CF-3
1.97(2)
0.369
1.71(2)
0.426
Note: Estimated with year dummies (not reported).
Note: Heteroskedasticity consistent standard errors in parentheses.
Note: Instrument set expands with time to include all valid lags in the panel.
20
In order to obtain consistent estimates of the cash-flow coefficient in Equation 2 , we
follow the research strategy suggested by Griliches and Hausman (1986, p.114). First, firm
effects are removed by first differencing so that Equation 3 becomes
R D a
—
R D it-
1=
P cf(C F it
—
C F it-i)
+ (, —
e*
+
(3cf w i t
—
(4)
where year dummies have been suppressed for clarity. Next, consistent estimates of (3cf
in Equation 4 are obtained using instrumental variables.34 The natural instruments are
lags of cash flow, which are highly correlated with the first difference of current cash flow,
but uncorrelated with the composite error term under the assumption that the transitory
component w a is independently distributed. In this case, all lags of cash flow dated t —2
and earlier are valid instruments. This specification appears in Column 4 of Table 5. In
order to allow for the possibility of serial correlation in the transitory component of cash
flow, we consider an alternative specification in which this component follows an MA( 1 )
process. In this case, lags of cash flow dated t —3 and earlier are uncorrelated with Wtt-i,
and are therefore valid instruments. This specification appears in Column 5.
The results in Columns 4 and 5 are computed using the generalized method of moments
(GMM) estimator developed by Hansen (1982) and White (1982). This estimator is efficient
and allows for conditional heteroskedasticity in the errors.35 For comparison, Column 3 re
m uch sm aller.
34 We n o te th a t the use o f lags of cash flow as instrum ents also provides som e a d d ition al assurance against
the possib ility of reverse cau sation from current R&D to current cash flow.
35T he estim ator applied to Equation 4 is
p =[ W
z h ~ lz ' w
y ' w
' z t i~ lz
x
where Y is the vector of the dependent variable, W is the m atrix of explanatory variables (including year
dum m ies) and Z is the m atrix of instrum ental variables, which includes all available lags o f valid instrum ents
in the panel. T h e rows of Y and W are first stacked by cross section, and then each cross section is stacked
by tim e period. T h e m atrix o f instrum ental variables is block diagonal, where each block is the m atrix
of instrum ental variables corresponding to the respective tim e period (See H oltz-E akin, N ew ey and Rosen
(1988) for further d etails). A consistent estim ate of the elem ent rs of ( Q / N ) is given by
N
( h / N ) r,
= 53(e.veijZ1rZ,-,)/Ar.
?
1=
1
for all r ,s , where the eh are consistent estim ates of the residuals obtained using a first-sta g e, instrum ental
variable estim ate of ft. T h is procedure is valid when E ( e i r e JS) = 0 for all t , i , r ,s such th a t i ^ j, that is,
when the error term is assum ed to be independent over cross sectional u n its. H ence, th is procedure adm its
21
estimates the OLS specification in Column 2 for the shorter panel required for implementing
the GMM estimator.
The instrumental variables estimator in Column 4 is 0.362, which is nearly twice the
within-firm estimate, but very close to the between-firm estimator. Using the summary
statistics in Table 2, this coefficient implies a cash flow elasticity for R&D of 0.670. Thus,
the results in Table 5 indicate that the within-firm results underestimate the effect of cash
flow on R&D, but that the between-firm results do not.
For physical investment, the
estimated coefficient and implied elasticity are 0.476 and 0.822, respectively. In contrast to
the R&D results, these estimates are very close to the within-firm estimates, implying that
physical investment is relatively more responsive to transitory movements in cash flow.
The results in Columns 4 and 5 are essentially identical, which suggests that the tran
sitory component is well represented by an independent process. Additional evidence in
support of the specification and the validity of the instruments is provided by Hansen’s
(1982) chi-squared test of the model’s overidentifying restrictions. These test statistics and
their p-values appear in Rows 4, 5, 9, and 10 of Table 5. This test easily accepts the
specification for both R&D and physical investment.
The instrumental variables results reconcile the difference in magnitude between the
within-firm and between-firm estimates.
Collectively, these results indicate that firms
smooth R&D in response to transitory movements in cash flow because of high adjust
ment costs. For physical investment, adjustment costs do not appear to be as important.
Alternatively, adjustment costs may be important, but this downward bias is roughly offset
by the fact that the higher adjustment costs for R&D induce firms to smooth R&D at the
expense of physical investment.
