WHEN ARE NEW FIRMS MORE INNOVATIVE THAN ESTABLISHED FIRMS? Scott Shane. Riitta Katila

Transcription

1 WHEN ARE NEW FIRMS MORE INNOVATIVE THAN ESTABLISHED FIRMS? Scott Shane Riitta Katila Robert H. Smith School of Business University of Maryland College Park, MD Tel: (301) Fax: (301) Acknowledgments: The first author would like to thank Don Kaiser, Lita Nelsen, and Lori Pressman at the MIT Technology Licensing Office for access to the data on MIT patents and for answering many questions about the data and TLO policies and procedures. Jonathan Eckhardt, Anil Gupta, and Ken G. Smith provided useful comments.

2 WHEN ARE NEW FIRMS MORE INNOVATIVE THAN ESTABLISHED FIRMS? Abstract We propose a model in which industry conditions influence the relative advantages of new and established firms as innovators. Using a unique data set of commercialization attempts of inventions simultaneously at risk of commercialization by both new and established firms, we find that industry characteristics - including the number of firms in the industry, the size of the market, the availability of capital for new firms, and manufacturing intensity - influence whether new or established firms are more innovative. Running headline: When are new firms more innovative? Keywords: innovation, new firms. 1

3 INTRODUCTION When are new firms more successful at innovation than established firms? In this study, we examine an important contingency that influences the answer to this question the industry context in which innovation occurs. We argue that new firms are neither better nor worse than established firms at innovation across all industry contexts. Rather, the relative advantages of new and established firms at innovation depend on time-varying industry characteristics, including the number of firms in the industry, the size of the market, the availability of venture capital, and manufacturing intensity. We provide new empirical evidence for this proposition by examining a unique data set of technological inventions simultaneously at risk of commercialization by both new and established firms. At least since Schumpeter (1934, 1942), researchers have debated whether new firms are better than established firms at innovation, defined by Schumpeter (1934) as the development and commercialization activities that transform an invention into a product or service introduced to the marketplace. While researchers from the evolutionary and life-cycle perspectives have argued that as organizations age they become less and less adaptive to new environmental needs (Abernathy and Utterback 1978, Sørensen and Stuart 2000), other researchers propose that organizational aging increases innovativeness (Teece 1986, Rao and Drazin 2002). Researchers who have argued that new firms are more innovative than established firms have identified at least four factors that give new firms an advantage. First, existing organizations are imprinted by the characteristics present at their founding, and often find it difficult to respond to environmental changes quickly (Hannan and Freeman 1984, Sørensen and Stuart 2000). Since innovative activities are one of the main ways for organizations to adapt to changes in their environments (Schoonhoven et al. 1990), younger firms should be more successful at innovation 2

4 than older ones. Second, unlike established firms, new firms are not bound by established routines and structures that often perform poorly in identifying and developing new information (Henderson and Clark 1990). Third, new firms provide strong incentives for organization members to engage in innovation, whereas established firms often focus incentives on encouraging those activities that their firm is already engaged in (Holmstrom 1989, Burgelman 1994). Fourth, established firms seek to serve their existing customers, who prefer suppliers that focus on providing existing products (Christensen and Bower 1996). New firms are not constrained by existing customer expectations and thus should be more successful at innovation. In contrast, other researchers have argued that established firms are more innovative than new firms. First, unlike new firms, established firms possess complementary assets in manufacturing and marketing or more favorable structural positions that enhance their ability to generate new products from technology (Teece 1986, Hannan 1998). Second, established firms have access to internal cash flows that enable them to finance innovation without accessing capital markets, thereby reducing the cost of innovation (Greenwald et al. 1984). Third, established firms have accumulated routines that facilitate their ability to create new products more reliably and at a lower cost (Argote 1999, Eisenhardt and Tabrizi 1995). Fourth, established firms often possess scope advantages, which facilitate innovation because of the uncertainty, inappropriability, and indivisibility of innovation (Arrow 1962). Although both research traditions have made persuasive arguments and have generated strong empirical evidence in support of their perspective, this literature suffers from two omissions that make it difficult to determine if new firms are more innovative than established firms. First, contingency theory (Lawrence and Lorsch 1967) would suggest that neither new firms nor established firms have a universal innovation advantage, but, instead, that the 3

