vr my, . ad“ “‘9‘ xwfi 'u q ’aflméy 5;" .g v-fi «r 1" v.9 v ‘5 2t: - éfirfil 4; , ‘-' dé‘g‘i‘: 3/). .N < x I 4“.'~ l a". .«f ‘3‘ .fi... ,4, . jag rt‘uuufi ’" "333.13; 3‘»; %;K” £11,)“ 3?" if -‘ V ‘.F\ any: '31:}6'6‘"m£'?~r3‘-7u .42; . a»: V.‘ m ., . .. iawfifiwpggq‘gfim" agagbfi' 511 Em“...w“°§;£§:t£l$'3§“c' ' 4” ‘03., V. -A-.- . as \‘r wry; ,. ‘ . . . E11550; {'5‘1‘5. - "1 '*“f,".-.'A‘ 1.” l .‘1 fig Il'” MICHIGAN I! 3;!!! !3!;!! 3IIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIIII 301037 9893 This is to certify that the dissertation entitled Firm Size and the Returns to R&D Investment in High-Technology Industries presented by Teresa M. Doyle has been accepted towards fulfillment of the requirements for Economics Ph- 1). degree in JMAM Major professor Date w MSU is an Affirmative Action/Equal Opportunity Institution 0-12771 LIBRARY Mlchlgan State Unlverslty PLACE IN RETURN BOX to remove this checkout from your record. TO AVOID FINES rotum on or before date duo. DATE DUE DATE DUE DATE DUE , \ MSU I. An Affirmative Action/Equal Opponunlty Institution mun-a. FIRM SIZE AND THE RETURNS TO R&D INVESTMENT IN HIGH-TECHNOLOGY INDUSTRIES By Teresa M. Doyle A DISSERTATION Submitted to Michigan State University in partial fulfillment of the requirements for the degree of DOCTOR OF PHILOSOPHY Department of Economics 1994 a"; n.“ 1" ~4-u—. 1 ". . b', 3“- \5 I0 ‘I L. , ‘ s '9 v. .1. A All". “4‘". >1 " ABSTRACT FIRM SIZE AND THE RETURNS TO R&D INVESTMENT IN HIGH-TECHNOLOGY INDUSTRIES By Teresa M. Doyle One of the recurring arguments in the policy discussions has been that domestic enterprises are too small to maintain technological leadership in an increasingly global marketplace. This argument is not new. Schumpeter (1942) postulated that there exists a positive relationship between firm size and technological progress, and that static economic efficiency should be foregone in order to achieve long-run technological progress. This study tests the Schumpeter hypothesis for five high-technology industries in which R&D investment is especially important: the computer, machinery, medical equipment, pharmaceutical and semiconductor industries. The methodology used in this paper is a standard asset valuation framework in which the firm chooses an investment strategy to maximize its expected present discounted value. Returns to R&D, or technological success, is measured by the change in the market’s valuation due to R&D investment over a five year period. In testing the Schumpeter hypothesis, a continuous relationship between firm size and retums to R&D investment is first assumed. Results indicate that only in the computer and medical equipment industry is Schumpeter supported. The second part of the empirical analysis does not assume this continuous relationship and tests for a threshold effect between firm size and R&D profitability. I test for parameter stability of the estimated coefficients, and find that parameter stability can be rejected for all five industries. This implies that the coefficients are not stable over firm size. By using a maximum likelihood approach, points of structural instability are determined and the model is reestimated by creating firm size-dummy interaction terms. By allowing for endogenously determined thresholds, this methodology is able to determine the definition of large and small firm size. Results are consistent with the first part of the empirical analysis; only in the computer and medical equipment industry is Schumpeter supported. In the other three industries (machinery, pharmaceutical, and semiconductor) Schumpeter cannot be supported. These results suggest that Schumpeter’s hypothesis needs to be qualified by closely analyzing industry conditions. g. .2 }. To my parents, pm Janice and Michael-P. Doyle t“. ' ‘ l-‘ 'I- 1' ' i ILA: ' ' . . ‘ n \ ' BUN-Walt. um i w w = ' 2 ' .5 . ID “if grew-n". ' :1'.!.‘ :-'.- =z'.‘u . ~n.’ . i' " r it; s Stat: 2 rut-mum {mam-ave: 8.7' a! . em NF. 1'. Im‘aifi L‘-‘ 3m ”Hfr'ie‘uv M swarm. 'urfltrh ’z.'.;'~. r » 3 . 2' ‘07.. r-"v‘i Q“ 1‘. . '1 V‘ '. . ' _‘ . 5354’ . rnr . l . t) ' l ." £~ . "v it 2“. - . ‘ I E' t v. ‘ . . .l l' ‘t_' l? .- A.d‘- “5'. «“31 ._‘_. .‘g 13-“ in“ ‘ ' ' .., r _ . ..,, . . l ' ' - . . . A ("V“gt‘n‘a‘ ‘K :Tmf ‘rksf'n‘ >t-k' {Iotl‘v'tL '. u #35 ml’vkls'd ',v_«’ .w‘k‘ .3 \ 7t t.“ ‘ é 9 ALT; sK‘ “’3’“ ~o. N*.“‘1-. Q ”a. I a” - \. ..g '7“.- C “ l“! 3 ' “‘La'. . K ”up 0 H \4‘. .‘~. . H; ., A N.l_§ ACKNOWLEDGEMENTS The completion of my dissertation would not have been possible without the assistance of my dissertation committee. The Chair of my committee, Professor Kenneth Boyer, provided me with invaluable advice and encouragement from the beginning of the process until the end. His insights greatly improved the quality of my dissertation. Professor Mary Bange and Professor Bruce T. Allen offered helpful comments and‘ encouragement throughout the many stages of the dissertation. Professor Mary Bange was always timely in responding to my drafts. I would especially like to thank Dr. Bruce Allen for his support of my work through one of his research grants. The Center for International Business and Educational Research helped support this study. Numerous people helped me at different stages of my PhD. program. My gratitude goes to the professors in the Economics Department for their guidance. and to my fellow Ph.D. students for their friendship and support. I am indebted to Jim LeBay from the Michigan State University Computer Center for his help in obtaining the data for this dissertation. A congenial office setting and employment was provided by Lake Forest College while I completed this dissertation. Finally, I would especially like to thank my parents for their support. My deepest gratitude goes to my husband. Thomas Klier, for his advice, patience, and encouragement. This dissertation would not have been possible without him. vi LIST OF TABLES LIST OF FIGURES INTRODUCTION TABLE OF CONTENTS 1. REVIEW OF THE LITERATURE A. B. C. D. MEASUREMENT OF TECHNOLOGICAL PROGRESS ARE INDUSTRIES HOMOGENEOUS? CRITICISMS OF EMPIRICAL TESTING R&D AND FIRM PERFORMANCE II. DESCRIPTION OF THE INDUSTRIES A. INDUSTRY DEFINITIONS 1. 2. 3. 4. 5. COMPUTER INDUSTRY MACHINERY INDUSTRY MEDICAL EQUIPMENT INDUSTRY PHARMACEUTICAL INDUSTRY SEMICONDUCTOR INDUSTRY INDUSTRY CONDITIONS vii page 1 11 12 21 23 28 28 29 30 32 33 34 35 TABLE OF CONTENTS (CONT) III. THEORETICAL RELATIONSHIP BETWEEN FIRM SIZE AND TECHNOLOGICAL PROGRESS A. INNOVATION AS A PROCESS 1. THE SERIAL MODEL 2. THE CHAIN-LINKED MODEL 3. THE SIMULTANEOUS MODEL 4. LINKAGES B. FRAMEWORK FOR EMPIRICAL ANALYSIS 1. THE FIRM’S MAXIMIZATION PROBLEM 2. TANGIBLE VS. INTANGIBLE ASSETS 3. RISK AND THE VALUATION MODEL 4. DIVERSIFICATION 5. RISK AND THE REQUIRED RATE OF RETURN C. MODEL 1. EMPIRICAL FORMULATION AND MEASUREMENT ISSUES 2. FIRM SIZE-INTANGIBLE ASSET INTERACTION TERMS 3. PARAMETER STABILITY viii 40 40 41 42 45 47 49 50 51 53 54 55 60 61 65 67 IV. REST] IV. RESULTS A. B. DATA TABLE OF CONTENTS (CONT) PARAMETER TESTS a. SPECIFICATION 1 b. SPECIFICATION 2 ESTIMATION RESULTS: FIRM SIZE-INTANGIBLE ASSET INTERACTION TERMS 1. 2. 3. 4. 5. COMPUTER INDUSTRY MACHINERY INDUSTRY MEDICAL EQUIPMENT INDUSTRY PHARMACEUTICAL INDUSTRY SEMICONDUCTOR INDUSTRY PARAMETER STABILITY l. 2. THE CUSUM AND CUSUM OF SQUARES TEST QUANDT’S LOG LIKELIHOOD RATIO TEST ESTIMATION RESULTS, PARAMETER STABILITY a. COMPUTER INDUSTRY b. MACHINERY INDUSTRY 70 7O 77 80 80 88 9O 91 91 92 92 99 99 107 110 112' APFENI APPEXI APPERI A‘DPC \7 9.). D. TABLE OF CONTENTS (CONT) c. MEDICAL EQUIPMENT INDUSTRY (1. PHARMACEUTICAL INDUSTRY 6. SEMICONDUCTOR INDUSTRY SUMMARY OF RESULTS V. SUMMARY AND CONCLUSIONS APPENDIX A: APPENDIX B: APPENDIX C: APPENDIX D: DEFLATION OF MARKET VALUATION TEST OF THE EQUALITY OF COEFFICIENTS OF THE FIVE HIGH-TECHNOLOGY INDUSTRIES HETEROSKEDASTICITY CORRECTION SPECIFICATION THREE RESULTS 113 114 115 145 154 156 161 163 5. SE) 10 l. LIST OF TABLES COMPUTER INDUSTRY: 10 Largest Firms Ranked by Sales (1989) page 30 MACHINERY INDUSTRY: 10 Largest Firms Ranked by Sales (1989) MEDICAL EQUIPMENT INDUSTRY: 10 Largest Firms Ranked by Sales (1989) PHARMACEUTICAL INDUSTRY: 10 Largest Firms Ranked by Sales (1989) SEMICONDUCTOR INDUSTRY: 10 Largest Firms Ranked by Sales (1989) INDUSTRY DESCRIP'IIVE STATISTICS INDUSTRIES USED IN ESTIMATION SUMMARY STATISTICS NUMBER OF OBSERVATIONS REPORTING VARIOUS STATISTICS SUMMARY STATISTICS BEFORE AND AFTER FILTER THE STOCK RESPONSE TO R&D INVESTMENT SALES-R&D INTERACTION TERMS PARAMETER STABILITY RESULTS: COMPUTER INDUSTRY SUMMARY OF COEFFICIENTS FOR THE COMPUTER INDUSTRY PARAMETER STABILITY RESULTS: MACHINERY INDUSTRY 31 33 34 35 39 81 82 85 97 117 118 121 LIST OF TABLES (CONT) SUMMARY OF COEFFICIENTS FOR SPECIFICATION 2 MACHINERY INDUSTRY PARAMETER STABILITY RESULTS: MEDICAL EQUIPMENT INDUSTRY SUMMARY OF COEFFICIENTS FOR SPECIFICATION 2 MEDICAL EQUIPMENT INDUSTRY PARAMETER STABILTY RESULTS: PHARMACEUTICAL INDUSTRY SUMMARY OF COEFFICIENTS FOR SPECIFICATION 2 PHARMACEUTICAL INDUSTRY PARAMETER STABILITY RESULTS: SEMICONDUCTOR INDUSTRY SUMMARY OF COEFFICIENTS FOR SPECIFICATION 2 SEMICONDUCTOR INDUSTRY SUMMARY OF RESULTS FROM SPECIFICATION 1 AND 2 STANDARD AND POOR’S COMPOSITE INDEX BASIC MODEL THE STOCK RESONSE TO R&D INVESTMENT SIZE-DUMMY INTERACTION TERMS THE STOCK RESPONSE TO R&D INVESTMENT SUMMARY OF COEFFICIENTS FROM SPECIFICATION THREE xii 122 123 124 125 126 127 128 141 154 158 170 171 TX». La. Ill TM Md. 12: 13: 14: LIST OF FIGURES THE SERIAL MODEL THE CHAIN LINKED MODEL THE SIMULTANEOUS MODEL 73>0 and y,>0, SCHUMPETER SUPPORTED y,>0 and y,<0, SCHUMPETER SUPPORTED y,>0 and y4<0, SCHUMPETER NOT SUPPORTED y,<0 and 74>0, SCHUMPETER NOT SUPPORTED, Scenario A y,<0 and y,>0, SCHUMPETER NOT SUPPORTED, Scenario B y,<0 and y,<0, SCHUMPETER NOT SUPPORTED SPECIFICATION 2, SCHUMPETER SUPPORTED SPECIFICATION 2, SCHUMPETER NOT SUPPORTED THE EFFECT OF SIZE ON R&D PROFITABILITY. COMPUTER INDUSTRY THE EFFECT OF SIZE ON R&D PROFITABILITY. COMPUTER INDUSTRY, EXCLUDING LARGE FIRMS THE EFFECT OF SIZE ON R&D PROFITABILITY. MACHINERY INDUSTRY THE EFFECT OF SIZE ON R&D PROFIT ABILITY . MEDICAL EQUIPMENT INDUSTRY xiii page 41 513 45 72 73 74 75 76 77 77 79 94 94 95 95 -1 _.. ' ’- j.- ' ud' ‘ A ‘V-m. "111* s. :0: ‘1' hi. v, 1.3. :4: 20: 21: 22: 23: 24: 25: 26: 27: 28: 29: LIST OF FIGURES (CONT) THE EFFECT OF SIZE ON R&D PROFITABILITY. PHARMACEUTICAL INDUSTRY THE EFFECT OF SIZE ON R&D PROFITABILITY. SEMICONDUCTOR INDUSTRY CUSUM TEST, PARAMETER STABHJTY CUSUM TEST, PARAMETER STABILITY REJECTED CUSUM OF SQUARES TEST-PARAMETER STABILITY CUSUM OF SQUARES TEST: PARAMETER STABILITY REIECT'ED CUSUM TEST-COMPUTER INDUSTRY CUSUM OF SQUARES TEST-COMPUTER INDUSTRY QUANDT’S LOG LIKELIHOOD RATIO TEST-- COMPUTER INDUSTRY THE STOCK RESPONSE TO R&D INVESTMENT-- COMPUTER INDUSTRY THE STOCK RESPONSE TO R&D INVESTMENT-- COMPUTER INDUSTRY (EXCLUDING FIRMS WITH SALES>$20 billion) CUSUM TEST-MACHINERY INDUSTRY CUSUM OF SQUARES TEST--MACHINERY INDUSTRY QUANDT’S LOG LIKELIHOOD RATIO TEST-- MACHINERY INDUSTRY xiv 96 96 103 103 105 106 129 129 130 130 131 132 132 133 S...“ Q. ‘1- CD. \‘. ~11: 42: 30: 31: 32: 33: 34: 35: 36: . 37: 38: 39: '41: 42: LIST OF FIGURES (CONT) STOCK RESPONSE TO R&D INVESTMENT—- MACHINERY INDUSTRY CUSUM TEST--MEDICAL EQUIPMENT INDUSTRY CUSUM OF SQUARES TEST-MEDICAL EQUIPMENT INDUSTRY QUANDT’S LOG LIKELIHOOD RATIO TEST-- MEDICAL EQUIPMENT INDUSTRY THE STOCK RESPONSE TO R&D INVESTMENT-- MEDICAL EQUIPMENT INDUSTRY CUSUM TEST--PHARMACEUTICAL INDUSTRY CUSUM OF SQUARES TEST--PHARMACEUTICAL INDUSTRY QUANDT’S LOG LIKELIHOOD RATIO TEST—- PHARMACEUTICAL INDUSTRY THE STOCK RESPONSE TO R&D INVESTMENT -- PHARMACEUTICAL INDUSTRY CUSUM TEST--SEMICONDUCT OR INDUSTRY CUSUM OF SQUARES TEST--SEMICONDUCTOR INDUSTRY QUANDT’S LOG LIKELIHOOD RATIO TEST-- SEMICONDUCTOR INDUSTRY THE STOCK RESPONSE TO R&D INVESTMENT-- SEMICONDUCTOR INDUSTRY XV 133 134 134 135 135 136 136 137 137 138 138 139 139 4‘0.) ,- U.-.c. .. .73. , V~i~AL rt r-O ‘ y l “H L‘. at. 1:. ‘- m L\z..s I ‘~&LI!Z“ s." 35- A. \Vu‘ ”\J.‘ am an H .1 a \ ~XI. )3? ‘C‘AI' 39.} I1"“: ‘33' Kg. V. INTRODUCTION Innovation is central to economic growth. Yet, innovation is a complex matter, differing from one industry to another. Knowing that technological innovation can improve economic growth, governments try to encourage it. However, little is understood about how technological advance occurs or, more important, why some companies are better than others at turning it to their advantage. This Study investigates an important aspect of the technological innovation process. Specifically, it examines whether firm size is a significant determinant of R&D profitability in high-technology industries. The relationship between firm size and technological progress has, over the years. come to be known as the Schumpeter hypothesis. Although classical economists recognized that technical advance facilitated economic growth, they failed to develop an economic theory of technological progress.1 Joseph Schumpeter was the first to derive an economic theory of technological progress. He noted that technology is not an exogenous phenomenon: technology is the result of innovation, and innovation is an endogenous activity (Schumpeter ( 1942)). An important theme of Schumpeter’s theory ‘ For example, Adam Smith (1937) identified the two main ingredients that make technical advance an economic theory: to gain a competitive advantage and the requirement of investment. Smith, however, did not combine these points into a theory. Marx (1919) analogized an economic system to a biological entity that is ever changing compared with a chemical reaction that is tending toward an equilibrium. Marx appeared, however, to view technical advance as a change in the environment that capitalists reacted to instead of initiated (Kamien and Schwartz (1982)). 1 D In: ~94e~<01' ALL...» nu"a‘ l~.Lb»~ u 3 a \n1-II: . ‘. 5 t‘ t 2 is that there exists an incompatibility between perfect competition and entrepreneurial activity. When considering long-run technological progress. the former should be rejected. Schumpeter defended the current capitalist system with its large firms and oligopolistic markets.2 While he conceded that perfect competition could theoretically assure efficient allocation of resources at a point in time, Schumpeter maintained that it would hinder long-run efficiency by stifling innovative activity: The large scale establishment or unit of control... has come to be the most powerful engine of...progress and in particular of the long—run expansion of total output not only in spite of, but to a considerable extent through...strategy which looks so restrictive when viewed in the individual case and from the individual point of time. In this respect, perfect competition is...inferior, and has no title to being set up as a model of ideal efficiency. (Schumpeter (1942), p. 106) That is, Schumpeter believes that the process of creative destruction is more important than price competition. In introducing new products or processes a firm must be able to engage in monopolistic practices in order to secure profits from its investments. Although these profits will eventually be eroded by competition, they must last long enough to make innovation worthwhile. Society will benefit from technological progress, and the supranorrnal profits will enable the firm to continue to invest in R&D.3 Schumpeter’s assertions were vague, and it was left to others to fill in blanks and omitted steps. 2 Schumpeter's conjectures on the relationship between firm size and innovation wavered in his writings. For example. in Theog of Economic Development, Schumpeter argued that innovation typically arose from smaller firms. See Scherer (1992) for an overview of Schumpeter’s writings. 3 This hypothesis has two distinct elements. First, there exists a positive relationship between firm size and technological progress and second, a positive relationship between market power and technological progress. Schumpeter did not distinguish between the two hypotheses; he considered size and market power to be inextricably linked. (Baldwin and Scott(l987), Schumpeter(l939), and Scherer(1992)) 3 Galbraith (1952) extended Schumpeter’s hypothesis and argued that because of the cost and uncertainty involved, R&D could only be undertaken by large firms: Most of the cheap and simple inventions have, to put it bluntly, been made. Not only is development now sophisticated and costly but it must be on a sufficient scale so that successes and failures will in some measure average out...Because development is costly, it follows that it can be carried on only by a firm that has the resources associated with considerable size. (Galbraith (1952), pp. 91-92). The importance of technical advance as an economic theory was helped by Solow’s (1957) study which found that 90% of the doubling of per capita output in the US. non- farm sector was the result of technical advance and only 10% was due to an increase in the capital-labor ratio. Since Schumpeter, economists have been testing for a relationship between firm size and technological progress. To date, the overall results have been inconclusive (see Baldwin and Scott(1987), Karnien and Schwartz(1982), and Scherer(1992)). From the point of view of this study, these papers are deficient in four respects. First, the majority of the studies undertaken have utilized insufficient proxies for technological progress. Tests of the Schumpeter hypothesis have primarily focussed on whether inputs to the innovation process increase more than proportionately with firm size, using R&D expenditures or R&D employment as proxies for innovative inputs (Comanor (1967), Scherer (1984) and Bound et al.(1984)). However, R&D expenditures and employment are insufficient proxies of technological progress. Expenditures and employment are inputs in the innovation process; however it is a large step to conclude that innovative inputs equal outcomes of the innovative process (innovative outputs). Other studies have addressed this problem by testing the relationship between firm size and the number of L. a r bum"; v.15“- dfiwamn .1154. Al 5‘ ’ 7T. LLOFAL AAA SILLCV E .1 “WW! *7 Beta; 11 14:, - Pa . Urgjbr‘ig sendifik‘r VII” I‘M "‘La\\ 1 4 patents registered, which is a measure of innovative output (see Bound et al.(1984) and Scherer(1965)). Unfortunately, simply adding the number of patents neglects qualitative differences among them; many patents reflect minor improvements and have little economic value, while others are extremely valuable (Griliches (1990)). Instead, this study examines the change in a company’s market valuation due to investment in R&D over a five year period. It is argued that the stock market takes into account information about firms’ expected future profitability, and therefore expected outcomes from R&D investment. This methodology avoids the problems of using innovative inputs and/0r patents as a proxy for technological progress. Second, the approach of pooling different industries takes a simplistic view of the Schumpeter hypothesis; technological opportunity and appropriability conditions are assumed to be identical across the industries. Technological opportunity refers to differences in the opportunities that industries face for research. Appropriability conditions refer to a firms ability to appropriate sufficient returns to make the innovation worthwhile. However, there is evidence that small and large firms’ innovative activity responds to different technological and economic regimes (Acs and Audretsch (1987, 1988, 1991)). In order to test the Schumpeter hypothesis, five high-technology industries are analyzed: the computer, machinery, medical equipment, pharmaceutical, and semiconductor industries. These industries are similar in that technological opportunity is high in all five, but they differ in appropriability conditions. This paper tests for a relationship between firm size and R&D profitability and relates these results back to industry conditions. A- i a! 111.431.?“ the "n ‘v s LA. I fifl'nqun.‘. fiVUAAIHA I I Lr-A\r~ ‘. u».,t.'\ 11 C0 .IIZ'JI EEK; . 1;; It not r omit tr Pirate: mini 5 Third, previous studies use current R&D expenditures or employment data as a measurement of technological progress or innovation. Yet. innovative effort is not only the knowledge generated in any one period, but the accumulated stock of knowledge. This study employs panel data and is therefore able to test a hypothesis using an accumulated stock of R&D knowledge. And fourth, it has been argued that Schumpeter’s theory must be viewed as a threshold theory. That is, technological progress is not necessarily an increasing and continuous function of market power or business size (Markham (1965)). Most all of the previous studies use empirical methods in which they assume a continuous function and do not explicitly test for a threshold effect. This study uses empirical methodology in order to test for a threshold effect. First, a test of parameter stability is employed. If parameter stability is rejected, statistical analysis is able to determine the break point(s) for the size regimes, which allows "large" and "small" firm size to be statistically defined instead of assuming a priori what these size regimes are. Separating firms into different size regimes, a model is estimated to test whether firm size is a significant determinant to R&D profitability. A. ADVANTAGES OF SMALL FIRMS VS. LARGE FIRMS Three general arguments have been presented in support of Schumpeter’s assertion of a positive relationship between firm size and technological progress. First. there are economies of scale in R&D expenditures, which implies that a larger R&D staff operates more efficiently than a small one. This may be due to researchers being more i. {“1 awn "‘ »S\ ..l .. T36 I‘ " ~~J LA “IAAAL 6 productive when they have more colleagues with whom to interact, or when a large research group allows for division of labor. Second, an R&D program of a given size operates more efficiently in a larger firm. Mansfield et al.(1971) identified three critical probabilities associated with economically successful R&D projects, only one of which deals with technical success. The other two critical probabilities depend upon development and successful commercialization. This leaves room for other activities taking place in a firm to aid in turning successful research into an economically valuable product or process. For example, marketing and financial planning are better developed in large firms, which allows these firms to have an advantage over their smaller counterparts. Furthermore, Nelson (1959) argues that a large firm with an established name may be in a better position to exploit the development of innovation because it is easier for a large firm to enter into a new market compared to a firm without an established name. Another reason why large firms may have an advantage is due to diversification. Since it is impossible to predict the outcome of an R&D project, a given unit of R&D expenditures may or may not produce knowledge beneficial to particular project. If a firm is engaged in only one project, there is a higher risk involved in producing knowledge the firm is not able to use. compared with a firm engaged in a large number of projects. Thus, the more diversified a firm, the less risky is R&D, and the more a firm will gain, on average, from a given unit of R&D expenditures.4 Third, economies of scale in the financial market exist. There may be capital market imperfections in which large firms have advantages in securing finance for R&D projects. This may be due to size of the firm being ‘This argument assumes there is no market for knowledge per se. vwa 1.)” 1 L01 .Ladl'ss 0 105.131 I A dzsclcsur: H. ;—vym ‘ H fIYLPUIAIu't‘ \L ‘ [C‘ LAI‘I _l:\': ‘ Cl [.713 "2.: I - §‘,V"!“ ». ~I:...‘ 1\Q I“ . {INA ‘ .k‘aJu U” T C1114“ .‘kxl- . it I 7 correlated with the availability and stability of internally generated funds? Even with a perfect capital market, self-financing may be more desirable because borrowing requires disclosure of the very information that constitutes the innovation. Large firms are presumably better able to undertake internal financing than smaller firms. Yet, the Schumpeter hypothesis of a positive relationship between firm size and technological progress is not without significant challenge. One of the most influential of the early studies is that of Jewkes, Sawers, and Stillerman (1969). They studied 61 significant inventions made in Great Britain and the United States from 1900-1950, and found that: More than one-half of the cases can be ranked as individual invention in the sense that much of the pioneering work was carried through by men who were working on their own behalf without the backing of research institutions and usually with limited resources and assistance or, where the inventors were employed in institutions, these institutions were, as in the case of universities, of such a kind that the individuals were autonomous, free to follow their own ideas without hindrance.‘S (p. 82) In fact, they found that only 12 of the 61 inventions were attributed to large corporations. There are many reasons why small firms may have an advantage in innovation. First, small firms may have a greater incentive to innovate when they can expect to capture substantial portions of the market by gaining first mover advantages.7 Second, there may be disadvantages to size. Diseconomies of scale may occur if research ’With the growth of venture-capital institutions, this has probably become less important. ‘Jewkes, Sawers, and Stillerman (1969) did concede that development of an invention could be costly and could require the resources of a large firm. 7 This is due to a smaller firm being able to capture a larger absolute amount of the quasi-rents due to cannibalization. See Arrow(1962), Scherer and Ross(1990), and Rosen(1991). ECIUI‘I R&Dpr Sniafl It Q’:~QF o bit-Nuns- 5 un' L .4 ”SET 1.. . «l _,, 13171155 starts .11 15:53:31 F?‘ .3; MAWU “131:5: 8 decisions pass through layers of management. Different perceptions of risk may stifle ambitious innovation. For example, Rosen (1991) shows that large firms will invest in R&D projects which are safer and that supplement rather than replace current technology. Small firms tend to invest in riskier R&D projects. This may result in small firms earning a higher return on their R&D investment. Still, diseconomies of scale in R&D, rather than the economies mentioned earlier, could occur if a large number of people are involved in a project. Scientists and entrepreneurs may become less motivated in large firms as their ability to capture the benefits from their efforts diminishes.s Also, higher status and pay in a large firm often leads to management positions, instead of research (Scherer and Ross (1990)). In sum, theory does not predict an unambiguous relationship between firm size and technological progress. The relationship may vary by industry; it is left to empirical work to determine. B. INDUSTRY CONDITIONS Underlying factors in industries may explain advantages of firm size in innovation. These are usually classified into two broad groups: technological opportunity and appropriability conditions. Levin et al.(1987) investigate the extent to which appropriability conditions of both product and process patents differs across industries. They found that the effectiveness of patents in preventing duplication are limited. The major reason given for the limited effectiveness of patents is that competitors are able to 3 There are numerous examples of new enterprises founded by discouraged scientists from large US firms such as Hughes Aircraft, IBM, Sperry Rand(Unisys), Texas Instruments, and Western Electric. (See Scherer and Ross(1990)) 9 "invent around" patents. Yet, the survey data Show other mechanisms, such as investments in complementary sales and service efforts, first mover advantages, and the ability to move down the learning curve to be quite effective in appropriating the returns from innovation. The limited effectiveness of patents indicates that it is not enough that the research activity itself is successful; a firm must also be able to organize the other parts of the process efficiently and effectively for successful commercialization. That is, research must be combined with other aspects of the process, so-called complementary assets. These complementary assets could be considered services, such as after-sales support in the case of computers, or marketing in the case of medical equipment, in which dissemination of information about the equipment is essential. When these complementary assets are not present, competitors and/or imitators are likely to profit from the innovationuat times more than the firm that was first to innovate. In addition, large returns can be made from a modest technological advance if the company has the appropriate complementary assets. Because of the high cost and uncertainty of research, and the importance of complementary assets, firms are most likely going to require access to capabilities which lie outside their organization in the process that starts with the concept of an idea and eventually might lead to the successful marketing of a new product. Examples include university research consortia, as well as horizontal and vertical linkages. Most important, especially for small firms, are the vertical linkages. That is, two or more firms may address particular aspects of the innovation process and therefore not undertake all of the steps in-house. Thus, firms need not perform all activities in-house for innovation to occur and to be successful. 