Financing R&D in mature companies: An empirical analysis

Econ. Innov. New Techn., 2003, October, Vol. 12(5), pp. 425–447 FINANCING R&D IN MATURE COMPANIES: AN EMPIRICAL ANALYSIS ASHER A. BLASSa,* and OVED YOSHAb,y a Research Department, Bank of Israel; bBerglas School of Economics, Tel Aviv University, 69978 Tel Aviv, Israel (Received 14 December 2001; Revised 20 August 2002; In final form 12 September 2002) We study financing patterns of publicly traded R&D-intensive manufacturing firms in Israel. We further characterize R&D-intensive firms by size, physical capital intensity, and whether they issued stocks in the United States, asking whether these features are associated with particular financing patterns. To address these issues, we present, for the first time, adjusted flow of funds charts that treat R&D expenses as a capital outlay (rather than an operating cost that reduces profits, as standard accounting principles prescribe). We also address the question of how R&D inputs should be measured – using R&D expenses or R&D personnel. We construct both expenditure- and personnel-based R&D measures for each firm in our sample, and investigate to what extent these measures are mutually consistent. Keywords: Cash flow; Equity; Financing; Flow of funds; Government grants and subsidies; R&D JEL Classification: D21, H25, G32, G38, L6, O32 1 INTRODUCTION We investigate whether financing patterns of publicly traded R&D-intensive firms differ from those of comparable firms that are not meaningfully engaged in R&D. There are three major reasons for expecting such differences. First, R&D activities are hard to monitor and thus entail high potential for misalignment of interests between the founding manager-owners and providers of external financing. As a consequence, R&D expenditures are more likely to be financed with internally generated funds. Second, R&D capital often takes the form of knowledge, which is not a physical asset that can be used as collateral. This limits the ability of R&D-intensive firms to borrow, and suggests that external financing to such firms is likely to be provided in exchange for equity. Third, because the accumulated knowledge from R&D cannot be kept secret (and the effectiveness of patents has its limits) the returns on R&D expenditure cannot be fully appropriated by firms, suggesting that the government might provide financing for such investment, for example in the form of grants and cheap external funds. For a lucid exposition of these arguments see Himmelberg and Petersen (1994), Hall (2002), and references therein. * E-mail: [email protected] Corresponding author. E-mail: [email protected] y ISSN 1043-8599 print; ISSN 1476-8364 online # 2003 Taylor & Francis Ltd DOI: 10.1080=1043859022000029249 426 A. A. BLASS AND O. YOSHA Hall (2002) further assesses the empirical evidence on the financing of R&D. Most evidence on this issue is from the United States, with some evidence from the United Kingdom, Japan, France, Germany, and Ireland. In recent years, Israel has become a world center of R&D and technological innovation, and it is of interest to investigate financing patterns of R&D firms there. Much of the R&D activity in various countries is carried out in relatively mature firms that are often publicly traded on a stock exchange. These firms have access to a variety of financing sources and, in addition, are eligible for government R&D subsidies. Our sample consists of such firms (as is the case in several studies cited in Hall, 2002). We ask whether the R&D-intensive firms in the sample are special and whether they are financed differently from other firms in the sample. Our study is unique in several respects. First, we construct firm-by-firm and year-by-year flow of funds charts that allow us to track the relative importance of all the financing sources, and all the uses of funds including R&D expenses. Second, we present, for the first time, adjusted flow of funds charts that treat R&D expenses as a capital outlay (rather than an operating cost that reduces profits, as standard accounting principles prescribe). Third, we compare the financing patterns of publicly traded Israeli R&D-intensive firms that issued stocks in New York to those that issued stocks only in Israel.1 Fourth, we tackle the question of how R&D intensity should be measured. The OECD has addressed this important issue in the Frascati Manual (OECD, 1993) that establishes criteria for measuring R&D inputs using R&D expenses and R&D personnel.2 It should be noted that the relative merits of these different methodologies for measuring R&D inputs are not fully understood. We partially address this issue by constructing both expenditure and personnel-based R&D measures for each firm in our sample. We investigate to what extent these measures are mutually consistent, and conclude that for the purposes of a study like ours, R&D expenses are more meaningful. In many countries, the government subsidizes R&D in young companies.3 Israel is no exception in this respect, although the Israeli government also subsidizes several established firms in the form of R&D grants. A secondary goal of this paper is to follow Griliches and Regev (1999, 2001) and characterize firms in the sample that receive substantial government assistance for R&D.4 A systematic evaluation of the effectiveness of such policy is beyond the scope of this paper, but the information we report raises some potentially important questions. A branch of the empirical literature on R&D financing concentrates on determinants of R&D expenditures and, in particular, on whether they are sensitive to cash flow. Following Fazzari, Hubbard, and Petersen (1988), a positive answer is interpreted as indication of liquidity constraints. A related question in such studies is whether liquidity constraints have a larger effect on R&D expenditure than on other types of investment. (This might be the case if there is greater information asymmetry in the financing of R&D projects.) Prominent contributions on this issue are Hall (1992), Himmelberg and Petersen (1994), Bond, Harhoff, and Van Reenen (1999), and Mulkay, Hall, and Mairesse (2000). The robust finding in these studies is that cash flow affects R&D expenditure. Our sample of Israeli publicly traded manufacturing firms in the 1990s is not ideal for studying financing constraints because part of the 1990s were a ‘‘hot-issue’’ market for public 1 See Blass and Yafeh (2001). The Manual is summarized in OECD (1994). 3 ‘‘Examples of such programs are the U.S. Small Business Investment Company (SBIC) and Small Business Innovation Research (SBIR) programs. Together, these programs disbursed 2.4 billion in 1995 . . . In Germany, more than 800 federal and state government programs have been established in the recent past . . . ’’ (Hall, 2002, p.14) Hall also describes such programs in Sweden and the United Kingdom. 4 Much aid for R&D is directed to universities and start-ups. We do not study these important channels of R&D subsidy; see Trajtenberg (2000). 2 MATURE COMPANIES 427 equity offerings on the Tel Aviv Stock Exchange. In addition, international capital movements were liberalized in the early 1990s and firms were permitted to issue equity abroad. As a result, most of the firms in our sample raised considerable amounts of equity capital in Tel Aviv and New York (see Blass and Yafeh, 2001), and were abundant in cash and liquid assets. It is unlikely that they felt liquidity constrained during these years. When firms raise external equity financing, the relevant question is whether ‘‘internally generated funds plus external equity financing’’ is associated with higher R&D investment.5 We briefly address this issue for the sake of comparison with the above-mentioned literature. Our data set consists of the population of manufacturing firms listed on the Tel Aviv Stock Exchange (about 250 firms),6 and is augmented here to include an additional 70 Israeli firms traded on U.S. exchanges. In terms of the type of firms under consideration, the sample is comparable to U.S. data employed in Himmelberg and Petersen (1994) and Cohen and Klepper (1992), and U.K. data employed in Bond, Harhoff, and Van Reenen (1999) – in all these studies the data pertain to relatively mature companies traded on a stock exchange.7 Bond, Harhoff, and Van Reenen acknowledge that for both the U.K. and the German data they use, ‘‘the samples are not representative of firms in either country’’. (Appendix I.) This qualification pertains to all the above mentioned studies including our own.8 Griliches and Regev (1999, 2001) use an impressive sample of over 24,000 Israeli manufacturing firms, the overwhelming majority of which are not traded on a stock exchange. Their sample is considerably more representative of the population of firms but lacks financial data and is, therefore, inappropriate for studying the financing of R&D.9 In recent years, small start-up firms, typically financed by venture capital, are becoming increasingly important. There are several studies that focus on such firms, as surveyed in Bottazzi and Da Rin (2002) and Hall (2002).10 Ideally, we would want a comprehensive sample that incorporates all types of firms – young start-ups, more mature but privately held firms, and mature publicly traded companies – with real and financial data. Unfortunately, such a sample is not available for any country. Obviously, this limits the ability to draw policy conclusions from any single study. In our main analysis, we characterize the firms in the sample that are R&D-intensive in terms of industry, age, size, geographical location, and profitability. To address the issue of R&D financing, we use the flow of funds to calculate, for each firm-year observation, the ratios of internal funds, debt, equity, and government financing (capital grants, R&D grants, and deferred taxes) to total sources and, similarly, the ratios of capital expenditure, investment in inventory, investment in liquid assets, investment in R&D, and dividend payouts to total uses. We further compute, for each firm-year observation, standard accounting profitability ratios such as net profits to sales and net profits to equity, as well as year-by-year Tobin’s q. We then examine whether R&D-intensive firms rely more on equity financing, bank credit, government assistance, or internal funds, whether they differ in terms of non-R&D uses of funds, and whether they exhibit higher accounting profitability. We also ask whether (lagged) Tobin’s q and liquid funds (internally generated 5 Internally generated funds correspond to ‘‘cash flow’’ conventionally measured as profits plus depreciation. See Blass and Yosha (2002). 