5 innovation advantages of new or established firms depend on the appropriateness of their organizational attributes to the environmental conditions facing the firm at a given point in time. The tendency of scholars on each side of the debate to draw empirical evidence from different industries suggests that industry conditions may indeed be one important contingency that influences whether new or established firms are more innovative (Acs and Audretsch 1990). For example, researchers that have shown that established firms are more innovative than new firms have tended to study industries such as automobiles (Utterback and Abernathy 1975) and pharmaceuticals (Gans and Stern 2000); whereas researchers that have shown that new firms are more innovative than established firms have tended to examine industries like minicomputers (Romanelli 1989) and hard disk drives (Christensen and Bower 1996). Second, researchers have been unable to directly test whether new or established firms are more innovative 1 because they have been unable to observe technological inventions at risk of innovation by the two groups of firms. To examine this question, researchers need to observe a set of inventions prior to efforts by new and established firms to commercialize them. Simple observation of innovations by new and established firms fails to accomplish this goal because such an effort confounds the invention with its commercialization. Data on innovations made by firms that have invented the underlying technology raises the question of whether new (established) firms are more inventive than established (new) firms, more innovative than those firms, or both. Moreover, innovations often emerge simultaneously with the new firms that create them, making it difficult to examine the innovation process in new firms without sampling 1 To date the closest that empirical researchers have come has been to compare large and small firms, as Acs and Audretsch (1990) and Pavitt, Robson and Townsend (1987) have done. While hugely informative, comparison of small and large firms explores a different theoretical question than a comparison of new and established firms. Small, established firms can exist for a long period of time and possess many of the same assets and capabilities as large, established firms, just on a smaller scale. As a result, a comparison of large and small firms can directly assess the effects of size on innovation, but not the effects of prior existence, which is the focus of our study. 4

6 on the dependent variable. Finally, new and established firms rarely are at risk of innovating technologies drawn from the same distribution of inventions. Because new technology firms are often born as spin-offs from established firms (Klepper 2001), the technologies that they seek to innovate are often a selected sample of inventions that were born in established firms, but passed over by those organizations. This selection bias makes it very difficult to draw inference from a comparison of new technology commercialization efforts of new and established firms. In this paper, we empirically investigate the proposition that time varying industry characteristics affect whether new firms are more innovative than established firms. To do this, we examine the performance of both new and established firms at commercializing inventions discovered at the Massachusetts Institute of Technology between We argue that four industry characteristics disadvantage new firms as innovators relative to established firms: small numbers of firms in the industry; lack of venture capital flowing to the industry; a high manufacturing intensity; and large market size. THEORY DEVELOPMENT Following prior research, we define innovation as a process that begins with an invention, proceeds with the development of the invention, and results in the introduction of a new, product, process or service to the marketplace (Edwards and Gordon 1984:1, see also Schumpeter 1934, 1942). In other words, innovation begins when an invention is chosen for further development, with the ultimate goal of introducing it to the market (Kuznets 1962). This study explores the industry conditions under which new organizations are better at this process than organizations already in existence. To do this, we must identify the key characteristics required for innovation, and establish how new and established firms differ in possessing these characteristics. 5

7 Successful innovations make demands upon ingenuity, technical knowledge, and ability, require a great deal of material capital and skilled resources (Kuznets 1962: 34), and cannot influence firm performance until they have been successfully introduced to the market (Kogut and Zander 1992). First, successful innovation requires innovative capabilities, or a set of skills and routines to transform inventions into new products. Second, successful innovation requires capital. Capital allows the innovator to follow the normal path of developing the new product before selling it to others. Third, innovation requires information about manufacturing and marketing of products. Because new products must be produced and sold after they are developed, successful innovation often requires the incorporation of information about manufacturing and marketing those new products. Fourth, innovation requires the innovator to believe that the market size is sufficient to justify the investment of time and money in the uncertain activity of developing new products in place of investing in other activities. These four characteristics of innovation are important because new firms differ fundamentally from established firms on these dimensions. First, new firms lack the routines for innovation from existing operations that established firms possess. A firm s existing knowledge base, which is developed over time, can help an established firm to generate new products from technological inventions (Fleming 2001, Katila in press). Thus, any innovative activities that new firms undertake require the creation of new product and process development routines, rather than the exploitation of existing routines. Second, new firms lack cash flow from existing operations that existing firms possess. As a result, the financing of innovation requires the acquisition of capital from external investors. Third, new firms lack marketing and manufacturing assets that established firms possess from existing operations. Therefore, new firms must build or contract for these assets to undertake innovation. Fourth, new firms lack the 6