10 Results suggest that these high—technology industries are not homogeneous and that firm size is a significant determinant to R&D profitability. But since industries are not homogeneous, the relationship between firm size and R&D profitability differs across industries. For example, the stock response to R&D investment in the medical equipment industry for firms with sales greater than $400 million is positive and significant, while it is not significantly different from zero for firms with sales less than $400 million. In the Industrial machinery industry, small firms are found to be those with less than $79 million in annual sales, while large firms are those with greater than $308 million in sales. Results suggest that the stock response to R&D investment in both small and large firms is greater than that for the average size firms. These results aid us in understanding why previous studies of the Schumpeter hypothesis are inconclusive. Pooling industries ignores the fact that industries are heterogeneous; in some instances large firms are the "engines of technological progress", while in others, small firms appear to be more innovative. This paper is organized in the following manner. The literature review is presented in Chapter I, while Chapter 11 includes definitions and descriptive statistics of industry conditions in the five high-technology industries studied. Chapter III presents the theoretical relationship between firm size and technological progress. Empirical results are discussed in Chapter IV, and conclusions and prospects for future research are offered in chapter V. .‘3 ‘Ph “Anni ,_ VITA" H4 1‘ ‘ '1 C a rm: I. REVIEW OF THE LITERATURE Schumpeter argued that there exists a relationship between firm size and technological progress. However, it is difficult, if not impossible, to measure technological progress using an unambiguous, unidimensional scale. Even if a comprehensive list of all innovations were available, the contribution of an innovation to economically valuable knowledge is difficult to assess. Despite these difficulties, a variety of proxies for technological progress have been introduced. This chapter presents a review of the literature concerning the relationship between firm size and technological progress. I will first discuss the three major deficiencies in the literature: first, the majority of the studies utilize insufficient proxies for technological progress; second, many ignore industry conditions. And third, many of the studies do not test Schumpeter’s original hypothesis. Throughout this section I discuss what conclusions we are able to draw from these studies and I will explain the advantages of my study and how it differs from previous work. Last, I discuss how this study expands on some recent literature which has analyzed the effectiveness of R&D expenditures. "PM?" ulbh‘yw‘ mpug b:- muf“: Of 12 A. MEASUREMENT OF TECHNOLOGICAL PROGRESS The majority of the studies concerned with the relationship between firm size and technological progress typically test whether technological progress increases more than proportionately with firm size. These studies proxy technological progress using a measure of innovative inputs or innovative outputs. For example, regression studies typically regress innovative inputs or outputs (X), on a measure of firm size (S): log(X,) = a + blog(S,) + u,. The Schumpeter hypothesis would then be supported if b>l. Another method is to deflate innovational intensity by firm size: (X/S)i = a + bSi + u,. In this case, the Schumpeter hypothesis is supported if b>0. Some of the more recent studies include quadratic terms in the regression equation. Yet, the proxies that the literature use are insufficient in measuring technological progress. First, innovative inputs, such as R&D expenditures or personnel engaged in R&D are deficient to the extent that they indicate resources allocated toward enhancing technological progress, not their outcome (Karnien and Schwartz (1982)). Measuring inputs by the number of personnel employed in R&D assumes that they are the only source of innovation. It may be that someone directly involved in the production process, or someone in marketing initiates the new product or process. R&D employment and expenditure data are also subject to considerable error in reporting; firms are given considerable latitude in classifying activities used for financial reporting.1 1For example, Standard and Poor’s Computstat data, which obtain data mainly from 10—K reports to the Securities and Exchange Commission, permit a more liberal definition of R&D expenditures than the National Science Foundation survey data. Cohen and Mowery(1984) find that Compustat data indicates firms conduct 12% more R&D, on average, than indicated by the FTC Line of Business data which uses the more restrictive 35133-le IICLICI'S l rescued 13 Researchers have recognized this, and have employed measures of innovative outputs, such as patents and the number of significant innovations, as a measure of technological progress. Unfortunately, simply adding the number of patents neglects qualitative differences among them; many patents reflect minor improvements and have little economic value, while others are extremely valuable. For example, a survey of patent holders conducted in the 1960’s at the Patent and Trademarks Foundation finds that the reported economic gain is highly dispersed. The mean economic value for patents that were reported to be in use and have a positive economic gain was $577,000 per patent. Yet, the reported standard deviation was $1.5 million, and the median economic value per patent was approximately $25,000. A separate analysis, which includes the above group of patents plus all patents that were not in use and had no economic gain or economic loss, finds that the mean economic value falls to approximately $112,000($473,000 in 1988 dollars).2 Further, many important innovations are not patented, most often process innovations.3 The most difficult to obtain, but probably the best measure of technological progress is the number of significant innovations.4 Yet, like patents, innovations are still NSF data. 2The National Science Foundation conducted two studies in the 1980’s which attempted to evaluate the economic value of patents. In 1982, the Chemistry program of the NSF found the average economic value of a patent to be $500,000 (Cutler, 1984). An analysis of patents by engineering grantees found that the royalty potential was $73,000, but had a very large dispersion. 3See Griliches (1990) for a survey of patent statistics as economic indicators. ‘ Data on significant innovations have been assembled for selective industries such as the semiconductor (Tilton, 1971) and pharmaceutical industries (Pelzman (1973), Schwartzman (1976)). Data compiled by the US. Small Business Association identified an leg: the SJ. 5311 siz- lllCTSlxlIl 19.5 C) rtsPe-ct t I bClW°=q \ryt mare R5 conflict \ [DOSE of m~l~.4 - Eluucb A numb.- befits; \qn “Std“? 14 an imperfect measure of technological progress in that all innovations do not contribute the same amount to technological progress and differ in economic value. One of the earliest empirical studies relating firm size and innovational inputs utilized data for 387 Fortune 500 Companies in 1960. Hamberg (1964) found a positive but weak association between R&D expenditures and firm size, using employment as a proxy for firm size. Using the same data, but a slightly different model, Comanor(1967) found weak support for the Schumpeter hypothesis. Using a sample of 67 firms from 1957-1960, Mueller (1967) found that research intensity was negatively associated with firm size. More recently, Scherer (1984) found mild support for a Schumpeterian increasing returns hypothesis, using Line of Business data. Bound et a1. (1984) utilize 1976 Compustat data; their results suggest that the elasticity of R&D expenditures with respect to sales is close to unity, but also significant nonlinearities in the relationship between sales and R&D expenditures. This implies that both large and small firms are more R&D intensive than average sized firms. Bound et al. suggest that their results conflict with earlier studies because of the inclusion of smaller firms. They point out that most of the previous work is based on Fortune 500 Companies, while their data set includes a large number of small publicly traded firms. In sum, these studies appear to be inconclusive as a group: they neither support nor refute the Schumpeter hypothesis. A number of studies looked directly at innovative outputs; they test the relationship between firm size and patent counts or the number of significant innovations, which are used as proxies. Scherer (1965) used the number of patents issued to a firm in 1959 as a proxy for innovations introduced in 1982. 15 average inventive output 4 years earlier. His results indicate that smaller firms were responsible for a higher relative share of innovation: "If anything, the results show that fnms below a half billion dollar sales mark generate more inventions relative to their size than do giant firms" (p. 1114). Using Line of Business and patent data, Scherer (1984) found that the majority of the industries reveal constant returns to scale in innovation. Bound et al.(1984) found support for both Scherer’s earlier (1965) study and his 1984 study: the small firms have a larger ratio of patents per dollar of R&D expenditures than larger firms. Yet, after R&D expenditures reach $1 to $2 million, patents per dollar of R&D expenditures remain nearly constant except for the very large firms. This result supports Scherer’s later study. One reason why we may find conflicting results from Bound et al. and Scherer (1984) is due to the data employed. Bound et al. utilizes a more comprehensive data set which includes many small publicly traded firms, while the Line of Business data include only larger firms. Acs and Audretsch (1991) examine the relationship between firm size and the rate of technological change with a measure of innovative activity. Their data set includes 1,695 US. manufacturing firms for 1982. The authors find no evidence to support the hypothesis of increasing returns to innovative output’ In sum, there is little support for the hypothesis that innovative outputs increase more than proportionately with firm size. In fact. evidence suggests that innovative outputs may be inversely related to firm size, which supports the hypothesis of diseconomies of scale. On the other hand, studies using innovative inputs as proxies for 5Acs and Audretsch (1989) compare the relationships between the number of patented inventions with their innovation data. They found that patented inventions provide "a fairly good, although not perfect, representation of innovative activity". Etflffll‘g‘tm Swiss. rtsu' 1mm: prams used This 3 $165an we prom 511113.331 Ilg I'l‘rlIC [he Ill: “In; “C Lll'o are high}- he 8' ARE fad Some St SChu an. _,, Al‘p\ Lsr I Nah, \ KAI P0333, It‘ll l6 technological progress appear to be mixed. Thus, taking into consideration all of these studies, results are inconclusive. If size is a significant determinant of technological progress, it is not overwhelmingly so. Yet, again, these studies are deficient because the proxies used for technological progress are inadequate. This study, instead, examines the response of a company’s stock value to R&D investment. The stock market takes into account all information about firms’ expected future profitability, and therefore expected outcome from R&D investment. Thus, while still utilizing R&D expenditure data, R&D expenditures will be an independent variable, while the indicator of inventive output is the change in the stock market value of the firm. Thus, we avoid the problems associated with using patent data, in which economic value’s are highly heterogeneous, as well as the assumption that innovative inputs are identical to technological progress. B. ARE INDUSTRIES HOMOGENEOUS? The second criticism of the existing literature is that the majority of the studies pool high and low technology firms and take no account of differences in industries. In fact, some studies which do recognize industry differences find that any support for the Schumpeter hypothesis disappears after industry effects are taken into account. Nelson, Peck, and Kalachek (1967) noted that previous analyses which found a positive relationship between firm size and innovative inputs may have been misinterpreted. That is, in analyzing the entire economy larger firms were found to spend a higher proportion on R&D. Yet, Nelson et a1. find that this was due in large part to the l7 fact that certain industries, such as aircraft, electronics, and chemicals, were characterized by both greater than average R&D intensity and firm size. They found the relationship within individual industries to be much less pronounced. Soete (1979) utilized R&D data compiled by Business Week magazine for over 700 large corporations for 1975 and 197 6. Overall, results suggested that R&D expenditures as a percentage of sales increase with firm size, being particularly high for the six largest firms. Yet, no clear pattern of an increasing or decreasing relationship emerged when he separated firms into 17 industry groups.6 Unfortunately, his analysis on industries tests only for a linear relationship. Scherer (1984) combined Line of Business Data with a set of patent data. For each of 196 industries, he calculated elasticities of R&D with respect to firm size (sales). He found that the percentage of industries indicating constant returns to scale with respect to firm size and R&D spending and firm size and patenting to be 71.4% and 73.4%, respectively. The percentage of industries indicating decreasing returns with respect to patenting was slightly higher than those reporting increasing returns: 11.3% vs. 15.3%. On the other hand, the elasticity of R&D spending with respect to R&D spending revealed increasing returns for 20.4% of the industries, while decreasing returns occurred in 8.2% of the industries. Thus, in a majority of the industries size is not a significant factor for technological progress--industry differences are important. As Mansfield et al. notes (1971): “Soete found the scale coefficients to be significantly greater than one in the automotive, office equipment, and instrument industries. While the scale coefficient was significantly less than one in the textile industry. The aircraft, paper, machinery, fuel, rubber, drugs, electrical and chemical industries were all found to exhibit constant returns to scale. So which m Inttw'stlt Technolt for F333;; transfer: Conditio mics-:11; term i Expires aIPIIin "Em: 18 In some industries. like steel, the biggest firms have carried out relatively few innovations; in other industries, like chemicals, the biggest firm seems to have performed very well.... This diversity is strikingly at odds with some of the simple generalizations found frequently in the literature. Some recent studies have begun to investigate underlying factors in industries which may provide us with a greater understanding of why some industries are more innovative than others. These are usually classified into two broad groups. Technological opportunity refers to industries differing in the opportunities they face for research. That is, there are interindustry differences in production possibilities for transforming research into successful products and/or processes. Appropriability conditions refer to the firm’s ability to appropriate sufficient returns to make the investment worthwhile. A patent allows perfect appropriability for a limited period in return for public disclosure, which guarantees diffusion of knowledge after the patent expires. Levin, Cohen, and Mowery (1985) employ technological opportunity and appropriability survey data to test whether these variables are able to explain interindustry variation in R&D intensity.7 Results indicate that the size of the firm is positively associated with business unit R&D intensity. Yet, when controlling for industry effects, the effect of size is insignificant. The industry characteristics perform best in the chemical industry, where the vectors of opportunity and appropriability variables are each 7The technological opportunity variables are intended to capture three dimensions: closeness to science, the importance of technical knowledge, and industry maturity. Appropriability conditions are measured with two indices: (1) the effectiveness of six mechanisms used by firms to capture and protect the competitive advantages of new processes and products, and (2) the range of imitation costs and time lags for major and minor, process and product, and patented and unpatented innovations. 19 jointly significant. In the electrical and machinery industries, the technological opportunity variables are found to be significant, while the appropriability variables are not. In the food industry, neither the appropriability nor the opportunity variables are statistically significant. Using a unique approach in investigating technological opportunity, Jaffe (1986) constructs a technology based classification of technological opportunity. He utilizes the distribution of the firms’ patents over different patent classes to characterize the technological position of the firm. The vector of technological opportunity dummies is statistically significant in explaining interfirrn differences in patents, profits, and Tobin’s q.a Yet, Jaffe also finds that conventional industry dummy variables perform equally well. Levin et al.(1987) investigate the extent to which appropriability conditions of both product and process patents differs across industries, and thus could explain a portion of interindustry variation in patenting. They find that only five industries of a possible 130 lines of business regard product patents as highly effective. While twenty industries regard them as moderately effective. On the other hand, only three industries regard process patents as moderately effective. The major reason given for the limited effectiveness of patents is that competitors are able to "invent around" patents. Yet, the survey data show other mechanisms, such as investments in complementary sales and service efforts, first mover advantages, and/or the ability to move down the learning curve, to be quite effective in appropriating the returns from innovation. These studies indicate the "Tobin’s q is defined as the market value of the firm divided by the replacement cost of the firm’s tangible assets. 20 importance of industry conditions in explaining technological progress. Acs and Audretsch (1987, 1988) suggest that the innovation activity of large and small firms is affected by different technological environments. Their results suggest that large firms (more than 500 employees) tend to have the relative innovative advantage in industries which are capital intensive, concentrated, highly unionized, and produce a differentiated good. On the other hand, small fnms (fewer than 500 employees) tend to have a relative advantage in industries which are highly innovative, utilize a large component of skilled labor, and have a relatively high proportion of large firms. Yet, Acs and Audretsch recognize that innovation counts do not measure the economic value of these innovations, and suggest that large firms may introduce innovations with higher market values. In sum, the evidence indicates that the Schumpeter hypothesis relating innovation to size appears to be dependent on industry conditions. That is, evidence suggests that industries are not homogeneous. By combining firms into one group, conclusions about the relationship between firm size and innovative activity for the economy as a whole may be made. But this would not aid in determining why these results are found, since the assumption of homogeneous industries cannot be made. This study analyzes five high-technology industries: the computer, industrial machinery, medical equipment, pharmaceutical, and semiconductor industries. By analyzing the relationship between innovation and firm size for each industry separately, knowledge in the area of the relationship between firm size and technological progress can be made. L _ LIC fir: Schurr. 13% Origin: Infield 353515113 21 C. CRITICISMS OF EMPIRICAL TESTING There has been much discussion over Schumpeter’s theory and what the empirical studies in the Schumpeterian tradition have actually tested. Markham (1965) was one of the first in expressing dissatisfaction with the empirical tests. He states that the original Schumpeterian hypothesis "does not imply that the measurable innovational effort of a firm as expressed, say, in terms of its R&D expenditures should be a continuous and increasing function of market power, business size, or, for that matter, retained earnings." (p. 324) Instead, Schumpeter’s theory must be viewed "as a threshold theory; departure from perfect competition (or some form of monopoly) does not follow that twice this volume of departures, somehow measured, should lead to twice the volume of ' innovations." Fisher and Temin (1973) argue that empirical research testing whether innovative inputs increase more than proportionately with firm size does not correspond to the original Schumpeter hypotheses. Fisher and Temin interpret Schumpeter as proposing increasing returns to scale in R&D with respect to both the size of the research establishment and the firm. They identify two relevant elasticities: the elasticity of research inputs with respect to the firm size and the elasticity of R&D output with respect to the firm. Fisher and Temin argue that empirical studies claiming to refute Schumpeter only show that the first elasticity is less than unity. To support Schumpeter, one must also show that the second elasticity is greater than one.9 9Scherer (1973) criticized Fisher and Temin for not accumme representing empirical work that they attacked. Also, Rodriguez (1979) notes that in Fisher and Temin’s model, increasing returns to scale is inconsistent with profit maximizing equilibrium. I‘d-Wm | 5VAAAAA 'S-chut: 22 Kohn and Scott (1982) expand Fisher and Tieman’s argument and develop conditions under which increasing returns to R&D inputs imply that the original Schumpeter hypothesis holds.lo They also offer two definitions of the extent to which an industry may be considered "Schumpeterian". First, the greater the elasticity of (declining) marginal R&D cost with respect to the output of R&D, the more "Schumpeterian" the industry is in the cost sense. And second, the greater the marginal value added by R&D inputs with respect to the size of the firm, the more "Schumpeterian" the industry is in the productivity sense.11 In sum, the empirical methodology of testing the Schumpeter hypothesis has been challenged on the grounds that it is not testing Schumpeter’s original hypothesis. Schumpeter argued that a large corporation appears to be especially conducive to organized innovative effort, but he never argued that there should be an increasing and continuous relationship between firm size and innovation. This study will address these criticisms. First, I address Markham’s criticism by testing for a threshold effect using various statistical techniques. I also model Schumpeter’s hypothesis assuming a continuous function in order to compare results. Second, I address Kohn and Scott’s criticism by utilizing a cost based test, which they l"Ihey demonstrate that it is possible to have increasing returns to R&D inputs and hence decreasing marginal R&D cost at equilibrium, provided that the marginal value added by R&D is falling more rapidly than its marginal cost “ An essential methodological point is that empirical estimates of how "Schumpeterian" an industry is Q be made by measuring the elasticity of R&D expenditures with respect to firm size at equilibrium, provided that the firms face similar R&D cost and productivity conditions. However, without uniform conditions, that is, a cross sectional analysis, results from a test of the relationship between firm size and R&D activity does not infer the Schumpeter hypothesis. (Baldwin and Scott (1987)) 23 describe as similar to measuring R&D effectiveness. There are studies which measure R&D effectiveness, yet only two of these studies take into account firm size. This literature is discussed in the following section. D. R&D AND FIRM PERFORMANCE An alternative line of literature analyzes the effectiveness of R&D expenditures and patents on an indicator of firm performance, such as firm productivity or profitability. That is, instead of using R&D expenditures as a proxy for technological progress (that is, a dependent variable) R&D expenditures is included as an independent variable. Firm productivity or profitability is instead the dependent variable. Numerous studies have estimated the return to R&D expenditures using a Solow residual model, focussing on productivity growth. For example, Clark and Griliches (1984) focus on productivity growth in the 1970’s at the business level and find estimates of the rate of return to R&D investment to be almost 20%. In their model, they test for industry effects by including industry dummy variables. They conclude that estimation within industries has little effect on the results. Griliches and Mairesse (1984) analyze the relationship between output, employment, physical and R&D capital during the years 1966-1977. They find that in the cross sectional dimension, there is a strong relationship between firm productivity and the level of its R&D investments. Yet in the time dimension. this relationship comes close to vanishing. Griliches (1986), using NSF-R&D Census data from the 1970’s, confirms the earlier studies in that R&D contributes positively to productivity growth. Griliches also finds that privately financed R&D has a 51 321' .. \‘W‘ '3 3 SVDSULOL R&D. r~- ' anti ll ’3“!!n,~‘, uhSIIJL‘I ftnttio 24 a significantly larger effect on productivity than federally financed R&D, and that basic research appears to be more important as a productivity determinant than other types of R&D. Yet, firm productivity is a measurement of process innovation only. This fails to take into account product innovation which is also an important component of technological advance. Process innovation considers upward shifts in the production frmction; product innovation considers the introduction of new production functions.12 R&D effectiveness can also be determined using accounting profits or a market value based measure of profitability. This allows consideration of both product and process innovation. For instance, Grabowski and Mueller (1978) suggest biases in previous structure-performance studies which utilize accounting profits. These studies ignore investment in intangible capital, such as R&D expenditures. The authors find that nearly half of the variance in unadjusted profit rates was due to improper accounting treatment of intangible capital. Ravenscraft and Scherer (1982) find that the long-run effect of a unit change in R&D expenditures leads to a change of $3.075 in gross profits. Profits as an indicator of a firm’s performance has come under much criticism (Fischer and McGowan, 1983). The firm’s market value is an alternative measure of firm performance. Several studies have considered the value of R&D expenditures and patents using a market value approach. For example, Ben-Zion (1984) finds that a one dollar expenditure in R&D increases the firm’s market value by 2.6 relative to the market’s valuation of book value. Pakes (1984) investigates the dynamic relationships between the ‘2 In actuality, classification of product and process innovations depends on one’s perspective. To some, a new computer is a product innovation; to others, a process innovation. In met; the hm; CSIIIleII “I: ‘LJS. I 25 number of successful patent applications of firm’s investment in inventive activity (its R&D expenditures), and a measure of firm performance (its stock market value). He finds a significant correlation between the stock market rate of return and unexpected changes in both patent applications and R&D expenditures.l3 Pakes (1985) estimates that an "unexpected" patent leads to an increase in the firrrr’s market value by $810,000, while an unexpected increase of $100 of R&D expenditures is associated with an increase in the firms value of $1,870. Connolly and Hirschey (1988) use a Bayesian approach in estimating the effect of unexpected patents and R&D expenditures on the firms market value. The data employed are mainly large firms; the sample includes 390 firms from the Fortune 500 from 1977. They find that each unexpected patent adds $5.33 million to excess valuation.” They also find that an extra dollar of R&D expenditure adds $3.60 to excess valuation. Taking a slightly different approach, Cockbum and Griliches (1988) measure the value of patents by examining the stock market’s response to patent grants. They find that the estimated value of a patent is approximately $500,000. Yet, when R&D expenditures is included in the equation, this estimate is halved. They conclude that the stock market responds more strongly to changes in a firm’s R&D spending than to changes in a firms stock of patents. Yet, none of these studies examine the relationship between firm size and the stock response to R&D investment. If firm size is an important determinant in the stock response to R&D, failure to include this results in specification bias. 13Unexpected changes refers to changes that could not be predicted from the history of the variables included in the data set. 1‘Excess valuation is defined as the market value of common stock plus the book value of debt minus the book value of tangible assets, all normalized by sales. 