7 Himmelberg and Petersen (1994) use COMPUSTAT data for firms listed on stock exchanges in the United States, limiting the sample to firms with assets less than 10 million U.S. dollars. Cohen and Klepper (1992) use a sample of several hundred firms in Fortune’s top 1000 list. For the United Kingdom, Bond, Harhoff, and Van Reenen (1999) use a sample of manufacturing firms listed on a stock exchange. 8 In the next section, we compare sample averages in our data to those of Israeli national-level data published by the Central Bureau of Statistics and the Chief Scientist at the Ministry of Industry and Trade. 9 Bound et al. (1984) describe the construction and merging of various data sets of publicly traded manufacturing firms in the United States but they do not address the issue of financing. 10 See the international comparison of investments backed by venture capital in Germany, Israel, Japan, and the U.K. by Mayer, Schoors, and Yafeh (2001). 6 428 A. A. BLASS AND O. YOSHA funds and new equity financing) are determinants of such investment. Finally, we evaluate the relative stock market performance of R&D-intensive firms. 2 DATA CONSTRUCTION, VARIABLE DEFINITION, AND DESCRIPTIVE STATISTICAL ANALYSIS We describe the sample of firms, the raw data sources, and the procedure for constructing firm-level year-by-year flow of funds. We then explain our definition of R&D-intensity and compare with another method used in the literature. We characterize the financing patterns of the firms in the sample, and calculate firm-level year-by-year Tobin’s q, checking whether R&D-intensive firms are different. 2.1 General Description of the Sample The sample consists of about 250 Israeli manufacturing firms that are listed on the Tel Aviv Stock Exchange,11 and 70 Israeli firms traded on U.S. exchanges (a total of 321 firms). We follow the sample for 8 years, from 1990 to 1997. Almost two thirds of the firms in the sample went public during the 1990s, while many that had been listed prior to the 1990s issued more stock in this period, particularly in 1992 and 1993. The sample includes most large Israeli manufacturing firms in electronics and chemicals (including pharmaceuticals), the sub-sectors in which most R&D is performed.12 Of the 50 largest firms in these sub-sectors in 1997 (by sales), 36 are in our sample either directly or as subsidiaries of publicly listed firms. (The remaining firms are privately held, or are stateowned enterprises, or subsidiaries of U.S. firms such as Motorola.) Moreover, these 36 firms represent 73 percent of sales of the 50 largest firms in the above sub-sectors. Our sample of publicly traded manufacturing firms is not representative of the nation-wide manufacturing sector, but it is reasonably representative of the sub-population of mature large manufacturing firms.13 2.2 Data Sources Data were collected from four key sources: (1) financial statements obtained mostly from a COMPUSTAT-type database (‘‘Dukas’’) compiled by the Tel Aviv Stock Exchange from annual reports; (2) stock price data; (3) prospecti submitted by firms issuing equity (IPOs or seasoned offerings) during the 1990s; and (4) flow statements compiled by the Bank of Israel Research Department from annual reports. In addition, we collected data on geographic location and firm age (mostly from firm prospecti). 2.3 Financial Statements Data The Dukas database contains information on all the securities listed on the Tel Aviv Stock Exchange. Because Dukas contains just four or five years of data at any point in time, it was 11 In the official Tel Aviv Stock Exchange classification by industry, the category ‘‘manufacturing’’ includes venture capital firms and holding companies. To preserve the (relative) homogeneity of the sample, these firms are not included. 12 Electronics includes communication equipment, medical and scientific equipment, electronic components, hardware, and some types of software, while chemicals includes basic chemicals, fertilizers, pharmaceutical, refining, paints, and plastics. In our sample, 63 percent of R&D expenditure is by firms in electronics and 22 percent in chemicals. 13 It goes without saying that our sample of publicly traded manufacturing firms is not representative of the population of start-up firms that only perform R&D and are often backed by venture capital. MATURE COMPANIES 429 necessary to reconstruct early financial statements figures by matching current and older versions of Dukas. Since firms that go public are required to provide financial statements for twoyears prior to the IPO, our sample includes pre-IPO data for such IPOs (more than half of the sample). 2.4 Stock Price Data The Tel Aviv Stock Exchange provides daily data for all securities listed on the exchange. The data include price changes, number of shares outstanding, year of IPO and dividend information. The data have been stored and accumulated by the Bank of Israel Research Department and allow us to compute a firm’s market value at any point in time. Data for stocks listed in the United States have been mainly entered by hand from daily newspapers. We use these data for calculating firm-level Tobin’s q, stock ‘‘beta’s’’, and stock excess returns. 2.5 Prospecti We collected data from prospecti submitted to the Israeli Securities Authority (and the U.S. SEC) in which firms issuing equity are required to provide information about their lines of business, future prospects, business risks, ownership structure, geographic locations of plants and markets, R&D expenses, year of incorporation, and distribution of employees’ occupations. Since most firms issued equity during the 1990s, either for the first time as IPOs or as a seasoned offering, the data are available for the overwhelming majority of firms. 2.6 Flow Statements The Bank of Israel Research Department has collected the 1990 through 1997 annual reports for listed firms and entered by hand for each firm and each year the ‘‘Consolidated Flow Statement’’. The statement decomposes the change in a firm’s balances into flows derived from operating activities, investment activities, and financing activities (broken down into up to 50 sub-entries). This information allows us to construct a flow of funds for each firm-year in the sample. In order to summarize the information contained in the flow statements, the approximately 50 types of entries provided in the statements were combined into 10 broader groups. Four – internal funds, net stock issues,14 net increase in debt, and government financing – can be viewed as alternative sources of capital; six – capital expenditure, R&D expenses, the increase in inventories, in other working capital, in liquid assets, and dividend payments are ‘‘uses’’. For each firm-year observation, we compute the ratios of these sources and uses to total sources, as well as standard profitability ratios such as net profits to sales and net profits to equity.15 Some of the observations are obvious outliers (e.g., in terms of one or more of these ratios), and were removed; see the Appendix for a detailed description of the procedure used. 2.7 Measuring R&D-Intensity We follow the vast majority of studies in this area and rely on firm-level self-reported R&D expenses as a measure of R&D-intensity. These expenses are reported in the Dukas database for companies listed in Tel Aviv and by COMPUSTAT for companies listed in the United 14 Convertible bonds are treated here as equity, even though the interest on such bonds appears as an expense (and not as a dividend payout). 15 The calculation of the flow of funds is analogous to that in Blass and Yosha (2002). The differences arise in the treatment of R&D expenses, described later, that required construction of the entire flow of funds charts from scratch. 430 A. A. BLASS AND O. YOSHA States. Publicly traded firms adhere to very precise (and quite strict) guidelines for reporting expenses, issued by the Israeli Securities Authority and the U.S. SEC. Yet, R&D expenses reported in the profit and loss statements are not sufficiently precise for our purposes because they are net of government R&D subsidies. We, therefore, rely on the flow statements of these firms, as well as the notes to the financial statements, to construct firm-by-firm series of gross R&D expenses, government R&D subsidies, and R&D royalties paid to the government whenever relevant. We define R&D-intensity as the ratio of gross R&D expenses to total uses (or total sources). We calculate this ratio for every firm-year observation, but in most of the analysis we rely on the average ratio over time as a measure of a firm’s overall R&D-intensity. Using this average measure we classify firms into broad R&D-intensity groups. For the sake of robustness, and to be consistent with other studies, we also employ R&D-intensity calculated as the ratio of R&D expenses to non-financial uses, to sales, and to (lagged) fixed capital. 2.8 Another Measure of R&D-Intensity Alternatively, R&D-intensity can be measured by R&D personnel. Quoting from OECD (1994) regarding ways to measure R&D-intensity: ‘‘For statistical purposes two inputs are measured: R&D expenditures and R&D personnel. Both inputs are normally measured on an annual basis: so much spent during a year, so many person-years used during a year. Both series have their strengths and weaknesses, and, in consequence, both are necessary to secure an adequate representation of the effort devoted to R&D’’. (p.6).16 These strengths and weaknesses have not been studied systematically with firm-level data. The next few paragraphs constitute a modest first step. In Israel, data on R&D personnel and R&D expenses are collected by the Central Bureau of Statistics for a wide range of firms (including privately held firms) as part of the Manufacturing and Crafts Surveys (unreported) and the Survey of Research and Development in Manufacturing ( published).17 In this study, we restrict attention to publicly traded firms for which financial data are available.18 For these firms, the R&D personnel measure is based on the percentage of employees engaged in R&D activities, and is obtained from equity offerings prospecti. For each firm in our sample, we construct this measure, asking whether it is consistent with the expense-based measure of R&D-intensity. The findings are summarized in Exhibit A. The correlation of the personnel-based and the expenditure-based measures is reasonably high, with 94 firms (out of 238 firms for whom we have data on workforce composition) both reporting R&D expenses and employing R&D personnel, and 78 firms not reporting R&D expenses nor employing such personnel. Only 17 firms report R&D expenses without employing personnel specializing in R&D. It turns out that these firms did not employ R&D personnel at the time they went public, but in later years engaged in small amounts of R&D activity. Approximately one fifth of the firms (49 out of 238) report zero R&D expenses, yet employ engineers and other research-type personnel. This requires investigation. In particular, the question arises whether R&D spending may be understated because it would seem that such 16 See also OECD (1993). For certain years, such data are also available internationally in surveys conducted by the United States National Science Foundation; see Bound et al. (1984, p.26) for a discussion. 