8 opportunity cost of undertaking innovative activities that established firms possess as a result of having existing alternatives to a particular course of action (Dean et al. 1998). What we do not know, however, is whether new firms are more innovative in some industries, but not in others, because of differences in the importance of these four dimensions to innovation in that industry at that point in its evolution. For example, is the value of information from manufacturing less important to innovation in the software industry than it is to innovation in the automobile industry since software is not manufactured? Similarly, is access to internal capital less important to innovation in biotechnology than to innovation in the chemicals industry because of the interest of venture capitalists in investing in new biotechnology companies in the 1980s and 1990s? In the section below, we develop specific hypotheses about expected relationships between dimensions of industry and new firm superiority at innovation. HYPOTHESES Following the definition of industry employed in prior studies (Acs and Audretsch 1988, Dean et al. 1998), we define an industry as the group of firms that supplies a given market using similar inputs, and use three-digit Standard Industrial Classification (SIC) codes to measure this categorization (see Hitt et al. 1997). We develop four hypotheses about the relationship between industry characteristics and the relative advantage of new vs. established firms at commercializing inventions. First, new firms will be more likely than established firms to commercialize inventions in industries composed of a greater number of firms. Second, new firms will be more likely than established to commercialize inventions in industries that receive more venture capital. Third, new firms will be less likely than established firms to commercialize inventions in industries that are manufacturing intensive. Fourth, new firms will be less likely than established firms to commercialize inventions in larger markets. 7

9 Number of Firms Industries differ significantly in the number of firms that compete in them at a given point in time. Because new firms lack resources and routines for innovation, innovation by new firms is facilitated in industries with large numbers of firms for several reasons. First, industries with larger numbers of firms are industries in which greater variance in approaches to innovation exist. The variance in approaches to innovation is an increasing function of the number of firms in an industry because each firm, with innovation modes and routines specific to its history, will offer a different alternative approach (Dosi 1988). Because new firms introduce new approaches to innovation, they are likely to be more successful, on average, in industries in which high variance in approaches to innovation exist. Second, industry fragmentation increases market uncertainty (Cohen and Klepper 1992), thereby increasing the advantage of new firms, which have greater flexibility than established firms in shifting between markets. In contrast, established companies, by virtue of their existing customer base, are less able to shift markets and so find innovation under market uncertainty more difficult (Christensen and Bower 1996). On the other hand, when industries have few firms, the limited number of competitors reduces the number of potential new products. As a result, the uncertainty of which product will be accepted by the market is lessened. Third, product development by new firms is facilitated by entry into markets with a large number of firms because new firms in these markets are less likely than new firms in markets with few firms to face competition from a few large firms with the resources to defeat them before they have established a revenue base through successful product introduction (Romanelli 1989). When an industry is composed of a few firms, the new firm s efforts to commercialize 8