26 Link (1981) recognized the association between fu‘m size, innovative inputs, and firm performance, and tested whether a positive relationship between firm size and firm productivity exists. Concentrating on the chemical, machinery and trucking industries, Link found large firms earn a rate of return to R&D which is significantly greater than zero. However, he could not reject the hypothesis that the rate of return to R&D equals zero for small firms. Lichtenberg and Siegel (1991) extend Link’s analysis by using detailed confidential longitudinal microdata. Dividing firms into three different size regimes, the authors find returns to company-funded R&D to be significantly higher for the largest firms compared with the smallest firms. However, both Link and Lichtenberg and Siegel consider only process innovation. Process innovations are more likely to come from large established firms than are product innovations.” Therefore, it is not too surprising that the Schumpeterian hypothesis is supported. On the other hand, by analyzing the relationship between the market value of the firm and R&D inputs, the stock response to R&D investment in terms of both product and process innovation can be determined. In sum, much of the empirical literature proposing to test the Schumpeter hypothesis considers whether innovative inputs increase more than proportionately with firm size. Yet, increasing returns to R&D inputs do not necessarily imply increasing returns to technological progress, which is the core of Schumpeter’s theory. A test of the '5 Large firms have greater incentives to develop internal process improvements. This is due to incentives to develop process improvements internally. Assuming that a new process reduces costs by a given percentage, the larger a firms output, the larger the total amount of savings. (Rosen (1991), Scherer and Ross(1990)). relifiifis‘; III: 550111 M for is not W prcgftss 1115mm! 5 cf R&D EtLhC-d fans “33 is stable I'L‘- N I ~LIHDIUCI 27 relationship between firm size and innovative outputs, such as patents, is deficient since the economic value of patents is highly heterogeneous. Furthermore, previous literature tests for a continuous relationship between firm size and technological progress, which is not what Schumpeter argued. An alternative approach in measuring technological progress analyzes the effectiveness of R&D expenditures on firm performance. This literature has not, however, considered whether firm size influences the marginal value of R&D output in terms of both product and process R&D. This study provides a unique method of testing the Schumpeter hypothesis. It determines the stock response to the firms market value for five high—technology industries, and tests whether this relationship is stable over an index of firm size. This approach both separates out industry effects and considers process and product innovation. H. DESCRIPTION OF THE INDUSTRIES This section presents a description of the five industries studied. In order to evaluate results from the Schumpeter hypothesis, an understanding of industry conditions is necessary. First, I present the definitions of the 4—digit SIC industries, and list the 10 largest firms in each of the industries. Second, I describe and compare industry conditions, specifically baniers to entry and appropriability conditions, which have been postulated to influence small and large firms’ innovation rates(Acs and Audretsch (1987, 1988), Scherer and Ross(1990), Since appropriability and barriers to entry are not directly observable, I examine some of the measures which have been used as proxies. A. INDUSTRY DEFINITIONS The following section presents the definitions of the 4 digit SIC industries included in the Study as defined by the 1987 Census of Manufactures along with the 10 largest firms in the industry in 1989.1 ‘ This information is obtained from Dun’s Business Rankings (1991) which ranks businesses by sales volume within 4 digit SIC industry categories. 28 29 1. COMPUTER INDUSTRY The Computer industry includes SIC 3570, 3571, 3575, 3577, and 3578. SIC 3570 is the general computerSIC classification code. Firms classified into this SIC are usually the larger firms which produce products in several sectors.2 SIC 3571 is made up of establishments primarily engaged in manufacturing electronic computers. The 1987 Census of Manufactures defines electronic computers as machines which: (1) store the processing program or programs and the data irnmtadiately necessary for execution of the program; (2) can be freely programmed in accordance with the requirements of the user; (3) perform arithmetical computations specified by the user; and (4) execute, without human intervention, a processing program which requires them to modify their execution by logical decision during the processing run. Included in this industry are digital/analog computers. SIC 3572 includes establishments which manufacture computer storage devices, while computer terminals is classified under 3575. SIC 3577 is defined as computer peripheral equipment, n.e.c. which includes products such as printers, plotters, and graphic displays. The Census of Manufactures defines calculating and accounting equipment, SIC 3578, as: establishments engaged in manufacturing point—of-sale devices, fund transfer devices, and other calculating and accounting machines, except electronic computers. Included are electronic calculating and accounting machines which must be paced by operator intervention, even when augmented by attachments. These machines may include program control or have input/output capabilities. Last is office machines, n.e.c. (SIC 3579), which includes establishments producing such items as typewriters and word processing equipment. 2 The Census of Manufacturers does not provide data on this industry. However, the data used in this study classifies firms into this particular code. 30 TABLE 1: COMPUTER INDUSTRY: 10 Largest Firms, Ranked by Sales (1989) flrr_n Annual Sales 1. I.B.M. $54.20 billion 2. Unisys $9.71 billion 3. Digital Equipment $9.38 billion 4. N.C.R. $5.64 billion 5. Litton $4.86 billion 6. Apple $4.07 billion 7. Wang $3.06 billion 8. Pitney Bowes $2.25 billion 9. Harris $2.06 billion 10. Honeywell $2.06 billion 2. MACHINERY INDUSTRY Included in the machinery industry are SIC 3533, 3559, 3560, and 3561. Oil and gas field machinery (SIC 3533) includes those establishments primarily engaged in manufacturing machinery and equipment for use in oil and gas fields, or for drilling water wells, including portable drilling rigs. Special industry machinery, n.e.c., includes establishments producing equipment such as smelting and refining equipment, industrial sewing machines, and automotive maintenance machinery and equipment. SIC 3560 is 31 the general machinery SIC classification code. Firms classified into this SIC are usually the larger firms which are most likely producing equipment in more than one product litre.3 Last is SIC 356l--pumps and pumping equipment--which include pumps for either general industrial, commercial, or household use, except fluid power pumps and motors. TABLE 2: MACHINERY INDUSTRY: 10 Largest Firms, Ranked by Sales (1989) fig Annual Sales 1. Dresser Industries $3.11 billion 2. Combustion Engineering $3.04 billion 3. Baker Hughes $1.92 billion 4. Babcock & Wilcox Co. $1.70 billion 5. Hughes Tool Co. $1.15 billion 5. Tilling Thomas Inc. $1.15 billion 7. Baker Hughes Prd. $0.81 billion 7. Tools Inc. $0.81 billion 9. Big Three Industries $0.52 billion 10. Cameron Iron Works Inc. $0.50 billion 3 Again, the Census of Manufacturers does not provide statistics on this SIC code. 32 3. MEDICAL EQUIPMENT INDUSTRY Included in the medical equipment industry are SIC 3841, 3842, 3844, 3845, 3851. and 3861. Surgical and Medical Instruments (SIC 3541) is composed of establishments engaged in manufacturing medical, surgical, ophthalmic and veterinary instruments and apparatus. The Census of Manufactures defines SIC 3542, surgical appliances and supplies, as: made up of establishments primarily engaged in manufacturing orthopedic. prosthetic, and surgical appliances and supplies; arch supports and other foot appliances; fracture appliances, elastic hosiery, abdominal supporters, braces, and trusses; bandages: surgical gauze and dressings; sutures; adhesive tapes and medicated plasters: and personal safety appliances and equipment. SIC 3844--X-Ray Apparatus and Tubes--is composed of firms which manufacture radiographic X-ray, fluoroscopic X-ray, and therapeutic X-ray apparatus and tubes for medical, industrial, research and control applications, or in manufacturing other irradiation equipment including gamma and beta—ray equipment. SIC 3845 is defined as electromedical equipment and SIC 3851 is ophthalmic goods such as ophthalmic frames, lenses, and sunglass lenses. SIC 3861, photographic equipment and supplies, is made up of establishments primarily engaged in manufacturing: (1) photographic apparatus. equipment, parts, attachments, and accessories, such as still and motion picture camera and projection apparatus; (2) photocopy and microfilm equipment; or (3) sensitized film, paper, cloth, and plates, and prepared photographic chemicals. 33 TABLE 3: MEDICAL EQUIPMENT INDUSTRY: 10 Largest Firms, Ranked by Annual Sales (1989) Elm; Annual Sales 1. Eastman Kodak Co. $13.30 billion 2. Xerox $10.30 billion 3. Minnesota Mining $9.42 billion & Mfg. Co. 4. Johnson & Johnson $8.01 billion 5. Baxter Intematl. $6.22 billion 6. Polaroid Corp. $1.76 billion 7. Becton Dickinson $1.58 billion 8. Kendall CO. $1.00 billion 9. Nashua Corp. $0.87 billion 10. Bausch & Lomb $0.84 billion 4. PHARMACEUTICAL INDUSTRY The 1987 Census of Manufactures defines SIC 2834 as: ...made up of establishments primarily engaged in manufacturing, fabricating, or processing drugs in pharmaceutical preparations for human or veterinary use. The greater part of the products of these establishments are finished in the form intended for final consumption, such as ampoules, tablets, capsules, vials, ointrnents, medicinal powders, solutions, and suspensions. Products of this industry consist of two important lines, namely: (1) pharmaceutical preparations 34 promoted primarily to the dental, medical or veterinary professions, and (2) pharmaceutical preparations promoted primarily to the public. TABLE 4: PHARMACEUTICAL INDUSTRY: 10 Largest Firms, Ranked by Annual Sales (1989) fig Annual Sales 1. American Home Products $5.02 billion 2. Pfizer $4.91 billion 3. Hoechst $4.61 billion 4. Abbott $4.38 billion 5. Smithkline, Beckman Corp. $4.32 billion 6. American Cyanarnid CO. $4.16 billion 7. Eli Lilly & Co. $3.64 billion 8. Wamer-Lambert $3.48 billion 9. Schering-Plough Corp. $2.69 billion 10. Upjohn CO. $2.52 billion 5. SEMICONDUCTOR INDUSTRY The 1987 Census of Manufactures defines SIC 3674 as: made up of establishments primarily engaged in manufacturing semiconductors and related solid-state devices. Important products of this industry are semiconductor diodes and stacks, including rectifiers, integrated microcircuits (semiconductor networks), transistors, solar cells, and light sensing and emitting semiconductors (solid State) devices. 35 TABLE 5: SEMICONDUCTOR INDUSTRY: 10 Largest Firms, Ranked by Annual Sales (1989) Fi_nn_ Annual Sales 1. Texas Instruments Inc. $5.59 billion 2. National Semiconductor $2.46 billion 3. Intel $1.90 billion 4. Advanced MicroDevices $1.00 billion 5. Western Digital Corp. $0.77 billion 6. Micron Technology Inc. $0.30 billion 7. Uniden Corp. of America $0.30 billion 8. Micropolis $0.29 billion 9. Everex Systems Inc. $0.27 billion 10. LSI Logic Corp. $0.26 billion 13. INDUSTRY CONDITIONS This section presents descriptive statistics on variables which have been asserted to influence the innovation rates of large and small firms (Acs and Audretsch (1987, 1 988), Scherer and Ross(1990), Karnien and Schwartz (1982)). The innovation rate is a 36 measure of the number of innovations relative to firm size.4 These statistics may provide information on why the Schumpeter hypothesis is supported or rejected once results are Obtained. As noted in the Introduction, an R&D program of a given size may Operate more efficiently in a larger firm. In other words, scale economies in production may provide economics for R&D. One measure of scale economies in production is the capital intensity of an industry. The higher the capital intensity, the more likely large firms are able to take advantage of the economic gains from innovation. Similarly, the extent of product differentiation through advertising may grant advantages to large firms while inhibiting small firms innovation rates.’ Therefore, the higher the capital intensity and/or advertising intensity of an industry, the more likely large firms will have an advantage in R&D. On the other hand, Acs and Audretsch have shown (1987, 1988) that SI”Ilaller firms are relatively more innovative the higher the percentage of large firms coIllprising an industry. One of the major tenants of the Schumpeter hypothesis is that market power and the potential for accruing economic rents is a necessary condition for innovation. A \ ‘ The absolute number of innovations must be standardized by a measure of size in Or(ier to compare relative rates of innovation. 5 The definition given by compustat for advertising expense is that "this item 1‘Qpresents the cost of advertising media (radio, television, newspapers, periodicals, etc.) (1 promotional expenses". This can be compared with the average intensity for anufacturing. Scherer and Ross note that the median manufacturing industry devoted Q~ 66 percent of its sales revenue to advertising in 1977 (Federal Trade Commission. Statistical Report: Annual Line of Business Report, 1977 (Washington: April, 1985). I\Iowever, if total selling expenses are taken into account (cost of sales representatives, Qint of sales displays, coupons, samples, advertising allowances to retailers, and trade lowances as well as media advertising expenses), the median value increases to 7.75% QI: manufacturers’ sales in 1977. ‘5': #094 ”I" I. It.- M- ‘v '1... Ins-.- ,, Afiw fir. "A“: . r m m “. 3'11 ‘r I} 37 Common measure of the extent of imperfect competition is the degree of concentration. While this study examines the relationship between firm size and technological progress, there is a considerable literature on the relationship between concentration and technological progress (see Baldwin and Scott (1987) and Karnien and Schwartz (1982) for an overview). The argument relating concentration to technological progress is similar to that of the advantages of large firms: Innovation can occur only in the presence of market power and "...only by a firm that has the resources which are associated with c onsiderable size." (Galbraith) Therefore, if results indicate that large firms are more profitable in R&D in a particular industry, we can observe if there is a correlation between supporter of Schumpeter and concentration. Concentration measures have E 1‘ equently been included in regression equations which test the Schumpeter hypothesis. Yet, once industry conditions are incorporated, these measures are Often insignificant(see I(a-l'nien and Schwartz, Baldwin and Scott(1987)). However, the majority Of the studies control for industry effects at the 2 digit level whereas this study disaggregates industries further than 2 digits. As stated in the previous Chapter, underlying factors in industries may be able to explain interindustry differences in R&D investment and innovative performance. Yet, all important question remains to be answered: if firm size is a significant determinant 0f R&D profitability, is there a pattern between appropriability conditions and support for the Schumpeter hypothesis? For example, patent effectiveness might be extremely itl1])ortant for small firms. in part because other means of appropriation may not be I:‘§€asible(Levin et al.(1987)). Table 6 shows the effectiveness of product and process E) atents as a means of capturing and protecting the competitive advantage of new 38 processes or products. On a 7 point scale, the average for all industries is 3.5 for process patents and 4.3 for product patents. Appropriability conditions are below average in the computer industry and above average in the pharmaceutical industry, and approximately average for the other 3 industries. The mean scores infer that patents are not considered to be a very effective means in protecting the competitive advantages of new or improved precesses and products. In fact, in Levin et a1. survey, they fmd that lead time, learning curves, and sales/service efforts were regarded as substantially more effective than patents in protecting product innovations.‘5 And secrecy and the mechanisms listed above were more effective in protecting new processes, than were patents. These statistics, while useful, do not provide the entire picture of appropriability conditions. \ 6 Unfortunately, the authors’ only publish patent effectiveness broken down by iIndustry, not the others. 39 TABLE 6: INDUSTRY DESCRIPTIVE STATISTICS Computer Machinery Medical Pharma- Semicond Equipment ceutical uctor Advertising 1.9% 0.7% 4.1% 10.9% 0.7% Intensity Capital 27.6% 42.5% 37.2% 34.9% 86.3% Intensity CR4 43 16 43 26 40 HHI 793 146 854 318 597 Process 3.3 3.2 3.2 4.9 3.2 Patents Product Patents 3.4 4.4 4.7 6.5 4.5 Advertising Intensity is advertising expenditures divided by sales. These values are calculated from the Compustat Tapes, including all available observations (1982-1989). Papital intensity is gross assets divided by value-Of-shipments. Data are obtained from c U.S. Dept. of Commerce, Bureau of the Census, Annual Survey of Manufactures, I 982, Industry Profiles. " Q .0ncentration ratios: CR4 is the four firm concentration ratio. HHI is the Herfmdahl- ITilt-schman Index. Data are obtained from "The U.S. Dept. of Commerce, Bureau of the Qelrsus, Annual Survey of Manufactures, I 982, Concentration Ratios. I r(acesslProduct Patents: the effectiveness of patents as a means of capturing and lDrotecting the competitive advantages of new or improved processes or products, based On a 7 point scale (1=not an effective means). These scores are obtained from Levin et a4- (1987)). 1H. THEORETICAL RELATIONSHIP BETWEEN FIRM SIZE AND TECHNOLOGICAL PROGRESS "Innovation is the search for and the discovery, development, improvement, adoption and commercialization of new processes, products, and organizational structures md procedures. It involves uncertainty, risk taking, probing and reprobing. cthperimenting, and testing. It is an activity in which "dry holes" and "blind alleys" are the rule, not the exception." (Jorde and Teece (1992)). This chapter develops a theory f 01‘ the relationship between firm size and innovation. First, models of the process of innovation are described, including a discussion of the differences between the serial 111()dels and the simultaneous model of innovation. Second, a framework for empirical a“Ilalysis is developed. And last, an econometric model is established. A — INNOVATION AS A PROCESS Schumpeter did not specifically discuss the organizational aspects of the inIlovational process. Yet, in order to understand the role of firm size in innovation, it i S crucial to consider the process aspect of innovative activities (see Kline and Rosenberg ( 1 986), Teece (1989)). There have been a number of attempts to impose a conceptual bl‘der on the innovation process. However, just as inventions are typically very crude in 40 k — 41 their early Stages. so are the general models of innovation (Teece (1986)). Three models 0f the innovation process are discussed: the traditional serial model, the chain-linked model, and the simultaneous model. 1. THE SERIAL MODEL Traditional descriptions of the innovation process have focussed on the "serial . model", which is also referred to as the linear model. In this model, the innovation 1) rocess proceeds in steps from research to development to marketing. Only after RESEARCH completion of one stage is the process 8 able . . . to continue. A diagrammatic DEVELOPMENT exéunple of the serial approach is presented in Figure l. U . . . _ PRODUCTION AND DESIGN First, an Idea 13 conceived. a per ect is formulated. and expenditures on 8 bzisic research occur. This may or may . MARKETING 1 not lead to successful research. and/or 8 DO - Ssrbly a patentable product or process. FINAL CUSTOMER ‘ IQ\avever. if this stage is successful. urther rnvestrnent rs needed in order to t FIGURE 1: THE SERIAL MODEL hansform this basic research into a ‘3 eruct or process. that is. the 42 development stage. The next Stage is production and design. Here, an acceptable design is formulated and the manufacturing process is implemented. Only after the successful completion of this stage does marketing of the innovation occur. This model has been widely criticized for a number of reasons. First, there are no feedback mechanisms to and from the different areas of operation, or from users. However, feedback represents an important component of innovation. Rarely is the design and optimization of the innovation correct from the Start. Jorde and Teece (1992) note another problem with the serial model in that "...it does not address the many small but cumulatively important incremental innovations that are at the heart of technological Change in many industries, especially well-established industries like semiconductors, Computers, and automobiles." And third, this model does not characterize process innovations as opposed to product innovations, which usually do not require marketing. 01' some other step of the process. Yet, the serial model may be a relevant Characterization of the innovation process when large idiosyncratic fixed investment is required for commercialization. Some examples of innovation which follow the serial model include the NASA Mercury and Apollo programs, Department of Defense projects. the IBM 360, and the Xerox 9000 family of high speed copiers (Teece (1989)). 2. THE CHAIN-LINKED MODEL One possible alternative to the linear model is the chain-linked model.1 This model is depicted in Figure 2. In this model, the authors consider five alternative paths ‘ lSee Kline and Rosenberg (1986) for a detailed discussion. I‘b "Nu ‘un H, ; a G) a? RESEARCH R (P I 31‘. 3 II all. \\ \ ( I ' KNOWLEDGE i D Kf'l K ‘4 K“ . S \ J ' "L. - l I |\ ”'12; I ”‘12 | “‘2": - C ‘ INVENT V AND OR \DETAILED yREDESIGN j DISTRIBUTE POTENTIAL l PRODUCE | DESIGN c I AND c\_t_\ND MARKET l ANALYTlC I AND I PRODUCE I MARKET I DESIGN I TEST 1 | CD: (P: @r (P: Q? l L ¥ ' é 1 FIGURE 2: THE CHAIN-LINKED MODEL Kline and Rosenberg (1986) .— of the innovation process. instead of just one as in the serial model. The first, called the central-chain-Of innovation (indicated by the arrows labeled "C"), proceeds from left to right: beginning with design, proceeding to development and production, and finally marketing. This path is similar to the traditional serial model, although it doesn’t necessarily begin with research. The second path is an extension of the first and incorporates feedback mechanisms (marked by "f' and "F"). That is, at any step. the innovation cycle is able to feedback information to improve on previous stages. The third path addresses the small incremental innovations by modeling science alongside the development process (S). This stage recognizes that research encompasses the entire process, and may enter into the process at any particular point. or at all points of the process. For example. Kline and Rosenberg state that "When we confront a problem in 44 technical innovation, we call first on known science, stored knowledge, and we do so in serial stages."2 The link between research and problems in invention and design is the fourth course. This path is relevant when science produces radical innovations which create new industries (for example, semiconductors, lasers, and genetic engineering). And the last course is the feedback from products to science (arrow "1") in which innovation spurs innovation; the process becomes circular.3 In sum, there are three important differences to note when comparing the chain- linked model to the traditional serial model. First, there are five paths of activity instead of just one. Second, instead of research being pertinent only at the beginning of a project. this model realizes that research may encompass the entire process. And third, the importance of feedback is recognized in this model. R&D personnel must be closely connected to the end user a_ng to the marketing personnel. In addition, development requires feedback from both marketing and research. Yet, just like the serial model, the chain-linked model assumes a particular sequence of activities, from research to design to production and then marketing. 2Kline and Rosenberg (1986) illustrate this path with an example. In order to reduce pollution and improve mileage, you are interested in innovating an improved carburation- induction system for the spark ignition automobile engine. In order to accomplish this you must mix fuel and air at the molecule-to-molecule level, which conventional carburation systems fail to do. You first ask yourself whether you know of a current device that will do the job. If your answer is "no", you consult your colleagues and perhaps existing literature. If that does not yield a solution, you consult with experts in the field. Only if this is not fruitful do you initiate research. 3For example, "Without the microscope, one does not have the work of Pasteur, and without that work there is no modern medicine. Without the telescope, we would not have the work of Galileo, and without that work we would not have modern astronomy and cosmology, nor would space exploration with its various innovations have been possible." (Kline and Rosenberg, (1986) p. 293). c-rl‘J. _..J've~ I. ". .,. P—v .. p 7 if. "wueu . --. mule; U \i. ‘ 1“» \s1 45 3. THE SIMULTANEOUS MODEL In many cases innovation may not begin with research nor is it serial. In some instances time is of the essence and in order to succeed commercially the different stages must act simultaneously. Such a model is portrayed in Figure 3. It allows for the simultaneous work on all parts of the project at a given point in time, instead of progressing with the project only . after one stage has been completed. The simultaneous model does not necessarily assume that all advantages accrue to the winner of a R&D or patent race. but it does recognize the importance of time. Because the different activities of the firm occur concurrently, marketing may be undertaken before development and design are finalized. Thus. close contact is necessary between f . | U {I I D D. r-: E s v r l r: c M .L N. A o n P a K M E e P r N a r ‘r o N D G u c 'r I o N I CUSTOMER 1 FIGURE 3: THE SIMULTANEOUS MODEL the different activities for projects to be commercially successful. 46 In sum, the traditional serial model greatly oversimplifies the organizational challenges of innovation. A result of this is that the research aspect of the process is overestimated while other factors are not discussed. The chain-linked model and the simultaneous model point out that it is not enough for the research activity itself to be successful; a firm must also be able to organize the other parts of the process efficiently and effectively for successful commercialization. That is, research must be combined with other parts of the process, the so-called complementary assets. These complementary assets could be considered services, such as after-sales support in the case of computers, medical instruments or machinery, or marketing in the case of pharmaceuticals, in which dissemination of information about a new drug is essential. When these complementary assets are not present, competitors and/or imitators are likely to profit from the innovation--at times more than the firm that was first to innovate.4 In addition, large returns can be made from a modest technological advance if the company has the appropriate complementary assets in place.5 ‘ The classic example of the importance of complementary assets is Electrical Musical Industries Ltd. (EMI) computerized axial tomography (CAT) scanner. Within six years of introduction in the U.S., the company had lost market leadership and by the eighth year it had dropped out of the CAT scanner business. The CAT scanner required a high level of training, support, and servicing because of its technological sophistication. EMI had none of these services and did not contract with another company to provide these services. Unfortunately for EMI, once the product was available it could be reverse engineered and copied quite easily. 5 The classic example in this case is the IBM PC which was introduced in 1981. IBM used the new 16-bit microprocessor developed by Intel (the 8088) and a new operating system (DOS) which was developed for IBM by Microsoft. Apart from those new innovations, IBM used existing standards. The key to IBM’s success was therefore not revolutionary innovation. but IBM’s complementary assets: services and marketing. Another key was IBM’s realization that in order for the PC to be successful, software was needed-in a short period of time. Instead of attempting to develop software itself, IBM made the operating system information publicly available and software firms proliferated. 