4
Conclusion
Contrary to previous studies, we find a substantial effect of internal finance on R&D expen
ditures for the firms in our panel. This result is robust to a variety of estimators and control
variables, however the estimated magnitude is sensitive to the econometric specification. In
arbitrary autocorrelation in c ,t .
22
particular, we argue that the conventional within-firm estimator for R&D is downward bi
ased if, because of adjustment costs, firms do not respond to transitory movements in cash
flow. We correct for this problem by following the research strategy outlined by Griliches
and Hausman (1986), and obtain cash flow elasticities for R&D and physical investment of
0.670 and 0.822, respectively. These results are consistent with the view that the principal
determinant of investment for small, high-tech firms is internal finance.
We suspect that in addition to the “adjustment cost” bias noted above, an important
reason why previous studies found no efFect is that they examined large firms that were
unlikely to face significant internal finance constraints. This is because large firms may have
better access to external finance, and typically generate cash flows in excess of investment
needs. For these reasons, our study examines small, high-tech firms. Having done so, it is
important to point out that “small” firms are very important in U.S. manufacturing. Acs
and Audretscli (1988, 1990) demonstrate this point with a set of findings for firms with less
than 500 employees, which is close to our size cutoff of 10 million in capital stock .36 They
report that firms in this size range accounted for 94.2 percent of all firms, 21.4 percent of
sales, and 28.9 percent of employment in manufacturing in 1982. Firms with less than 100
employees accounted for 1 2 . 1 percent of sales and 16.5 percent of employment. Of more
importance for assessing the results of our study, Acs and Audretsch (1988) find that firms
with less than 500 employees accounted for approximately 40 percent of all innovations in
manufacturing in 1982.
We end by mentioning a few avenues for future research. It would be interesting to
estimate the ex p o s t return to R&D for firms of the type we have studied. Several studies
have found high private rates of return to R&D .37 While appropriability problems can
explain high public rates of return, we suggest that financial constraints may explain the
puzzle of high private rates of return. Another avenue is to reconsider the internal finance
effects on R&D for larger firms using panel data techniques. The difficulty in designing
and interpreting the results of such a study is that most large firms are not likely to face
binding finance constraints. We have run exploratory regressions for a panel of Compustat
36In particular, for our sam ple o f firms, the mean level of em ployees is 237 in 1983 and 407 in 1987.
37See for exam ple Griliches (1986) , Jaffe (1986) and Bernstein and Nadiri (1988).
23
firms covering the same time period and the same high-tech industries. We find that while
estimated cash flow coefficients decline quite dramatically for large firms, they do remain
statistically significant.38
38For exam ple, for firms w ith capital stock s betw een $10 and $100 million, the cash flow coefficient
estim ated in first differences w as approxim ately 0.08, slightly more than on e-h alf the size o f the coefficient
reported in Table 5. For firms over 100 million in assets, the casli flow coefficient was approxim ately 0.04,
but still significant. W hile these estim ates are only exploratory, they su ggest th a t s o m e large firms m ay also
face financing constraints for R& D.
24
References
Acs, Zoltan J. and David B. Audretsch (1990), I n n o v a tio n a n d S m a ll F ir m s . Cambridge:
The MIT Press.
Acs, Zoltan J. and David B. Audretsch (1988), “Innovation in Large and Small Firms: An
Empirical Analysis” A m e r ic a n E c o n o m ic R e v ie w 78, 678-690.
Akerlof, George A. (1970), “The Market for ‘Lemons’: Quality Uncertainty and the
Market Mechanism” Q u a r te r ly J o u r n a l o f E c o n o m ic s 84, 488-500.
Arrow, Kenneth J. (1962), “Economic Welfare and the Allocation of Resources for
Invention” in R. R. Nelson, (ed.), T h e R a te a n d D ir e c tio n o f I n v e n tiv e A c t i v i t y :
E c o n o m ic a n d S o c ia l F a c to r s . Princeton: Princeton University Press, 609-625.
Bernanke, Ben and Mark Gertler (1989), “Agency Costs, Net Worth and Business
Fluctuations” A m e r ic a n E c o n o m ic R e v ie w 79, 14-31.
Bernstein, Jeffrey I. (1986), R e s e a r c h a n d D e v e lo p m e n t, T a x I n c e n tiv e s , a n d th e S tr u c tu r e
o f P r o d u c tio n a n d F in a n c in g . Toronto: University of Toronto Press.