18 Of the 50 largest Israeli manufacturing firms in electronics and chemicals (including pharmaceuticals), the two manufacturing sub-sectors with the highest R&D-intensity, 36 firms are in our sample, of which 25 report R&D expenses. The largest firm, the Oil Refineries, does not report R&D expenses, but the next largest 11 firms (4 chemical, 1 pharmaceutical, and 6 in electronics) all report R&D expenses. Two excluded firms are very large and spend considerable resources on R&D (Israel Aircraft – the second largest, and Motorola Israel – fifth). 17 MATURE COMPANIES 431 firms might actually be conducting R&D. If so, some of the 57 firms for whom no employment data are available and report no R&D spending might also be conducting R&D. Exhibit A – Reporting of R&D Expenses in Annual Reports versus Percentage of Employees in R&D from Prospecti R&D employees in prospectus YES (# of firms) NO (# of firms) Unknown (no prospectus) Reporting R&D expenses in annual reports YES (# of firms) NO (# of firms) 94 17 26 49 78 57 To address this issue, we study in detail the type of activities undertaken by the 49 firms employing R&D personnel but not reporting R&D expenses. We first discuss the accounting and legal guidelines according to which publicly listed firms report R&D. Israeli accounting practices generally conform to the accounting standards of the Financial Accounting Standards Board (FASB) in the United States.19 According to FASB Statement 2, R&D costs are charged as an expense (rather than an investment) when incurred. Research is defined as a ‘‘planned search or critical investigation aimed at discovery of new knowledge with the hope that such knowledge will be useful in developing a new product . . . or a new process or in bringing about significant improvement to an existing product or process’’. Development is defined as the ‘‘translation of research findings . . . into a plan or design for a new product or process or for a significant improvement to an existing product or process . . . ’’ According to the FASB, the types of activities that typically would be included as a research and development expense include: 1. laboratory research aimed as discovery of new knowledge; 2. searching for applications of new research findings or other knowledge; 3. formation, design, modification, engineering activity and testing of product or process alternatives; 4. design, construction and testing of prototypes, models, pilot plants and tools (involving new technology). By contrast, other related activities would not be recorded as an R&D expense, but as a cost of production: 1. 2. 3. 4. 5. 6. engineering follow through in an early phase of commercial production; quality control and trouble shooting during commercial production; improvement upon the qualities of, or periodic changes to existing products; adaptation of capabilities to a customer’s needs; routine design of tools; design and construction engineering related to start-up of facilities or equipment.20 19 These accounting standards are very similar to those enforced by the Israeli Securities Authority and the U.S. SEC which, in turn, are very close to the guidelines in OECD (1993, 1994). 20 The banning of the capitalization of R&D expenses occurred in 1975. Horwitz and Kolodny (1981) discuss the effects of this decision. Among these are a relative reduction in book equity, higher leverage, and effects on firms meeting security listing requirements. Horwitz and Kolodny further found that the decision caused a reduction in R&D spending by small firms. In many countries, R&D capitalization is allowed and even required. In the United Kingdom, for example, SSAP requires that development expenditures be deferred (but not pure and applied research). An international comparison would be an important project in its own right. 432 A. A. BLASS AND O. YOSHA As mentioned, we analyze the activities of the 49 firms that report R&D employees, yet do not report R&D spending in their annual reports (Exhibit A). A number of interesting results are observed: 1. These firms are not engaged in research or product development but rather in activities such as quality control that should not be recorded as R&D expenses as indicated by the FASB. We combed through the prospecti of 39 of the 49 firms and spoke to the management in a few cases where questions remained: in 33 cases, the R&D-type personnel were engaged in quality control or improvement, technical supervision, or periodic changes to existing products as well as adaptation of capabilities to a customer’s needs. Six of the firms indeed engaged in R&D and in their prospecti (but not in their annual reports) R&D expenses data are in fact provided in the appended notes. In these cases, however, the expenses are miniscule – less than 3 percent of total sources.21 2. The number of employees engaged in such activities in these firms tends to be low, both in absolute numbers (approximately 15) and as a percentage of the entire workforce. 3. The breakdown of these firms by industry is different than that of firms reporting R&D expenses – whereas 81 out of 94 firms employing R&D-type personnel and reporting R&D expenses are in electronics, only 11 out of the 49 not reporting R&D expenses but with R&D-type personnel are in electronics. It should also be noted that according to the Securities Law (Chapter 6, par. 47), Israeli corporations are required to itemize R&D expenses as a separate expense and not to co-mingle them with ‘‘other expenses’’ if they are greater than 5 percent of net profits. Overall, the evidence suggests that firms that do not report R&D spending, to the extent that they do in fact engage in R&D, do so to a very limited extent. We conclude, that for the purposes of this study, reported R&D expenses are more meaningful.22 In light of the extensive government support to R&D-intensive firms in many countries, it is important to measure correctly R&D-intensity. Whether our conclusion, that the expense-based measure is better, carries over to other settings and other countries is a worthy topic for research with obvious policy implications. 2.9 R&D Spending and Government R&D Subsidies In our sample, 109 firms report R&D expenses at least in one annual report of these all but ten report R&D expenses at least half of the time, while 80 percent of R&D spending in our sample is conducted by one third of the firms. As mentioned, most of the R&D spending is recorded in electronics (63 percent) and chemicals and pharmaceuticals (22 percent).23 In 26 percent of firm-year observations for these firms, government financing (net of royalties) amounts to more than 30 percent of R&D expenses, while no government assistance is provided in 23 percent of the observations. Over time, the percentage of firms receiving government financing in excess of 30 percent of R&D expenses has declined. On average over 21 Each of these cases was recorded prior to 1993. As mentioned, the R&D expenses are from annual reports whereas the R&D employment numbers are from prospecti. If, for example, a firm does not employ R&D personnel at the time of a securities offering but later hires such personnel, our data for this firm several years after the offering will say that it reports R&D expenses but does not employ R&D personnel. If indeed this is the source of the discrepancy between the two R&D-intensity measures, the measure based on R&D expenses reflects more closely the firm’s year-by-year R&D activities. 23 The dominance of electronics in R&D expenditure is consistent with the data on the entire manufacturing sector reported in the Survey of Research and Development in Manufacturing, 1998, published by the Central Bureau of Statistics. 22 MATURE COMPANIES 433 the sample years, 34 percent of government R&D subsidies are granted to the firms in our sample. 2.10 U.S. R&D Firms, Local R&D Firms, and Zero R&D Firms In light of the large differences between Israeli firms that issue securities in the United States and those that issue only in Tel Aviv (see Blass and Yafeh, 2001), we chose to perform the analysis in terms of three groups: firms that report R&D expenses and issued stocks in the United States (U.S. R&D firms), firms that report R&D expenses and issued stocks only in Tel Aviv (local R&D firms), and firms that do not report R&D expenses throughout the sample (zero R&D firms).24 On average over the sample years, 21 percent of total government R&D subsidies are granted to the U.S. R&D firms and 13 percent to the local R&D firms in our sample. There are striking differences among these groups as a straightforward descriptive statistics analysis reveals. 2.11 Descriptive Statistics Table I displays descriptive statistics.25 It is immediately apparent that (within the class of publicly traded firms) R&D-intensive firms are larger and younger than average, especially those that issued stocks in the United States. This is an unusual pattern – typically, size TABLE I Manufacturing Firms Publicly Traded on the Tel Aviv Stock Exchange and Israeli U.S.-Listed Firms – Descriptive Statistics, 1990 – 1997 (Unweighted Averages Over All Firms and All Years;* Standard Error of the Mean in Parentheses**). Zero R&D firms Local R&D firms U.S. R&D firms Size (total assets – Dec 1998 million NIS) 216 (13) 458 (78) 584 (73) Firm age (years since incorporation) 28.8 (0.5) 20.8 (0.8) 11.6 (0.6) 6.9 (0.3) 0.040 (0.002) 0.094 (0.006) 0.524 (0.005) 0.178 (0.012) 6.9 (0.6) 0.058 (0.007) 0.098 (0.006) 0.564 (0.010) 0.240 (0.024) 4.8 (0.6) 0.098 (0.014) 0.124 (0.017) 0.367 (0.022) 0.078 (0.021) 0.111 (0.009) 1057 0.175 (0.023) 317 0.508 (0.061) 166 Years since IPO Profitability (net profits=sales) Return on equity (net profits=equity) Leverage (total assets minus equity=total assets) Percentage of firms in Development Areas Sales growth Number of firm-years*** *‘‘Profitability’’ is weighted by sales, ‘‘Return on Equity’’ by equity, ‘‘Leverage’’ by total assets, and ‘‘Sales Growth’’ by sales. ** Each variable for each firm is averaged over time, and then the standard error is calculated. ***In this table, and all subsequent tables, outliers are not included; see the Appendix. The number of firms after removing outliers varies from about 120 in 1990 to over 250 in 1997. 