R&D investment and internal finance: the cash flow effect

Econ. Innov. New Techn., 2005, Vol. 14(3), April, pp. 213–223 R&D INVESTMENT AND INTERNAL FINANCE: THE CASH FLOW EFFECT CARTER BLOCH∗ The Danish Centre for Studies in Research and Research Policy, University of Aarhus, Finlandsgade 4, 8200 Aarhus N., Denmark (Received 24 February 2004; Revised 22 September 2004; In final form 4 October 2004) This paper investigates the cash flow effect on R&D investments for firms in Denmark. Evidence is found that internal funds are important in explaining R&D investments, indicating that R&D investment decisions are affected by credit market imperfections. Cash flow sensitivities are larger both for smaller firms and for firms with low debt relative to assets. Furthermore, this effect is also present after controlling for cash flow’s potential role as a predictor of future profitability. Keywords: R&D; Cash flow; Internal finance; Financial constraints; Credit market imperfections JEL Classification: G32; O32 INTRODUCTION It is often argued that there is too little investment in R&D. An important reason for this is that much of the ideas and knowledge created by R&D is non-rival; its use by one firm does not preclude its use by another. Firms may then be unable to fully appropriate the gains from their investments, which then leads to underinvestment. However, there may also be another reason why firms’ R&D investment is suboptimal. The nature of R&D investments may affect the access to and the cost of external funding. According to the Miller–Modigliani (Modigliani and Miller, 1958) theorem, optimal levels of investment should be indifferent to capital structure. However, economic theory on information problems offers a number of reasons why this may not be the case. R&D, as with other types of investments, may be characterized by problems of asymmetric information.1 The firm has better information on the likelihood of success and the nature of the project than the investor. The investor may then have difficulties in determining which projects are most promising. Furthermore, reducing information problems in this case is problematic since firms are reluctant to provide more information on their research due to strategic considerations. ∗ Tel.: 1 (45) 8942 2398; Fax: (45) 8942 2399; E-mail: [email protected] Classical references are Akerlof (1970) and Stiglitz and Weiss (1981). ISSN 1043-8599 print; ISSN 1476-8364 online © 2005 Taylor & Francis Group Ltd DOI: 10.1080/1043859042000312710 214 C. BLOCH Myers and Majluf (1984) argue that managers’inside information on the value of investment projects will make them reluctant to seek external finance, instead relying on internal funds. As with other types of investment, external financing of R&D also creates problems of moral hazard;2 due to limited liability, firms may be willing to take on greater risk than otherwise when projects are financed by external funds. These problems are potentially greater for R&D, given that R&D is very difficult to collateralize. An extensive literature3 has examined the presence of financial constraints on investment in fixed capital, most often by examining whether cash flow affects investment. Theories on credit market imperfections show that firms’ net worth may affect both the access to and the cost of external financing. Cash flow functions both as a proxy for firms’ net worth and as a measure of their internal funds available for investment. A smaller number of papers4 have also considered the role of cash flow on R&D investment. R&D and fixed capital investments differ in a number of important ways. First, since a large portion of R&D expenditure is wages and salaries, and the output is mainly knowledge and ideas, R&D is difficult to collateralize. Second, much of the knowledge created is tacit; the firm’s knowledge base sits with its researchers, and in order to maintain it, the firm must keep their staff employed. This implies that there are high adjustment costs for R&D, which also motivates the smoothing of R&D spending. Third, the returns to R&D investment are much more uncertain for investments in fixed capital. The majority of work in this area has been on US data. However, consideration of these issues in other countries is interesting both in its own right and also in comparison with US results. Financial systems and relationships between firms and banks may vary considerably from country to country. In terms of broad characteristics, the Danish financial system resembles most the Continental European model, in which banks have the central role in providing financing for businesses.5 The stock market, although having grown both in size and importance in recent years, is still relatively small in Denmark. Bonds issued by the government and mortgage credit institutions are the most dominant type of financial assets in Denmark, though the issuance of bonds by private businesses is very limited. By far the most prevalent form of financing for firms is bank loans. Though, there are important differences between the Danish financial system and universal banking systems in, e.g., Germany. In