Bernstein, Jeffrey I. and M. Ishaq Nadiri (1988), “Financing and Investment in Plant and
Equipment and Research and Development” in M. H. Preston and R. E. Quandt,
(ed.), P r ic e s , C o m p e titio n , a n d E q u ilib r iu m . New York: Phillip Allan, 233-248.
Bernstein, Jeffrey I. and M. Ishaq Nadiri (1989), “Rates of Return on Physical and R&D
Capital and Structure of the Production Process: Cross Section and Time Series
Evidence” in B. Raj, (ed.), A d v a n c e s in E c o n o m e tr ic s a n d M o d e llin g . Dordrecht:
Klewer Academic Publishing, 169-185.
Bester, Helmet (1985), “Screening vs. Rationing in Credit Markets with Imperfect
Information” A m e r ic a n E c o n o m ic R e v ie w 75, 850-855.
Calomiris, Charles W. and R. Glenn Hubbard (1990), “Firm Heterogeneity, Internal
Finance, and Credit Rationing” E c o n o m ic J o u r n a l 1 0 0 , March, 90-104.
Devereux, Michael and Fabio Schiantarelli (1989), “Investment, Financial Factors and
Cash Flow: Evidence from U.K. Panel Data” mimeo.
Elliot, J. W. (1971), “Funds Flow vs. Expectational Theories of Research and
Development Expenditures in the Firm” S o u th e r n E c o n o m ic J o u r n a l 37, 409-422.
Fazzari, Stephen M. and Michael J. Athey (1987), “Asymmeteric Information, Financing
Constraints, and Investment” R e v ie w o f E c o n o m ic S tu d ie s 69, 481-487.
25
Fazzari, Steven, R. Glenn Hubbard and Bruce C. Petersen (1988), “Financing Constraints
and Corporate Investment” B r o o k in g s P a p e r s o n E c o n o m ic A c t i v i t y 1 , 141-195.
Gertler, Mark (1988), “Financial Structure and Aggregate Economic Activity: An
Overview” J o u r n a l o f M o n e y , B a n k in g a n d C r e d it 2 0 , 559-596.
Grabowski, Henry G. (1968), “The Determinants of Industrial Research and Development:
A Study of the Chemical, Drug and Petroleum Industries” J o u r n a l o f P o l i t i c a l
E c o n o m y 76, 292-306.
Griliches, Zvi (1979), “Issues in Assessing the Contribution of Research and Development
to Productivity Growth” B e ll J o u r n a l o f E c o n o m ic s 1 0 , Spring, 92-116.
Griliches, Zvi, (ed.) (1984), R & D , P a te n ts , a n d P r o d u c tiv ity . Chicago: University of
Chicago Press.
Griliches, Zvi (1986), “Productivity, R&D, and Basic Research at the Firm Level in the
1970’s” A m e r i c a n E c o n o m ic R e v ie w 76, 141-154.
Griliches, Zvi and Jerry A. Hausman (1986), “Errors in Variables in Panel Data” J o u r n a l
o f E c o n o m e tr ic s 31, 93-118.
Griliches, Zvi and Jacques Mairesse (1984), “Productivity and R&D at the Firm Level” in
Zvi Griliches, (ed.), R & D , P a t e n t s a n d P r o d u c tiv ity . The University of Chicago
Press, 339-374.
Hall, Bronwyn and Fumio Hayashi (1989), “Research and Development as an Investment”
NBER Working Paper #2973, Cambridge: National Bureau of Economic Research.
Hansen, Lars (1982), “Large Sample Properties of Generalized Method of Moments
Estimators” E c o n o m e tr ic a 50, 1029-1054.
Hayashi, Fumio and Toluu Inoue (1989), “Implementing the Q Theory of Investment in
Micro Data: Japanese.Manufacturing 1977-1985” Osaka University.
Hoshi, Takeo, Anil Ivashyap, and David Scharfstein (1991), “Corporate Structure,
Liquidity and Investment” Q u a r te r ly J o u r n a l o f E c o n o m ic s 106, 33-60.
Henderson, James M. and Richard E. Quandt (1971), M ic r o e c o n o m ic T h e o r y . New York:
McGraw-Hill.
Holtz-Eakin, Douglas, Whitney Newey and Harvey S. Rosen (1988), “Estimating Vector
Autoregressions with Panel Data” E c o n o m e tr ic a 56, 1371-1395.
Hsiao, Cheng (1986), A n a l y s i s o f P a n e l D a ta . Cambridge: Cambridge University Press.
26
Hubbard, R. Glenn and Anil Ivashyap (1990), “Internal Net Worth and the Investment
Process: An Application to U.S. Agriculture” mimeo, Columbia University.