24 The zero R&D group includes nine firms that did not report R&D expenses in most years and for whom R&D expenditures (whenever reported) represented less than 1.5 percent of all uses. 25 Here, and in what follows, outliers are not included; see the Appendix. The number of firms after removing these outliers varies from about 120 in 1990 to over 250 in 1997. 434 A. A. BLASS AND O. YOSHA and age are highly correlated – that highlights the success of R&D-intensive firms. Indeed, such firms – in particular those that issued stocks in the United States – exhibit faster than average sales growth.26 It is also interesting that R&D-intensive firms are much more profitable, in particular those that issued stocks in the United States. This finding is in sharp contrast to the popular view that regards young R&D firms as currently non-profitable, but with high expected profits. This view is probably inspired by start-up firms; the results in Table I indicate that R&Dintensive firms that make it to the initial public offering stage (especially if the offering is in New York) exhibit higher than average profitability. In calculating profitability, R&D is not regarded as a capital outlay (rather than as an expenditure) as we do for calculating flow of funds. This suggests that the true profitability of R&D-intensive firms (i.e., when R&D outlays are not expensed but, instead, are depreciated over time) is even greater than that reported in the table. U.S. R&D firms exhibit a lower leverage ratio, which is consistent with the interpretation that Israeli firms regard equity financing in the United States as an effective alternative to domestic bank financing. This is not so for local R&D firms possibly because Israeli banks are still quite dominant in local financial markets, including the stock market; see Yafeh and Yosha (1998), Blass, Yafeh, and Yosha (1998), Ber, Yafeh, and Yosha (2001), and Blass and Yosha (2002).27 A striking finding is that local R&D firms are heavily concentrated in outlying development areas (for simplicity defined in our analysis as firms located in telephone area codes 06 and 07 during the sample period) whereas U.S. R&D firms are conspicuously not located in these areas. Location of R&D-intensive firms seems to be responsive to financing opportunities: the Israeli government subsidizes investment in the periphery areas attracting capital intensive – see Blass and Yosha (2002) – and local R&D firms, whereas U.S. investors are more impressed with centrally located firms.28,29 2.12 Constructing Flow of Funds Adjusted for Investment in R&D In order to summarize the information contained in the flow statements we combined the approximately 50 types of entries provided in the statements into 10 broader groups. Four sources of funds: internally generated cash, net stock issues, net increase in debt, and net government aid (capital grants plus R&D grants plus deferred taxes less royalties); and six uses of funds: capital expenditure, R&D expenses, the increase in inventories, in working capital (other than inventory), in cash (or cash equivalents) and securities, and dividend payments. The R&D expenses reported in the financial statements (Dukas) are net of government aid for R&D and, therefore, underestimate the actual expenditure on R&D. To correct for this, we scanned the notes to annual reports by hand for firm-level year-by-year government aid for R&D and added this amount to R&D expenses as reported in Dukas obtaining gross R&D 26 This is broadly consistent with Griliches and Regev (1999) who find that R&D expenses per worker are positively correlated with firm size and negatively correlated with firm age. On the issue of R&D and firm size, see Cohen and Klepper (1996) and references therein. 27 The lower leverage ratio is not a consequence of differences in usage of corporate bonds – Israeli firms raise negligible amounts of financing in this manner (a puzzle in its own right); see Yafeh and Yosha (1998) and Blass and Yosha (2002). 28 It should be kept in mind that firms seeking financing in the United States benefit from reduced tax payments. 29 This brings to mind the issue of spillovers. Suppose that R&D-intensive firms benefit from clustering. If some firms seek financing from U.S. investors but others do not, they would not locate in the same area. Those who do not seek financing abroad will move to development areas to benefit from government assistance whereas those who seek financing abroad will remain in the center to accommodate the tastes of U.S. markets. This may reduce the degree of geographic clustering and diminish the benefits from inter-firm spillovers. MATURE COMPANIES 435 expenses, which is defined as a use. Since government aid for R&D is a source, we added such aid to total sources. To be consistent with the treatment of other sources of funds, where net flows are used (e.g., net increase in debt), and also to be consistent with total uses (that are equal to total net sources), we added net government aid to total sources. This is the main adjustment of the flow of funds constructed in Blass and Yosha (2002) where R&D is treated as an expense that reduces profits (and hence reduces total sources), and government aid for R&D is ignored (and therefore omitted from total sources).30 2.12.1 Sources Sources are displayed in Table II. Internally generated cash is calculated by adding depreciation (a non-cash expense) to net income and then further adding and subtracting several additional components. We add expenses such as deferred employee benefits, that are included in the calculation of net income but do not involve cash flows. Most important, because in the financial statements (Dukas) R&D is treated as an expense that reduces profits (and total sources), we further add R&D expenses. (As mentioned, this is the main adjustment of the flow of funds constructed in TABLE II Flow of Funds in Manufacturing Firms Publicly Traded on the Tel Aviv Stock Exchange and in Israeli U.S.-Listed Firms – Average Sources , 1990–1997 (Averages Over All Firms and All Years Weighted by Total Sources; December 1998 Million NIS; Fraction of Total in Space Brackets; Standard Error of the Mean in Parentheses*). Zero R&D firms Local R&D firms U.S. R&D firms 18.0 [0.595] (1.2) 7.1 [0.233] (0.7) 42.3 [0.451] (7.3) 10.9 [0.117] (1.9) 84.3 [0.536] (11.2) 49.3 [0.313] (8.7) Net increase in debt2 3.8 [0.124] 1.2 25.8 [0.276] (11.6) 14.9 [0.095] (7.9) Net government aid3 1.4 [0.046] (0.2) 14.6 [0.156] (3.6) 8.8 [0.056] (1.9) Total sources 30.3 (2.0) 93.7 (20.1) 157.3 (17.7) Number of firm-years** 1057 317 166 Internally generated funds1 Net stock issues *Each variable for each firm is averaged over time, and then the standard error is calculated. **See the notes to Table I. 1 Including R&D expenses (to correct for the fact that in the financial statements R&D is treated as an expense). 2 Bonds and other debt primarily from banks. 3 Includes grants, R&D subsidies net of royalties, and deferred taxes. 30 The following example should clarify the exact adjustments made. Consider a firm that in a given year spent 140 on R&D and paid 20 in royalties to the government (because R&D that had been subsidized prior to that year generated profits). Further assume that, in that year, the firm received 60 as a government R&D subsidy. Our adjusted flow of funds data would show that the firm used 100 of its own (internally generated) funds plus 40 in government net aid to finance 140 in R&D spending. By contrast, in the Dukas database (or in COMPUSTAT), R&D expenses would appear as 100, namely 140 spent in practice less the net transfer from the government. As mentioned in the text, we do not add the entire 60 received from the government to total sources. Instead, we add the net amount received (60 minus 20 paid in royalties) to total sources. We do so in order to be consistent with the treatment of other sources of funds, where it is conventional to use net flows: net increase in debt, net equity issued (i.e., stocks issued net of stocks repurchased), and so forth. 436 A. A. BLASS AND O. YOSHA Blass and Yosha, 2002.) We then subtract revenue items, such as unrealized gains on marketable securities, which are included in the calculation of net income but do not involve cash flows. U.S. R&D firms rely considerably more on equity issues (defined as total proceeds received from floatation of common stock, convertible bonds,31 and warrants as well as proceeds received from exercise of warrants and options). This is consistent with the view that stock market finance is more suited for financing R&D due to ‘‘diversity of opinion’’ in a large decentralized capital market (Allen, 1993). Local R&D firms do not rely on stock floatation and, instead, rely on debt (mostly bank debt) and government aid considerably more than firms that do not perform R&D. An intriguing result is that R&D expenses are not more likely to be financed with internally generated funds. This is in contrast to other countries where a positive relation between R&D spending and internally generated funds was consistently found; see, e.g., Himmelberg and Petersen (1994) and Hall (2002). As mentioned, part of the 1990s were a ‘‘hot-issue’’ market for public equity offerings on the Tel Aviv Stock Exchange and, in addition, many Israeli firms issued equity in New York. As a result, most of the firms in our sample raised considerable amounts of external equity capital on a stock exchange and were abundant in cash and liquid assets. It is, therefore, unlikely that they felt liquidity constrained during these years.32 We will investigate this issue further in the regression analysis (Tab. VII). 