particular, legal regulations for financial institutions greatly restrict the influence that banks may have on firms.6 Thus, the close relationships between banks and firms that are encountered in other bank-based systems are absent. On the basis of this, it may be expected that credit market imperfections are more prevalent in Denmark than in other bank-based systems. The objective of this paper is to investigate the cash flow effect on R&D investments in Danish firms. This is done by examining two issues. The first is whether there is a relationship between cash flow and R&D investments. If firm borrowing is subject to financial imperfections, then R&D expenditures will be dependent on firms’ own cash flow. Thus, cash flow’s ability to explain R&D investments may provide evidence of financial imperfections. The second issue involves isolating the effects of credit market imperfections from demand shifts, or changes in investment opportunities. For example, cash flow may also function as an indicator of future profitability (for example, an increase in cash flow may generate 2 See, e.g. Holmstrom and Tirole (1997) and Townsend (1979). Examples are Fazzari et al. (1988), Gilchrist and Himmelberg (1995). 4 Among them, Hao and Jaffe (1993), Himmelberg and Petersen (1994), Hall (1992), Mulkay et al. (2001), Boughaes et al. (2003). A review of literature on the financing of R&D can be found in Hall (2002). 5 See, e.g. Østrup (1994) and Andersen et al. (2001). 6 See, e.g. Østrup (1994). 3 R&D INVESTMENT AND INTERNAL FINANCE 215 expectations of greater profits in the future), which suggests that cash flow may help explain R&D investments simply because it forecasts future earnings, and not due to financial constraints. This second issue is addressed in two ways. First, following Fazzari et al. (1988) and Hao and Jaffe (1993), subsets of the sample are examined that, a priori, are considered more likely to be subject to financial constraints. If cash flow effects are stronger for these subsets, then this provides evidence that at least part of cash flow effects on R&D investment are due to credit market imperfections. Second, both sales and an empirical measure of Tobin’s Q are used to control for demand effects. In addition, using a simplified approach based on that used to examine capital investment in Gilchrist and Himmelberg (1995), an estimate of expected future profitability, ‘Expected Q’, controlling for, among other things, cash flow’s predictive power for future earnings is constructed. This makes it possible to distinguish between the two interpretations. A finding that cash flow still helps to explain R&D investments after controlling for its ability to forecast future earnings provides strong evidence of a cash flow effect due to credit market imperfections. These questions are examined using a panel of Danish firms listed on the Copenhagen Stock Exchange. The dataset draws on a comprehensive database compiled by the Danish Centre for Studies in Research and Research Policy on R&D activities for Danish firms from 1989 to 2001. This R&D data has then been combined with accounting and stock market data for all non-financial firms listed on the Copenhagen Stock Exchange.7 The cash flow approach for measuring credit market imperfections has been the subject of criticism, most notably by Kaplan and Zingales.8 Among the most important of these are first, that subgroups should not be fixed over the entire period since financial constraints may not be binding in all periods. Related to this is the problem of endogeneity; whether financial constraints are binding will influence sample selection criteria. Second, Kaplan and Zingales argue that there is no solid theoretical argument for a monotonic relationship between cash flow and investment. For example, financially distressed firms are likely to have lower cash flow sensitivities. This paper uses a priori classifications of firms (i.e. fixed subgroups) for the following reason. Although it may well be the case that financial constraints are not always binding, there is good reason to believe that credit market imperfections may affect firms’ behavior even when they are not binding. A forward-looking firm will try to avoid costly adjustment to its investments as a result of financial constraints and may maintain a buffer stock of cash and other liquid assets. In this way, credit market imperfections may be equally relevant both when constraints are actually binding and when they are not. Furthermore, by this reasoning, the endogeneity problem is not an issue, as this rests on the argument that subgroups should be allowed to vary over time (i.e. selected according to whether constraints are actually binding). The paper also considers the issue of financially distressed firms and monotonicity empirically, and does not find this to be a problem for the data here, since there are only a small number of distressed firm observations in the sample,9 and the removal of these firms from the sample does not have a significant effect on the results.10 7 Annual account data are from the Account Database from the Copenhagen Business School. Stock market data is obtained from the Danish Stock Database from the Center of Analytical Finance at the Aarhus School of Business. A more detailed description of the R&D data is given in the appendix. 