47 4. LINKAGES Up to now, all activities have been assumed to take place in-house. Yet, because of the high cost and uncertainty of research, innovation is most likely going to require access to capabilities which lie beyond the organization of a single firm. Examples include universities, as well as horizontal and vertical linkages. University research has played an important role in the innovation process through numerous channels, including publications, research contracts and technical conferences.<5 The link to universities may be important in gaining access to graduate students, to attract faculty as partners in business ventures, and to encourage faculty to become familiar with their equipment. With the university-firm connection, firms may be more aware and attentive to changing conditions. The great extent of horizontal linkages--research consortia--is a relatively new phenomenon. Because of successful foreign competition in the past two decades, cooperation among U.S. companies was advocated for the development of basic technologies was advocated. Yet until the antitrust laws were relaxed it was too risky for companies to enter into research agreements.7 The escalation of R&D costs represents 6 However, there has been some debate on the commercial applicability and value of university technology. See Teece (1989). 7 The National Cooperative Research Act was passed in 1984 which provides a "rule of reason" for evaluating the legality of each cooperative R&D venture on a case-by-case basis instead of declaring them "per se" illegal. This act also limits potential liability of these research consOrtia to actual damages compared with treble damages. 48 another reason why companies are forming R&D consortia.8 By law, R&D consortia can only perform basic research and development up to a prototype stage and are prohibited from manufacturing and marketing products. Thus, the member companies are not cooperating with one another throughout the entire process.9 Horizontal linkages are beneficial to firms in that they can help firms to internalize part of the spillover effects to the consortium, because the firms receiving the benefits are likely to include a greater portion of firms which incurred the corresponding R&D costs. The effect of greater appropriability is, of course, to encourage greater investment in new technology. Markets tend to underinvest in new technology because those innovative firms which support R&D do not appropriate all the benefits of innovation-~imitators are often more profitable than the innovating firm (See Levin et al.(1985)). Another argument for linkages is that they help reduce unnecessary duplication of research efforts. and may also assist in the definition of technical standards for systemic innovation (Teece (1986)).10 Most important, especially for small firms, are the vertical linkages. That is, two or more firms may address particular aspects of the innovation process and therefore not 8 For example, Oncogen is a three company research venture formed by Genetic Systems, Syntex, and Bristol-Meyers to develop diagnostic and therapeutic products for cancer treatment which would be too costly for one company to develop (see Davis(l985)). 9 While many of these R&D consortia contain large firms, one research consortium (MCC) has formed an Associates Program through which smaller companies can gain access to the consortium’s results. Also, the Midwest Technology Development Institute (MTDI) is a consortium which pools the resources of 10 midwestern states in order to make technology available to small business. ‘° Of course, the uncertainty of innovation often requires pursuing multiple technological paths. 49 undertake all of the steps in-house. For example, one fum’s research may be patented and licensed to another firm for production and marketing. That is, research occurs in one fum, while the remaining stages occur in another firm. Or, research, development. production and design are all successful and take place within one firm. However, the firm does not have the marketing resources necessary so the product is licensed to another firm. Numerous other scenarios are possible. The point is that firms need not perform all activities in-house for innovation to occur and to be successful. In addition, contractual relationships can bring added credibility to the innovator, especially if the innovator is relatively unknown and the contractual partner is established and viable.‘1 Teece (1986) discusses the factors that determine whether a contractual solution or integration is the optimal solution. To summarize, the process nature of innovation is difficult to characterize because of its numerous dimensions. This study recognizes that organizational features of the firm are important, but does not seek to analyze all of them. Instead, it looks at one particular aspect: does the size of the firm affect R&D efficiency? That is, are small firms more profitable in turning research inputs into successful innovations or vice versa? B. FRAMEWORK FOR EMPIRICAL ANALYSIS The theoretical relationship between firm size and technological progress is analyzed by means of a profit maximization-market valuation model. It is argued that “For example, Cipher Data Products, Inc. contracted with IBM to develop a low- priced version of IBM’s 3480 0.5 inch streaming cartridge drive, which became the industry standard. . .rfl 5! “J 50 the stock market takes into account information about firms’ expected future profitability and therefore factors in expected outcome due to R&D investment. Accordingly, the effectiveness of R&D investment in high-tech firms is analyzed by evaluating the stock response to R&D inputs. The stock response measures the expected outcome of investment. If this response is significant, it indicates that these investments are deemed profitable by the market. If the stock response is not significantly different from zero, one can presume that the market does not value these investments and believes that it will n_ot improve economic growth. After briefly introducing the market valuation model, I discuss the firm’s assets-- both tangible and intangible--and incorporate these into the valuation model. A description of how risk factors into the model is then examined. And finally, the model is extended to allow for a test of whether firm size is a significant determinant of R&D efficiency. 1. THE FIRM’S MAXIMIZATION PROBLEM The firm maximizes its value: V=f(X,, X2 XN) where V=the value of the firm and X,...XN are the assets held by the firm. The value of the firm at time t is a function of the present value of the firrn’s future income streams: INCOME:1 INCOME2 INCOMEc T INCOME (1) = + + ..... —— = 2 —— (1+I) (1+I)2 (1+Ilt 1:21 (1+rlt 51 Where "r" is the discount rate and "Income" is the profit in each time period. Therefore, the firm’s value at time t is a function of its future income streams, which in turn are a function of its assets: V = F {INCOME STREAMS (assets)}. The firm’s assets can be separated into 2 broad categories: tangible assets and intangible assets (TA and 1A, respectively): V = f(income Streams (TA, IA)) (2) The firm invests in these assets to the point at which the marginal increase in firm valuation is equal to the marginal cost of investment. 2. TANGIBLE VS. INTANGIBLE ASSETS The firm’s tangible assets are its plant, property and equipment, which can be measured using standard accounting techniques. Intangible assets can be disaggregated into the firm’s knowledge stock of capital and marketing stock of capital (KS, MS, respectively). For a high-technology firm, the extent of its knowledge capital is especially important. V = f(income streams (TA, KS, MS)) (3) where K8 is the firrn’s knowledge stock and MS is the firm’s marketing stock. 52 The firm’s intangible assets are more difficult to measure than the firm’s tangible assets. Since research is the primary component of a firrn’s intangible assets in high- technology industries, R&D investment can be a proxy for knowledge stock (See Griliches (1981), Connolly and Hirschey (1984, 1990), Hall(1993)). Advertising expenditures are also a form of intangible assets, which represent the degree of product differentiation. Although R&D investment and tangible asset investment are based on similar considerations so that one could use the discounted present value of future income streams to evaluate the desirability of R&D investment, there is an important difference between these two types of investments: the future net income stream resulting from an R&D project is subject to more uncertainty than investment in plant and equipment. Tangible assets can often be sold, whereas it is difficult to put a price-tag on a stock of firm knowledge. One example of this is Texas Instruments’ investment in the "magnetic bubble memory" technology. Texas Instruments (TI) invested in this technology for seven years, hoping that its efforts would lead to a revolutionary computer chip, and would capture a large share of the market held by ordinary electronic memory chips (DRAMS). Unfortunately for TI, the company’s engineers underestimated the speed with which the older memory chip technology would evolve. DRAMS had become so cheap and capable by the time the new technology--magnetic bubbles--were ready for the market that the computer makers refused to switch technologies. By that time, TI had sunk tens of millions of dollars into the project. Although there is most likely a correlation between R&D investment and the net present value that can be attributed to R&D, this example illustrates that it is not guaranteed and is much more variable than that in tangible assets. 53 3. RISK AND THE VALUATION MODEL A consideration which must be taken into account when analyzing market valuation models is uncertainty. By definition, innovation implies creating a new product or process, and this new product or process includes elements that we do not understand and are skeptical about in the beginning. A firm is uncertain how this new product will fare in the market, or whether R&D is successful even to make it to the market. That is, innovation implies uncertainty, or risk. Therefore, an understanding of risk, and how risk affects a firm’s value is required. Risk can be separated into two broad categories: systematic and unsystematic risk. Systematic (or nondiversifiable) risk refers to that portion of the variability of a security’s return caused by factors affecting the market as a whole.12 These sources affect all securities and cause the returns from all securities to vary more or less together. The second type of risk is unsystematic risk, and is risk that is unique to the firm. An example of this is variability in the firrn’s return that is due to its own management capabilities.’3 Another type of unsystematic risk, which is especially important in high technology firms, is R&D risk. R&D risk can be regarded as the risk of achieving success in all of the areas necessary to making a return on investment, for example, research, development, and marketing. Some of these sources of R&D risk are unique ‘2 Examples of systematic risk include: the change of the interest rate, changes in purchasing power (inflation), and changes in investor expectations about the overall performance of the economy. 13Other examples of unsystematic risk include such factors as management decisions, the availability of materials, foreign competition, and government regulation. :3 nuns. :3 ...I M)“. F affirms to all attired to l : are risk it‘- lzlflTlS 2H: “.3 lines at raise one prt -..r produc are more Elton achie l.,. segment whi tare of many In addit "<. 22 in b, .- fi. 54 to the industry, and one would expect risk to affect firms in one industry approximately the same way. For example, in the semiconductor industry, foreign competition reduced the returns to all U.S. companies. Yet, much of the "R&D risk is specific to the firm and is considered to be unsystematic risk. The following section describes how firms are able to reduce risk through diversification. 4. DIVERSIFICATION Firms are able to reduce risk through diversification. By diversifying into different product lines which have different risk-retum characteristics, overall risk could be reduced because one product line may achieve high returns at the same time that the returns from another product line are low, and vice versa. The combined returns over time are therefore more stable and are subject to less uncertainty. Of course, the amount of risk reduction achieved depends on the degree of correlation between the returns of the individual product lines. The risk of failure is another determinant of the variability of returns, important especially in high-technology firms. Failure can be defined as an investment which generates a negative rate of return. The risk of failure explains the desire of many companies to diversify. In addition to reducing risk by diversifying across product lines, a firm can also reduce risk by investing in different projects within the same product line. A well known example is oil and gas exploration.l4 Oil and gas exploration is very Similar in its 1" As an example, assume that the firm knows that any wildcat well that they drill will cost approximately $2 million and have only a 10% chance of success. Successful wells will result in profits of $24 million. In this case, the expected return of the firm . ‘ fl .1 y ,: 3114355 [L “h ,. '7‘ C “W «ANS--..UI I'D; f£~55531' L7 3:: a: man (II ears-:1 rem :33: Villa; I be g as: “a "TWA-d rat ; the fir t -“ V .' 31g] . “:33. “I tr f‘ 55 riskiness to the high-tech industries considered in this study, except that high-technology companies complicate the analysis due to the number of steps involved. In other words, successful research doesn’t necessarily mean a successful product, as in oil exploration; there are many more steps involved. 5. RISK AND THE REQUIRED RATE OF RETURN Risk is important because it affects the required return on investment: the higher the uncertainty of returns, the higher the required return for investors. To understand this. risk and the expected return for a firm are analyzed. This is then translated into a required return for investors, and this required return is incorporated into equation 1, the market valuation model. I begin by generalizing the valuation model to include uncertainty of returns. The expected return for a firm depends upon the weighted average of the expected return for all of the firms product lines. Let the expected return for a firm at time t be: would be 30% (Expected Retum=24(.10)-2(.9)=$0.6m). If a firm drilled only one well, there would be a 90% chance that the firm would fail, and the total project risk would be very high. Unsuccessful wells will result in no profits, and the entire investment will be a loss, or failure. However, if one firm drilled 100 wildcat wells, the risk of failure from all wells would be very low(.1 ‘00). In order for the firm to break-even, only 7.7% of the wells need be successful. This return is achieved with very little risk relative to that facing a firm drilling a single well(Moyer, McGuigan, and Kretlow (1992)). As this example illustrates, the risk of drilling any individual well can be diversified away very effectively by expanding production. An analogy of this example for high-tech firms is that one firm can focus on more than one product line or can focus on alternative research paths to achieve a breakthrough in R&D. This is one example of how firm size can affect risk and therefore return on investment. TE: st. £th :'.-'_r;-O-! '\‘~ 61 ‘ «ULJJLrlh‘b C ‘_ '77- any one 56 N’ T Em“) =; z WMRM (4) '1 t'l where Wm is the expected return for firm i at time t, R1,; is the expected return from an individual product line: W, = 1 and) S Wi S l. The expected return is translated into a required rate of return for investors by using a conventional approach-~the risk adjusted discount rate model. Let the required return from any one investment j, denoted kj, be equal to the following: k,- = rt + Q (5) where rf is the risk free rate and 9,- is the risk premium required by investors. A common method in determining the risk premium is by measuring the systematic risk: Beta(B). Usually betas are computed for firms; therefore equation 5 is extended to focus on the overall required return for the firm. Beta is a measure of the volatility of a firm’s returns relative to the returns of a broad-based market portfolio. It is defined as the ratio of the covariance of returns for firm j and the market portfolio to the variance of returns for the market. A Beta (B) = 1.0 indicates that the firm is of average systematic risk (that is, the same risk characteristics as the market as a whole). A B greater than 1.0 indicates that a firm has greater than average systematic risk; likewise, a [3 less than 1.0 indicates that the returns to the firm have less than average systematic risk. The beta coefficient tenets sttndud detidti ;, 32th ensured over a By diversifying. eit Izstptttduct line. the 1 ..t-"ifl. Aftnn's beta can tee r.J represents the it radiating the return it: item a stock inde eteptis indicated bi 33': . ' .tptnieular samt N3] .3 , ...CL dunng [he ta, ll immittm. G). that 57 The beta coefficient is a function of the firm’s standard deviation of returns (03), the markets standard deviation (cm), and the i‘h firm’s correlation with the markets return (pm) each measured over a particular sample period: _ Cov(r1,rm) _ 01 - Omz — p1,m-a; (6) I31 By diversifying, either across product lines or by expanding the number of projects within a product line, the firm’s returns will be less variable, and therefore a firm’s B will be lower. A firm’s beta can be determined by estimating the following regression equation: r1,c=a1+flirm.c+ei.c (7) Where rLt represents the return for the im firm in the tth period. This is usually determined by calculating the return for the firm over a one month period. The analogous one-period return from a stock index (usually Standard and Poor’ 3) is denoted by rm. The regression intercept is indicated by oq. [3, is the beta systematic risk index for the im firm estimated over a particular sample period. It is customary to use 60 monthly data points to calculate betas (Kolb and Rodriguez (1992)). Last, 8.", is an unexplained residual for the ith firm during the t‘h time period: E(ei.,)=0. Given B, it is now possible to compute the risk premium, 9, that is applicable to the firm. Rearranging equation 5 I obtain: 4w? 9 is [l ‘..~.~ V . ' ' . a; "‘ 31'" fl , ll. Usin N ‘ . \ 1 37-.;:ct w (9=k*-rf (8) where G is the risk premium for the firm, k* is the investor’s required return, and rf is the risk-free rate of return. Next, let kIn be the expected rate of return on the overall market portfolio and rf be the expected risk free rate. Then the average market risk premium is equal to:15 szkm'rf (9) Using the firm’s beta, the investor’s required return can now be determined. For a project with average risk ([3 equal to 1.0), the risk premium should be equal to the market risk premium. If a firm’s beta is 2.0, this implies that the firm’s return is twice as risky as the market average, so its risk premium should be twice the market risk premium. 9 = [3(km ' rt) (10) Rearranging equation 8, the required return is: k"‘=r,+(~)f (11) ‘5 Based on historic stock market data over the time period 1926-1989, the average market risk premium has been 8.6%. (Moyer, McGuigan, and Kretlow (1992)). . - . 1?" ' (@311?th In J» :e;.:':.es imt "‘ ’15" Jan»; GI ll "fs’ the «a. M. ffql‘ Ineoi ,4 g t . as rent 59 And substituting in equation 9: k* = r, + mic,n - r,) (12) Thus, the required return is a function of the risk-free rate and the market return which are the same for all firms, and the firm’s beta, which is a measure of the volatility of a firm’s returns as compared with the market. This is the point at which diversification becomes important. Recall that the more stable the returns, perhaps due to a large number of investments or product lines, the lower the firm’s beta will be. This implies that the required return (k*) is lower. Incorporating this into equation 1 (the firm’s net present value), and substituting k*, the required return on investment, for the interest rate (r): _ INCOME1 + INCOME2 + INCOMET _ T INCOME (1 3) (1+kat) (1+k*)2 """ (1+k*)T m (1+k*)‘ The implication of the above equation is that the higher the perceived risk of the firm by the market, the higher the required rate of return (indicated by k*), and the lower the firm’s net present value, all other things equal. i. MODEL In this Section -tu test the s. ;::s:etietted in C 1.21;: mere than 31d. This is th :titiiures over a ‘ innun about f‘ 7.133: due to R3: nil-D inputs ar 5 7nd high-teem steely on five k glitzeutieal, an 513353361 in order 3‘5‘313' Simple f '4 lid teehnolog “' 1nd large f1 60 C. MODEL In this section, an empirical model is formulated from the market valuation model in order to test the Schumpeter hypothesis. It differs in three key assumptions from the studies reviewed in Chapter II. First, instead of focusing on whether innovational inputs increase more than proportionately with firm size, the outcome of these inputs is analyzed. This is done by examining the response of a company’s stock value to R&D expenditures over a five year period. It is argued that the stock market takes into account information about firms’ expected future profitability and therefore factors in expected outcome due to R&D investment. Using this method, the problems associated with using both R&D inputs and outputs (patent data) do not occur. Second, instead of pooling both low and high-technology firms, industry conditions are given recognition by focusing separately on five key high-tech industries: the computer, machinery, medical equipment. pharmaceutical, and semiconductor industries. And third, various statistical methods are employed in order to test for the effect of firm size. This is done, first, by employing a relatively simple functional form which assumes a continuous relationship between firm size and technological progress. Second, I specifically test for the precise definition of small and large firm size: that is, a continuous relationship is not assumed to exist. 61 l. EMPIRICAL FORMULATION AND MEASUREMENT ISSUES I begin by generalizing equation 3 in order to take risk into account: V = f(income streams (TA, KS, MS, Risk)) (14) where V is the market valuation of the firm, TA is the tangible assets, KS is the knowledge stock, and MS is the marketing stock. R&D investment is used as a proxy for knowledge capital and advertising expenditures for marketing capital. As stated in the previous section, the firm’s beta is commonly used as a measure of systematic risk. Yet, the true functional form of the relationship between firm valuation and tangible and intangible assets is unknown. One possible method of analyzing the relationship between firm valuation and tangible and intangible asset investment is by means of applying a Taylor expansion to an unknown functional form: V1,c = 111' t + VITALc + YzRDLc + “ADVL: + 74R15k1,c+€1,c (15) where V, TA, RD, and ADV are defined above. A firrn’s beta is used as a proxy for risk. The coefficients (7,, Y2» and Y3) measure the effect of a change in the different types of capital held by the firm. The coefficients on the different types of capital represent the investors’ perceived effectiveness of the firm for turning different types of 62 capital investment into future income streams. The coefficient on "Risk" (74) represents the effect of risk on market valuation. This framework will be used in this study to analyze whether R&D efficiency differs across firm size. Measuring the effectiveness of different types of capital at a given point in time has potential problems. First, stocks of knowledge are being measured; however, the amount of the knowledge capital held by the firm prior to a given point in time is unobservable, and therefore the total amount held at any particular time is therefore also unobservable.“ Second, the firm’s knowledge base affects firm value (due to expected future income streams), but the value of the firm may also affect the firm’s stock of knowledge (a higher firm valuation may increase R&D investment, all other things held constant). One method of mitigating these problems is by first differencing. First differencing can eliminate the importance of the unobservable levels of intangible capital held by the firm by only considering the change during the relevant time period. In addition, first differencing diminishes the importance of the endogenous setting of capital levels. Entering beta directly into the model would require computing the firms’ betas at the beginning and end of the time period in order to calculate the change during that time period. Empirical investigations by Baesel (1974), Blume (1971), and Roenfeldt et al. (1978) examine the stationarity of betas over time. These studies all conclude that the stability increases as the length of the estimation period increases--betas are relatively ‘6 For example, this data set spans from 1982-1989. There is no way to measure the firm’s stock of knowledge capital before 1982. Since the knowledge stock is unobservable before 1982. it must also be unobservable after 1982. iii: in periods of 4 01 :5: due of beta at 1116 3: then first differene libeta is a relev :teec'itied.19 In this acres due to R&D :nzdttres. not absolu .. expect R&D inter :tflaedwith beta. T taste the coefficients \ ' A test of beta stat gaping them into qui :tuin period on tl fires of subsequent ~hug in the initial p U dill. iI’liten' - lf ' Silt m firm beta h nuhen the “~35 longer the u tom . 1MP}: Smce lm't Minna ' is. cPOrtfolios 63 stable for periods of 4 or more years.17 For this reason, I assume that firms have the same value of beta at the beginning and end of the time period (five years). This means that, when first differencing, the beta (risk variable) drops out of the equation.18 If beta is a relevant variable and it is omitted, then the regression equation is misspecified.19 In this case. R&D expenditures may be correlated with beta since outcomes due to R&D investment are subject to great uncertainty. Yet, it is relative expenditures, not absolute expenditures, that would suggest a higher beta. Therefore, we may expect R&D intensity (R&D expenditures as a percentage of firm size) to be correlated with beta. This may be important if R&D intensity varies over size classes, because the coefficients on R&D which measure the perceived effectiveness of R&D ‘7 A test of beta stationarity usually entails ranking firms by their beta coefficient and grouping them into quintiles. To evaluate the influence of the length of the subsequent estimation period on the stability of beta coefficients, betas are estimated for various lengths of subsequent periods. Transition matrices are established based on the firms ranking in the initial period and its ranking in subsequent periods (see Roenfeldt et a1. (1978)). 18Even if firm betas were to be calculated, Harrington (1983) states that "betas vary greatly, even when they are calculated on the basis of historical returns." She also notes that the longer the time period in which beta is calculated, the better the forecast accuracy. Since investors are assumed to take all information into account when determining portfolios, one assumes that they analyze the most accurate beta available. Since the most accurate beta is one which utilizes a longer time horizon (as stated previously, usually 5 years), this, by definition, assumes that beta is constant over that period of time. This is not to say that beta is constant over a five year period; in fact, it is most likely that the change in beta is not equal to zero. This assumption implies that beta’s are not expected to change drastically over this period. ‘9 The implication of this is that biased and inconsistent estimates of the coefficients are obtained if there is a correlation between the omitted variable and the independent variables. If the variables are uncorrelated, the estimator will be unbiased, however, the Variance of the coefficient will be biased (overstated). As such, the hypothesis that a Coefficient is significantly different from zero may be rejected when that coefficient is Significant. 64 expenditures may also include the effect of risk. I find no evidence that R&D intensity varies across size classes. The proxy for risk--beta--is a measure of systematic risk. Alternatively, unsystematic risk can be taken into account by deflating firms’ market value by a market index, which is discussed in detail in Appendix 1. This allows for the separation of increases in market value due to overall increases in the general level of the market compared with increases due to tangible and intangible asset investment. To sum up, while risk does play a major role in market valuation, this effect can be factored out by assuming that unsystematic risk for a firm is constant over a 5 year period. In addition, industry conditions are taken into account by estimating coefficients for each industry separately. And unsystematic risk is taken into account by deflating the fir‘rn’s valuation by a market index (see Appendix 1). Beginning with equation 15 and first differencing: t'T C‘T V1,c‘V1,c-s =a +71 (Assetsi't-Assetsi'cfi) +72 2 1201,: +73 2 ADV1 C'T-S C‘T‘S (16) where: Vin-Vi“: the change in the market value of the firm over a five year period. Assets-m-Assetswz the change in the value of the firm’s tangible assets. 