2.12.2 Uses Uses are displayed in Table III. Capital expenditures include investments (net of sales) in property, plants, and equipment. Investment in working capital (other than inventory) is defined as an increase in current assets, such as trade accounts receivable, less current liabilities, such as trade accounts payable. Investment in cash and securities is defined as a net change in cash holding plus a purchase of publicly traded securities and mutual funds. The local R&D firms invest heavily in fixed capital, devoting four times as many resources to capital expenditures than R&D expenditures. Indeed, in every year of our sample these firms devoted greater resources to capital expenditures. Is this an indication that R&D and capital expenditure are complements (an unlikely interpretation in light of the very different numbers for U.S. R&D firms, and the fact that zero R&D firms also invest heavily in fixed capital)? Was capital intensity in Israeli firms sub-optimal in the early 1990s due to the influx of immigration and labor from ex-Soviet countries, resulting in massive subsequent investment in physical capital (in that event, U.S. R&D firms must have a different production structure that is inherently less capital intensive)? Or do zero and local R&D firms simply ‘‘specialize’’ in obtaining government aid of all sorts (capital grants and subsidies as well as R&D subsidies)? We have no hard evidence regarding this issue, but its potential policy implications are interesting. U.S. R&D firms differ significantly from other firms as they devote 34 percent of funds to increasing cash balances and securities holdings, reflecting the large amounts of cash raised during the stock offerings.33 The three groups differ in their dividend payout polices, with the zero R&D group paying out the largest fraction of total uses.34 31 Convertible bonds contain both debt and equity features and are treated here as equity, even though the interest on such bonds appears as an expense (and not as a dividend payout). 32 Nonetheless, it should be noted that internally generated funds as a percentage of all sources net of stock issues was lower on average for the zero R&D firms than for the local R&D firms. 33 In fact, Blass and Yafeh (2001) find that these firms dilute ownership much more than firms that issue stocks in Israel. 34 The variation over time in this category is very small. MATURE COMPANIES 437 TABLE III Flow of Funds in Manufacturing Firms Publicly Traded on the Tel Aviv Stock Exchange and in Israeli U.S.-Listed Firms – Average Uses, 1990–1997 (Averages Over All Firms and All Years Weighted by Total Sources; December 1998 Million NIS; Fraction of Total in Space Brackets; Standard Error of the Mean in Parentheses*). Zero R&D firms Local R&D firms U.S. R&D firms Capital expenditure 16.6 [0.548] (1.3) 53.1 [0.566] (12.7) 29.1 [0.185] (4.0) Investment in inventory 2.7 [0.088] (0.5) 4.2 [0.045] (1.5) 10.4 [0.067] (3.2) Investment in working capital 3.3 [0.108] (0.6) 1.6 [0.017] (2.5) 11.2 [0.071] (2.9) Investment in cash and securities 3.6 [0.118] (0.7) 13.9 [0.149] (6.6) 53.4 [0.340] (12.2) Dividends 4.2 [0.138] (0.4) 7.7 [0.082] (1.6) 6.6 [0.042] (2.3) R&D expenditure (gross) 0.0 [0.000] (0.0) 13.2 [0.141] (2.3) 46.5 [0.296] (5.3) Total uses 30.3 (2.0) 93.7 (20.1) 157.3 (17.7) Number of firm-years** 1057 317 166 *Each variable for each firm is averaged over time, and then the standard error is calculated. **See the notes to Table I. 2.13 Investment in Physical Capital in the Entire Manufacturing Sector As mentioned, in our sample of publicly traded manufacturing firms, 63 percent of the R&D spending is recorded in electronics (and 22 percent in chemicals and pharmaceuticals). Table III indicates that the local R&D firms in our sample also invest heavily in physical capital. We want to know whether this is similar to the investment patterns in the entire manufacturing sector (publicly traded and privately held firms). Aggregate data by sub-industry for the entire manufacturing sector are obtained from various publications in the framework of the Survey of Manufacturing, published by the Central Bureau of Statistics. These data confirm that the bulk of R&D expenditure is in electronics (and to some extent in chemicals and pharmaceuticals), as in our sample of publicly traded firms.35 We cannot mimic the numbers displayed in Table III for these sub-industries because there are no flow of funds data for the entire manufacturing sector (not to speak of subindustries or 35 The following information is from Survey of Research and Development in the Business Sector, 1998 (C.S. 26=2001), published by the Israeli Central Bureau of Statistics, Table XXII:‘‘International Comparison – Manufacturing Intramural Expenditure on R&D’’. The table displays the share of each industry in total national R&D expenditure of this type. Switzerland’s chemical (including pharmaceutical) products is responsible for 86 percent of expenditure in the country, compared to an OECD average of 27 percent. In Hungary, chemical (including pharmaceutical) products is responsible for 66 percent of expenditure in the country. In Israel and Finland, the electronic communication equipment is responsible for 78 and 52 percent, respectively, of expenditure in the country, compared to an OECD average of 17 percent. In the Netherlands, electrical equipment is responsible for 42 percent of expenditure in the country, compared to an OECD average of 7 percent. In Italy and South Korea, the bulk of R&D expenditure (31 percent) is in transport equipment. Overall, countries seem to be quite specialized in R&D in specific sub-industries. 438 A. A. BLASS AND O. YOSHA individual firms). Instead, we performed the following back-of-the-envelope exercise. We calculated the following ratio for each sub-industry (in 1997):cumulative investment in physical assets=(cumulative investment in physical assets and other long-term assets plus current assets less current liabilities). The denominator is a proxy for the stock of accumulated net assets in the sub-industry, and the ratio thus measures the extent to which resources are allocated to the accumulation of physical assets rather than working capital, for example. This ratio varies a lot across sub-industries: for example, in food manufacturing and in building materials – low-tech industries engaging in very little R&D – the ratio is 0.65 and 0.67, respectively, whereas in ‘‘electrical and electronics’’ it is much lower, 0.37. This pattern – low capital intensity in the R&D-intensive sector – is not consistent with the results displayed in Table III, suggesting that privately held and publicly traded firms (not traded in the U.S.) use their funds differently, possibly due to different financing patterns and opportunities, and size effects. This issue is beyond the scope of our article, and it serves as a further qualification for the applicability of the findings to the entire manufacturing sector. 2.14 Calculation of Tobin’s q We measure firm-level year-by-year Tobin’s q as the market value of assets divided by their replacement value. Replacement values are calculated assuming that fixed assets and inventories appreciate at a rate equal to that of the consumer price index (CPI). The market value of assets equals the market value of common equity plus the value of debt. The latter is calculated as the replacement value of assets less the (CPI adjusted balance sheet) sum of the book value of common equity and deferred taxes and employee benefits.36 (We do not construct an estimate of R&D stock, and ‘‘replacement value’’ refers to physical capital.) Table IV displays the year-by-year unweighted average of firm-level Tobin’s q for the firms in the sample. For local R&D and zero R&D firms, Tobin’s q so calculated rises dramatically in 1992 and 1993 reflecting the stock price run-up in those years. For zero R&D firms it declines in 1996, and when market conditions improved in 1997 it rises again. TABLE IV Tobin’s q for Manufacturing Firms Publicly Traded on the Tel Aviv Stock Exchange and Israeli U.S.-Listed Firms, 1990 – 1997* (Unweighted Averages Over All Firms; Standard Error in Parentheses; Number of Observations). Average: year ending 1990 1991 1992 1993 1994 1995 1996 1997 Average: all firms and all years Zero R&D firms 1.04 (0.24) 66 1.23 (0.37) 104 1.73 (0.66) 124 2.20 (0.96) 139 1.16 (0.38) 154 1.05 (0.37) 159 0.90 (0.24) 139 0.97 (0.33) 135 1.28 (0.67) 1020 Local R&D firms 1.15 (0.38) 17 1.51 (0.86) 31 2.28 (1.19) 41 2.66 (1.34) 38 1.11 (0.36) 43 1.11 (0.49) 50 1.09 (0.64) 44 1.25 (0.58) 44 1.47 (0.96) 308 U.S. R&D firms 1.82 (0.76) 4 6.75 (5.25) 9 3.88 (1.17) 10 3.01 (1.29) 16 2.09 (1.80) 22 2.51 (1.45) 31 1.64 (1.19) 28 1.85 (1.49) 20 2.52 (2.11) 140 *The number of firms ranges from 87 in 1990 to 199 in 1997; these are the firms in the sample for which stock price data were available. 36 See Blass and Yosha (2002) for further details. They did not study firms that are traded in New York. For these firms, the adjustment takes into account year-by-year fluctuations in the U.S. dollar=NIS exchange rate. MATURE COMPANIES 439 On average (and in most years), Tobin’s q of local R&D firms exceeds that of zero R&D firms. In every year, Tobin’s q for U.S. R&D firms was above that of the zero R&D and local R&D firms. This is consistent with Hall’s (1992) findings for a sample of U.S. firms, and with the well-established positive relation between R&D spending and stock market value; see, e.g., Griliches (1981), Jaffe (1986), Hall (1992), and Johnson and Pazderka (1993). This empirical regularity suggests that R&D-intensive firms (and in particular the U.S. R&D firms in our sample) possess many intangible assets that are valued by investors. It also suggests that the treatment of R&D as an expense, rather than an asset, understates the value of corporate assets. 3 EMPIRICAL ANALYSIS In this section, we investigate more systematically the regularities suggested by the descriptive statistics presented in the previous section. Using simple linear regression, we study the determinants of the various source and use ratios calculated from the unique panel of firmlevel flow of funds data. Then, to compare with studies in the literature for other countries, we briefly study the determinants of R&D-intensity. Finally, using simple linear regression with dummy variables for R&D-intensive firms, we ask whether the performance of the stocks of these firms is different than average. 