8 See Kaplan and Zingales (1997, 2000) and the reply to their arguments in Fazzari et al. (2000). Schiantarelli (1996) and Hubbard (1998) contain general discussions of a number of the methodological issues concerning credit market imperfections and investment. 9 A firm was considered distressed if it had either a negative or very low cash flow to assets ratio. 10 These changes are not reported in the paper, though they can be obtained from the author. 216 C. BLOCH If is found that in general, cash flow has a positive, significant effect on R&D investments. Furthermore, when subsets are considered that, a priori, are more likely to be subject to credit market imperfections, this effect is stronger (i.e. the coefficient on cash flow increases). This is the case, both when comparing small and large firms and when comparing firms with low and high levels of debt. A bootstrapping exercise provides support that differences in coefficient estimates for these subsets are statistically significant. In addition, when Expected Q is used instead of Tobin’s Q in the regressions, cash flow still has a significant effect on R&D investments, in particular for small and low debt firms. These results give a strong indication both of the presence of a cash flow effect on R&D, and that it reflects the presence of credit market imperfections. THE MODEL The empirical model that will be used in this paper is RDit CFit Sit = β0 + β1 Qit + β2 + β3 + eit TAit TAit TAit (1) where RDit is R&D investment for firm i in period t, TA is the book value of total assets, Q is an empirical proxy of Tobin’s Q, CF is cash flow,11 S is sales and e is an error term. I use a standard empirical measure of Tobin’s Q, defined as the ratio of the market value to the book value of the firm’s total assets.12 Q and TA are values at the beginning of the period, whereas CF, RD and S are values during the period. As indicated by Eq. (1), cash flow, sales and R&D are normalized by total assets. As noted above, the finding that firms’ cash flow helps explain R&D investment can be interpreted in two ways: • The effect of cash flow on R&D investment is due to firms’limited access to external finance. • R&D investment’s positive association with cash flow is not due to credit market imperfections, but is instead due to cash flow’s role as a predictor of firms’ investment opportunities. Hence, having found a ‘cash flow effect,’ it is equally important to identify the role of cash flow in affecting R&D investment. The inclusion of an empirical measure of Tobin’s Q is designed to capture the effect of investment opportunities; given that Tobin’s Q theoretically is, under certain conditions, a sufficient statistic13 for the profitability of investment, any additional effects of cash flow after controlling for Q are argued to be due to financial constraints. Adjustment costs and other factor specific characteristics imply that investments in fixed capital, working capital and R&D may respond to changes in different ways. And, in efforts to equate expected marginal returns, investment decisions for each factor may interact. Although the Tobin’s Q model has mainly been used to model fixed investment,14 empirical measures of Tobin’s Q function as an indicator of the value of all assets and thus can be used to control for demand effects for both fixed and R&D investment. 11 Cash flow is defined as operating income before R&D expenditures, plus liquid assets. More precisely, Q is calculated as the market value of equity plus the book value of long-term debt, divided by the book value of total assets. 13 That is, average Q, which empirical approximations are a proxy of, is equal to marginal Q. See Hayashi (1982). 14 Though, see Himmelberg and Petersen (1994). 12 R&D INVESTMENT AND INTERNAL FINANCE 217 There may also be concerns on the quality of empirical approximations of Tobin’s Q. For a number of factors, such as imperfect competition, result of Hayashi (1982) may not hold, resulting in a wedge between marginal and average Q. In addition, if empirical measures are a poor proxy of the theoretical measure of Q, then cash flow effects can be argued to supplement the empirical measure of Q as a predictor of future investment opportunities. This issue is addressed in the following ways.The addition of sales to the model is intended to capture the wedge between average and marginal Q.15 In addition, as noted earlier subsamples of the dataset are examined to determine whether cash flow effects are stronger for firms that, a priori, are considered likely to be subject to credit market imperfections. In order to examine the significance of coefficient differences among subsets, the author employs a bootstrapping procedure to estimate the sampling variance of coefficient estimates which are used to test whether cash flow coefficients are different.16 A low empirical p-value provides support that the observed differences