2RD: sum of R&D expenditures over a five year period. EADV: the sum of advertising expenditures over a five year period. 65 The model above characterizes the profitability of different types of capital investment on market valuation. but it does not indicate the relationship between firm size and capital effectiveness--specifically R&D profitability. Again. the true functional form of this relationship is unknown. I present two possible methods of modelling this relationship. The first is a relatively simple approach which involves including sales- intangible asset interaction terms. The second approach statistically tests for points of structural change. 2. FIRM SIZE - INTANGIBLE ASSET INTERACTION TERMS A relatively simple way of testing the Schumpeter hypothesis is by applying a Taylor expansion to an unknown functional form: SPECIFICATION 1: t-T Vin-Vise-s=a1;e+71 (ASSETSi,t-ASSET51,C_5) +72 2 RDLt+ t'T-S t'T t'T t'T 73(Sales* Z RD)i.t+y4(Salesz* Z: RD)Lt+'y5 2 ADV1OC+ t‘T-S t'T-S t'T-S C'T tIT 16(Sales* 2 ADV) 1't+y7(Salesz* 2 ADV) “gem t'T-S t‘T-S where: Vm-Vm,5: the change in the market value of the firm over a five year period. Vm-VM is deflated by the change in a market index over the relevant time period. This controls both for systematic risk and general trends in the stock market (see Appendix 1). 66 Assetsm-Assetsm,5: the change in the value of the firrn’s tangible assets over a five year period. 2RD: sum of R&D expenditures over a five year period. Sales*}:RD: a firm size-R&D interaction term. Sales is a proxy for firm size. Sales2*£RD: controls for non-linearities in the relationship between firm size and the rate of return to R&D. EADV: the sum of advertising expenditures over a five year period. Sales*XADV: a firm size-advertising interaction term. Sales is a proxy for firm size. Salesz*EADV: controls for non-linearities in the relationship between firm size and the rate of return to advertising. Size enters into the relationship between firm size and the stock response to R&D investment by means of the sales-R&D and salesz-R&D interaction terms. The coefficient 71 represents the return to tangible assets. The Schumpeter hypothesis will be tested by means of evaluating the coefficients on the Sales-R&D interaction terms. Including the sales-squared term forces the relationship between firm size and the stock response to R&D to be quadratic. There is evidence that there exist non-linearities in the relationship between firm size and R&D expenditures and patents (Scherer (1984), and Bound et al.(1984)). By including the squared term, I am able to test for this non- linear relationship. This specification is a variant of those used by previous studies which regress innovative inputs or outputs on a measure of firm size (see Chapter 1). Note that . “WA 67 this specification assumes a continuous function.20 Yet, it may be the case that the relationship between firm size and R&D profitability is not continuous. In other words. there may be a particular firm size such that the stock response to a firm with sales greater than that amount is significantly greater than (or less than) the stock response to R&D for a firm with sales less than that particular firm size-~in other words, a threshold effect. In addition, this specification in not able to determine the definition of large and small firm size. 3. PARAMETER STABILITY Ideally, one would like to first statistically determine whether there exist different size regimes. If this is found to.be the case, one can then test for differences in R&D profitability by separating firms into these different size groups. In order to arrive at Specification 2, I relax the assumption of a continuous quadratic relationship. I first specifically test for points of structural change. One method of testing for parameter stability is to perform the CUSUM and CUSUM of Squares statistical test (Brown, Durbin, Evans (1975)).21 However, this test 2° An alternative "simple" specification would be to divide firms into 2 or 3 equal size groupings. One could create dummy size-R&D (and Advertising) interaction variables. A t-test is then able to determine whether the stock response in large firms is significantly different from that in small firms. Yet, this specification a priori assumes a definition of large and small firms. Only by chance would one choose the "true" break points. 2‘ An alternative test of the stability of the regression coefficients is "Chow’s Predictive Test" (Kmenta (1986)), which is based on the F-statistic. However, this test is similar to Specification 3 (see Appendix 4), which assumes particular break points. This may lead to the rejection of the null hypothesis of different size regimes when more than one regime actually exists. 68 does not provide information on the point(s) at which parameter instability occurs, or which variable(s) cause the instability. In the case in which parameter stability is rejected. appropriate break points can then be determined by applying Quandt’s (1958) log likelihood ratio test. Finally, the Schumpeter hypothesis can be tested by creating dummy size-interaction terms for n-l of the n different size groups as determined by the log likelihood ratio test. For example: SPECIFICATION 2: V1,: — V1,” =a1, t ”751,;- +DL1' t+ yl (Assetsht-Assetsi' t_5) + DSM2 (Assetsi' t -AssetsL b5) +DL=w3 (Assetsi, t-Assetsi' t_5) + t... m c=r (18) Ya z RDi,e+YSDS* z RDi,c+YsDL* 2 R01.“ t=T-5 t=T-5 t=T-5 t-T t-T CUT y, z Anvmwens: 2 ADvagDL: 2 ADvm+em C'T-S CIT-5 cur—5 where: Vm-VM: the change in the market value of the firm over a five year period.Vi.,-Vi.,_5 is deflated by the change in a market index over the relevant time period. This allows us to control for general trends in the stock market. See Appendix 1. Assetsm-Assetswz the change in the value of the firms tangible assets over a five year period. An alternative test of parameter stability, which is also based on recursive residuals, is proposed by Harvey and Collier (Harvey (1981)). The switching regressions model determines appropriate break points, but does not specifically test for parameter stability. l‘dz sum of R&D exp 33“»? the sum of ads IS‘YRDaDd DS‘EAD‘ Edwiflflcd value. r ETERDnnd DUEAI .73; then a specified Note that two. specification tee teas. etc. may be f0. For medium si iii for small firms WERE“. For lat ’3 mu + DUZRI 2a. tmns is detern S ta.1 t 13 [ht sum. :‘efiflfim a - {5- Slmi in . “N by Subseir e‘efinine wheth- 69 RD: sum of R&D expenditures over a five year period. ZADV: the sum of advertising expenditures over a five year period. DS*ZRD(and DS*ZADV): Dummy interaction variables. DS=l if annual sales are less than a specified value, which is different for each industry. DL*2RD(and DL*2’.ADV): Dummy interaction variables. DL=1 if annual sales are greater than a specified value, which is different for each industry. Note that two break points (or three size regimes) are assumed to exist in the above specification (equation 18). This need not be the case--one break point, two break points, etc. may be found to exist. For medium size firms, the stock response to R&D is “Y4- The stock response to R&D for small firms is the sum of the coefficients "Y4 + y,"--the coefficients on 2RD... + DS*ZRDL,. For large firms, the stock response to R&D is the sum of the coefficients on 2RD” + DL"‘22RDi',--y4 + 76. The standard error for the stock response to R&D in small firms is determined by: Standarderror(?4+?5) =fif+sé+2Cov(?4,?5) (19) where s4 is the standard error for the coefficient Y4, and s, is the standard error for the coefficient 75. Similarly, the standard error for the stock response in large firms can be computed by substituting $6 for $5, and Cov(y4,76) for Cov('y4,'ys). A t-test can then be used to determine whether the stock response to R&D in large firms is significantly different 7min small f‘ gamer. . 4. PAR5 lhe Schumpt :23in is tester a. SPE The effect t 33133111211 deri‘ ‘3 um size (33} :‘Wn ' ‘r‘wtatton of u \ . lab-*3. - lLdSIng (Y3>( C or de. 70 from that in small firms. A similar analysis can be performed on the advertising coefficients. 4. PARAMETER TESTS: The Schumpeter hypothesis of a positive relationship between firm size and R&D profitability is tested by means of analyzing the effect of firm size on R&D profitability. a. SPECIFICATION l The effect of firm size on R&D profitability is tested by means of analyzing the mixed partial derivative (denoted Z) of the change in market valuation with respect to both firm size (sales) and the stock of R&D capital (2RD): _ 62 (V1,c'V1,c-s) Z- = 2 Sal (20) OSaleSGYRD 73+ 7‘ es Z signifies the effect of a one dollar increase in the stock of R&D holding sales constant; or the effect of a one dollar increase in sales holding the stock of R&D constant. It is used to show how this effect changes as firm size changes. In order to do this. interpretation of the signs on the independent variables-q3 and Y4--is necessary. The sign on 73 indicates whether the change in valuation (Vu-VM) as R&D increases is increasing or decreasing (73>O and y3<0, respectively) as firm size increases. The rate of change of this increase or decrease is represented by 74- If y4<0, this implies that the change in 71 market valuation with respect to R&D is changing at a decreasing rate. On the other hand, 74>O indicates that the change in market valuation with respect to R&D is changing at an increasing rate. If Z changes sign from positive to negative (73>0 and y4<0), this means that the effect of firm size on R&D profitability changes from being a positive effect to a negative effect. Z could change from a negative effect to a positive effect (y3<0 and 74>0) as well. In testing the Schumpeter hypothesis, I evaluate how Z changes as firm size changes and whether Z is significantly different from zero. In order to evaluate the significance of Z, the standard error of Z--denoted S--is computed: 1 5= [Var(Y3) +4*Salesz*Var(y4) +4*Sa1es*Cov(y3, Y4) ] 3 (21) Next, 95% confidence intervals are determined for Z (Zi2S). This information is used in testing the Schumpeter hypothesis. If the upper confidence bound always lies above the zero axis, while the lower confidence bound always lies below the zero axis then Schumpeter is not supported regardless of the signs of 73 and 74. This scenario indicates that the effect of size on R&D profitability is not significantly different from zero. In order for size to have a significant effect, the lower and upper confidence bounds must either both lie above or below the zero axis for a range of firm size (not necessarily the entire range). Next I describe the four combinations of signs on 73 and Y4 and explain the conditions in which Schumpeter is supported. 1. Y3>O and 74>O. This implies that as firm size increases the change in market valuation as R&D changes is increasing (73>O) at an increasing rate (74>O), see Figure 4. In Figure 4, both the upper and lower confidence bounds lie above the zero axis for the entire range of firm size. which indicates that the effect of size on R&D profitability is l--\. “'1 72 increasing at an increasing rate (and is significantly different from zero).22 $0.035 g 0.03 Therefore, Schumpeter is go 025 supported. g 0'02 EQOIS 5 a 0.01 2- Y3>09 Y4<0' The 20.005 effect of size on R&D profitability is increasing (790): but at 3‘ dccmaSing FIGURE 4: y,>o, mo; SCHUMPETER SUPPORTED rate (y4 <0). If Z changes sign from positive to negative, then this would imply a threshold effect of firm size with respect to R&D profitability. Whether or not Schumpeter is supported depends upon the confidence bounds. For example, in Figure 5, for firms with sales less than the size corresponding to point X (approximately $7 billion in annual sales), size has a positive and significant effect on R&D profitability; as firm size increases, R&D profitability increases, albeit at a decreasing rate. However, for firms with sales greater than $7 billion, size does not have a significant effect on R&D profitability since the confidence bounds straddle the zero axis. Point Y corresponds to the firm size at which the effect of size on R&D profitability changes from a positive effect to a negative effect (in other 22Instead, if the lower confidence bound is below the zero axis while the upper confidence interval above, then Schumpeter would not be supported--size does not have a significant effect on R&D profitability. In other words, the rate of change in market valuation with respect to the sum of R&D expenditures and firm size does not significantly change as firm size increases. 73 words, as size increases R&D profitability now decreases). Figure 5 g 0.015 . . ,1 0.01 portrays a case in which g a. the Schumpeter hypothesis E was 2 O is supported. R&D g o 3 profitability increases as g4” E --001 size increases, but after a ' o 2 4 o a 10 12 i4 FWSIZE(SALES. NW” certain point, size is not a Significant “Emma“ 0f FIGURE 5: y,>o, y,0 and y4<0 (see Figure 6). The primary difference between Figures 5 and 6 is that the upper confidence bound in Figure 6 lies below the zero axis over a range of firm size. The interpretation of Figure 6 is similar to that of Figure 5. For firms with sales less than point X, R&D profitability is increasing at a decreasing rate, and this effect is tat-5.33333 dlilél‘tm :5 Bgn'eeu poin ,; licorreSPOndi“ finitely $9 bil 312515011 in sale :2 of silt 0“ ;:;:'itability 1! ...atly differ: 3: For fun zesales great: genuinely S' 12ml sales), 1 Eaten from 1: find to have a 760 am iiihlng rats | If and low 35313 7). In 3?tir5iti\'e am "SJUVC and S 74 significantly different from zero. Between points X and Y(corresponding to E 0"" at 0.003 ( approximately $9 billion to ’5’ 0.006 . . . E 0.004 $12 brllron 1n sales), the E g 0.002 effect of size on R&D a ° 34.002 profitability is not Em,“ . . . -0.006 srgmficantly drfferent from o 2 4 o a 10 12 14 msrzusnss. warmest) zero”. For firms with annual sales greater than Y FIGURE 6: 790, ”<0; SCHUMPETER NOT SUPPORTED (approximately $12 billion in annual sales), the effect of firm size on R&D profitability is negative and significantly different from zero. Therefore, Schumpeter is not supported in Figure 6 since size is found to have a negative and significant effect on R&D profitability. 3. y3<0 and 74>0. The effect of size on R&D profitability is decreasing at an increasing rate (see Figures 7 and 8). The Schumpeter hypothesis is supported only if the upper and lower confidence bounds lie above the zero axis for a range of firm size (Figure 7). In other words, Z changes sign and the effect of size on R&D profitability is positive and significantly different from zero. In reference to Figure 7, size has a negative and significant effect on R&D profitability for firms with annual sales less than the firm size associated with point X (approximately $4 billion in annual sales). At point 23This is because the upper confidence interval lies above the zero axis and the lower confidence interval lies below the zero axis. 7. approximately gig-manual sales). I it: of size on Re“ hi changes fr- t negative to positi 'r the effect of f iiiD profitability is 22‘.ch between pt fail Only when f: It larger than the it attained with j 51¢ puint at which mmmmui titre influence of 331313 effect Sehur )3: i n. @068 nor cross ten: size has a ne0 mm 0f Size on a -' tile on R&D pm ~ sigmfi . Cantly dl.’ Y (approximately $9 billion in annual sales), the effect of size on R&D profitability changes from being negative to positive. However, the effect of size on R&D profitability is not significant between points X and Z. Only when firms are larger than the firm size associated with point 75 0.0003 0.0002 0.0001 2 4 s s 10 12 14 16 musrzusumuumnss) FIGURE 7: y,o; SCHUMPETER NOT SUPPORTED, SCENARIO A. Y (the point at which the lower confidence bound crosses the horizontal axis) is this effect both positive and significant. In Figure 7, there exists a threshold effect to the positive influence of size on R&D profitability. Since size has both a positive and a negative effect, Schumpeter is not supported. Now suppose that the lower confidence bound does not cross the zero axis (see Figure 8). This scenario also implies a threshold effect; size has a negative and significant effect on R&D profitability up to point X, then the effect of size on R&D profitability is not significantly different from zero. The effect of size on R&D profitability changes from negative to positive at point Y, however, it is not significantly different from zero. Since, as firm size increases, R&D profitability 233$ 01‘ 15 mt ,nh different froI the SchumP'C“ ;;;;thesis i5 n‘ :35“: 3 t pd) and ‘{,<(). T‘. :t:i::‘tion 0f 5129 cities that t iinpeter hypothesis I: summed-regard if he position of etidenee bounds. Y hcoufidenee bound: «t1 in Figure 9. in in annual 511: 333m from zero. "he point at whic' thti.‘35 ' - $12: more Ctr-a #501120 76 decreases, or is not significantly different from 0.001 zero, the Schumpeter g 0.0003 , , ; 0.0000 hypothesrs rs not E 0000‘ 0.0002 supported.”4 a 4. <0 and <0. This a '°-°°°2 73 Y4 "' 43.0004 combination of signs 5 '°‘°°°5 ‘E’ 41.0008 U Impllcs that the .0’0010 2 ‘ 5 8 1° 12 1‘ mu srzr: (SALES. lN minus 3) Schumpeter hypothesis is not supported-~regardless FIGURE 8: 734), Y4>03 SCHUMPETER NOT , , SUPPORTED, SCENARIO B. of the posrtron of the confidence bounds. Yet. the confidence bounds provide an explanation why Schumpeter is not supported. For example, in Figure 9, for firms with annual sales less than point X (approximately $1 billion in annual sales), the effect of size on R&D profitability is not significantly different from zero. However, for firms with annual sales greater than X but less than Y (the point at which the upper confidence bound crosses the zero axis for the second time), as size increases R&D profitability decreases. Therefore, Schumpeter is not supported. 2" An alternative scenario is if the upper confidence bound lies above the zero axis and the lower confidence bound below. This implies that the effect of size on R&D profitability is not significantly different from zero, and therefore Schumpeter is not supported. 77 If th e u p p e r confidence bound always lies below the zero axis, the Schumpeter is not supported because size has a negative effect on R&D profitability. And if the EFFECT OF SIZE ON R&D PROFITABILITY . 1 upper confidence bound 0 1000 2000 3000 4000 5000 5000 7000 8000 msrzr (SALES. N 11.101153) lies above the zero axis, Schumpeter is again not FIGURE 9. 734) 744). , SCHUMPETER NOT SUPPORTED supported; srze has no significant effect on R&D profitability. b. SPECIFICATION 2: In this specification, size enters into the regression equation through dummy 2 interaction terms (refer to equation 1.5 *____ 18). A t-test can be used to determine ' .,._.__. 0.5 O ._._*_ i 3 .2 whether the stock response to R&D in E large firms is significantly different 3 -0.5 M f ' - tom that m small firms' The '0 200 400 000 1100 1000 12001400 100011100 2000 imam-(sen. hula-S) Schumpeter hypothesis is supported if FIGURE 10: SPECIFICATION 2. the stock response to R&D in the SCHUMPETER SUPPORTED largest size class is Significantly greater than the smaller size class(es). For example, if this log likelihoc 32220111201116" ”.725 significantly ”1* from zero)- tt'dei fall out of equ gsyiicantly less .. n be shown etch axis and fu 1;... we can test are the stock res titer size class however, 4": “CI’SLM or ( '31 C1'3th of S that 1101 do IE‘ies that th $213111 fIOm they Signific 78 Quandt’s log likelihood ratio test indicates the existence of 3 size regimes, then the Schumpeter hypothesis would be supported if n+7,5 is significantly greater than 74+75, and 74+}; is significantly greater than 74(in other words, 76 is positive and significantly different from zero). If two size regimes are found to exist in an industry (76DL*ZRD would fall out of equation 18), then Schumpeter’s hypothesis would be supported if 74+75 is significantly less than “(4 (that is, 75 is negative and significantly different from zero). This can be shown graphically by plotting the stock response to R&D investment on the vertical axis and firm size on the horizontal axis. Confidence bounds are plotted as well so that we can test for significance. For example, Schumpeter is supported in Figure 10 since the stock response to R&D is higher in larger firms than in both the medium and smaller size classes. However, if size does not confer an advantage to R&D profitability, then either the CUSUM or CU SUM of Squares test cannot reject parameter stability or the CUSUM or CUSUM of Squares test rejects parameter stability, but the different size regimes that exist are not due to the stock response to R&D investment. In the case of the latter, this implies that the stock response to R&D investment in large firms is not significantly different from that in medium or small (or both) size regimes. That is, 75 and 76 are not jointly significant. The parameter instability may be due to the long-term profitability of advertising investment. Alternatively, the Schumpeter hypothesis may not be supported because the long term profitability of R&D investment in small firms is greater than that in large firms. In this case both parameter stability would be rejected and the stock response to R&D in the largest size class would mt be significantly greater than the smaller size class(es). For mtg. it Quandt‘s log ages determines that n exist then sma‘. :33 profitable in R&C {gently greater the in size firms are m :liD then large rim ...ntlt greater th. “herds. y, is negati it of equation 18 @Etmtly greater th; 3: a e ' ”trample of thrs so example, if Quandt’s log likelihood ratio test determines that three size i 2' i regimes exist, then small firms are g ‘-5H—'—'—H—'—* more profitable in R&D if 74+ys is E ‘-—-—-—-—-—-—~—$ ' * * 0.5 significantly greater then 74+76. If 3 o ' . . a h—H—fi—H medlum Size firms are more profitable 4.50 m moo mm mm mm mo Hausa-(sun. hunt) in R&D then large firms, then 74 is FIGURE 11: SPECIFICATION 2, significantly greater than Y4+Ys~ In SCHUMPETER NOT SUPPORTED other words. Va is negative and significant If two size regimes exist (76DL*ZRD would fall out of equation 18), then small firms would be more profitable in R&D if “(#75 is significantly greater than Y4« That is, y, is positive and significantly different from zero. An example of this scenario is shown in Figure 11. This Ci‘t'; Tzaeter stat V'“\ 1‘ 5 east?“ Sill ”Alsatian k DAT Esti :33 I'Ilree F1; CUVer x3, YDrle “fer {he c :1“: (P « k“: Ltd]- ck IV. RESULTS This chapter begins with a description of the data. Next, I present regression results from specification 1 (firm size-intangible asset interaction terms). Third, I test for parameter stability using the CUSUM and CUSUM of Squares statistical test. If structural stability is rejected, I determine points of structural change and estimate specification 2, or a variant thereof. A. DATA: Estimation was performed on annual data from 1982-1989. Data were obtained from three Standard and Poor’s Compustatdata tapes: Industrial, Over the Counter, and Full Coverage. The Industrial file includes many large firms listed on the American and New York stock exchanges. The Over the Counter (OTC) tape includes companies traded over the counter, with inclusion criterion being that these firms must command significant investor interest. The Full Coverage tape includes companies traded on regional exchanges, wholly owned subsidiaries, and privately held companies. The firms are divided into five industries according to their Standard Industrial Classification (SIC). Table 7 presents a listing of the industries and the SIC code(s) for each industry. For 80 _____ I Fill?“ [WEST I hDUSTRY l——--—""" ‘lrrtter l :zmn Machinery l .l'eiital Equipment l'!‘ - . :antteutreal l itronductor .2: if he tire sectors. zetiittrres. advertisin a‘ maximum firm siz 1t: and median firm El skewed to the 81 TABLE 7: INDUSTRIES USED IN ESTIMATION ' INDUSTRY src CODE(S) ‘ Computer ' 3570-3572 3 3575-3579 1 Industrial Machinery 3533. 3559. 3560, 3561 | Medical Equipment 3841, 3842, 3844, 3845, 3851 3861 1 Pharmaceutical 2834 Semiconductor 3674 each of the five sectors, Table 8 presents summary statistics: the mean and median R&D expenditures, advertising expenditures and market value, and the minimum, median, mean, and maximum firm size and R&D intensity (R&D expenditures/Sales). Comparing the mean and median firm size in each of the five industries, note that the distribution of firm size is skewed to the left. For example, in the computer industry the mean firm size is $2.9 billion, while the median is $131.5 million; in the semiconductor industry the mean is $607.6 million, while the median is only $121.1 million. The average R&D intensity for all sectors pooled is 9.4%. The lowest mean of R&D intensity occurs in the machinery industry (5.8%), while the highest mean occurs in the pharmaceutical industry (12.4%). Information was obtained on a firrn’s market value, tangible assets, sales, R&D expenditures, and advertising expenditures. The criterion used in including observations is that all observations must report sales, assets at time t and time t—S, market value at time t and time t-5, and R&D expenditures over a five year period. Table 9 presents the total observations listed in the respective industries between 1987 and 1989 which report R&D Expenditure l‘, . )ifl'a \‘J'A . 5353143. 1 Advertising '. expenditure - lien: :11. '. 5‘ 0 MC “- . r llarlcet Va I View Median; ‘Firm Size yrimIIturn: iiiédian 82 TABLE 8 SUMMARY STATISTICS, IN MILLIONS S l l 1 Comp Mach Medi Phar Semi All R&D | Expenditures $224.3 $26.0 $46.5 $222.7 $60.7 $65.9 i Median: 8.9 7.2 1.8 144.0 9.9 8.6 I Advertising 1 expenditures 1 Mzean 23.3 9.5 27.7 176.5 5.0 28.5 1 Median: 1.3 2.4 0.5 49.1 1.9 2.2 1‘ Market Value I Mean. 1435.9 597.8 759.6 6351.6 821.5 1400.2 i. Median. 97.2 108.3 45.1 2448.5 100.1 117.0 ‘ Firm Size Minimum: 0.4 4.9 0.2 2.3 1.7 0.2 l Median: 131.5 129.7 42.8 1153.2 121.1 118.1 1 Mean: 2877.6 715.4 176.0 2367.1 607.6 863.4 Maximum: 14000.0 7100.0 1810.0 9700.0 6500.0 23000 R&D Intensity Minimum: 0.6% 0.5% 0.7% 0.6% 0.3% 0.3% Median: 7.7% 4.2% 5.6% 9.3% 8.5% 8.3% Mean: 9.1% 5.8% 7.4% 12.4% 9.8% 9.4% . 1 Maximum: 110.1% 17.9% 9.0 3% 42.1% 24. 8% 110. 0% Comp: Computer Industry; Mach: Machinery Industry; Medi: Medical Equipment Industry; Phar: Pharmaceutical Industry; Semi: Semiconductor Industry; Firm Size: Annual Sales; R&D Intensity: R&D Expenditures/Sales 83 sales, the total number of observations included in the sample, and the number of observations reporting each of the statistics above. When comparing the number of one year statistics available (market value. assets, and R&D expenditures) with the number of 5 year statistics available (Va-V“, Assetsi'rAssets-M, and 2RD) it appears as if many observations are excluded because statistics are not available five years prior. This criterion is more likely to exclude relatively more smaller firms compared to larger firms. Table 10 presents the percentage of observations included in the sample with less than $1 million, $10 million, and $100 million in annual sales, R&D expenditures and firm size, before and after the filter process. From this table I find that the 5 year criterion does indeed exclude many so-called "small" firms, regardless of how small is defined. However, the data set still includes a relatively large percentage of small firms (less than $100 million in annual sales).1 1 Many previous studies only included Fortune 500 firms. / 1111.19: NU Number Of 23:-803mm reportingi Sales 111111 Value Vn'Vn-s Assets Arena-Assets”, R&D 2RD i- .liter Filter \_' ' ‘ Q . . .1313: . taeiiitnres. L RD lie nnrn‘oer of r 84 TABLE 9: NUMBER OF OBSERVATIONS REPORTING VARIOUS STATISTICS Number of Comp Mach Medi Phar observations reporting: Sales Market Value Vi.t'Vi.r-5 Assets Assetsm-Assetsm R&D 2RD After Filter Vail/1.1.5 is the change in market valuation over a 5 year period. Assetsm-Assetsw is the change in tangible assets over a five year period. R&D is R&D expenditures, 2RD is the sum of R&D expenditures over a five year period. After filter is the number of observations included in the sample. C011 1 W L______ li0 and y4>0); see Figure 13. However, size is a significant determinant to R&D profitability only when sales are between $2 billion and $4.5 billion. Since size has a positive and significant effect on R&D profitability for a range of firm size, Schumpeter is supported. 2. MACHINERY INDUSTRY Size has a negative effect on R&D profitability, as indicated by the vertical intercept; see Figure 14. At a firm size of approximately $2.7 billion in annual sales, this effect changes from a negative to a positive effect. This means that, for firms with annual sales less than $2.7 billion, as size increases R&D profitability decreases, at an increasing rate(y4>0). R&D profitability attains a minimum at $2.7 billion, then R&D profitability increases as size increases. However, since the value of the upper confidence bound is greater than zero while the value of the lower confidence bound is less than zero, the effect of size on R&D profitability is not significantly different from zero and Schumpeter is not supported. 