10 technology immediately threaten the customer base of - and provoke a response from - larger and wealthier competitors that control marketing and distribution networks. Thus, we propose: H1: New firms will be more successful than established firms at commercializing inventions in industries with larger numbers of firms. Venture Capital Availability New firms lack the capital from ongoing operations that existing organizations can use to finance innovation. As a result, they often finance innovation by raising private equity capital. By selling equity, new firms can raise the cash necessary to obtain equipment and to hire people for commercialization activities (Schoonhoven et al. 1990). External capital is more costly to obtain than internally generated capital because of information asymmetry problems between new firms and external investors (Greenwald et al. 1984). This information asymmetry makes it difficult to monitor creative activities, such as innovation (Holmstrom 1989). Venture capitalists are entities that specialize in gathering and evaluating information on new companies (Amit et al. 1998). The structure of venture capital arrangements allows these organizations to overcome many of the information asymmetry problems that plague external financing (Gompers and Lerner 1997). 2 Venture capitalists tend to invest in new firms rather than in established firms. To obtain a comparative advantage over other types of investors, venture capitalists focus on investments in which the skills that they develop in choosing and monitoring investments offer the greatest potential returns (Amit et al. 1998). These skills involve identifying high potential entrepreneurs and venture opportunities, monitoring investments by sitting on boards of directors, helping new 2 Empirical research supports this proposition. When venture capital is not available, Roberts and Hauptman (1987) found that new firms are less likely to develop innovations. In a larger sample study, Kortum and Lerner (2000) found that venture capital has a positive and significant effect on patenting activity of new firms. 9

11 firms obtain additional financing, recruiting a management team, identifying customers and suppliers, and strategic and operational planning (Gorman and Sahlman 1989). Because investments in established companies involve a different set of skills than venture capitalists possess, including rigorous financial analysis of past financial performance of companies and an understanding of financial market dynamics, venture capitalists rarely invest in established companies, particularly publicly traded ones (Gompers and Lerner 1999). Established firms also are often reluctant to invest in opportunities in which venture capitalists invest. Established companies develop capabilities that help them serve particular markets. Thus, established companies are often reluctant to make investments that would lead them to exploit a different market from the one that they currently serve (Christensen and Bower 1996), and are often less successful than new firms if they decide to do so. In contrast, venture capitalists prefer to invest in new companies that are designed to exploit new markets because such a focus allows the new companies to avoid competing with established firms. Moreover, established companies, constrained by their incentive structures and existing reputations often tend toward more conservative investments than venture capitalists (Holmstrom 1989). As a result, industries differ in their access to venture capital (Amit et al. 1998). Because venture capital investments are important to new firm innovation, in industries in which venture capital is plentiful, new firms are better able to commercialize inventions than in industries in which venture capital is not plentiful. In contrast, in industries in which venture capital is less plentiful, new firms are not able to offset the innovation advantage of established firms that comes from access to internal cash flows. This argument leads to the second hypothesis: H2: New firms will be more successful than established firms at commercializing inventions in industries in which venture capital is more plentiful. Manufacturing Value-Added 10

12 The ability of new firms to commercialize inventions is influenced by the manufacturing intensity of the value chain in the industry. In some industries, much of the value-added comes from manufacturing; whereas in others, much of the value-added comes from other activities, such as product development or marketing. For example, innovation in industries that rely on complex, capital intensive continuous production processes, like automobiles, may demand greater manufacturing capabilities for product development than innovation in other industries (Cohen et al. 2000). In contrast, other industries, like software, have products that do not need to be manufactured, thereby reducing the importance of manufacturing capabilities in the process of product development (Teece 1998). We propose that new firms will be worse at commercializing inventions in industries that are more manufacturing intensive. First, Gort and Klepper (1982) have found that established firms have an advantage at innovation when accumulated knowledge from existing operations facilitates innovation. As Teece and Pisano (1994: 540) explain, entrepreneurial activity cannot lead to the immediate replication of unique organization skills through simply entering a market and piecing the parts together overnight. Replication takes time, and the replication of best practice may be illusive. New firms lack specialized complementary manufacturing assets needed for developing new products and therefore find it more difficult to innovate in manufacturing intensive industries (Teece 1986). Thus, when complementary assets in manufacturing are important to innovation in an industry, new firms have to create these assets from scratch or contract for them, placing new firms at a disadvantage relative to established firms. The creation of specialized manufacturing assets from scratch is costly, especially for new firms that are cash constrained. These firms may also lack the skills to build these capabilities efficiently or effectively. Particularly when the knowledge itself is tacit, established firms develop manufacturing capabilities that are difficult 11