are in fact significant. In order to address potential shortcomings in the empirical measure of Q, the expected value of Q17 is estimated using cash flow, sales, and other variables. If cash flow still affects R&D investment after controlling for its effect on expectations of future profitability, then this provides further evidence of a cash flow effect on R&D investment. THE DATA The dataset comprises non-financial firms listed on the Copenhagen Stock Exchange over the period, 1989–2001. In all, between 250 and 300 firms are listed on the stock market, with a total market capitalization of DKK 738 billion (approximately EUR 100 billion) at end-2001, which is equivalent to ∼55% of Denmark’s GDP in 2001.18 R&D expenditures among publicly traded firms in Denmark have increased steadily over the period, more than doubling.19 In order to focus on firms that are active in research and development, only firms with average R&D expenditures >1% of the book value of total assets are included. This yields an unbalanced panel consisting of 63 firms with a total of 390 observations. The impact of credit market imperfections on R&D investments may vary from sector to sector. Concerning firms’ innovative activities, there may be a number of characteristics that are specific to individual sectors,20 such as the degree of uncertainty for R&D investments, time lags between research and final products, the share of basic vs. applied R&D and appropriability. These factors may play a significant role in affecting firms’ access to external financing. Table I shows some summary statistics for the sample. Here, firms are classified by sector according to the General Industry Classification System.21 One can see that mean R&D intensities are much higher for Biomedicals and IT/Telecommunications, as are mean values for Tobin’s Q and cash flow ratios. Though, there is very large variation in values within each sector. 15 See, e.g. Fazzari et al. (1988) and Hubbard (1998). See also Cleary (1999). 17 That is, the expected value of Q in period t + 1 based on information in period t. 18 As a reference, the market capitalization of the NYSE (source: NYSE’s web site) was slightly higher than US GDP in 2001 (GDP data from OECD). 19 R&D among firms listed on the CSE amounts to a little under half of total private sector R&D in Denmark, which was DKK 21.9 billion. See Danish Centre for Studies in Research and Research Policy (2003). 20 See, e.g. Malerba and Orsenigo (1997). 21 Where, due to the number of observations, some sectors have been combined [industrials (20), materials (15) and utilities (55); consumer discretionary (25) and consumer staples (30); information technology (45) and telecommunications (50)]. 16 218 C. BLOCH TABLE I Variable Market capitalization∗ Sales R&D investment† Tobin’s Q† Long-term debt† Cash flow† Sector Mean Standard error Mean Standard error Mean Standard error Mean Standard error Mean Standard error Mean Standard error Number of firms Number of observations Summary statistics. Industrials, materials, utilities Consumption Biomedical IT/Telecommunications 1801 (4324) 1.277 (0.618) 0.022 (0.017) 1.066 (0.977) 0.170 (0.096) 0.264 (0.089) 34 213 2146 (5050) 1.368 (0.258) 0.025 (0.022) 0.886 (0.284) 0.118 (0.050) 0.255 (0.103) 10 67 12217 (20482) 0.661 (0.485) 0.158 (0.141) 2.682 (1.433) 0.075 (0.064) 0.520 (0.152) 9 58 7998 (19974) 0.822 (0.413) 0.089 (0.076) 2.565 (2.496) 0.121 (0.103) 0.390 (0.329) 10 52 ∗ In millions of DKK. Normalized by the book value of total assets. Tobin’s Q and cash flow as defined above. Sector means and standard errors calculated from firm averages and not from individual observations. † TABLE II Correlations for selected variables. Cash flow Cash flow (−1) R&D Earnings Earnings + R&D Q Q (+1) Sales R&D 0.481 −0.128 −0.360 0.717 0.020 0.139 0.456 0.351 0.535 0.422 0.218 0.471 0.201 −0.169 −0.141 0.397 0.299 Earnings is operating income before taxes and extraordinary expenses. Earnings + R&D is earnings before the deduction of R&D expenditures. Cash flow, earnings, R&D and sales are normalized by (beginning of period) total assets. (Beginning of period) Q is as defined above. (−1) and (+1) indicate year before and year ahead values, respectively. Table II shows the correlations of cash flow, earnings, sales and R&D. Of particular interest here is the association of cash flow with future performance in terms of earnings and sales. As Table II indicates, cash flow is positively correlated with R&D, Q and same-year earnings and sales, but has either almost no or a negative correlation with year-ahead earnings or sales. These correlations thus do not provide evidence that cash flow may function as an indicator of future profitability for firms in the sample. THE RESULTS The choice of estimation method should consider two potential sources of bias: individual firm effects and the correlation of explanatory variables with the error term. For example, a shock to R&D investments may also be assumed to affect cash flow and sales. The potential correlation between current variables and the error term argues for the use of generalized method of moments