3.1 Determinants of Source and Use Ratios Tables V and VI display cross-section regressions of source ratios and use ratios, averaged over time, on firm characteristics (also averaged over time).37 The regressions are Generalized Least Squares (GLS) and the weighting is by firm (log-) total sources. We include as regressors dummy variables for U.S. R&D firms and local R&D firms. These dummy variables are based on the average (over time) R&D-intensity of the firms in the sample. They can, therefore, be regarded as reasonably exogenous. A year-by-year analysis would be less appropriate for the empirical specification in Tables V and VI because year-by-year R&D expenditure is more likely to be endogenous. The coefficients on the R&D dummies in the first column of Table V indicate that, in contrast to other countries (see, e.g., Himmelberg and Petersen 1994 and Hall 2002), R&Dintensive firms are not more likely to be financed with internally generated funds.38 An inspection of the coefficients on the R&D dummies in the other columns of Table V reveals that once firm characteristics are controlled for, the financing patterns of R&D-intensive firms are not different. The descriptive statistics in Table II indicate that R&D-intensive firms that issued stocks in the United States rely on external equity financing, and local R&D firms on (bank) debt financing and government aid. Table V suggests that if firm characteristics are controlled for, these differences in financing patterns no longer obtain. We were intrigued by this result. Further scrutiny revealed that it is driven by the inclusion in the sample of ten large successful firms that paid R&D royalties to the government, as a percentage of sales of products developed with government aid (63 firm-year observations). When these firms were removed from the sample, the coefficient on the U.S. R&D variable in the second column of Table V becomes statistically significant.39 We conclude that U.S. R&D 37 For example, in the first column of Table V, the dependent variable is the firm-by-firm ratio of internally generated cash to total sources, averaged over time. 38 This result is consistent with the descriptive statistics in Table II. 39 The t-statistic is 1.93. The other coefficients in the regression are not affected in a meaningful way. 440 A. A. BLASS AND O. YOSHA TABLE V Factors Affecting Flow of Funds in Manufacturing Firms Publicly Traded on the Tel Aviv Stock Exchange and Israeli U.S.-Listed Firms – Sources, 1990–1997* (GLS Regressions Using Firm-Level Data Averaged Over Time; Standard Error in Parentheses). Dependent variable: Internally generated cash** Net stock issues Net increase in debt Size (log-total assets) 0.037 (0.042)
0.036þþ (0.0114) 0.010 (0.037) Profitability (net profits=sales) 0.087 (0.257)
0.519þþ (0.0694) 0.206 (0.228) Return on equity (net profits=equity) 0.061 (0.117) 0.040 (0.0316)
0.130 (0.104) 0.029 (0.037) Industry dummies Greater Tel Aviv area dummya YES 0.333þþ (0.136) YES
0.145þþ (0.0366) YES
0.057 (0.121) YES
0.131þþ (0.043) Development area dummyb 0.070 (0.148)
0.003 (0.0398)
0.176 (0.131) 0.108þþ (0.046) Multi-plant firm dummyc 0.150 (0.153)
0.068 (0.0412)
0.588þþ (0.136)
0.023 (0.480) 0.008þþ (0.003) 0.379þ (0.227)
0.026 (0.175)
0.056 (0.144) 0.140
0.002 (0.0011)
0.119þ (0.0612) 0.049 (0.0473)
0.043 (0.0387) 0.475
0.006þ (0.004)
0.228 (0.202)
0.004 (0.156) 0.060 (0.127) 0.118
0.001 (0.001)
0.030 (0.071)
0.021 (0.055) 0.039 (0.045) 0.216 Firm age (years since incorporation) Seasoned security**** U.S. R&D firm dummy Local R&D firm dummy R-squared Government aid***
0.012 (0.013) 0.227þþ (0.081) *Regressions are weighted by total sources. Number of firms: 258. Zero R&D firms are the control group. **Including R&D expenses (to correct for the fact that in the financial statements R&D is treated as an expense). ***Includes grants, R&D subsidies, and deferred taxes. ****Firms that go public are required to provide financial statements for two years prior to the IPO, and our sample includes pre-IPO data for about half the firms in the sample. For firms that went public, say, in the 1980s the variable ‘‘Seasoned security’’ takes the value 1. For a firm that went public, say, in 1996 it takes the value 0.5 (2 observations prior to the IPO, and two observations, 1996 and 1997, following the IPO). (The correlation between ‘‘Seasoned security’’ and ‘‘Firm age’’ is very low.) þ Significant at the 10 percent level. þþ Significant at the 5 percent level. a Equals 1 if firm is located in area codes 03, 09, and 0 otherwise. b Equals 1 if firm is located in area codes 06, 07, and 0 otherwise. c Equals 1 if firm has plants in more than one region, and 0 otherwise. Area codes 02, 04, 08 are the control region. firms do rely more on external equity financing, even after controlling for firm characteristics.40 It should be noted that we control for the fact that U.S. R&D firms tended to go public more recently. This is captured by the variable ‘‘Seasoned security’’defined as the fraction of observations for each firm in years following the IPO.41 Thus, the very successful firms that pay royalties to the government, being very large, can greatly affect the estimated coefficients in GLS regressions where weighting is by firm size (total sources). 40 As mentioned, this is consistent with the view that stock market finance is more suited for financing R&D than bank finance due to ‘‘diversity of opinion’’ in a large decentralized capital market (Allen, 1993). 41 As mentioned, firms that go public are required to provide financial statements for two years prior to the IPO, and our sample includes pre-IPO data for about half the firms in the sample. For firms that went public, say, in the 1980s the variable ‘‘Seasoned security’’ takes the value 1. For a firm that went public, say, in 1996 it takes the value 0.5 (2 observations prior to the IPO, and two observations, 1996 and 1997, following the IPO). Thus, to each firm in the sample corresponds one number between 0 and 1. The correlation between this variable and ‘‘Firm age’’ is very low, reflecting the fact that some but not all firms went public at a very young age. MATURE COMPANIES 441 TABLE VIA Factors Affecting Flow of Funds in Manufacturing Firms Publicly Traded on the Tel Aviv Stock Exchange and Israeli U.S.-Listed Firms – Uses, 1990–1997* (GLS Regressions of Ratio to Total Uses Using FirmLevel Data Averaged Over Time; Standard Error in Parentheses). Dependent variable: Investment in working capital Investment in cash and securities Dividends Capital expenditure Investment in inventory
0.013 (0.026)
0.034 (0.026) 0.045 (0.028) 0.011 (0.045) 0.009 (0.011) 0.389þþ (0.159)
0.170 (0.156) 0.004 (0.174)
0.580þþ (0.273)
0.007 (0.069)
0.0118þ (0.072) YES
0.066 (0.084)
0.002 (0.071) YES
0.039 (0.082) 0.056 (0.079) YES 0.230þþ (0.092) 0.048 (0.125) YES 0.186 (0.144) 0.003 (0.032) YES
0.011 (0.037) 0.295þþ (0.091) 0.116 (0.089)
0.180þ (0.100)
0.689 (0.157) 0.0001 (0.039)
0.009 (0.094) 0.081 (0.092)
0.030 (0.103) 0.090 (0.162)
0.028 (0.041) 0.002 (0.002)
0.005þþ (0.002) 0.001 (0.003) 0.001 (0.004) 0.001 (0.001) Seasoned security**
0.032 (0.140)
0.249þ (0.137)
0.036 (0.154) U.S. R&D firm dummy
0.110 (0.108)
0.154 (0.106) Local R&D firm dummy 0.058 (0.089) 0.277 Size (log-total assets) Profitability (operating profits=sales) Return on equity (net profits=equity) Industry dummies Greater Tel Aviv area dummya Development area dummyb Multi-plant firm dummyc Firm age (years since incorporation) R-squared 0.485þþ (0.241)
0.095 (0.061) 0.030 (0.119) 0.109 (0.186)
0.095þþ (0.047)
0.119 (0.087)
0.066 (0.097) 0.044 (0.152)
0.086þþ (0.038) 0.061 0.094 0.091 0.164 *Regressions are weighted by total sources. Number of firms: 258. Zero R&D firms are the control group. **Firms that go public are required to provide financial statements for two years prior to the IPO, and our sample includes pre-IPO data for about half the firms in the sample. For firms that went public, say, in the 1980s the variable ‘‘Seasoned security’’ takes the value 1. For a firm that went public, say, in 1996 it takes the value 0.5 (2 observations prior to the IPO, and two observations, 1996 and 1997, following the IPO). (The correlation between ‘‘Seasoned security’’ and ‘‘Firm age’’ is very low.) þ Significant at the 10 percent level. þþ Significant at the 5 percent level. a Equals 1 if firm is located in area codes 03, 09, and 0 otherwise. b Equals 1 if firm is located in area codes 06, 07, and 0 otherwise. c Equals 1 if firm has plants in more than one region, and 0 otherwise. Area codes 02, 04, 08 are the control region. The second column of Table V reveals that larger and more profitable firms rely less on stock offerings.42 An interpretation is that shareholders of such companies are not eager to dilute their ownership stake and share profits with outside investors. Of course, being profitable, they are also in less need for external equity financing. We turn to the fourth column, determinants of government aid (investment grants, R&D subsidies, and deferred taxes). There are two main justifications for providing R&D aid – subsidizing spillover externalities and correcting for capital market failures due to asymmetry of information (see Hall, 2002 for a discussion). It is interesting that profitable firms rely more on government aid. On the one hand, this indicates that 42 The estimated coefficients are not meaningfully affected by the omission of the firms that paid R&D royalties to the government. This was the case for most coefficients in the regressions displayed in Tables V and VI. From now on, if we do not say otherwise, it should be understood that results are not affected by the omission of these firms. 442 A. A. BLASS AND O. YOSHA TABLE VIB Factors Affecting R&D Expenditure in Manufacturing Firms Publicly Traded on the Tel Aviv Stock Exchange and Israeli U.S.-Listed Firms – Uses, 1990–1997* (GLS Regressions Uses Firm Level Data Averaged Over Time; Standard Error in Parentheses). Dependent variable: R&D expenditure (gross)=total uses net of increase in cash and securities** R&D expenditure (gross)=total uses net of increase in cash securities and dividend payments** 0.102 (0.083) 0.094 (0.085)
0.242 (0.444)
0.246 (0.458) R&D expenditure (gross)= total uses** log [R&D expenditure (gross)= total uses]** Size (log-total assets) 0.012 (0.060) 0.227þþ (0.065) Profitability (operating profits=sales) Return on equity (net profits=equity) 0.506 (0.320)
2.122þþ (0.586) 0.023 (0.183) 0.129 (0.220)
0.009 (0.039) 0.131 (0.254) 0.151 (0.262) Greater Tel Aviv area dummya
0.458þþ (0.199)
0.068 (0.241) 0.009 (0.043)
0.222 (0.277)
0.204 (0.285) Development area dummyb Multi-Plant firm dummyc Firm age (years since incorporation) Seasoned security***
0.455þ (0.235)
0.248 (0.251)
0.002 (0.007)
1.676þþ (0.271) 0.131 (0.288)
0.016þþ (0.008)
0.969þþ (0.326)
0.471 (0.349) 0.010 (0.010)
0.996þþ (0.336)
0.433 (0.359) 0.012 (0.010)
0.118 (0.343) 0.203 (0.392)
0.150þþ (0.074) 1.222þþ (0.476) 1.314þþ (0.491) 0.117 (0.191) 0.039 (0.213) 0.026 (0.041) 0.311 (0.265) 0.272 (0.273) 0.128 0.712 0.489 0.292 0.292 U.S. R&D firm dummy R-squared R&D expenditure (gross)= sales**
0.009 (0.013)
0.388þþ (0.069)
0.063 (0.051) 0.019 (0.054)