13 for new firms, that have not yet engaged in manufacturing, to replicate (Teece 1998). Furthermore, contracting for these assets is problematic for new firms because they lack the reputation to provide non-contractual safeguards against opportunism. Second, innovation often depends on the joint exploitation of manufacturing capability and new technology (Galbraith 1982). Thus, the established firms have an advantage in industries where knowledge of the manufacturing processes is necessary to correctly specify new product characteristics (Teece 1992) because the procedures and equipment for scale-up and process development are often developed through interaction between manufacturing engineers and research and development personnel (Pisano 1991). In contrast, when manufacturing is not important to innovation in an industry, manufacturing routines disadvantage established firms at innovation. Technology develops in an evolutionary manner, constrained by existing paradigms and trajectories (Dosi 1988). These constraints have strong exclusionary effects, leading engineers and scientists to focus on solutions to technical problems that lie within an existing evolutionary path. To better innovate within this path, firms develop routines for problem solving. These routines have powerful negative effects on organizational efforts to make use of information that alters the existing technological architecture, disrupts existing communication channels, or changes existing organizational routines (Henderson and Clark 1990). The exclusionary effect of existing routines reduces the variance in information about innovation that can be introduced into the organization. Because innovation is uncertain, this reduction in the variety of information employed, lowers the probability of innovation. As a result, when manufacturing is not important in an industry, the countervailing advantage of having manufacturing routines does not offset the constraining presence of existing routines and established firms are disadvantaged at innovation. We propose: 12

14 Market Size H3: New firms will be less successful than established firms at commercializing inventions in industries that are more manufacturing intensive. The ability of new firms to commercialize inventions is influenced by the size of the product market. First, new firms are often more successful as innovators in smaller, niche markets because of their greater speed and flexibility (Dean et al. 1998). The existing customer bases of established firms often make them less adaptive to the needs of niche markets. Second, small markets often cannot provide sufficient return for established firms to invest, leading independent entrepreneurs with low opportunity cost to exploit them. However, when markets are large, the size of the market justifies investment in product development by established organizations, generating competition to the new firm innovator (Shane 2001). Third, larger market size also implies potential for scale economies in product development activity, which provides an advantage to established firms. These arguments lead to the fourth hypothesis: METHODOLOGY Sample H4: New firms will be more successful than established firms at commercializing inventions in industries in which markets are smaller. This study explores 966 attempts by private firms to commercialize inventions discovered at MIT between 1980 and This sample includes all inventions patented by MIT and licensed during the seventeen-year observation period. These data were collected over a two-year period from the records of the MIT Technology Licensing Office. This office is the administrative unit responsible for management of intellectual property assigned to MIT. The records included information about all licensing agreements between MIT and private sector entities. Because MIT generates significant revenues from its technology licensing, it carefully documents the 13

15 information, including dates during which particular patents are covered by licensing agreements, the legal status of the licensee, and the stage of commercialization of products that use the invention. As a result, we are able to create with a high level of accuracy annual spells for each invention from the date of first license until first sale, termination of the license, or censoring; identify the type of entity that licensed the invention; and measure the year of first commercial sale of products developed from the invention. The sample of MIT inventions is uniquely appropriate for this study for several reasons. First, MIT patents provide a documented population of new technologies that can be identified after invention, but before commercialization. Second, MIT inventions provide a documented population of new technologies at risk of commercialization by both new and established firms. Because our argument depends upon showing that new firms will be better at commercializing inventions (net of the tendency to produce those inventions) for the same population of inventions, we need to identify inventions that are at risk of exploitation by both sets of firms after invention and before commercialization. Unpatented technologies do not provide these features because researchers cannot identify them after invention but before commercialization. Patents produced by firms do not provide these features because firm patents, by definition, require the existence of the firm. Therefore, attempts to identify new firm commercialization efforts for patents produced by firms either entails sampling on the dependent variable or selection bias. The former occurs when the commercialization effort is first identified and then the new firm that undertakes that effort is tracked. The latter occurs when commercialization efforts by new firms founded to exploit patents that established firms have chosen not to exploit are compared with commercialization efforts by established firms to exploit their patents. Analysis 14