as an estimation method. However, the small size of this sample and the large variation in the number of observations for each individual firm makes this approach problematic. It may therefore be difficult to attain a reasonable degree of precision for coefficient estimates, giving large standard errors. This was indeed what was found to be the case for the sample here. R&D INVESTMENT AND INTERNAL FINANCE 219 This paper will instead focus on results using a fixed effects estimation method.22 This allows for individual firm effects, but does not control for potential bias with the explanatory variables. As a control, the model was also estimated using the Hausman–Taylor (HT) instrumental variables approach.23 The advantages of the HT approach relative to a fixed effects model are that it captures within firm variation and it allows the examination of sector effects. The main disadvantage is that if there are substantial individual firm effects, it may product biased and inconsistent estimates. The Hausman specification test provides a method for assessing whether this bias exists.24 The results from the HT approach were fairly similar to those with fixed effects, giving an indication that the fixed effects estimates are fairly reliable. Results using the HT approach are included in the appendix for comparison. Table III shows the results for estimation of the model using the fixed effects method. The model is estimated for the full sample and for subsets of firms that, a priori, may be considered likely to be constrained. The model is also estimated using the estimate of expected future profitability, ‘Expected Q’. In constructing this estimate, future values of Tobin’s Q are registered on current and lagged values of Q, cash flow, sales and long-term debt, and the predicted values are then saved. Expected Q(EQ) then incorporates, among other things, cash flow’s ability to forecast expected future earnings. To compare coefficient estimates across subsamples, simulations were performed for each of the regressions. A bootstrapping procedure was used to form empirical estimates of the sampling variance of the cash flow coefficients. These variances were then used to examine whether differences in cash flow coefficients are significant. p-Values from these tests are shown in Table IV. In the full sample, all variables are positive and significant in the regressions with Q. In the regressions with EQ, the coefficient of cash flow falls both in size and significance, and is insignificant at the 10% level. However, cash flow is still significant for the regression using the HT approach (see Appendix A), where sector dummies for both biomedicals and IT are significant.25 This suggests that sector differences may be important in terms of cash flow effects on R&D investment. Small firms can be considered more likely to suffer from the information problems that cause credit market imperfections and may be more dependent on banks for external financing.26 Additionally, low levels of debt may be an indication that a firm has limited access to external funding. A comparison of results for small firms with those of large firms (or the full sample) indicates both that cash flow effects are stronger for smaller firms and that this effect is, at least in part, due to credit market imperfections. Coefficient estimates for cash flow increase (compared with the full sample) when considering small firms. Furthermore, when EQ is included instead of Q, the coefficient on cash flow declines but is still significant at the 10% level. Coefficient estimates of cash flow for large firms decline greatly both in size and in significance. Neither cash flow, Q, nor Expected Q are significant at the 10% level in regressions for large firms using the fixed effects model, while sales, in contrast to that for small firms, is 22 The regression method used is least squares with dummy variables. Mulkay et al. (2001) also use a fixed effects estimator for similar reasons. In addition, fixed effects estimation is the most commonly used method for cash flow analyses using the Tobin’s Q model (although GMM is often used for analysis using the Euler equation approach). Among these are Bougheas et al. (2003), Kaplan and Zingales (1997), Fazzari and Petersen (1993), Himmelberg and Petersen (1994), Cleary (1999) and Hao and Jaffe (1993). An exception is Gilchrist and Himmelberg (1995), who use GMM. 23 See Hausman and Taylor (1981). This approach involves estimation of a random effects model where instruments are used for those variables that are assumed to be correlated with the error term. 24 See Hausman (1978) or discussions in Hsiao (1986) and Greene (2003). 25 Coefficient estimates for sector dummies are not shown, though they are available on request from the author. 26 Where as larger firms, for example, may be able to issue their own bonds. 