JafFe, Adam B. (1986), “Technological Opportunity and Spillovers of R&D” A m e r ic a n
E c o n o m ic R e v ie w 76, 984-1001.
Kamien, Morton I. and Nancy L. Schwartz (1978), “Self-Financing of an ll&D Project”
A m e r ic a n E c o n o m ic R e v ie w 6 8 , 252-261.
Kamien, Morton I. and Nancy L. Schwartz (1982), M a r k e t S tr u c tu r e a n d I n n o v a tio n .
Cambridge: Cambridge University Press.
Levin, Richard C., Alvin K. Klevorick, Richard R. Nelson, and Sidney G. Winter (1987),
“Appropriating the Returns from Industrial Research and Development” B r o o k in g s
P a p e r s o n E c o n o m ic A c t i v i t y 3, 783-831.
Long and Malitz (1985), “Investment Patterns and Financial Leverage” in B. M.
Friedman, (ed.), C o r p o r a te C a p ita l S tr u c tu r e s in th e U .S . Chicago: University of
Chicago Press, 325-351.
Mairesse, Jacques and Mohamed Sassenou (1991), “R&D and Productivity: A Survey of
Econometric Studies at the Firm Level” NBER Working Paper #3666, Cambridge:
National Bureau of Economic Research.
Meyer, John R. and Edwin Kuh (1957), T h e I n v e s tm e n t D e c is io n : A n E m p ir ic a l S tu d y .
Boston: Harvard University Press.
Mueller, Dennis C. (1967), “The Firm’s Decision Process: An Econometric Investigation”
Q u a r te r ly J o u r n a l o f E c o n o m ic s 81, 58-87.
Mundlak, Yair (1978), “On the Pooling of Time Series and Cross Section Data”
E c o n o m e tr ic a 46, 69-85.
Myers, Stewart C. and Nicholas S. Majluf (1984), “Corporate Financing and Investment
Decisions When Firms have Information that Investors Do Not” J o u r n a l o f
F in a n c ia l E c o n o m ic s 13, 187-221.
Pakes, Ariel and Shmuel Nitzan (1983), “Optimum Contracts for Research Personnel,
Research Employment, and the Establishment of “Rival” Enterprises” J o u r n a l o f
L a b o r E c o n o m ic s 1 , 345-365.
Pakes, Ariel and Mark Schankerman (1984), “An Exploration into the Determinants of
Research Intensity” in Zvi Griliches, (ed.), R & D , P a t e n t s a n d P i'o d u c tiv ity . The
University of Chicago Press, 209-232.
27
R avenscraft, D avid and F. M. Scherer (1982), “T he Lag Structure o f R eturns to Research
and D evelopm en t” A p p l i e d E c o n o m i c s 14, 603-620.
Salinger, Michael A. and Lawrence H. Summers (1983), “Tax Reform and Corporate
Investment: A Microeconomic Simulation Study” in Martin S. Feldstein, (ed.),
B e h a v io r a l S im u la tio n M e th o d s in T a x P o lic y A n a ly s is . University of Chicago
Press, 247-287.
Scherer, F. M. (1965), “Firm Size, Market Structure, Opportunity, and the Output of
Patented Inventions” A m e r ic a n E c o n o m ic R e v ie w 55, 1097-1125.
Scherer, F. M. (1980), I n d u s tr ia l M a r k e t S tr u c tu r e a n d E c o n o m ic P e r f o r m a n c e . Boston:
Houghton Mifflin Company.
Schumpeter, Joseph A. (1942), C a p ita lis m , S o c ia lis m , a n d D e m o c r a c y . New York: Harper
and Row.
Spence, Michael A. (1979), “Investment Strategy and Growth in a New Market” B e ll
J o u r n a l o f E c o n o m ic s 1 0 , 1-19.
Stiglitz, Joseph E. and Andrew Weiss (1981), “Credit Rationing in Markets with
Imperfect Information” A m e r ic a n E c o n o m ic R e v ie w 71, 393-410.
Twiss, Brian (1986), M a n a g in g T e c h n o lo g ic a l I n n o v a tio n . New York: Longman.
W hite, Halbert (1980), “A Ileteroskedasticity-Consistent Covariance Matrix Estimator
and a Direct Test for Heterskedasticity” E c o n o m e tr ic a 48, 817-838.
W hite, Halbert (1982), “Instrumental Variables Regression with Independent
Observations” E c o n o m e tr ic a 50, 483-500.
28
@FedFRASER