3. MEDICAL EQUIPMENT INDUSTRY The effect of size on R&D profitability is positive up to a threshold point of $800 million in annual sales (see Figure 15). This is shown by the positive vertical intercept and by both the value of the upper and lower confidence bounds greater than zero. The 92 effect of size on R&D profitability does turn negative at $900 million in annual sales (the point at which Z changes sign). However, since the value of the upper confidence bound is greater than zero while the value of the lower confidence bound is less than zero, size is not a significant determinant of R&D profitability. In short, R&D profitability increases as firm size increases, but only up to a point. Since size has a positive effect, Schumpeter is supported. 4. PHARMACEUTICAL INDUSTRY Since the vertical intercept (73) is positive and the slope (74) is negative, this implies that as firm size increases, R&D profitability is increasing at a decreasing rate(see Figure 16). Z changes sign at a firm size of approximately $3.25 billion in annual sales which implies that the effect of size on R&D profitability becomes negative-~as firm size increases, R&D profitability now decreases. Since the value of the upper confidence bound is greater than zero while the value of the lower confidence bound is less than zero, size is not a significant determinant of R&D profitability for firms with sales less than $4.8 billion. However, size has a negative and significant effect when firm size is between $4.8 billion and $6.3 billion (when the value of the upper confidence bound is less than zero). Size, again, does not have a significant effect on R&D profitability for firms larger than $6.3 billion. Since size has either a negative or insignificant effect on R&D profitability, Schumpeter is not supported. Results :ecativel‘ff aenneae. 2. as size ir trident fro In so a sigmf; i'lil'u'SIries a $9.611 ln ’3 compu included"). ‘i-l ‘1 didgn {BS 93 5. SEMICONDUCTOR INDUSTRY Results from Specification 1 indicate that the effect of size on R&D profitability is negative(y3<0), see Figure 17. At a firm size of approximately $3 billion in annual sales this effect becomes positive (this is the point at which Z changes sign). This means that as size increases R&D profitability decreases for firms with less than $3 billion in sales, but increases for larger firms. However, since the value of the upper confidence bound is greater than zero while the lower bound is less than zero over the entire range of firm size, Schumpeter is not supported since the effect of size is not significantly different from zero. In sum, the results from specification 1 indicate that in some industries, firm size is a significant determinant of the stock response to R&D investment. However, industries are not homogeneous; the influence of firm size varies across the 5 industries studied. In only one industry, however, is the Schumpeter hypothesis supported (two if the computer industry excluding firms with annual sales greater than $20 billion is included). One possible reason for the lack of support is the assumption of a continuous effect. It may be the case that the relationship between firm size and R&D profitability is not continuous, but is instead discontinuous. Therefore, I will try an alternative form which tests for stability in the parameters and is able to test take a discontinuous functional form into consideration. Sales‘IRD Salesil‘ERD ZADV Sales‘ZADV Sales:*ZADV $7315.: d . . | “Al errors g: l LililgD -- 1 W ‘ii‘pn ex»), 11356331511”; 94 TABLE 11: THE STOCK RESPONSE TO R&D INVESTMENT SALES-R&D INTERACTION TERMS Computer Computer‘ Machinery constant 75.93” 83.78 -l9.02** ((0.66) (126.24) (4.33) ASSCISm-ASSCISLM 1 .42* * 1.54** 0.85 (0.08) (0.08) (0.15) 2RD -0.48 -0.53 1.26” (0.29) (0.57) (0.24) Salcs*XRD 0.3E-3 0.6E-3** -0.5E-2* (0.15-3) (0.3E-3) (0.03) Salesz*ERD -0.5E-8 -0.5E-7 * 0.1E-5 * (0.2E-8) (0.3E-7) (0.9E-6) ZADV -1 .67 0.28 -0.34 (1.27) (2.05) (1.27) Sales *XADV -0.1E-2 -0.2E-2 0.02" (0.9E-3) (0.2E-2) (0.8E-3) Saleszi‘ZADV 0.2E-7 0.2E-6** -0.5E-5* (0.2E-7) (0.1E-6) (0.3E-5) 1 164 158 73 R2 .76 .78 .69 F 75.57 82.63 23.60 7.152 Standard errors given in parentheses. *Significant at the 10% level **Significant at the 5% level Variable Definitions: Assets-,J-Assetsi'gs: Change in the book value of tangible assets over a five year period. ZRDIZADV:Sum of R&D/advertising expenditures over a 5 year period. Sales*ZRD/ZADV: Sales-R&D/advertising expenditure interaction terms. Sales2*2RD/2ADV: Salesz-R&D/advertising expenditure interaction terms. r4 [‘4 95 TABLE 11, CONTINUED Medical Pharmaceutical Semiconductor Equipment constant -0.64 69.80** -24.67** (3.86) (34.99) (8.8) Assets-u-Assetsw 1 .07 * * 0.50** 0.83?" (0.24) (0.04) (0.14) 2RD -0.91 2.14** 0.31 (0.60) (0.70) (0.47) Sales*ZRD 0.8E—2** 0.6E-3 -0.8E-3 (0.2E-2) (0.6E-3) (0.8E-3) Sales2*2RD -0.4E-5** -0.8E-7 0.1E—6 (0.1E—5) (0.7E-7) (0.1E-6) ZADV -1.29 -1.65 6.09* (1.16) (1.27) (3.80) Sales*XADV -0.5E-2 0.2E-3 0.3E-2 (0.4E-2) (0.7E-3) (0.7E-2) Sales2*2ADV 0.5E-5* 0.1E-7 -0.5E-6 (0.3E-5) (0.7E-7) (0.1E-5) N_ 111 79 79 R2 .75 .90 .31 F 47.51 88.6 6.09 df 7,103 7,71 7.71 Standard errors given in parentheses. *Significant at the 10% level “Significant at the 5% level Variable Definitions: AsseterAssetsM: Change in the book value of tangible assets over a five year period. 2RD/2ADV:Sum of R&D/advertising expenditures over a 5 year period. Sales*ZRD/ZADV: Sales-R&D/advertising expenditure interaction terms. Salesz*£RD/EADV: Salesz-R&D/advertising expenditure interaction terms. 96 P o o o M EFFECT OF SIZE (W R&D "(TITAHJTY to 20 so 40 so so me 512: (SALES. IN muons 3) FIGURE 12: THE EFFECT OF SIZE ON R&D PROFITABILITY, COMPUTER INDUSTRY 0.005 EFFECT or SIZE ON R&D PROFITABIJTY 4 5 11101214151820 rmsrzr(sAL£s.1Naw0Ns$) FIGURE 13: THE EFFECT OF SIZE ON R&D PROFITABILITY, COMPUTER INDUSTRY (EXCLUDING FIRMS WITH ANNUAL SALES >$20 BILLION) 0.015 0.01 0.005 1.73472E-18 -0.005 -0.01 -0.015 0 1000 2000 3000 4000 5000 5000 7000 11000 mszusuts. neurons” FIGURE 14: THE EFFECT OF SIZE ON R&D PROFITABILITY, MACHINERY INDUSTRY 0.015 0.01 0.005 1.73472E-18 -0.005 -'0.01 -0.015 0 200 400 600 800 1000 1200 1400 1600 1800 mszusxtcs. newness) FIGURE 15: THE EFFECT OF SIZE ON R&D PROFITABILITY, MEDICAL EQUIPMENT INDUSTRY 98 -0.0005 -0.001 -0.0015 ._ -0.002 0 E -0.0025 0 1000 2000 3000 4000 5000 5000 7000 3000 900010000 mu 5121: (SALES. N “wows 3) FIGURE 16: THE EFFECT OF SIZE ON R&D PROFITABILITY, PHARMACEUTICAL INDUSTRY 0.003 0 EFFECT OF SIZE ON R&D PRNITABIJTY o o 'o 'o o o .- H -0.001 -0.002 -0.003 0 1000 2000 3000 4000 5000 6000 7000 mszHSALEs. 111101101153) FIGURE 17: THE EFFECT OF SIZE ON R&D PROFITABILITY, SEMICONDUCTOR INDUSTRY 99 C. PARAMETER STABILITY Results reported above need to be qualified in regard to the limitations imposed by assuming that the stock response to R&D investment is continuous over a ranking of firm size. In addition, specification 1 does not specifically test for a statistical definition of large and small firm size. Ideally, one would like first to determine statistically whether there exist different size regimes. If this is the case, one can then test for differences in R&D profitability by separating firms into these different size groups. This allows a determination of firm size regimes from the data without assuming a priori knowledge of these regimes. 1. THE CUSUM AND CUSUM OF SQUARES TEST Is the relationship between the change in firms’ market value and tangible and intangible asset investment stable over firm size? The definition of stability refers to the estimated coefficients of the explanatory variables, and whether they remain constant over a ranking of firm size. Since the regression model assumes constancy of the regression coefficients for all sample observations, the model will be misspecified if these coefficients are not stable. and the model does not correct for this instability of the parameters. The relatively simple procedures for testing model stability are the F-test or dummy variables. However, these tests require prior knowledge of the point at which the change in the estimated coefficients is presumed to occur, knowledge which may not be obtainable. Another relatively simple method of detecting observations which diverge 100 from the model specification is an examination of residuals. However, a plot of OLS residuals, or a plot of their squares against the ordered variable‘5 is not a sensitive indicator of small or gradual changes in the parameters (Brown et a1. (1975)). The CUSUM and CUSUM of Squares statistical tests developed by Brown, Durbin, and Evans (1975) can be used to test parameter stability over an index of firm size. These procedures and the procedures used to test for the points of structural change are preferred over F-tests or dummy variables because they do n_ot require a priori knowledge of the point in which the function shifts. These statistical tests have usually been applied to time series data. However, they can be used to test for model stability in any data set that can be ranked (see Link (1981)). The null hypothesis (Ho) that all of the regression coefficients are stable over an index of firm size will be tested against‘the alternative (HA) of parameter instability: 711 _ 712 = 7110 721 = 722 = '72 H0: 731 = 732 = 73 (25) where the superscript refers to the observation number and the subscript refers to the coefficient in the regression equation. 7 The ordered variable could be time in days, months or years in time-series studies. In this study, the ordered variable is sales, a proxy for firm size. 101 The procedure is as follows. First, the observations must be ranked in some manner. Because I am testing for parameter stability across firm size, the observations are sorted from the smallest firm size to largest. Next, a series of regressions are run in which the number of observations increases from k+1 to N, where k is the number of regressors and N is the total number of observations. The CUSUM and CUSUM of Squares require the calculations of one-period prediction residuals which are obtained by applying the regression computed with r-l observations in order to predict the 1m observation using k explanatory variables. Let er be the forecast error: er: yr - x,’y,,l, where Y“ is the least squares coefficient vector computed using all observations up to (but not including) y,x,. For each regression the forecast error is calculated and normalized by its standard deviation: 3 Yr-xll'vx-l = Yr-x;?r-1 I (26) w r=k+1, . . .N «1 +x; (le‘Xr-r) A": r 8 where X’,_l = [X,,..X,,,] The variable w,.,d,...wN is a sequence of approximately normal variables such that E(w,) = 0 (Brown et al. (1975)). The set of scaled residuals are independent and distributed N(0,0'z). The CUSUM test is based on the cumulative sum of these recursive residuals: w,= 2 _' r=k+1,...N <27) 62 102 The significance of the departure of the sample path of Wr from its mean value ’line (E(W,) = 0) is tested. If the structure of the equation changes over a ranking of firm size, this will result in a shift of the residuals when compared with a model assuming constant coefficients. If a structural shift occurs, a disproportionate number of recursive residuals would tend to have the same sign; this implies that the population errors are no longer centered at zero. In other words, if y, is constant up to observation r~=ro and different from then on, Wr will have a mean zero up to ro and nonzero mean from then on. If the CUSUM of residuals fall outside the confidence bounds, then H0 is rejected. These confidence bounds lie symmetrically above and below the line Wr such that the probability of crossing one or both lines is or, the significance level. Upper and lower confidence bounds are given by: +6 N- +2a(r-k) (28) _ m where N is the total number of observations in the sample, k is the number of regressors and a = 0.948 at the 5% significance level (Harvey, p. 152). Two graphical examples of the CUSUM test are shown in Figures 18 and 19. The values on the horizontal axis correspond to the observation number (not firm size). The values on the vertical axis correspond to the value of W,--the cumulative sum of recursive residuals. Figure 18 portrays an example in which structural stability cannot be rejected by the CUSUM test. 103 On the other hand, in Figure 19 parameter stability is rejected; the test statistic lies outside its confidence bound (observation l3). V I q 7 q I 4 5—1 4-1 3-1 2-1 3,“ A m -3—1 -4- -34 -8-1 -7-l "‘1 " 1 t 1 1 I r 1 1 I I I I 14” r 3 6 7 a 9 1O 11 12 13 14 15 1B 17 1O 1! 20 OB‘EHVATION NUMBER FIGURE 18: CUSUM TEST-PARAMETER STABILITY O-J 7-1 0 fl 5 A 4.. 3-1 ad A /I O -3: -3 d -4 d -3- -.d -14 A —.—1 I I l I r I r r I 1 T r I r 6 7 O I 10 11 12 13 14 13 1B 17 1. 19 20 OBEIIVATIONIIIIEN FIGURE 19: CUSUM TEST-~PARAMETER STABILITY REJECTED 104 The CUSUM of squares test is based on the normalized cumulative sum of squared residuals from a recursive estimation model. The difference between the CUSUM and CUSUM of squares tests is that the CUSUM of Squares is stronger and more reliable than the CUSUM test (Link (1981) and Garbade(l977)). In addition, the CUSUM test is aimed at detecting mainly systematic and gradual movements in the coefficients, while the CUSUM of squares test detects mainly haphazard types of movements. In other words, the CUSUM of Squares test is better at detecting parameter instability if the relationship between firm size and the stock response to R&D is not continuous. The term haphazard refers to a discontinuous function. Although it is possible that both the CUSUM and CUSUM of Squares tests reject parameter stability. The two tests are independent and it is possible for one test to reject the null hypothesis while the other does not (for examples, see Brown, Durbin, and Evans (1975), and Stern, Baum and Greene (1979)). The CU SUM of Squares test statistic (8,) equals, the value of the cumulative sum of w} up to observation r, divided by the cumulated sum of w,2 over the entire period: s,=££L r=k+1, . . .N (29) S, follows a Beta distribution with mean: (r-k)/(N-k). The value of S, will lie between zero and one: S,=O if r $700 million, and specification 2 is estimated. The estimation results are presented in Tables 18 and 19 and shown graphically in Figure 38. The results indicate that firms smaller than $700 million in annual sales have a significantly higher stock response to R&D than larger firms. The stock response to R&D in smaller firms is between 4.25 and 6.41(s=0.9). The results also suggest that the stock response to advertising investment is not significantly different from zero for either of the firm size classes. By statistically determining whether parameter stability exists, and the break points for different size regimes, firm size is found to be a significant determinant of R&D profitability. In sum, Schumpeter’s hypothesis does not hold in the pharmaceutical industry. 6. SEMICONDUCTOR INDUSTRY Both the CUSUM and CUSUM of Squares tests cannot reject the hypothesis of structural stability, while the CUSUM of squares test rejects structural stability at the 5% leve1(see Figure 39 and 40). Results from Quandt’s log likelihood ratio test reveals two possible points of structural change: $40 million and $42 million (see Figure 41). Both points are significant by themselves, but it is impossible to test whether structural change exists at both $40 million and $42 million due to insufficient degrees of freedom. 2° Test statistic: Ho: -21n 0=718.93, x2,(.01)=86.92. Therefore the null hypothesis of no structural change is rejected. 116 Therefore, results for structural change occurring at $40 million are presented.21 See Tables 20 and 21 and Figure 42. Results indicate that smaller firms, those with annual sales less than $40 million, that the stock response to R&D that is not significantly different from zero. On the other hand, in regression number 2 and 3, the stock response to R&D in larger firms is negative and significantly different from zero. These results conform to the reality of the semiconductor industry in the 1970s when foreign competition almost wiped out the industry. By performing the CUSUM test, statistical definitions of "large" and "small" firm size in this particular industry has been determined. Specification 1 found the stock response to R&D to be insignificant over the entire range. By testing for parameter stability and statistically determining break points the statistical definition of "large" and "small" firm size is determined and it is found that different size regimes exist. Thus again, in the semiconductor industry, the Schumpeter hypothesis is not supported. 21Results from Quandt’s test rejects the null hypothesis of no structural change at this point: -21n ¢=171.ll, 78586.92. PARA 00051 DS DL Assetsuaa SrAssctsn-r‘ UAssets,,-2 ZR] SIR {Al 32A LEA //.- -/ madam err ”ileum“ mink111; 0 ., standard errors are in parentheses. respectively. DS and DL are dummy variables. DS=l if annual sales are less than $180 million; 0 otherwise. DL = 1 if annual sales are greater than $1900 million; 0 otherwise. TABLE 12 PARAMETER STABILITY RESULTS: COMPUTER INDUSTRY 1 2 3 constant -23.48 £21.24 54.02 (265.4) (227.7) (132.4) DS 28.08 27.37 (369.8) (317.4) DL 1991.57** 1843.27** (346.6) (327.8) Aswan-Assets“, 1.84** 0.84** 1.23** (0.34) (0.1) (0.1 ) S(Assets-,,—Assets,',_5) -0.60 (3.6) L(Assets-,.,-Assets,,,5) -1.06** (0.3) 2RD -004 1.20** 0.80** (0.5) (0.3) (0.2) szRD -1.96 -3.17 -3.67 (6.1) (5.6) (4.0) LZRD -1.93** -3.08 -o.94** (0.5) (0.4) (0.2) ZADV -2.80 -2.53** -3.63** (1.4) (0.9) (0.7) SEADV -0.78 (23.2) LEADv 2.04 (1.9) R2 .81 .75 .80 F 59.46 69.57 124.77 **/* indicates significance at the 5%/10% level, E11 (11 () ("J Ac l I C. f-o (T 118 TABLE 13: SUMMARY OF COEFFICIENTS FOR SPECIFICATION 2 ‘ COMPUTER INDUSTRY 1 Stock Response to: 1 2 3 constant--small 4.60 6.14 firms (257.04) (222.3) constant--average -23.48 -21.24 54.03 { firms (0.09) (227.8) (132.4) l constant--large firms 1967.52** 1822.07" 1 (222.8) (268.2) I tangible assets: 1.24 1 small firms (3.59) 1 tangible assets: 1.84** O.84** 1.24" 1 average firms (0.34) (0.1) (0.1) : tangible assets: 0.78** i large firms (0.05) i R&D: small firms -2.00 -l.97 -2.87 l (6.03) (5.6) (4.0) ! R&D: average -004 1.20** 080* ‘ firms (0.51) (0.3) (0.2) . R&D: large firms -3.89** -l.88** -o.14 i (0.18) (0.2) (0.2) I Advertising: small -3.58 . firms (23.16) Advertising: average -2.80* -2.53** -3.43** firms (1.43) (0.9) (0.7) ‘ Advertising: large -0.76 1 firms ' (1.21) . F 59.46 69.57 124.77 ! df 11,148 7,152 5,154 ' R2 .81 .75 .80 J To ‘ 7"?“ Ts inwei WMT__—u a" ”rms (the coefficient is constrained to be zero).Small firms: <$180 million: Average: $180m.- $1.9b.; large: >$l.9b. F 133 01h, 119 TABLE 12a: COMPUTER INDUSTRY i EXCLUDING FIRMS WITH ANNUAL SALES >320 BILLION l 1 2 3 ; constant -152.16 -176.39 86.74** 1 (245.8) (175.6) (35.7) : Ds 38.61 193.67 I (252.2) (173.5) ' DL -5068.83** -6763.70** . (1541.7) (1279.9) i Assets,_, - Assets,5 0.84 0.68** 1.22** l d (0.6) (0.2) (0.1) I S(Assets,, - Assets-M) 0.97 I (0.6) L(Assets,, - Assets,,_,) -0.48 i (0.6) 2RD 0.73 0.73 0.02 (1.1) (0.9) (0.9) SERD -3.08** -1.95* -2.60** (1.2) (1.0) (1.1) LZRD 3.97** 4.16" 2.29** (1.3) (1.0) (0.9) )2on -1.13 0.46 -2.77** (2.0) (1.2) (1.1) szADv 5.57** (2.7) LEADV -0.89 (4.1) R2 0.78 0.91 0.86 F 46.94 230.20 197.42 I df 11,146 7,150 5.152 1 N 158 158 158 l DS and DL are dummy variables. DS=1 if annual sales are less than $360 million; 0 otherwise. DL = 1 if annual sales are greater than $1900 million; 0 otherwise. 120 ____ ___ __ —_—_.. .—____._.*__,.~k—___ .1 FTABLE 13a: SUMMARY OF COEFFICIENTS FOR COMPUTER INDUSTRY (EXCLUDING FIRMS WITH SALES>$20 BILLION) I I 1 2 3 I constant--small -113.55** 17.28 I I (56.0) (234.8) I l I constant--average -152.16 -l76.39 86.74” I I firms (245.8) (175.6) (35.7) I I constant--large firms -5221.99** -6940.08** I I (1522.0) (1269.7) I tangible assets: 18]" I I small firms (0.3) I I tangible assets: 0.84 0.68** 1.22“ average firms (0.6) (0.2) (0.1) tangible assets: 0.36 I large firms (0.2) I R&D: small firms -2.35** -1.22** -2.58** I (0.6) (0.4) (0.6) I R&D: average 0.73 0.73 0.02 I firms (1.1) (0.9) (0.8) I R&D: large firms 4.70** 4.89** 2.31** I (0.8) (0.4) (0.3) I Advertising: small 4.44M . firms (1.8) Advertising: -1.13 0.47 -2.77** I average firms (2.0) (1.2) (1.1) I Advertising: large -2.02 I firms (3.5) I F 46.94 230.2 197.42 df 11,146 7,150 5,152 I R2 0.76 0.91 0.86 Note: blank cells indicate that the stock response is the same as that to average firms (the coefficient is constrained to be zero). Small firms: <$360m.: Average: $360m.- $1.9b.; large: >$l.9b. SI A. LIA // 121 TABLE 14 PARAMETER STABILITY RESULTS: MACHINERY INDUSTRY 1 2 3 Constant 4284* 42.55M 49. 58** (25.6) (24.1) (23. 7) DS -64.33** -64.07** -70. l 1** (26.4) (25.2) (24. 5) DL -125.60 -125.34 -130. 69 (223.8) (222.1) (226. 5) Assets'rAssets-M 0.50" 0.50** 0. 50** (0.1) (0.1) (0.1) S(Assets,,-Assctsi,..5) 0.24 (0.2) L(Assets,',-Assets,,_5) 0.00 (0.1) 2RD -l.11* -1.11* -l. 10* (0.6) (0.6) (0. 6) SZRD 2.00** 1.94** 2.01 ** (0.7) (0.7) (0.7) LZRD 2.39** 2.39** 2.38** (0.8) (0.8) (0.7) ZADV —0.54 -0.52 - l .29 (1.5) (1.56) (0.9) SZADV 0.10 1.85 (4.8) (4.9) LZADV -0.78 -0.80 (2.1) (2.2) R2 .55 .57 .58 F 9.12 11.56 15.32 df 11,62 9,64 7,66 DS and DL are dummy variables. DS=1 if annual sales are less than $79 million; 0 otherwise. DL = 1 if annual sales are greater than $308 million: 0 otherwise. I Stoc C1 C01 C005 R8 R. Ad F 3» 527/ I MACHINERY INDUSTRY Stock Response to: 1 2 3 constant--small -21.49** -21.52** -21.53** fums (6.5) (7.3) (7.0) constant--average 42.84”“ 42.54“ 49.58” firms (25.6) (1.8) (24.5) constant--large firms -82.79 -82.80 -81.11 (222.3) (222.3) (220.4) tangible assets: O.75** small firms (0.2) tangible assets: 0.50** 0.50** 0.50** average firms (0.1) (0.1) (0.1) tangible assets: 0.50** large firms (01) R&D: small firms 0.84** 0.83** 1.10** . (0.3) (0.4) (0.4) R&D: average -1.11* -1.11* -l.10** firms (0.6) (0.6) (0.6) R&D: large firms l.28** l.28** 1.28" (0.5) (0.5) (0.5) Advertising: small -0.44 1.33 firms (4.6) (4-7) 5 Advertising: average -0.54 -0.52 -1.29 ' firms (1.5) (1.6) (0.9) Advertising: large -1.32 -132 firms (1.4) (1-4) F 9.12 11.56 15.32 (If 1 1,62 9,64 7,66 R2 .55 .57 .58 ‘ote: t . I ce ‘cmlcate at e stoc response 3 e sfle “to ave ' TABLE 15: SUMMARY OF COEFFICIENTS FOR THE 122 Small firms: <$79m.; Average: $79m.-$308m.; large: >$308m. I i . I I rm . 123 TABLE 16 I PARAMETER STABILITY RESULTS: MEDICAL EQUIPMENT I INDUSTRY I 1 2 3 Z constant 985.34“ 985.49“ -041 (294.5) (294.2) (6.7) DS -989.40** -989.30** (294.4) (294.6) . Assets,,-Assets,,_5 0.12 0.12 0.65** (0.2) (0.2) (0.1) I S(AssetsI,-Assets....) 1.16** 1.16** I (0.3) (0.3) i 2RD 1.33** 1.33M 1.06** I (0.3) (0.6) (0.4) . SZRD -1.43* -1.55** -103" i (0.7) (0.6) (0.5) I EADV -0.80** 081 -070 I (1.1) (1.1) (0.5) ' SEADV -0.61 -0.44 I (1.2) (0.8) I . R2 .78 .78 .55 ; I F 67.92 79.86 32.79 I (if 7,124 6.125 5.126 I standard errors are in parentheses. **/* indicates significance at the 5%/10% level, respectively. DS is a dummy variable. DS=1 if annual sales are less than $400 million; 0 otherwise. \\ Sum' CC CODSI Rétl list zidv 1\d\ “and. **/* Non: COCI 124 TABLE 17 I SUMMARY OF COEFFICIENTS FOR SPECIFICATION 2: ; MEDICAL EQUIPMENT INDUSTRY I Stock Response to: 1 2 3 I constant--small -4.12 -3.85 I firms (7.3) (7.4) I constant--large firms 985.34** 985.49”MI -0.41 (294.5) (294.3) (6.7) I tangible assets: 1.2s** l.27** I 1 small firms (0.2) (0.2) I I tangible assets: 0.12 .12 0.65" i I large firms (0.2) (0.2) (0.1) I I R&D: small firms -0.10 -0.22 0.03 I I (0.3) (0.4) (0.3) I I R&D: large firms 133* 1.33** 1.06" I (0.6) (0.6) (0.4) Advertising: small -1.42 -1.l4** I firms (1.4) (0.6) ' Advertising: large -O.80 -0.81 -044 I firms (1.2) (1.1) (0-8) I I F 67.92 79.87 75.53 df 7.124 6 125 5.126 R2 ; __i ‘..__n,, ..i m. _ _ standard errors are given in parentheses. **/* indicates significance at the 5%/10% level, respectively. Note: blank cells indicate that the stock response is the same as that to large firms (the coefficient is constrained to be zero). Small firms: <$400m.; large firms: >$400m. PAR SIA lf' sun **/1 DS 1 2 3 Constant 244.09 49.72 39.48 (1 163.1) (44.4) (47.0) DS -1 13.06 (1 163.6) AssetsIII-Assetsm 0.79 0.91 0.39“ (1.02) (0.8) (0.2) S(Assetsm-Assetsi_,,5) -O.38 -O.S4 (1.0) (0.8) 2RD 1.16 1.03 1.44“ (1.0) (0.8) (0.6) SZRD 3.09** 5.28“ 4.97** (1.3) (1.2) (1.0) ZADV 0.07 0.07 0.13 (0.8) (0.7) (0.7) SZADV -3.87** -2.15 -2.21 (1.1) (1.4) (1.5) standard errors are in parentheses. **/* indicates significance at the 5%l10% level, respectively. DS is a dummy variable. DS=1 if annual sales are less than $700 million; 0 otherwise. ll 126 TABLE 19 SUMMARY OF COEFFICIENTS FOR SPECIFICATION 2: PHARMACEUTICAL INDUSTRY Stock Response to: 1 constant--small 1 13.03" firms (1 1.7) constant--large firms 244.09 (1 163.1) tangible assets: 041’” small firms (0.2) tangible assets: 0.79 large firms (1.0) R&D: small firms 4.25** (0.9) R&D: large firms 1.16 (1.0) Advertising: small -3.80** firms (03) Advertising: large 0.07 (0.8) 98.03 standard errors are given in parentheses. **/* indicates significance at the 5%l10% level, respectively. Note: blank cells indicate that the stock response is the same as that to large firms (the coefficient is constrained to be zero). Small firms: <$700m.; large firms: >$700m. ll II 818 *t 127 I TABLE 20 I PARAMETER STABILITY RESULTS: SEMICONDUCTOR INDUSTRY I I 1 2 3 I constant -187.91** 405.35" —50.06** I I (108.9) (5.3) (5.9) I I I I DS 183.47 43.21 | I (480.0) (35.6) I I Assam-Assess: 0.46** 1.84" 069*... I I (0.1) (0.2) (0.1) I I S(Assets,I,-Assets.III5) -8E—3 I I (27.8* I I RD -122 -O.68** -0.32** I = (45.6) (0.1) (0.1) I I szRD 0.74 3.96* -1.76 I I (0.7) (2.4) (3.34) I I ZADV 9.77 -1.81 2.49** I I (60.2) (1.5) (1.4) . I SZADV 3.50 17.49 16.54 I I (3.0) (41.0) (42.84) I I R2 .80 .73 .87 I I F 54.07 50.89 102.85 I I df 7 84 5,86 6,85 . standard errors are in parentheses. **/* indicates significance at the 5%l10% level, respectively. DS is a dummy variable. DS=1 if annual sales are less than $40 million: 0 Otherwise. __ __ _. _ _______ LAHLAI TABLE 21 I SUMMARY OF COEFFICIENTS FOR SPECIFICATION 2: SEMICONDUCTOR INDUSTRY Stock Response to: 1 2 3 I constant--small 4.44 —6.84 fums (467.5) (35.2) constant--large firms -l87.9* -105.35** -50.10** (108.9) (5.3) (5.9) tangible assets: 0.45** I small firms (27.8) tangible assets: 0.46** 1.84** 069" large firms (0.1) (0.2) (0.1) R&D: small firms -l.96 3.35 -2.08 (45.5) (2.7) (3.4) I R&D: large firms -0.74 -0.68** -0.32** I (0.7) (0.1) (0.1) I Advertising: small 13.27 16.32 19.03 I firms (602.9) (45.4) (42.8) Advertising: large 3.50 -202 2.49** I firms (3.0) (1.6) (1.4) I F 54.07 50.89 102.85 df 7,84 5,86 6.85 R2 .80 .73 .87 Standard errors are given in parentheses. **/* indicates significance at the 5%l10% level, respectively. Note: blank cells indicate that the stock response is the same as that to large firms (the coefficient is constrained to be zero). Small firms: <$40m.; large firms: >$40m. 129 CUSLM TEST STATISTIC 0 20 40 60 80 100 120 140 150 lIEIRVAUONIIIEER FIGURE 22: CUSUM TEST--COMPUTER INDUSTRY L2 5’ an 5’ p 7' CUSW OF SQUARES TEST STATISTIC O O L: ha 0 O 20 40 60 80 100 120 140 160 llfiIRVAHONIIIIER FIGURE 23: CUSUM OF SQUARES TEST-- COMPUTER INDUSTRY 1Q30 -501: -I -570 -I -530 .I .330 -1 -.oo .4 -I10 -1 .030 -I an -I m 'I L” le‘llm htlo -IS° -I -0011 -I I 470 I ”(I .In (.l-) FIGURE 24: QUANDT’S LOG LIKELIHOOD RATIO TEST-COMPUTER INDUSTRY 1 (II I d» J. o STOCK RESPCNSE TO R&D NVESTIENT s 0 1000 2000 3000 4000 5000 6000 7000 8000 900010000 m 512: (SALES N muons 3) FIGURE 25: THE STOCK RESPONSE TO R&D INVESTMENT--COMPUTER INDUSTRY 1.31 I ‘I H ll 11 II I A g L L A A STOCK RESPONSE TO R&D NVESDENT .0 500 1000 1500 2000 2500 3000 3500 4000 4500 5000 FIN 5122 (SALES N “LENS 3) FIGURE 26: THE STOCK RESPONSE TO R&D MESTMENT-COMPUTER INDUSTRY (EXCLUDING FIRMS WITH SALES > $20 BILLION) 132 15 GJSIM TEST STATISTIC o -10 -15 0 10 20 30 40 50 60 70 llEERVAflONlflIIER FIGURE 27: CUSUM TEST--MACI-IINERY INDUSTRY ‘ -. La 5’ a O C N CUSIM OF SQUARES TEST STATISTK: .0 .0 ‘ CI 0 0 10 20 30 40 50 60 70 80 (IEIRVAUONIIIIER FIGURE 28: CUSUM OF SQUARES TEST-- MACHINERY INDUSTRY 133 a. 'r r!» 3 s s 8 '- Oumdt': Log likclillood Rollo '0 O I I l V N - O O O . -720 Firm Siz- (Sdcs) FIGURE 29: QUANDT'S LOG LIKELIHOOD RATIO TEST--MACHINBRY INDUSTRY I O -‘ :4 M-‘UINU' l a a I I J. - 1 UI .. STOCK RESPGISE TO R&D NVESTIENT 1 1° 1 U N o 500 1000 1500 2000 2500 mu s12: (SALES 01 muons 3) FIGURE 30: THE STOCK RESPONSE TO R&D INVESTMENT--MACHINERY INDUSTRY 134 20 15 10 CUSlM TEST STATISTIC -15 -20 20 30 4O 50 60 7O 80 90 100 110 120 (XEIRVAUONIMAIER FIGURE 31: CUSUM TEST--MEDICAL EQUIPMENT INDUSTRY d N 8 g : U! z; 0.3 E ‘“ 0.5 3 '6 g 0.2 ° 0 0 20 40 so 30 100 120 (IEIRVADONDIIEER FIGURE 32: CUSUM OF SQUARES TEST--MEDICAL EQUIPMENT INDUSTRY 135 -600 d 0‘10 -t -420 . -430 .1 -440 4 -430 -1 -4” .1 Gum‘s Loo leolmooo title -470 -! i -‘90 rTw vvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvvv Fl?! III. Chi”) FIGURE 33: QUANDT’S LOG LIKELIHOOD RATIO TEST--MEDICAL EQUIPMENT INDUSTRY 3 'i a! 12.5 4‘ 1‘ ’I" ‘ ”¢’ OL” #‘i* g 2 1.5 53 dI-II~ ill—‘* II: II 1|» II£I °‘ 1 (3 .— 3: (J 5 (J 95 moo—I 34.5 it £3 -1 Irltuw-In-n+-—qk m -1.5 0 200 400 600 800 1000 1200 1400 1600 1800 2000 FIRMSIZE(SAL£SINH.UONS$) FIGURE 34: THE STOCK RESPONSE TO R&D INVESTMENT--MEDICAL EQUIPMENT INDUSTRY 136 CUSLM TEST STATISTIC O ‘1 "fi 10 20 30 40 50 60 70 BO (IEERVANONImAIER FIGURE 35: CUSUM TEST--PHARMACEUTICAL INDUSTRY CUSIM OF SQUARES TEST STATISTK: o a: 10 20 30 40 50 60 70 80 (IEIRVAUONIIIIIR FIGURE 36: CUSUM OF SQUARES TEST-- PHARMACEUTICAL INDUSTRY 137 -314 -310 d dd. -1 -m .1 an -4 .m .. -m d -33. J -)fl -1 -m -4 -3” q M's Loo “hum htlo 3...