220 C. BLOCH TABLE III Cash flow Q EQ Regression results. Sales Constant 0.0081∗ (0.0034) 0.0111∗ (0.0035) −0.0026 (0.0091) 0.0013 (0.0091) 0.0135∗ (0.0017) 0.0142∗ (0.0018) 0.0065 (0.0054) 0.0105∗∗ (0.0056) −0.0024 (0.0050) −0.0026 (0.0051) 0.0112∗∗ (0.0057) 0.0177∗ (0.0060) 0.0050 (0.0036) 0.0042 (0.0122) 0.014∗ (0.0046) 0.0075 (0.0081) 0.0241∗ (0.0112) −0.0028 (0.0157) 0.0095∗ (0.0025) 0.0052 (0.0052) 0.0202∗ (0.0083) −0.0061 (0.0140) 0.0231∗ (0.0065) 0.0287∗ (0.0122) 0.0270∗ (0.0083) −0.0157 (0.0158) 0.0099∗ (0.0036) 0.0224∗∗ (0.0136) Number of observations df (adj) R 2 390 315 0.828 390 315 0.831 197 150 0.822 197 150 0.826 193 153 0.890 193 153 0.890 177 134 0.833 177 134 0.838 213 169 0.724 213 169 0.723 191 147 0.807 191 147 0.816 211 168 0.728 179 135 0.835 Dependent variable: R&D investment Full Full Small Small Large Large Low debt Low debt High debt High debt High R&D High R&D Low cash High cash 0.0330∗ (0.0109) 0.0179 (0.0120) 0.0692∗ (0.0221) 0.0439∗∗ (0.0240) 0.0070 (0.0061) 0.0040 (0.0068) 0.0570∗ (0.0215) 0.0381∗∗ (0.0225) 0.0098 (0.0092) 0.0148 (0.0136) 0.0359∗ (0.0174) 0.0067 (0.0193) 0.0506∗ (0.0150) 0.0331∗ (0.0164) 0.0056∗ (0.0021) 0.0230∗ (0.0065) 0.0091∗ (0.0038) 0.0369∗ (0.0115) 0.0012 (0.0014) 0.0048 (0.0041) 0.0051∗∗ (0.0030) 0.0255∗ (0.0091) 0.0030 (0.0049) −0.0042 (0.0134) 0.0099∗ (0.0035) 0.0417∗ (0.0105) −0.0006 (0.0025) 0.0095∗ (0.0034) Notes: Estimation method: fixed effects (least squares with dummy variables), using both firm and time dummies (coefficients for fixed effects not shown). Heteroskedasticity consistent standard errors are in parentheses. ∗ Indicates significance at the 5% level, and indicates significance at the 10% level. EQ is Expected Q in the next period, based on current values of cash flow, sales, long-term debt, and Q. Cash flow, R&D investment and sales are normalized by the book value of total assets. Full is the entire sample. Small is all firms with average sales less than one billion DKK. Low debt consists of firms with average long-term debt (relative to total assets) less than the sample median. High (low) cash are firms with average cash flow ratios greater (less) than the sample median value. High R&D is firms with an average R&D to Sales ratio greater than median. ∗∗ TABLE IV Empirical p-values for differences in cash flow coefficients. Small minus large Small minus full Low debt minus high debt Low debt minus full Low cash minus high cash High R&D minus low R&D 9.533(0.000) 5.000(0.000) 5.105(0.000) 3.073(0.001) 9.540(0.000) 4.198(0.000) 4.341(0.000) 3.286(0.000) 0.324(0.373) 5.194(0.000) 1.498(0.068) 0.034(0.487) Regressions including Q EQ Notes: ‘Q’ and ‘EQ’ indicate regressions using Tobin’s Q and Expected Q, respectively. Shown are t-statistics for differences in cash flow coefficients for subsamples, e.g. ‘small minus large’ is the difference between the cash flow coefficient for small firms and that for large firms. Statistics are t-distributed with 198 degrees of freedom. p-Values in parentheses indicate the probability that coefficient differences are less than or equal to zero (i.e. a low p-value indicates the difference is positive and significant). Statistics are based on bootstrapping exercise with 100 replications, where an empirical sampling variance is estimated. highly significant. Though using HT estimations, which account for sector differences, both Q and EQ are significant.27 Empirical p-values from Table IV indicate that the difference between coefficient estimates for cash flow is significant both between small and large firms, and between small firms and the full sample. A potential explanation for lower cash flow sensitivities for large firms is that if these firms experience less volatility in sales and R&D investment, then R&D may be relatively less sensitive to changes in cash flow. This would imply that larger cash flow coefficients for small 27 See Table AI. R&D INVESTMENT AND INTERNAL FINANCE 221 TABLE V Volatilities for small and large firms. Large firms Small firms R&D Sales Total assets Cash flow 0.375 (0.258) 0.358 (0.242) 0.213 (0.204) 0.247 (0.190) 0.309 (0.229) 0.263 (0.218) 0.389 (0.325) 0.494 (0.309) Notes: Volatility is calculated for each firm as variance over the sample period divided by mean values. Values shown in the table are sample means (with standard deviations in parentheses) for these firm level measures of volatility. firms may not be due to credit market imperfections. To examine this, Table V shows average volatility for R&D, sales, cash flow and total assets. Volatility for R&D, sales and total assets are very similar for small and large firms, whereas cash flow volatility is higher for small firms. Thus, there is no indication from this that large firms have less volatility in sales and R&D investments. Results for subsamples according to debt levels are similar to those for firm size. For firms with low debt, the coefficient on cash flow is larger compared with the full sample, and is still significant (at the 10% level) when EQ is included. For the ‘high debt’ subsample, none of the variables are significant. The fact that cash flow coefficients increase (compared with the full sample) for low debt firms and, more importantly, that the coefficient estimate is significant when EQ is included, provide support for the presence of a cash flow effect. Also here, empirical p-values indicate that the difference between