16 We use Cox proportional hazard event history models to predict the hazard of commercializing products developed from the 966 efforts to commercialize the inventions. We focus our analysis on the interaction between the characteristics of the industry in which the commercialization attempt takes place and whether the licensee of the invention was a new firm founded to commercialize the invention or was a previously existing firm. To control for invention-specific effects (the same invention can be licensed by several firms), we use a robust estimation procedure clustering by patent number (Lin and Wei 1989). Descriptive statistics and correlations for all variables are shown in Table Insert Table 1 about here --- The U.S. Patent and Trademark Office (USPTO) does not categorize patents by industry. Because patent classes can exist in more than one industry and firms can operate across industries, patents can only be classified to specific industries through the use of a concordance that specifies which industry SIC code is the most likely to have a particular patent classification. We use the USPTO s concordance to make this allocation. The concordance matches the six digit patent class to a three digit SIC code. Once this concordance was made, we were able to gather data on the industry characteristics that we measure. Dependent Variable - Commercialization We operationalize the dependent variable, success in the commercialization process, by measuring two competing outcomes: the likelihood that the licensee sells a product developed from the licensed technology for revenue (First Sale), and the likelihood that the licensee will abandon the license to the piece of technology (Licensing Abandonment). While the First Sale variable gives an external confirmation of the success of the commercialization effort, Licensing Abandonment gives an internal confirmation of failure at that effort. 15

17 Once an invention is licensed and the commercialization effort begins, three possible events can occur. First, a first sale of a product created from that invention can occur at any point in time. To measure this event, we construct annual spells that start when the invention is first licensed and end when a product using the invention generates its first dollar of sales to a third party. If first sale is achieved, the First Sale variable is coded as one in that year, otherwise it is coded as zero. We determine that a first sale has been achieved by examining records from the MIT technology licensing office, which indicates whether sales have occurred for products or services that employ the technology. We expect that the accuracy of the information on first sales is high. Not only are the licensees contractually obligated to inform MIT of the successful sale of products or services that use its technology, but MIT also has a strong incentive to verify this information since its royalty revenues depend on the successful sale of licensed technology. A second possibility is that at any point during the observation period the firm may decide to terminate the commercialization effort and abandon the license. To measure licensing abandonment we construct annual spells in which we use records of the licensing contracts maintained by the MIT Technology Licensing Office to determine if a license to a patent remains in force in a given year. If the patent license is terminated, the Licensing Abandonment variable is coded as one in that year, otherwise it is coded as zero. As above, we expect that the accuracy of the information on license abandonment is quite high. Because one of the major functions of the MIT Technology Licensing Office is to manage the Institute s licenses, it maintains excellent records on whether licensing agreements are in force, and what particular inventions they cover. The third possibility is that the event of interest (First Sale or Licensing Abandonment, respectively) does not occur during our observation period. In that case both First Sale and 16

18 Licensing Abandonment are coded as zero during the observation period, and as censored at the end of the observation period. Covariates New firm. We examined the MIT Technology Licensing Office records to identify the licensees of the patented inventions. If the licensee was a firm that did not exist prior to the year of the license, the licensee was coded with a dummy variable of one to indicate that it was a new firm. Number of Firms. We use a time varying covariate to capture the annual number of firms in the industry because previous technology strategy researchers have demonstrated that the number of firms in an industry is not fixed, but varies over the industry s evolution (Tushman and Anderson 1986, Utterback 1994). We use Census of Manufactures data to measure the number of firms in the industry because it is the most reliable data source for these data and has been used in many prior studies (e.g., D Aveni and Ravenscraft 1994, Dean and Snell 1996). Venture Capital. Data from the Securities Data Corporation s venture capital database were used to calculate the annual dollar value of venture capital funding in the industry. We use a time varying covariate to capture venture capital funding because researchers have shown that venture capital funding shifts significantly across industries over time (Shane and Stuart 2002). Securities Data Corporation is the leading provider of industry-level venture capital data and the accuracy and validity of its data has been shown in several studies (Lerner 1994, 1995). Manufacturing Value-Added. We use data from the Census of Manufacturers to calculate the annual dollar value of manufacturing and the annual dollar value of shipments in the industry. Manufacturing value-added is calculated as the ratio of these two numbers in the industry-year. We use a time varying covariate to capture manufacturing intensity because previous researchers have shown that manufacturing intensity changes as industries evolve 17