} 'TT'TTTTTTTTIITTIIU'UTTIT‘TTUTI IIIIIIII Hr- In. (hi-u) FIGURE 37: QUANDT’S LOG LIKELIHOOD RATIO TEST--PHARMACEUTICAL INDUSTRY 9 E'“ 7 g5“ 35 24“ 3 I; I I A A I I I L I 1 b—Q—h—u—c—c—c—H—O—C g0 :: : : .1; :: : = : : : 4 0 1000 2000 3000 4000 5000 0000 7000 0000 000010000 MSIZEGMLBNWT) FIGURE 38: THE STOCK RESPONSE TO R&D INVESTMENT--PHARMACEUTICAL INDUSTRY 138 CUSW TEST STATISTIC o 0 10 20 30 40 50 60 70 00 (IEIRVAHONIIIEER FIGURE 39: CUSUM TEST--SEMICONDUCI OR INDUSTRY cu. —- L: O O a F’ N CUSIM OF SQUARES TEST STATISTK: O O 3. b- O 0 10 20 30 40 50 60 70 80 (IEIRVAHONIIIEER FIGURE 40: CUSUM OF SQUARES TEST-- SEMICONDUCTOR INDUSTRY 139 D. SUMMARY OF RESULTS In order to compare the results from Specification 1 and 2, the stock response to R&D must be calculated for Specification 1. As explained in Chapter 4, Specification 1 tests the Schumpeter hypothesis by analyzing rates of change in market valuation since it is easier to see when the market valuation of R&D changes sign. The stock response to R&D investment in Specification 1 is calculated as follows: Stock Response = yz+y3Sales+y4Sales2 ----------- (33) The standard error of the stock response is: Warm) +SaleszVar(v,) +3016“ Varm) + (34) 2COV(12,Y3) +ZSales3COV(73.w + aleszCMYZJJI‘” The actual ranges in which the market positively values R&D investment and negatively values R&D investment in Specification 1 are not expected to be identical to the range(s) found in Specification 2 since the estimation process in which these were derived were not the same. Specification 1 assumes a continuous functional form whereas Specification 2 statistically tests for different size regimes. However, in checking for robustness of the 140 results for the two specifications, the qualitative pattern of the stock response is expected to be similar. That is, suppose that the stock response to R&D in Specification 1 is not significantly different from zero for small firms, positive and significantly different from zero for medium size firms and then not significantly different from zero for large firms. Then the stock response to R&D is expected to be higher in medium size firms compared to small or large size firms in specification 2 as well. However, the "definition" of small, medium, and large (that is, the actual range in which the stock response to R&D investment is positive, negative, or not significantly different from zero) may be different from that in Specification 1. Since Specification 2 statistically tests for firm size regimes. I put more weight on these results as compared to the results obtained in Specification 1. Table 22 summarizes the results of Specification 1 and 2 for each of the five high- technology industries analyzed. It indicates the range of firm size for which the market values R&D investment positively, negatively, or zero, respectively. In the computer industry, results are consistent across specifications: the stock response to R&D investment is not significantly different from zero for small firms. positive and significant for medium size firms, and for large firms, the market does not positively value R&D investment. As expected, the definitions of small, medium, and large differ across specifications. As mentioned previously, the effect of the differences in firm size regimes is most likely exacerbated by the very large firms in the sample. When the largest firms in the sample are excluded (firms with annual sales greater than $20 billion) the results indicate that market does not value R&D investment positively for smaller size firms while for larger firms the stock response is positive and significant. 141 m TABLE 22: SUMMARY OF RESULTS FROM SPECIFICATION 1 AND 2 Firm Size in which the Stock Response to R&D Investment is: Positive and Significant Negative and Not Significantly Significant Different from zero Spec. 1 Spec. 2 Spec. 1 Spec. 2 Spec. 1 Spec. 2 Comp $3.5b. to $180m. to --- >$l.9b. 0 to $3b. <$180m. $35b. $1.9b. & >$35b. Comp* $10.5b. to >$l.9b. --- <$390m. O to $390m. to $12.5b. $10.5b. & >12.5b. $1.9b. Mach <$160m. <$79m. & $1.1b.to $79m. to $160m. to --- >$308m. $3.3b. $308m. $1.1b. & >$3.3b. Medi $250m. to >$400m. --- --- <$250m. <$400m. $1.55b & >$l .55b Phar <$3 b. <$700m. --- --- >$3b. >$700m. Semi --- --- --- >$40m. entire <$40m. range E Comp: Computer Industry Comp*: Computer Industry, excluding firms with sales greater than $20 billion. Mach: Machinery Industry Medi: Medical Equipment Industry Phar: Pharmaceutical Industry Semi: Semiconductor Industry 142 Specification 1 also finds that for the largest firms in the sample (greater than $12.5 billion in annual sales) the stock response to R&D investment is not significantly different from zero. Quandt’s log likelihood ratio test would not have been able to detect a structural change at this point since there are not enough degrees of freedom. In the machinery industry the results from both Specifications indicate that the stock response to R&D is positive and significantly different from zero for small firms. The definition of small is less than $160 million in Specification 1 whereas Specification 2 determines that small is less than $79 million. Both Specifications indicate that the stock response to R&D in medium size firms is not valued positively by the market. In Specification 2, the stock response to R&D is not significantly different from zero for firms with sales between $160 million and $1.1 billion, and the stock response is negative and significant for firms with sales between $1.1 billion and $3.3 billion in Specification 1. In Specification 2, the market values R&D negatively for firms with sales between $79 million and $308 million.20 The stock response to R&D in large firms is not significantly different from zero in Specification 1, whereas it is positive and significant in Specification 2. Thus, the qualitative pattern of the stock response to R&D is similar in Specification 1 to that in Specification 2: it is positive and significant for small firms, negative and significant for medium size firms, and then the stock response to R&D investment increases again for larger firms. In the medical equipment industry both Specification 1 and Specification 2 find that larger firms are more profitable in R&D than smaller firms. In Specification 1 the 20Although the difference between $308 million and $1.1 billion appears rather large, there are only 4 observations (5% of the sample) in this category. 143 definition of small is less than $250 million, whereas in Specification 2 the definition of small firms is less than $400 million. These results are actually quite similar since there are only 4 observations (or 3% of the sample) in which sales are between $250 million and $400 million. Whereas Specification 2 finds that the stock response to R&D for firms with annual sales greater than $400 million is positive and significant, Specification 1 finds that it is positive and significant up to $1.55 billion in annual sales and after that it is not significantly different from zero. Again, in order to perform Quandt’s log likelihood ratio test, there were not enough degrees of freedom available to test for a point of structural change for these firms. This may explain the finding that the stock response to R&D in the largest firms is not significantly different from zero in Specification 1. The results are consistent across Specifications in the pharmaceutical industry in that small firms are more profitable in R&D than larger firms. But, the definition of small and large are different. In Specification 1, small is defined to be less than $3 billion and in Specification 2 it is less than $700 million. Again, the discrepancies in the firm size can be attributable to the assumption of a continuous relationship between firm size and the stock response to R&D investment in specification 1. The general pattern of the market’s valuation is similar: positive and significant for small firms and not significantly different from zero for larger firms. In the semiconductor industry, Specification 1 found that the stock response to R&D is not significantly different from zero for the entire range of firm size. Specification 2 is able to test for different size regimes and finds that for firms with sales less than $40 million, the stock response to R&D is not significantly different from zero / .— 144 (30% of the sample), whereas for larger firms, the stock response is negatively valued by the market. V. SUMMARY AND CONCLUSIONS Ever since U.S. industry began feeling the sting of successful foreign competition, there have been calls for more cooperation of research efforts of private industry. One of the recurring arguments in the ensuing policy discussion has been that domestic enterprises cannot muster enough resources to maintain technological leadership in an increasingly global marketplace. This view has also been voiced in other industrialized countries, such as France, Germany, and Japan (Scherer (1992)). In order to encourage a greater amount and greater efficiency of R&D investment, Congress passed the National Cooperative Research Act in 1984. It institutes a "rule of reason" for evaluating the legality of cooperative R&D ventures on a case-by—case basis, rather than declaring them per se illegal. In addition, this act limits the potential liability of such consortia to actual, rather than treble damages. Policies like these are referred to as Schumpeterian, in reference to Joseph Schumpeter. Schumpeter postulated that there exists a positive relationship between firm size and technological progress, and that static economic efficiency should be foregone in order to achieve long-run technological progress. Proponents of that view stress the importance of complementary activities which make it easier for large firms to turn successful research into successful products and processes. On the other hand, opponents of Schumpeter stress that small firms may have an advantage in R&D due to 145 146 diseconomies of firm size and greater motivation of .scientists and engineers in smaller frrrns. This study tests whether firm size is a significant determinant of R&D profitability. It estimates a stock market valuation model for five high-technology industries: computer, machinery, medical equipment, pharmaceutical, and semiconductor. However, it is not able to determine why large firms are, or are not, relatively more profitable in R&D. It offers a possible explanation for differences in large and small firms’ R&D profitability rates in terms of industry conditions. I first estimate a market valuation model of the firm assuming that there exists a continuous relationship between firm size and R&D profitability. In testing the Schumpeter hypothesis, the mixed partial derivative of the change in market valuation with respect to both firm size (sales) and the stock of R&D capital is analyzed. It indicates the effect of a one dollar increase in sales, holding the stock of R&D constant, on the change in the market value of the firm. I evaluate how this mixed partial derivative changes as firm size changes. Results indicate that Schumpeter is supported only in the computer and medical equipment industry. As firm size increases the change in market valuation with respect to both firm size and R&D profitability increases in both of these industries. In the semiconductor industry I find that firm size is not a significant determinant of R&D profitability across the relevant range of firm size. Schumpeter is not supported in both the machinery industry and the pharmaceutical industry; size has a negative effect on R&D profitability. One possible reason for lack of support of Schumpeter is the assumption of a continuous relationship between firm size and R&D profitability. In fact, Markham 147 (1965) first suggested that Schumpeter’s theory might better be viewed as a threshold theory because technological progress is not necessarily an increasing and continuous fimction of market power or business size. The second part of the empirical analysis allows for such a threshold effect. I test for parameter stability of the estimated coefficients by using the CUSUM and Cusum of Squares statistical test (Brown, Durbin, Evans (1975)). I find that parameter stability can be rejected for all five industries implying that the coefficients are not stable over firm size. Unfortunately, the CUSUM and Cusum of Squares tests are not able to indicate either the specific point(s) of structural change or the unstable parameter(s). In order to determine the point(s) of structural change, Quandt’s log likelihood ratio test was applied. The thus obtained point(s) of structural change are used to divide firms into different groups and dummy size interaction terms are created for each group. Schumpeter’s hypothesis was tested by comparing the stock response to R&D over the different size classes. If the stock response to R&D is found to be significantly greater in larger firms than that in smaller firms, Schumpeter can be supported. On the other hand, if the stock response to R&D in small firms is significantly greater than that in smaller firms, or if there is not a significant difference in the stock response to R&D investment for different size classes, Schumpeter cannot be supported. These results are consistent with the first part of the empirical analysis that assumes a continuous relationship between firm size and R&D profitability. I still find that Schumpeter is supported in just two of the five industries. I place more confidence in the results from the second type of analysis because it does not assume a continuous relationship between firm size and R&D profitability. In addition, by allowing for 148 endogenously determined thresholds, this methodology is able to determine the definition of large and small firm size. The Schumpeter hypothesis is supported both in the computer and medical equipment industry; large firms are more profitable in R&D compared with their smaller counterparts.‘ However, the "definition" of large firms is not homogeneous across the different industries. For example, "large" firms were found to be greater than $1.9 billion in annual sales in the computer industry and greater than $400 million in the medical equipment industry. Both small and large firms are more profitable in R&D than medium size firms in the machinery industry which implies that Schumpeter is not supported. In the pharmaceutical and semiconductor industries Schumpeter is not supported as well; large firms are less profitable in R&D then their smaller counterparts. Small firms are found to be firms with annual sales less than $700 million in the pharmaceutical industry and firms with annual sales less than $40 million in the semiconductor industry. One potential reason for differences to exist between large and small firm’s R&D profitability rates is the importance of complementary assets. In industries in which patents or copyright protection is highly effective in appropriating the returns from R&D investment, the innovator is almost assured of protection from imitators. Even if the innovator does not have the required amount of complementary assets, a patent protects the innovator’s intellectual property and gives the innovator time to acquire these assets either by contracting out with other firms or by acquiring the assets. In such industries it is ofien the case that smaller firms that are more profitable in R&D since size provides ' The Schumpeter hypothesis is supported once firms with annual sales greater than $20 billion are excluded from the sample. 149 no advantage in terms of marketing or sales. In fact, size may prove to be a hinderance due to decisions having to pass through layers of management as well as scientists becoming less motivated due to administrative and hierarchical constraints. However, industries in which patents are highly effective in appropriating returns are the exception rather than the rule. In these industries, a patent provides little or no protection in appropriating the returns to R&D investment, for example, because it is fairly easy to reverse engineer a product. In fact, Levin et al.(1987) found that secrecy, lead time, and sales and service efforts were often times more important than patents as a means of appropriating the returns to R&D investment. This result implies that "success" in innovation is not enough for commercial success. The simultaneous model of innovation used in this study (see Chapter 3) incorporates both technological success and the ability to make use of complementary assets as determinants of commercial success of R&D efforts. Accordingly, innovators must turn to business strategy to protect innovative rewards from imitators. Therefore, if patents are not effective in appropriating the returns from R&D investment, the firm needs other complementary assets to protect its investment. Large firms are more likely to possess these specialized assets which may better allow them to take advantage of their technology. Small firms are less likely to have incorporated relevant specialized assets within their organizational boundaries and so will either have to incur the expense of trying to build them, or of trying to develop coalitions with competitors/owners of the specialized assets, and may put them at a disadvantage. The results of the empirical analysis can be explained by applying the concept of appropriability conditions and complementary assets to the five high-technology industries. LL 150 In both the computer and medical equipment industry, complementary activities such as sales, service efforts, and distribution are especially important. Since the core technology is fairly easy to imitate, commercial success is conditional upon the terms and conditions upon which the required complementary assets can be accessed. These complementary assets give large firms advantages over smaller firms, which seems to suggest why large firms were found to be more profitable in R&D relative to their smaller counterparts (see Chapter 3). For example, in the case of cardiac pacemakers, the technology was easy to imitate and so competitive outcomes quickly came to be determined by who had easiest access to complementary assets--in this case specialized marketing (see Teece (1986)). The small pocket calculator, introduced by Bowmar, serves as another example. It was fairly easy to innovate and Bowmar was not able to withstand competition from Texas Instruments, Hewlett Packard and others, and went out of business. Had Bowmar relied more on business strategy (sales and marketing), it may have had a better chance. An example of an industry in which patent effectiveness is high is the pharmaceutical industry. In this industry small firms were found to be more profitable in R&D then large firms. In the pharmaceutical industry patents deny imitators access to the relevant knowledge and therefore the innovator is almost assured of translating its innovation into market value. Marketing is important in the pharmaceutical industry, however, if the innovating company does not possess the desirable endowment of complementary assets, protection of intellectual property will afford it time to access these assets either through a contractual relation or through building those assets in-house. For example, the first co-promotion in the industry occurred in 1981 when Glaxo did not have the necessary sales force to market its anti-ulcer drug Zantac. Glaxo signed an agreement 151 with Hoffrnann-La Roche in which their salespersons would market and sell the drug (Economist (Feb., 1989)). Although one might expect large firms to have an advantage in the semiconductor industry because distribution channels are extremely important and semiconductor chips are fairly easy to reverse engineer, the results indicate that large firms actually earn a negative return on their R&D investment. However, because of the rapid changes that took place during this decade, the 805 represent a special case. Many small firms were started by former employees of the larger semiconductor companies. A large percentage of these entrepreneurs had left their previous jobs because of frustrations over not having their ideas approved and/or not being able to reap the benefits from their innovative efforts. Since these employees took their research ideas from the larger firms, this would explain the result that larger firms actually earned a negative return on their R&D investment, whereas smaller firms did not. Since it is fairly easy to reverse engineer products in the machinery indusz, appropriability conditions are below average. Large firms have built up strong marketing departments and distribution channels are also important for commercial success. According to the above argument, one would expect Schumpeter to be supported, and therefore large firms to be more profitable than their smaller counterparts. Yet results indicate that both small and large firms are more profitable in R&D compared to medium size firms. One potential explanation for this finding is that small firms have tended to serve the niche markets, sometimes even supplying the larger firms. This study represents a start in addressing a question recently posed by Scherer (1992): "... at what kinds of innovative activities do large organizations have (the) comparative advantage? Comparative disadvantage?" First, it finds that striking 152 differences exist in the relationship between firm size and R&D profitability for a group of high-technology industries. In some, Schumpeter is supported whereas in others, the Schumpeter hypothesis cannot be supported. The reasons for this may be attributable to appropriability conditions and complementary assets. This suggests that the Schumpeter hypothesis needs to be qualified in the sense that it must take into account an industry’s appropriability conditions. Second, the defrrrition of what constitutes large and small firms is found to differ across the industries studied. Previous studies have not accounted for this which might explain why they have not found that firm size is a significant determinant of technological progress. These results represent a contribution to the V ongoing debate on how to design policies to foster R&D efforts. Even in thehigh- technology sector of the economy it seems that Schumpeter’s general hypothesis needs to be qualified by closely studying industry specific conditions. Further research on the relationship between appropriability conditions and large and small firms’ R&D profitability will clarify the type of relationship between appropriability conditions, firm size and R&D profitability. It is conceivable that the observation of higher R&D profitability for small firms in some of the high-technology industries analyzed was influenced by the length of the time period analyzed. Small firms might use their resources more effectively over a period of approximately five years, however, it need not be the case that such a relationship continues to uphold over a longer period of time. They might not follow through with their efforts in continuously investing in development work (Ferguson (1988)). Further research needs to be done on the relationship of innovation and firm size over the long run. For example, it would be 153 interesting to know how results within and across industries differ if a longer time period, such as a twenty year interval, is analyzed. APPENDICES 4 ' ‘F.Vo-T) APPENDIX A APPENDIX A DEFLATION OF MARKET VALUATION The change in the market value of the firm is deflated by the change in the Standard and Poor’s 500 Composite index. This allows for the separation of increases if in market value that are due to overall increases in the general level of the market. compared with increases due to tangible and intangible asset investment. Data are obtained from the Dow Jones-Irwin Business and Investment Almanac (1991). Since first differences are computed, I deflate the difference by the change in the index in the current period compared to the change during a given base period. For example, I use three different time periods: 1982-1987, 1983-1988, and 1984-1989. First, the change in the Standard and Poor’s Industrial Average is determined for each period, then I divide this by the change in the index during the base period: 1982-1987. [ TABLE 23: STANDARD AND POOR’S COMPOSITE INDEX Year: 1982 1983 1984 1985 1986 1987 1988 1989 I S&P: 158 186 186 234 270 286 320 398 l 1 AS&P(,987,,982)=128 AS&P(,,,,,,,,,=134 AS&P,,,,,,,,,,=212 154 155 Thus, I deflate by 1.047 for 1983-1988, and 1.682 for 1984-1989.1 These weights indicate that the change in the Standard and Poor’s index was 4.7% greater during the time period 1983-1988 than from 1982 to 1987. Similarly, the change in the Standard and Poor’s index was 68.2% greater during the time period 1984 to 1989 than from 1982 to 1987. ' Two additional general stock market indexes are available: the S&P 500 average of daily closing price, and the Dow Jones Industrial Average. Similar regression results are obtained when using the S&P average of daily closing price. On the other hand, in some cases, results obtained using the Dow Jones Industrial Index are slightly different from those employing the Standard and Poor’s Index. I attribute this to the small number of companies included on the Dow Jones Index. Since my data set includes a large number of small firms, I utilize the broader index: the S&P 500. Industry market indexes for each industry are also available. However, these indexes contain only a small number of companies. For example, in the semiconductor industry, only 5 firms are included, while just 15 are included in the computer industry index. Furthermore, the firms included are very large firms, which may not be representative of overall industry performance. 1 asapumwm, = 1.047; AS&P,,,,,,,,,,_ = 1.682 AS &P( 1987-1982) AS&P(1987-1982) APPENDIX B APPENDIX B TEST OF THE EQUALITY OF COEFFICIENTS OF THE FIVE HIGH-TECHNOLOGY INDUSTRIES The null hypothesis of constant coefficients for the five industries is tested using three models. The basic model includes only three independent variables: the change in tangible assets over a five year period (Assetg',-Assets,,_5), and the sum of R&D and advertising expenditures over a five year period (2RD and ZADV, respectively). The second model is equivalent to Specification 1; independent variables include sales- intangible asset interaction terms (Sales*ZRD, Sales*2ADV, Sales2*2RD, Salesz*ZADV). And the third model is similar to Specification 3(which is presented in Appendix 4); the difference being that in this case. all industries have identical break points. That is, firms are divided into 3 equal size groups and dummy size-interaction terms are created. On the other hand, in Specification 3 I separate firms into 3 equal groups for each industry, therefore the break points differ across industry. For example, here, since industries are pooled, the computer industry has a large number of firms categorized as "small", with only a few "large" firms, whereas in the pharmaceutical industry, most firms are categorized as "large". 156 III 41.1111 1|IIII 157 Table 24 presents results for the basic model. Not surprisingly, the change in tangible assets is the primary determinant of the change in the market value of the firm. Overall, R&D expenditures are a significant determinant of the change in a firm’s value. Only in the computer industry is this coefficient not significant. Pooling industries, the stock response to R&D investment is l.09(s=0.14), yet this varies significantly across industries. For example, in the pharmaceutical industry, the stock response to R&D investment is 4.18(s=0.49), while in the semiconductor industry this value is - 0.67(s=0.15). The negative coefficient indicates that, in general, R&D investment was not profitable during this time period. The stock response to advertising expenditures is positive and significantly different from zero when industries are pooled: 1.27 (s=0.13). Yet, when each industry is examined separately, only in the computer and semiconductor industry is the coefficient on advertising expenditures significant. The hypothesis of constant coefficients across the five industries is rejected at the 1% significance level (F=150.78 Fc=2.37). The regression results for model 2 (Sales-intangible assets interaction terms) is shown below. Separate results for each industry are presented in Table 11, see Chapter 4. Again, the hypothesis of constant coefficients across the industries is rejected at the 1% significance level (F=80.53 r=,,=1.94).2 2 Pooling all data, the stock response to R&D increases as firm size increases. it attains a maximum at approximately $5 billion in annual sales. then it begins to decrease. The stock response to R&D is not significantly different from zero for fums with sales less than $150 million. For the average firm size of $830 million, it is .46(s=.l4), while for the largest firm size of $22 billion. it is negative and significantly different from zero. EA 3D. flutter.