cash flow coefficients for low and high debt firms (and between low debt firms and the full sample) is significant. Highly R&D intensive firms may be considered riskier investments. This could then imply that financial constraints are greater for these firms. Results are mixed here. The coefficient for cash flow is only moderately higher using Tobin’s Q, and for Expected Q it is insignificant. A possible explanation here is that the group of highly R&D intensive firms also includes a number of large, well-established firms that may be much less affected by credit market imperfections. Also included in Table III are subsamples divided according to cash flow ratios. The critique of Kaplan and Zingales (1997) rests in part28 on the argument that firms with large cash flow may be considered less likely to be financially constrained, yet they find that these firms have the highest cash flow sensitivities. It can be noted from Table III that cash flow coefficients are in fact lower here for firms with high cash flow positions. This suggests that the results discussed earlier for ‘a priori constrained’ firms (i.e. small or with low debt) are not due to the fact that these groups have higher levels of cash flow. CONCLUSION This paper has examined the cash flow effect on R&D investments for Danish firms using the ‘Tobin’s Q approach.’ It has examined whether cash flow helps to explain R&D investment when Tobin’s Q is used as a measure of future investment opportunities. Two approaches have been used to overcome the problem that empirical measures of Tobin’s Q may not be adequate proxies for the theoretical measure: considering subsets of ‘a priori constrained firms’, and controlling for cash flow’s ability to predict future profitability. Evidence is found of a cash flow effect on R&D investments both for smaller firms and for firms with low debt. Furthermore, this cash flow effect is still present when including Expected Q, which controls for cash flow’s role as a predictor of future investment opportunities. These results are based on R&D intensive firms that are publicly traded. Though, the results suggest that financial 28 Their classifications take into account other factors, among them management statements on liquidity positions in annual reports, though firms’ cash positions are a central factor. 222 C. BLOCH restrictions may be even greater for smaller firms. These results are also qualitatively similar to studies using US data. However, owing to the fact that model specifications are not the same, a straightforward comparison of quantitative estimates is not fully feasible. Acknowledgements Comments from Ebbe Graversen, the editor and two anonymous referees are gratefully acknowledged. References Akerlof, G.A. (1970) The Market for ‘Lemons’: Quality, Uncertainty, and the Market Mechanism. Quarterly Journal of Economics, 84, 488–500. Andersen, T.M., Dalum, B., Linderoth, H., Smith, V. and Westergård-Nielsen, N. (2001) The Danish Economy: An International Perspective. DJØF Publishing: Copenhagen. Bougheas, S., Gorg, H. and Strobl, E. (2003) Is R&D Financially Constrained? Theory and Evidence from Irish Manufacturing. Review of Industrial Organization, 22, 159–174. Cleary, S. (1999) The Relationship Between Firm Investment and Financial Status. The Journal of Finance, 54, 673–692. Danish Centre for Studies in Research and Research Policy (2003a) R&D Statistics 2001: Research and Development in the Private Sector (in Danish). Fazzari, S. and Petersen, B.C. (1993) Working Capital and Fixed Investment: New Evidence on Financing Constraints. Rand Journal of Economics, 24, 328–342. Fazzari, S.R. Hubbard, G. and Petersen, B.C. (1988) Financing Constraints and Corporate Investment. Brookings Papers on Economic Activity, 1, 141–195. Fazzari, S.R. Hubbard, G. and Petersen, B.C. (2000) Investment–Cash Flow Sensitivities are Useful: A Comment on Kaplan and Zingales. Quarterly Journal of Economics, 115, 695–705. Gilchrist, S. and Himmelberg, C.P. (1995) Evidence on the Role of Cash Flow for Investment. Journal of Monetary Economics, 36, 541–572. Greene, W. (2003) Econometric Analysis, 5th ed. Prentice Hall Englewood Cliffs: New Jersey. Hall, B. (1992) Investment and Research and Development and the Firm Level: Does the Source of Financing Matter? Working Paper 4906, NBER. Hall, B. (2002) The Financing of Research and Development. Oxford Review of Economic Policy, 18, 35–51. Hao, K.Y. and Jaffe, A.B. (1993) Effect of Liquidity on Firms’ R&D Spending. Economics of Innovation and New Technology, 2, 275–282. Hausman, J.A. (1978) Specification Tests in Econometrics. Econometrica, 46, 1251–1271. Hausman, J.A. and Taylor, W.E. (1981) Panel Data and Unobservable Individual Effects. Econometrica, 49, 1377–1398. Hayashi, F. (1982) Tobin’s Marginal q and Average q: A Neoclassical Interpretation. Econometrica, 50, 215–224. Himmelberg, C.P. and Petersen, B.C. (1994) R&D and Internal Finance: A Panel Study of Small Firms in High-Tech Industries. The Review of Economics and Statistics, 76, 38–51. Holmstrom, B. and Tirole, J. (1997) Financial Intermediation, Loanable Funds, and the Real