0.002 (0.002) *Regressions are weighted by total sources. **Sample includes only U.S. and local R&D firms (85 firms, in log regression 74 firms). For this reason, industry dummies are omitted in this regression. ***Firms that go public are required to provide financial statements for two years prior to the IPO, and our sample includes pre-IPO data for about half the firms in the sample. For firms that went public, say, in the 1980s the variable ‘‘Seasoned security’’ takes the value 1. For a firm that went public, say, in 1996 it takes the value 0.5 (2 observations prior to the IPO, and two observations, 1996 and 1997, following the IPO). (The correlation between ‘‘Seasoned security’’ and ‘‘Firm age’’ is very low.) þ Significant at the 10 percent level. þþ Significant at the 5 percent level. a Equals 1 if firm is located in area codes 03, 09, and 0 otherwise. b Equals 1 if firm is located in area codes 06, 07, and 0 otherwise. c Equals 1 if firm has plants in more than one region, and 0 otherwise. Area codes 02, 04, 08 are the control region. such aid is not wasted on losing companies; on the other hand, it raises the question whether aid reaches those firms who really need it, unless one is willing to argue that more profitable firms generate greater spillover externalities. Firms in development areas depend on government aid (as we would expect) and those in the Tel Aviv area do not rely on such aid. The industry dummies in all the regressions are mostly insignificant (not shown).43 Older firms rely more on internal sources and less on bank debt. Surprisingly, older firms do not rely less on government aid. This is consistent with subsidization of spillover externalities 43 A notable exception is textiles for which a greater reliance on debt at the expense of internally generated funds is observed. MATURE COMPANIES 443 (that should not differ by firm age) but not with correcting capital market failures due to information asymmetry (that should be greater for younger firms). Finally, notice that R&D-intensive firms – both U.S. and local – do not rely more on government aid for R&D (see the coefficients on the U.S. and local R&D firm dummies in the fourth column). This is also an intriguing result, and it too is driven by the inclusion of the ten large successful firms that paid R&D royalties to the government. When these firms are removed from the sample, the coefficients on the R&D dummy variables in the fourth column of Table V become positive and statistically significant.44 Thus, R&D-intensive firms do rely more on net government R&D aid, although the royalties paid to the government by the few highly successful R&D firms are sufficiently substantial to obscure this finding in the regression.45 Table VIA displays the results of an analogous exercise for use ratios, excluding R&D expenditure ratios that are displayed separately in Table VIB. Recall that in Table III we saw that, on average, local R&D firms are also intensive in physical capital. The first column of Table VIA reveals that once other firm-level characteristics are controlled for this no longer holds. The first column of Table VIA also indicates that because subsidies are given to investment in development areas, not surprisingly, firms in these areas invest relatively more in fixed assets (capital expenditure), as do profitable firms. Combining this finding with those reported in Table V, we obtain the following pattern: firms in development areas and profitable firms rely on government subsidies, and use these funds for investment in physical assets. Another interesting finding, consistent with the results displayed in Table III, is that zero R&D firms pay out more dividends than R&D-intensive firms. We now turn to R&D expenditure. In the first column of Table VIB, R&D expenditure is normalized by total uses (as in Tables V and VIA), but for the sake of robustness and comparison with the literature, in other columns we normalize R&D expenditure by other variables. The normalization in the two last columns of the table (by uses less increase in cash and securities, and uses less increase in cash and securities and dividend payments, respectively) is intended to create a measure of R&D spending as a ratio of non-financial uses. In the regressions displayed in this table, only U.S. and local R&D firms are included because for the other firms in the sample R&D expenditure is zero. For this reason, the dummy variable for local R&D firms is omitted. The regressions use only 85 observations, mostly in electronics, and, therefore, industry dummy variables are also omitted. In Tables V and VIA, we found that profitable firms, and firms in development areas rely on government aid and use these funds for investment in physical assets. Do these firms also invest more in R&D? For firms in development areas, the answer is clearly no – the coefficient on the development area dummy is negative and significant in all the columns but one of Table VIB.46 The coefficient on profitability varies across specifications, and we cannot draw conclusions regarding the empirical relation of profitability and R&D expenditure. In the second column of Table VIB, the left-hand variable is the log of the R&D expenditure ratio to total uses. The coefficient on log-size is positive and significant, which is consistent with findings in, e.g., Bound et al. (1984) reporting that R&D-intensive firms are large. In the other specifications, log-size is not significant.47 44 The t-statistics are 1.66 and 2.46 for U.S. and local R&D firms respectively. The other coefficients in the regression are not affected in a meaningful way. 45 It is harder to argue that a positive relation between R&D-intensity and R&D grants and subsidies indicates that R&D aid stimulates R&D because we lack information on the counterfactual – what would R&D-intensity have been without such aid. Lach (2000) studies this issue obtaining mixed results. Some of his results point to partial crowding out (about one half) and some to full crowding out of R&D expenditure by government R&D subsidies. 46 Comparing with Table I is not meaningful due to the different sample used. 47 See Cohen and Klepper (1996) and references therein. 444 A. A. BLASS AND O. YOSHA The coefficient on the dummy variable for U.S. R&D firms is insignificant in all the columns of Table VIB. That is, as a fraction of total sources, U.S. R&D firms do not invest more in R&D than local R&D firms. This does not contradict Blass and Yafeh’s (2001) finding that Israeli companies that issue stocks in the United States are more R&D-intensive than Israeli companies that issue stocks in Tel Aviv. Their findings indicate that, conditional on a firm not being R&D-intensive, the data would predict that the firm would list in Tel Aviv. The results here, as evidenced by the insignificant U.S. R&D dummy coefficient, indicate that, conditional on a firm being R&D-intensive, the fact that it is listed in the United States (and not in Tel Aviv) does not mean that it is more R&D-intensive. 3.2 Determinants of Year-by-Year R&D Intensity We turn to the determinants of expenditure on R&D. As mentioned, a branch of the empirical literature on R&D financing concentrates on determinants of year-by-year investment in R&D and, in particular, on whether liquidity matters for such investment; see Hall (1992), Himmelberg and Petersen (1994), Bond, Harhoff, and Van Reenen (1999), and Mulkay, Hall, and Mairesse (2000). Until now, we ran cross-sectional regressions using firm-level variables averaged over time. For the sake of comparison with this literature, we now present panel regressions using year-by-year variables, normalizing R&D spending by (lagged) fixed assets. Table VII displays these regressions for R&D-intensive firms in electronics and in the entire sample, respectively. We perform GLS panel regressions (weighted by total sources) where the dependent variable is R&D expenditures normalized by lagged fixed assets (property and equipment).48 The left-hand variable and the right-hand variables (including TABLE VII Factors Affecting Year-by-Year R&D Expenditure in Manufacturing Firms Publicly Traded on the Tel Aviv Stock Exchange and Israeli U.S.-Listed Firms, 1990–1997* (GLS Regressions; Dependent and Independent Variables Measured as Deviation from Mean Over Time; Standard Error in Parentheses). R&D-intensive firms in electronics** dependent variable: R&D expenditure= lagged property and equipment All R&D-intensive firms dependent variable: R&D expenditure= lagged property and equipment
0.328 (0.077)þþ
0.298 (0.062)þþ Lagged Tobin’s q 0.057 (0.038) 0.058 (0.030)þ Leverage (total assets minus equity=total assets)
0.164 (0.177)
0.147 (0.144) Year dummies Investment in cash and securities=total uses YES
0.027 (0.035) YES
0.029 (0.027) 0.202 196 0.171 280 Lagged size (log-total assets) R-squared*** Number of firm-year observations *Regressions are weighted by total sources. The regressions allow for firm-fixed effects. **Including firms traded in the U.S. (most are in electronics). ***R-squared of the weighted regression. þ Significant at the 10 percent level. þþ Significant at the 5 percent level. 48 Ideally, we would want to normalize by the lagged stock of R&D capital, but we do not have a good measure of this stock. Hall (2002) surveys various studies for other countries that use measures of R&D stock. MATURE COMPANIES 445 Tobin’s q) are measured as deviations from their mean over time. In this specification any firm-fixed effects wash out. To control for potential liquidity effects on R&D expenditure, we include on the right-hand side the variable ‘‘Investment in cash and securities=total uses’’, which is a measure of the firm’s liquid assets. This variable should be greater the more internally generated cash is generated and the more external equity financing is raised. As mentioned, most of the firms in the sample raised external equity financing during the sample period. Therefore, this variable captures well the abundance in liquid assets. We find, in both regressions, that the variable is not a statistically significant determinant of R&D expenditure. This suggests that the majority of firms in the sample were not constrained by availability of cash and liquid assets in their R&D outlays in any given year. The coefficient on size is negative and significant in both regressions, indicating that as a firm grows larger (recall that this regression allows for firm-fixed effects), expenditure on R&D, relative to fixed assets, declines.49 In both regressions, we obtain a positive coefficient on Tobin’s q (statistically significant at the 10 percent level in the second column) suggesting that R&D expenditure (as a ratio of fixed assets) responds positively to increases in a firm’s market value. In this sense, for our sample of established firms, investment in R&D behaves like investment in physical capital and responds positively to Tobin’s q, in line with numerous studies for other countries surveyed in Hall (2002).50 3.3 The Stock Market Performance of R&D Intensive Firms Did investors on the Tel Aviv Stock Exchange benefit