- .s. - 158 TABLE 24: BASIC MODEL 1 constant AAssetsmM 2RD ZADV N R2 F df All -5171 0.65** 1.09** 1.27** 503 .61 388.38 1 (77.12) (0.06) (0.14) (0.13) 3,499 . - Comp -78.68 1.22** 0.13 1.26" 164 .82 44.09 1 ' (80.52) (0.06) (0.12) (0.37) 3,160 | 1 Mach -3779 0.54** 1.09" -1.26 73 .58 38.98 1 3 (26.66) (0.06) (0.24) (0.98) 3,75 ; lMedi 121.44** 0.49" 1.20** -0.60 111 .80 280.12 1 . (68.19) (0.05) (0.29) (0.42) 3,107 1 1 Phar -138.5** 0.64** 4.18" 0.19 79 .69 77.23 i 1 (35.0) (0.27) (0.49) (0.64) 3,75 1 1 Semi -88.34* 1.13" -O.67** 4.65“ 79 .74 76.90 1 1 (52.22) (0.12) (0.15) (1.43) 3,75 1 Standard errors are in parentheses. **/*statistically significant at the 5%l10% level, respectively. All: All 5 high-technology industries, pooled; Comp: Computer industry, including IBM; Mach: Machinery industry; Medi: Medical equipment industry; Phar: Pharmaceutical industry; Semi: Semiconductor industry; Assetsimgz Change in the book value of ZRD/ZAdv: Sum of R&D/advertising tangible assets over a five year period; expenditures over a 5 year period, respectively. F .54. a. . MP. Friarfl 159 POOLED RESULTS: MODEL TWO vu-VM = -5353 +.85AAssets +.292RD (13.98)** (0.19)** (0.05)** + 0003211652141) -.2E-7SaleSZZRD +3.042An‘v (.000006)** (0.4E-8)** (0.51)** - 0.00098a1esZADV +.8E-7 88.168sz 4» 51.: (.0002)** (0215-7)M N=502 F = 3428.43 R’=O.76 df = 7,497 standard errors are in parentheses. ** indicates significance at the 5% level. POOLED RESULTS: MODEL THREE V1.1.vi,t-5 = '0.82 '1’ 1.15MSSCIS '1' 0.19% (0.57) (0.12)** (0.67) + 1.64 13912110,, - 1.51 DL*2RD - 1.852ADV (0.99)M (0.75)** (2.61) - 2.37 135*):on +4.47 DL*Z‘.ADV + 8... (4.99) (2.64)* N=502 ‘ F = 251.21 R2=0.68 df = 7,497 standard errors are in parentheses. **/* indicates significance at the 5%l10% level, respectively. 160 The third model separates firms into three equal groups based on firm size.3 The break points are $29.40 million and $195.22 million in annual sales, with each group containing either 167 or 168 observations (see results above). For the smallest size category, the stock response to R&D is l.83(s=.19), while for the largest size category it is -l.32(s=.42). On average, the stock response to R&D is not significantly different from zero.‘ The null hypothesis of constant coefficients across all five high-technology industries is again rejected at the 99% significance level (F=73.36, Fc=l.94). Since all three models reject the hypothesis of constant coefficients over the five industries, each industry is analyzed separately. 3 Results exclude International Business Machines. ‘ At first glance. this appears to refute the previous model. However. in this model we are defining large to be firms with sales greater than $180 million, which according to the previous analysis. combines firms in which the stock response to R&D is positive and significant. negative and significant, and not significantly different from zero. Therefore. while the two models may be difficult to reconcile because of the differences in break points, one does not refute the other. The stock response to advertising expenditures is not significantly different from zero for the smallest and average size categories, while it is positive and significantly different for the largest'category. For the smallest size category, the stock response to advertising expenditures is -4.22(s=1.98), while for the largest size category it is 2.62(s=1.07). APPENDIX C APPENDIX C HETEROSKEDASTICITY CORRECTION I assume an additive form of heteroskedasticity in which the estimated variance of the disturbance (62,) takes the form: 01 = a + b(Sa1651) + C(Salesf) (35) where a,b and c are constants to be estimated. In some cases, c=0. To obtain asymptotically efficient estimators of the regression coefficients I perform the following steps: (a) estimate the least squares equations (Specifications 1 and 2). (b) Apply the least squares method to: 912 = a + b(Salesi) + C(Salesf) + v, (36) where e" are the ordinary least squares residuals from the equation estimated in part (a) above. The "first round" estimate of 01,2 is: 161 162 iii a + 5(sa1es,) + etsaresf) (37) (c) These first round estimates of a. b. and c are not asymptotically efficient because vi is heteroskedastic. Therefore the "second round" estimators of a,b,and c are obtained by applying the least squares method to: 2 2 1 . 91 = a 12 +bSa as, + cSalea + W (38) a; a, a: 6: These estimators a,b. and c are asymptotically efficient. Then. the second-round estimator of of is: 012 = a + Esalesi + C(Sales) 12 (39) (d) Asymptotically efficient coefficients of Specification 2 are estimated using 0'1 as weights: 1' M 555938,, c-Assets, , H, 11;, R01: :1 11’ t .1" 5 I ‘ 0 I 7’__0 150 1' 130 1': f 1;: (5‘19”: 30’ 15519-9”): an) (2 ADV...) (4 0 1 4' Y3 M + 1‘____£_ ‘0' Y, H 0": f axlc r 6": + T. 0 - + 77 a + a": 11‘ "‘ The heteroskedasticity correction procedure for Specification 2 is similar to that of Specification 1. APPENDIX D APPENDIX D SPECIFICATION THREE RESULTS Specification 3 entails dividing firms into 3 equal size groups and then creating dummy size-interaction variables. For example, Specification 3: T 1‘ V1: C " V1, c_5 =G1 +62 (3.8861331; C 'ASSGIZS'“ c_5) +a32 RD1' E +¢4D1* 2 R01. C T '1' T T—S T T-S 11519342: RDLC+a62ADVLt+a7DlIkzADVLt+aeD31k£ADVLt+eht T-S r-s r-s T-S (41) where: Vin-Vi“: the change in the market value of the firm over a five year period. Vin-Vim is deflated by the change in a market index over the relevant time period. This allows us to control for general trends in the stock market. See Appendix 1 for details. Assetsm-Assetsizgsz the change in the value of the firms tangible assets over a five year period. XRD/EADV: sum of R&D/advertising expenditures over a five year period, respectively. 163 164 D1*ZRD(and D1*2ADV): Dummy interaction variable. Dl=l if annual sales are less than a specified value. which is different for each industry. D3 *ZRD(and D3 *ZADV): Dummy interaction variable. D3=1 if annual sales are greater than a specified value, which is different for each industry. This specification is similar to Specification 2, however, here break points are a priori assumed instead of statistically determining whether or not they exist. This may lead to the rejection of the hypothesis that different size regimes exist when in fact size regimes do exist.’ For medium size firms, the stock response to R&D is 05. The stock response to R&D for small firms is the sum of the coefficients "0t, + 014"--the coefficients on 2RD, 5 For example, suppose that the break point between "medium" and "large" firms is $1 billion in annual sales according to specification 3 above, but the "true" break point, as found in specification 2, is somewhat larger, say $1.5 billion. Also suppose that the true definition of large firms includes approximately one half of the firms that specification three classifies as large (firms with annual sales between $1 billion and $1.5 billion). Next, assume that the stock response to R&D for the "true" medium firms (sales<$1.5 billion) is not significantly different from zero, while for the "true" large firms it is positive and significant. Then, by combining "true" medium size firms into the "true" large firm size category (as specification 3 does), the hypothesis that the stock response to R&D is positive and significant in large firms may be rejected. Similarly, the hypothesis that the stock response to R&D investment in large firms is not significantly different from that in medium firms when it actually is different. Thus, a significant difference between the stock response in different size regimes may not be found when there actually is a difference. Note that conflicting results may be obtained from specifications 1 and 3 because of their limitations. For example, results from specification 1 could indicate that the stock response to R&D in the mean size firm is positive and significantly different from zero, but because of the classification scheme used in specification 3, results may indicate that the stock response to R&D investment for the mean firm size is not significantly different from zero. Or, because specification 1 assumes a continuous relationship, one could come to the conclusion that medium size firms earn a positive and significant return to R&D, when actually there is a jump between small and large firms returns. This particular functional form is not able to compensate for that particular scenario. 165 + Dl*ZRD,. For large firms. the stock response to R&D is the sum of the coefficients on 2RD, + D3*ZRD,, or 01.3 + 01.5. Just as in Specification 2, a test of the Schumpeter hypothesis entails testing whether the stock response in large firms is significantly different from that in small firms. A similar analysis is performed on the advertising coefficients.6 I now present regression results for each of the 5 industries (see Tables 25 and 26). A. COMPUTER INDUSTRY Firms are divided into three equal groups based on firm size: (1) annual sales<$70 million (2) $70 million 5 annual sales < $400 million, and (3) annual sales 2 $400 million. Dummy interaction variables (D1*2RD, D3*ZRD, D1*2‘.ADV, D3*2‘.ADV) are constructed and the omitted group is the medium size group, category 2. On average, the stock response to R&D is negative and significant, implying that many of these firms experienced unsuccessful projects during the 1980s. Yet, these results vary with the size of the firm. For the largest category of firms (sales > $400 million), the stock response to R&D is 1.13 (s=0.51), whereas the stock response to R&D investment for the smallest group of firms is not significantly different from zero. Stderror (a, +0,) =i/s32 +342 +2Cov(a3, 84) Where s.I is the standard error for the coefficient 7,. "1‘ 166 These results are consistent with the previous specifications--Iarger firms are more profitable R&D. Yet, each of the two specifications use a different definition of large firm size.’ B. MACHINERY INDUSTRY Results from Specification 3 substantiate the previous two specification results in that the stock response to R&D investment for small firms is positive and significant. The break points for the three groups are at $83.5 million and $268 million in annual sales, which are similar to the "true" break points in Specification 2: $79m. and $308m.. The stock response to R&D for both the smallest and the largest group of firms is positive and significantly different from zero, while on average it is negative. For small firms the response is 1.44 (s=.24), while for the largest firms, it is 1.00 (s=0.55). The null hypothesis that the response is equal in both small and large-firms can not be rejected 7 These results are difficult to reconcile with results from specification 2 due to the differences in break points. This example highlights the problems associated with this specification 2: the a priori assumption of particular break points. It appears that this specification combines "true" small firms (sales < $180 million) in the medium size category. Thus results indicate that the stock response to R&D for firms with sales between $70 million and $400 million is negative and significant, when it may be due to those firms with sales less than $180 million. Also, here I combine true medium size firms ($180 million < sales < $1.9 billion) in the large firm category. Thus suggesting that large firms (greater than $400 million), earn a positive and significant return to R&D when it is only due to those firms with sales between $180 million and $1.9 billion--the true "medium" firms. 167 (t=.59). In small firms and large firms. the stock response to advertising expenditures is negative but not significantly different from zero.8 Thus, results from both specifications reveal that the stock market values R&D investment positively in small and large firms. C. MEDICAL EQUIPMENT INDUSTRY Again, firms are split into different categories based on frrm size: (1) sales<$18 million (2) $18 million 5 sales < $100 million, and (3) sales 2 $100 million, in which the excluded category is group two. Results from this specification corroborate previous results. The stock response to R&D in the smallest and medium size category is not significantly different from zero; whereas for the largest size category it is positive and significant (stock response: 1.47 (s=O.6l)). Advertising expenditures do not appear to play a significant role in determining the change in a firm’s market value. In general, for the average firm, the stock response to advertising investment is negative and significantly different from zero. Thus, in all 3 specifications the Schumpeter hypothesis is supported. D. PHARMACEUTICAL INDUSTRY Break points for the firm size categories in are assumed to be $385 million and $2970 million in sales. The stock response to R&D investment is positive and 11The stock response to advertising in small firms is -.38 (s=2.62), while in large firms it is -1.08 (s=2.29). 168 significantly different from zero for all three size regimes. However, there is not a significant difference in the stock response in any of the size regimes. The stock response in small firms is 4.29(s=1.40), while in large firms it is 2.29 (s=0.53). The stock response to advertising is not significantly different from zero for all 3 size regimes.9 Thus, these results highlight the problems with this‘specification: assuming a priori a definition of large and small size regimes. Based on these results, the hypothesis of differences in R&D profitability in different size regimes is rejected, but, based on the CUSUM test and Quandt’s log likelihood ratio test there exist different size regimes. E. SEMICONDUCTOR INDUSTRY The break points for Specification 2 are assumed to be $50 million and $220 million in annual sales. Results suggest that the stock response to R&D in the smallest size category is not significantly different from zero, whereas it is negative in firms with sales larger than $50 million. The stock response in the smallest size category is .43(s=.49), and in the average and largest categories it is -1.54/-0.81(s=.50/.41), respectively. The stock response to advertising expenditures in small firms is positive and significantly different from zero whereas in firms larger than $50 million, it is not significantly different from zero.10 9The stock response to advertising expenditures for small/large firms is -7.8/0.03 (s=6.84/0.32), respectively. loThe stock response to advertising in the smallest/largest size category is 4.68/4.30 (s=2.26/3.75), respectively. On average, the stock response to advertising investment is 2.68(s=4.17). 169 All three specifications indicate that the stock response to R&D in small firms is not significantly different from zero, whereas in larger firms it may be either negative or not significantly different from zero. Although Specification 1 finds that the stock response to R&D decreases as firm size increases, it is found to be insignificant over the relevant range. Specification 3 finds that firms with less than $50 million in annual sales earn a stock response to R&D that is not significantly different from zero, while it is negative and significant in larger firms. Thus, these results are consistent with Specification 2 in that the stock response to R&D in large firms is negative and significant. While the difference between $40 million and $50 million in annual sales (the break point between small and medium size firms in Specification 2) may sound minor based on the magnitudes that we have been discussing, there are 26 observations (32% of the sample) which fall into this category. Therefore by arbitrarily determining break points, a structural change may have only been determined by pure luck. gum” I. . .315". 5'3 _ 170 TABLE 25: THE STOCK RESPONSE TO R&D INVESTMENT SIZE-DUMMY INTERACTION TERMS Computer Machinery Medical Pharma— Semicon- Equipment ceutical ductor constant -1 17.01 ** -l7.52** -2.66 -22.92 -35.02** (18.57) (2.72) (2.22) (30.86) (10.39) “ AAssets 1.24** 0.71" 1.35" 0.98** 1.18" (0.17) (0.13) (0.31) (0.36) (0.14) DIZRD 8.08" 2.52** -0.04 2.49 1.98“ (1.32) (0.43) (1.74) (1.60) (0.62) 2RD -1.26 -1.08** -1.39 1.80* -1.54** (1.25) (0.38) (1.53) (1.00) (0.50) D3ZRD 1.99 2.08“ 2.86 0.49 0.74 (1.33) (0.67) (1.66) (1.01) (0.64) DIZADV 13.1 1* -1 1.85" 2.41 -8.62 2.00 (5.70) (3.02) (3.15) (6.53) (4.72) ZADV -8.1 l 1.47 -0.20 0.82 2.68 (5.09) (1.31) (0.09) (1.44) (4.17) D32ADV 7.43 -2.55 -0.38 -0.79 2.30 (6.19) (2.66) (2.40) (1.48) (5.64) " a 164 73 111 79 79 R2 .69 .61 .57 .69 .57 F 50.2 14.2 21.50 25.2 16.06 7.157 7.66 7,103 7.72 Standard errors are in parentheses **/* Indicates significance at the 5%l10% level, respectively. AAssets: change in the tangible assets of the firm over a 5 year period. DlZRDzdummy variable-Sum of R&D expenditures interaction term. D1£Advzdummy variable-Sum of advertising expenditures interaction term. D32RDzdummy variable-Sum of R&D expenditures interaction term. D32Adv:dummy variable-Sum of advertising expenditures interaction term. 171 SUMMARY OF COEFFICIENTS FROM SPECIFICATION THREE TABLE 26: THE STOCK RESPONSE TO R&D INVESTMENT COMPUTER INDUSTRY (break points: $70 mil. and $400 mil. Stock Response to R&D: small firms: 0.33 s=2.9l large firms: 1.13 s-0.51 Stock Response to Adv.: small firms: -11.81 8:11.10 large firms: -1.62 s=1.8l _—_.—_ ____ ______._—.____ —_._.._—___— _fi MACHINERY INDUSTRY (break points: $83.5 mil. and $268 mil.) Stock Response to R&D: small fums: 1.44 s=0.24 large firms: 1.00 s=0.55 Stock Response to Adv.: small firms: -0.38 s=2.62 large firms: -l.08 s=2.29 MEDICAL EQUIPMENT INDUSTRY (break points: $18 mil. and $100 mil.) Stock Response to R&D: small firms: -l.43 s=l.04 large firms: 2.86 s=0.61 1 Stock Response to Adv.: small firms: 2.21 s=2.11 large fums: -0.58 s=1.19 PHARMACEUTICAL INDUSTRY ' (break points: $385 mil. and $2.97 bil.) 1 1 I Stock Response to R&D: small firms: 4.29 s=l.40 1 large firms: 2.29 s=0.51 1 Stock Response to Adv.: small firms: -7.80 s=6.84 1 large firms: 0.03 s=0.32 1 SEMICONDUCTOR INDUSTRY 1 (break points: $50 mil. and $220 mil.) 1 Stock Response to R&D: small firms: 0.43 s=0.49 1 large firms: -0.81 s=0.4l 1 Stock Response to Adv.: small firms: 4.68 s=2.26 1 large firms: 4.30 s=3.75 LIST OF REFERENCES LIST OF REFERENCES Acs, A.J. and Audretsch, DB. 1987. "Innovation, Market Structure and Firm Size." Review of Economics and Statistics. 71: 567-574. Acs, A.J. and Audretsch, DB. 1988. "Innovation in Large and Small Firms: An Empirical Analysis." American Economic Review. 78: 678-690. Acs, A.J. and Audretsch, DB. 1991. "Innovation and Size at the Firm Level." Southern Economic Journal. 57: 739-744. Arrow, Kenneth J. 1962. "Econonric Welfare and the Allocation of Resources for Invention" in R.R. Nelson (Ed.) The Rate and Direction of Inventive Activity: Economic and Social Factors. Princeton: Princeton University Press. Baesel, Jerome B. December, 1974. "On the Assessment of Risk: Some Further Considerations." Journal of Finance. 1491-1494. Baldwin, W.L. & Scott, J.T. 1987. Market Structure & Technological Change. New York: Harwood Academic. Ben-Zion, Uri. 1984. "The R&D and Investment Decision and Its Relationship to the Firms Market Value: Some Preliminary Results." In Griliches, ed. R&D, Patents, and Productivig. Chicago: University of Chicago Press. Blume, Marshall B. March 1971. "On the Assessment of Risk." Journal of Finance. 1- 10. Bound, John et al. 1984. "Who Does R&D and Who Patents?" In Griliches, ed. R&D, Patents, and Productivity. Chicago: University of Chicago Press. Brown, R.L., Durbin, J., and Evans, J.M. 1975. "Techniques for Testing the Constancy of Regression Relationships Over Time." Journal of the Royal Statistical Socieg (Series B). 149-163. l72 173 Clark K. and Griliches, Zvi. 1984. "Productivity Growth and R&D at the Business 1 Level: Results from the PIMS Data Base." In Griliches, ed. R&D Patents and Productivity. Chicago: University of Chicago Press. Clarkson, Kenneth W. 1977. Intangible Capital and Rates of Return. Washington: American Enterprise Institute. Cockbum and Griliches, Zvi. 1988. "Industry Effects and Appropriability Measures in the Stock Market’s Valuation of R&D and Patents." American Economic Review. Vol. 78(2): 419-423. Cohen, W.M. and Levin, RC. 1989. "Empirical Studies of Innovation and Market Structure." In Schmalensee, R. and Willig, R.D., ed. Handbook of Industrial Organization Vol II. New York: Elseevier Science Publishers. Comanor, W.S. 1967. "Market Structure, Product Differentiation & Industrial Research." Quarterly Journal of Economics. 81: 639-657. Comanor, W. S. 1986. "The Political Economy of the Pharmaceutical Industry. " J__ournal of Economic Literature. 24: 1178-1217. * Connolly, RA. and Hirschey, M. 1984. "R&D, Market Structue, and Profits: A Value- Based Approach." Review of Economics and Statistics. 66: 682-686. Connolly, RA. and Hirschey, M. 1988. "Market Value and Patents." Economic Letters. 27: 83-87. Connolly, RA. and Hirschey, M. 1990. "Firm Size and R&D Effectiveness." Economic Letters. 32: 277-281. Davis, D. 1985. "R&D Consortia." _H_igh Technology. Vol. 5(10): 42-70. Dorfman, N. 1987. Innovation and Market Structure. Cambridge, MA: Ballinger. Dorow, Alan. Feb. 5, 1990. "A Trademark is Forever." Forbes. p. 176. Dun & Bradstreet Information Services. 1991. Dun’s Business Rankings. Parsippany, N.J.: Dun & Bradstreet. Ferguson, Charles H. May-June, 1988. "From the People Who Brought You Vodoo Economics." Harvard Business Review. 88(3): 55-62. Fischer, F .M. and Temin, P. 1973. "Returns to Scale in Research and Development: What Does the Schumpeterian Hypothesis Imply." Journal of Political Economy. 81: 56-70. 174 Fischer, Franklin M. and McGowan, John J. 1983. "On the Misuse of Accounting Rates . of Return to Infer Monopoly Profits." American Economic Review. 73: 82-96. Galbraith, J.K. 1952. American Capitalism: The Concept of Countervailing Power. Boston: Houghton Mifflin. Grabowski, H.G., and Mueller, DC. 1978. "Industrial Research and Development, Intangible Capital Stocks, and F inn Profit Rates." Bell Journal of Economics. 9: 328-343. Griliches, Zvi. 1979. "Issues in Assessing the Contribution of Research and Development to Productivity Growth." Bell Journal of Economics. 10: 92-116. Griliches, Zvi. 1981. "Market-Value, R&D, and Patents." Economic Letters. 7: 183- 187. Griliches, Zvi. 1986. "Productivity, R&D, and Basic Research at the Firm Level in the 1970’s". Amerigm Economic Review. Vol. 76(1): 141-154. Griliches. Zvi. 1990. "Patent Statistics as Economic Indicators: A Survey." Journal of Economic Literature. 28: 1661-1707. Jewkes, John, Sawers, David, and Stillerman, Richard. 1969. The Sources of Invention. New York: St. Martin’s. Hall, Bronwyn H. 1993. "The Stock Market Valuation of R&D During the 1980’s." American Economic Review. 83(2): 259-264. Hamberg, D. 1964. "Size of Firm, Oligopoly, and Research: The Evidence." Canadian Journal of Economics and Political Science. 30: 62-75. Hannay, N. Bruce. 1989. "Technology and Trade: A Study of U.S. Competitiveness in Seven Industries." in N. Rosenberg and R. Landau (eds), The Positive Sum Strategy. Washington DC: National Academy Press. Harrington, Diana R.. 1983. "Whose Beta is Best?" Financial Analysts Journal. July- August: 67-74. Harvey, A.C.- 1981. The Economiegic Analysis of Time Series. Southharnpton: Camelot Press. Jaffee, AB. 1986. "Technological Opportunity and Spillovers." American Economic Review. 76: 984-1001. 175 Karnien, Morton I. and Schwartz, Nancy. 1982. Market Structure and Innovation. New York: Cambridge University Press. Kline, SJ. and Rosenberg, N. 1986. "An Overview of Innovation." in N. Rosenberg and R. Landau(eds.), The Positive Sum Strategy. Washington, DC: National Academy Press. Kmenta, Jan. 1986. Elements of Econometrics. New York: MacMillan. Kohn, MG. and Scott, J.T. 1982. "Scale Economics in Research & Development: The Schumpeterian Hypothesis." Journal of Industrial Economics. 30: 239-249. Kolb, Robert W. and Rodriguez, Ricardo. 1992. Principles of Finance. Lexington, MA: D.C. Heath. Levin, R.C., Cohen, W.M. and Mowery, DC. 1985. "R&D, Appropriability, Opportunity and Market Structure: New Evidence on Some Schumpetn'an Hypotheses." American Economic Review Proceedings. 75: 20-24. Levin, R.C., Klevorick, A.K., Nelson, RR, and Winter, S.G. 1987. "Appropriating the Returns from Industrial R&D." Brookings Institute on Economic Activig. 783- 820. Lichtenberg, F .R. and Siegel D. 1991. "The Impact of R&D Investment on Productivity -- New Evidence Using Linked R&D -- LRD Data." Economic Ingm’ . 29: 203-229. Link, Albert. 1981. R&D in U.S. Manufacturino. New York: Praeger. Mansfield, Edwin. 1968. Industrial Research & Technological Innovation. New York: W.W. Norton. Mansfield, E., Rappoport J ., Schnee, J., Wagner, S. and Hamburger, M. 1971. Research and Innovlation in the Modern Corporation. New York: Norton. Markham, J .W. 1965. "Market Structure, Business Conduct, and Innovation." American Economic Review. 55: 323-332. Marx, K. 1919. Capital. Chicago: Charles H. Kerr. Moyer, R. Charles, McGuigan, James R, and Kretlow, William J.. 1992. Contemmragy Financial Management. New York: West. Mueller, DC. 1967. "The Firm’s Decision Process: An Econometric Investigation." Quarterly Journal of Economics. 81: 58-87. l . 1'1;! p. ’TIIJ 176 Nelson. RR. 1959. "The Simple Economics of Basic Scientific Research." Journal of Political Economy. 67: 297-306. Nelson, Richard R. 1990. "U.S. Technological Leadership: Where Did It Come From and Where Did It Go?" Research Policy. 19: 117-132. Nelson, RR, Peck, M.J., and Kalacheck, ed. 1967. Technology, Economic Growth, and Public Policy. Washington: Brookings Institution. ----- Feb. 4, 1989. "The New World of Drugs". The Economist. pp. 63-64. Pakes, Ariel. 1984. "Patents, R&D, and the Stock Market Rate of Return: A Summary of Some Empirical Results." In Griliches, ed. R&D Patents, and Productivig. Chicago: University of Chicago Press. Pakes, Ariel. 1985. "On Patents, R&D, and the Stock Market Rate of Return." Journal of Political Economy. 93: 390-409. Pelzrnan, S. 1973. "An Evaluation of Consumer Legislation: The 1962 Drug Amendment." Journal of Political Economy. 81: 1049-1091. Quandt, Richard E. 1958. "The Estimation of the Parameters of a Linear Regression System Obeying Two Seperate Regimes." Journal of the American Statistical Association. 53: 873-880. Quandt, Richard E. 1960. "Tests of Hypothesis That a Linear Regression System Obeys Two Seperate Regimes." Journal of the American Statistical Association. 55: 324-330. Ravenscraft, D. and Scherer, F .M. 1982. "The Lag Structure of Returns to Research and Development." Applied Economics. 14: 603-620. Rodriguez, GA. 1979. "A Comment on Fischer & Temin on the Schumpeterian Hypothesis." ournal of Political Economy. 87: 383-385. Roenfeldt, Rodney L., Griepentrog, Gary L., and Pflaum, Christopher C. 1978. "Further Evidence on the Stationarity of Beta Coefficients." Journal of Financial and Quantitative Analysis. 117-121. Rosen, Richard. 1991. "Research and Development with Asymmetric Firm Sizes." Rand Joupnyal of Economics. 22: 411-429. Scherer, F .M. 1965. "Firm Sizes, Market Structure, Opportunity, and the Output of Patented Inventions." American Economic Review. 55: 1097-1125. 177 Scherer, F.M. 1973. Research & Development, Returns to Scale and the Schummterian vaothesis: Comment. Berlin: International Institue of Management. Scherer, F .M. 1984. Innovation & Growth: Schumpeterian Pergmctives. Cambridge, MA: MIT Press. Scherer, F .M., and Ross, David. 1990. Industrial Market Structure and Economic Performance. Boston: Houghton Mifflin. Scherer, F.M. 1992. "Schumpeter and Plausible Capitalism." Journal of Economic Literature. Vol. 30(3): 1416-1433. Schumpeter, Joseph A. 1939. Business Cycles. New York: McGraw Hill. Schumpeter, Joseph A. 1942. Capitalism. Socialism. and Democracy. New York: Harper. Schwartzman, D. 1976. Innovation in the Pharmaceutical Indpm. Baltimore: Johns Hopkins University Press. Smith, Adam. 1937. The Wealth of Nations. New York: Modern Library. Soete, L.L.G. 1979. "Firm Size & Innovative Activity: The Evidence Reconsidered." European Economic Review. 12: 319-340. Solow, R. 1957. "Technical Change and the Aggregate Production Function." ReView of Economics and Statistics. 39: 312-330. Standard and Poor Corporation. 1989. The Compustat Annual, Industrial, and OTC Files. New York: Standard and Poor. Datatapes. Stern, Robert M., Baum, Christopher F ., and Greene, Mark N. 1979. "Evidence on Structural Change in the Demand for Aggregate U.S. Imports and Exports." Journal of Political Economy. 87: 179-192. Teece, David. 1986. "Profiting from Technological Innovation: Implications for Integration, Collaboration, Licensing, and Public Policy." Research Policy. 15: 285-305. Teece, David. 1989. "Inter-Organizational Requirements of the Innovation Process." Managerial and Decision Economics. 10: 35-42. Temin, P. 1979. "Technology, Regulation, and Market Structure in the Modern Pharmaceutical Industry." Bell Journal of Economics. 10: 429-446. 178 Thomadakis, Stavros B. 1977. "A Value-Based Test of Profitability and Market Structure." Review of Economics and Statistics. 59: 179-185. The U.S. Department of Commerce, Bureau of the Census. 1982. Annual Survey of Manufatures 1982 Concentration Ratios. The U.S. Department of Commerce, Bureau of the Census. 1982. Annual Survey of Manufatures. 1982. Industry Profiles. Vernon, John M. 1971. "Concentration, Promotion, & Market Share Stability in the Pharmaceutical Industry." Joum_afl of Industrial Economics. 19: 246-266. Williamson, Jeffrey G. 1991. "Productivity and American Leadership: A Review Article." The Journal of Economic Literature. 29: 51-68. Winter, Sidney G. 1984. "Schumpeteriam Competition in Alternative Technological Regimes." Journal of Economic Behflor and Organization. 5: 287-320.