Sector. Quarterly Journal of Economics, 112, 663–691. Hsiao, C. (1986) Analysis of Panel Data. Cambridge University Press: Cambridge. Hubbard, R.G. (1998) Capital–Market Imperfections and Investment. Journal of Economic Literature, 36, 193–225. Kaplan, S.N. and Zingales, L. (1997) Do Investment–Cash Flow Sensitivities Provide Useful Measures of Financing Constraints? The Quarterly Journal of Economics, 112, 169–215. Kaplan, S.N. and Zingales, L. (2000) Investment–Cash Flow Sensitivities are Not Valid Measures of Financing Constraints. The Quarterly Journal of Economics, 115, 707–712. Malerba, F. and Orsenigo, L. (1997) Technological Regimes and Sectoral Patterns of Innovative Activities. Industrial and Corporate Change, 6, 83–117. Modigliani, F. and Miller, M.H. (1958) The Cost of Capital, Corporation Finance and the Theory of Investment. American Economic Review, 48, 261–297. Mulkay, B., Hall, B.H. and Mairesse, J. (2001) Firm Level Investment and R&D in France and the United States. In Herrmann, Heinz and Rolf Strauch (eds) Investing Today for the World of Tomorrow. Springer-Verlag. Myers, S.C. and Majluf, N.S. (1984) Corporate Financing and Investment Decisions when Firms have Information that Investors Do Not. Journal of Financial Economics, 13, 187–221. Østrup, F. (1994) The Financial System in Denmark (in Danish). Institute of Finance, Copenhagen Business School. Schiantarelli, F. (1996) Financial Constraints and Investment: Methodological Issues and International Evidence. Oxford Review of Economic Policy, 12, 70–89. Stiglitz, J.E. and Weiss, A. (1981) Credit Rationing in Markets with Imperfect Information. American Economic Review, 71, 393–410. Townsend, R. (1979) Optimal Contracts and Financial Markets with Costly State Verification. Journal of Economic Theory, 21, 265–293. R&D INVESTMENT AND INTERNAL FINANCE 223 APPENDIX A A.1 R&D Data for Non-financial Firms on the Copenhagen Stock Exchange This section describes the compilation of R&D data for firms in the sample. The data are taken from the Danish Centre for Studies in Research and Research Policy’s (CFA) database for private sector R&D. For firms that have not reported any data to the survey, their annual reports were examined. If R&D expenditures were reported in their annual reports, then these numbers were used. In the case where firms did not report any statistics to CFA, nor were there any R&D expenditures listed on their annual reports, then these data were considered as missing. The R&D database at CFA contains data for every other year from 1989 to 1997 and annually from 1997 onwards. For the years missing between 1989 and 1997 (i.e. 1990, 1992, 1994 and 1996), R&D data was interpolated from years before and after. A.2 Additional Regression Results TABLE AI Results using the HT instrument variables approach. Cash flow Q EQ Sales Constant 0.0049 (0.0033) 0.0087∗ (0.0035) −0.0088 (0.0089) −0.0047 (0.0092) 0.0129∗ (0.0016) 0.0141∗ (0.0017) 0.0008 (0.0052) 0.0050 (0.0055) 0.0008 (0.0050) 0.0011 (0.0050) 0.0068 (0.0056) 0.0135∗ (0.0060) 0.0052∗∗ (0.0029) −0.0122 (0.0114) 0.0133 (0.0103) −0.0016 (0.0113) 0.0371∗ (0.0175) 0.0194 (0.0192) −0.0044 (0.0063) −0.0090 (0.0070) 0.0231 (0.0205) 0.0043 (0.0224) 0.0112 (0.0088) 0.0109 (0.0114) 0.0358∗∗ (0.0221) 0.0134 (0.0242) 0.0035 (0.0067) 0.0739∗ (0.0203) Number of observations Hausman (p-value) 390 4.169 (0.244) 15.447 (0.001) 7.518 (0.057) 34.905 (0.000) 2.254 (0.521) −9.537 (0.023) 11.256 (0.010) 84.289 (0.000) 0.384 (1.000) 0.344 (1.000) 9.145 (0.027) 50.307 (0.000) 0.094 (0.993) 22.999 (0.000) Dependent variable: R&D investment Full Full Small Small Large Large Low debt Low debt High debt High debt High R&D High R&D Low cash High cash 0.0465∗ (0.0107) 0.0277∗ (0.0118) 0.0769∗ (0.0210) 0.0527∗ (0.0234) 0.0097∗∗ (0.0060) 0.0040 (0.0067) 0.0839∗ (0.0210) 0.0671∗ (0.0222) 0.0071 (0.0090) 0.0066 (0.0143) 0.0523∗ (0.0172) 0.0233 (0.0194) 0.0498∗ (0.0143) 0.0537∗ (0.0166) 0.0088∗ (0.0020) 0.0338∗ (0.0065) 0.0114∗ (0.0035) 0.0442∗ (0.0115) 0.0025∗∗ (0.0013) 0.0094∗ (0.0039) 0.0074∗ (0.0028) 0.0320∗ (0.0093) 0.0016 (0.0048) 0.0018 (0.0149) 0.0129∗ (0.0033) 0.0501∗ (0.0108) 0.00004 (0.0022) 0.0140∗ (0.0032) 390 197 197 193 193 177 177 213 213 191 191 211 179 Notes: Estimation method: HT method (random effects with instrument variables). Time and sector dummies included in the regressions. Regressions performed under the assumption that sales and cash flow may be correlated with the error term (and were thus instrumented), while Q is strictly exogenous. Note that since EQ is based in part on current cash flow and sales, it cannot be assumed strictly exogenous. However, since at least one variable must be considered strictly exogenous in the regressions, EQ is not instrumented. Heteroskedasticity consistent standard errors are in parentheses. ∗ Indicates significance at the 5% level, and ∗∗ indicates significance at the 10% level. Hausman is the Hausman specification test statistic (chi-squared distributed under the null hypothesis, with degrees of freedom equal to the number of regressors). High values indicate bias and inconsistency for the random effects estimator.