from investing in R&D-intensive firms? To address this issue, we begin by studying the sales growth (that investors often care about) of zero R&D and local R&D firms. (In what follows we omit U.S. R&D firms.) For each firm, we calculated firm-by-firm and year-by-year sales growth, and averaged the firm-by-firm sales growth over time. We regressed this measure on firm characteristics (also averaged over time). The results are not very interesting, and are not displayed in a table. Moreover, unlike most of the results reported in Tables V–VII, many of them are sensitive to the inclusion of the firms that paid R&D royalties to the government. We summarize the main findings; unless we say otherwise, the reported results hold whether the firms that paid R&D royalties to the government are included or not. Not surprisingly, the sales of younger firms grow faster. We also find that the sales of profitable firms grow more slowly, which is a bit surprising. Maybe profitable firms do not make as much effort to expand whereas less profitable firms do. The coefficient on the local R&D firm dummy variable is negative and significant, suggesting that the sales of zero R&D firms grow faster than those of local R&D firms, but the result disappears when the firms that paid R&D royalties to the government are excluded. The relative stock market performance of the firms in our sample exhibits a similar pattern. For the years 1990–1997, we calculated firm-by-firm and year-by-year excess stock returns (stock returns net of the riskless interest rate) for zero R&D and local R&D firms. We computed market ‘‘beta’s’’ (relative to the general index of the Tel Aviv Stock Exchange) and, using these ‘‘beta’s’’, we calculated risk-adjusted excess stock returns adjusted for dividends. For each firm, we averaged these risk-adjusted excess returns over time and regressed them 49 This is not inconsistent with the result in Table I and with the result in the second column of Table VIB, that larger firms perform more R&D. Those results refer to a cross-section of firms, whereas the results in Table VII refer to the R&D expenditure of a firm as a function of its own size over time. 50 Blass and Yosha (2002) found, for a similar sample, a positive and significant relation between investment in physical capital and Tobin’s q. 446 A. A. BLASS AND O. YOSHA on time-averaged firm characteristics. As for sales growth, we find that the risk-adjusted excess returns of younger firms are higher. The coefficient on the local R&D dummy is positive and significant (at the 10 percent level), suggesting that the risk-adjusted excess returns of local R&D firms are larger than those of zero R&D firms, but the result disappears when the firms that paid R&D royalties to the government are excluded. 4 SUMMARY We studied the financing patterns of publicly traded R&D-intensive manufacturing firms in Israel, and presented, for the first time, adjusted flow of funds charts that treat R&D expenses as a capital outlay (rather than an operating cost that reduces profits, as standard accounting principles prescribe). We also addressed the question of how R&D inputs should be measured – using R&D expenses or R&D personnel. We constructed both expenditureand personnel-based R&D measures for each firm in our sample, and investigated to what extent these measures are mutually consistent. We found that firms that do not report R&D spending, to the extent that they do in fact engage in R&D, do so to a very limited extent, suggesting that the expenditure-based R&D-intensity measure is superior for a study like ours. This is an issue that requires further research. We also found that mature Israeli R&D-intensive firms that are traded on the Tel Aviv Stock Exchange exhibit a high market to book ratio (a high Tobin’s q). Another interesting finding is that, at least during the 1990s, R&D-intensive firms did not rely more on internally generated cash (unlike R&D-intensive firms in several OECD countries). This may be because they were able to raise inexpensive external equity financing on the booming exchange, especially in the first half of the decade. Acknowledgements This research project was conducted as part of the activities of the Science, Technology, and the Economy Program (STE) at the Samuel Neaman Institute for Advanced Studies in Science and Technology. Support for that project from the Institute is gratefully acknowledged. The authors further thank three referees, Saul Lach, and Robert Sauer for helpful comments, and Chanan Lavi and Renat Lokomet for excellent research assistance. Yosha thanks the Research Department, Bank of Israel for financial support. References Allen, F. (1993). Stock markets and resource allocation. In: Mayer, C. and Vives, X. (Eds.), Capital Markets and Financial Intermediation. Cambridge University Press, Cambridge=New York. Ber, H., Yafeh, Y. and Yosha, O. (2001). Conflict of interest in universal banking: Bank lending, stock underwriting, and fund management. Journal of Monetary Economics, 47, 189–218. Blass, A., Yafeh, Y. and Yosha, O. (1998). Corporate governance in an emerging market: The case of Israel. Journal of Applied Corporate Finance, 10, 79–89. Blass, A. and Yafeh, Y. (2001). Vagabond shoes longing to stray: Why foreign firms list in the United States. Journal of Banking and Finance, 25, 555–572. Blass, A. and Yosha, O. (2002). The reform in the Israeli financial system and the flow of funds of publicly traded manufacturing firms. In: Ben-Bassat, A. (Ed.), The Israeli Economy, 1985–1998: From Government Intervention to Market Economics. Essays in memory of Professor Michael Bruno, MIT Press, Cambridge, MA (forthcoming). Bond, S., Harhoff, D. and Van Reenen, J. (1999). Investment, R&D, and Financial Constraints in Britain and Germany. Institute for Fiscal Studies, Working paper Series No. W99=5. Bottazzi, L. and Da Rin, M. (2002). Venture capital in Europe and the financing of innovative companies. Economic Policy, 34, 229–269. MATURE COMPANIES 447 Bound, J., Cummins, C., Griliches, Z., Hall, B. and Jaffe, A. (1984). Who Does R&D and Who Patents? In: Griliches, Z. (Ed.), R&D, Patents, and Productivity. The University of Chicago Press, Chicago and London. Cohen, W. and Klepper, S. (1996). A reprise of size and R&D. Economic Journal, 106, 925–951. Cohen, W. and Klepper, S. (1992). The anatomy of industry R&D intensity distributions. American Economic Review, 82, 773–799. Fazzari, S., Hubbard, G. and Petersen, B. (1988). Financing constraints and corporate investment. Brookings Papers on Economic Activity, 1, 141–205. Griliches, Z. (1981). Market value, R&D and patents. Economics Letters, 7, 183–187. Griliches, Z. and Regev, H. (2001). R&D, government support, and firm productivity in Israeli industry, 1975–1994. In: Spivak, R. (Ed.), The Economic Evaluation of Technological Change. Proceedings of a Conference at NIST, Washington DC, US Government Printing Office. Griliches, Z. and Regev, H. (1999). Who is Doing R&D and Who is Getting R&D Grants. Harvard University, Mimeo. Hall, B. (2002). The financing of research and development. Oxford Review of Economic Policy (forthcoming). Hall, B. (2000). Innovation and market value. In: Barrell, R., Mason, G. and O’Mahony, M. (Eds.), Productivity, Innovation and Economic Performance, Cambridge University Press. Hall, B. (1992). Research and Development at the Firm Level: Does the Source of Financing Matter? NBER Working Paper No. 4096. Himmelberg, C. and Petersen, B. (1994). R&D and internal finance: A panel study of small firms. Review of Economics and Statistics, 76, 38–51. Horwitz, B. and Kolodny, R. (1981). The FASB, the SEC, and R&D. Bell Journal of Economics, 12, 249–262. Jaffe, A. (1986). Technological opportunity and spillovers of R&D: Evidence from firms’ patents, profits, and market value. American Economic Review, 76, 984–1001. Johnson, L. and Pazderka, B. (1993). Firm value and investment in R&D. Managerial and Decision Economics, 14, 15–24. Lach, S. (2000). Do R&D Subsidies Stimulate or Displace Private R&D? Science, Technology and the Economy (STE) Working Paper Series, no. 2, Samuel Neaman Institute, Technion, Israel. Mayer, C., Schoors, K. and Yafeh, Y. (2001). Sources of funds and investment strategies of venture capital funds: Evidence from Germany, Israel, Japan and the U.K. Mimeo. Said Business School, University of Oxford. Mulkay, B., Hall, B. and Mairaisse, J. (2001). Investment and R&D in France and in the United States. In: Bundesbank, D. (Ed.), Investing Today for the World of Tomorrow, Springer Verlag. OECD (1994). Main Definitions and Conventions for the Measurement of Research and Experimental Development (R&D). A Summary of the Frascati Manual 1993, OCDE=GD(94)84, Paris. OECD (1993). The Proposed Standard Practice for Surveys of Research and Experimental Development, Frascati Manual, Paris. Trajtenberg, M. (2000). R&D Policy in Israel: An Overview and Reassessment Science, Technology and the Economy (STE) Working Paper Series, no. 1, Samuel Neaman Institute, Technion, Israel. Yafeh, Y. and Yosha, O. (1998). Financial markets reform, patterns of corporate finance, and the continued dominance of large banks: Israel 1985–95. Economic Systems, 22, 175–199. APPENDIX: REMOVING OUTLIERS We removed from the sample firm-year observations in which we identified inconsistencies between the Dukas database and flow statements entered by hand. In this procedure, we used four key variables: net profits, flows from investment activities, flows from financial activities, and flows from operations. A discrepancy of 5 percent (provided it is greater than 5000 December 1990 NIS) in one or more of these variables led to the removal of the firm-year observation from the sample. 296 firm-year observations were removed for this reason. We also removed firms, which were extreme outliers in terms of the average (over time) of one or more source ratio, use ratio, or profitability ratio. Consider some ratio, x, compute the average of x over the sample years for a particular firm, and denote this average by x0 . Repeat for all the firms in the sample, and compute the average of x0 across firms, denoted x00 . Firms for which x0 minus x00 was larger than four standard deviations (of x0 ), for one or more source ratio, use ratio, or profitability ratio were removed from the sample. 49 firm-year observations were removed for this reason. We further removed firm-year observations for which total sources were negative (216 firm-year observations were removed for this reason), firms for which there are only one or two observations (9 firms), as well as firms for which the time-average of total sources was negative (13 firms).