19 (Teece 1992, Utterback 1994). We use Census of Manufactures data to measure manufacturing value-added in the industry because it is the most reliable data source for these data and has been used in many prior studies (e.g., D Aveni and Ravenscraft 1994, Dean and Snell 1996). Market Size. We use data from the Census of Manufacturers also to calculate the annual dollar value of shipments in the industry. We use a time varying covariate to capture market size because previous researchers have shown the size of the market changes as industries evolve (Christensen and Bower 1986, Utterback 1994). Control Variables Technical Fields. We control for the technical field in which the invention is found chemical, drug, electrical, mechanical (other is the base case) because commercialization of technologies by new and established firms should differ across types of technology. The existence of technological opportunities, the resource requirements and time needed to commercialize them, the tendency of inventors to disclose their inventions and other factors all vary across types of technology (Cohen and Levin 1989). By controlling for the technical field, we can partial out this type of variation from the data. Time Period. We use dummy variables for each of four periods ( ; ; ; and , with being the omitted period) to control for the time when the patent was filed because of changes in Federal law and MIT policy over the period. In 1980, the Federal government gave universities the property rights to federally funded inventions, increasing their incentives to license inventions. In 1984, the Federal government allowed universities to sell those property rights, further increasing their incentives to license. In 1987, MIT first allowed inventor-founded start-ups to provide equity to the university in return for the university s payment of patent costs, reducing capital constraints to inventor start-ups. In 18

20 1990, MIT permitted inventors to retain their share of royalties even if they held equity in a new company founded to exploit the invention, stimulating inventor-start-up activity. Selection Correction. Because a decision to abandon a licensed invention precludes the ability to sell a product developed from that invention, we control for license termination in the models that predict the achievement of First Sale. By including this control we avoid omitted variable bias that can lead to inconsistent estimates of the variables (Greene 2000). To generate this correction (Termination Correction), we use Lee s (1983) generalization of the Heckman selection model. In this correction predicted probabilities for license termination are used to generate a sample correction variable lambda: λ it φ = [ Φ ( ( ))] 1 F t 1 F ( t) i i where F i (t) is the cumulative hazard function for project i at time t, φ is the standard normal density function, and 1 Φ the inverse of the standard normal distribution function (Lee 1983). In constructing the termination correction variable, it is important to find at least one variable that should affect termination, but does not have a direct effect on the achievement of First Sale. We include the source of the original funding for the invention (whether the creation of the invention was partly financed by private firms or not) as an additional exogeneous covariate. This variable should predict license termination but not First Sale because firms willingness to terminate their licenses to technology depends on whether they and/or their competitors funded the development of the technology. The termination correction λ it is then included as a control in the models that predict First Sale. Another sample selection issue arises from the question of whether new and established firms license technologies in different types of industries. For example, new firms could potentially 19

21 self-select into licensing technologies in industries that established firms have left unexploited, perhaps because new firms are likely to be more successful than established firms at innovation in those industries. Such self-selection would prevent us from observing a full sample of new firm commercialization efforts. For this reason, we construct a second selection correction variable (Self-selection Correction). We estimate the likelihood that a new firm would license a given MIT invention, using the age of the patent technology class as an additional exogeneous predictor of new firm licensing. We select this variable because prior research has shown that inventions in younger technology classes are more likely to be licensed by new firms (Shane 2001) and because the age of patent technology class predicts neither First Sale nor License Abandonment. Data on the population of all patents issued to MIT between 1980 and 1996 and their licensees (if any) is used for this estimation. The Self-selection Correction is then constructed as above, and included in both First Sale and License Abandonment models. RESULTS Tables 2 and 3 report the results of the regression analysis on how the industry conditions affect (a) new firms success at achieving first sale (Table 2), and (b) the new firms decision to terminate the license (Table 3). In sum, three of our four hypotheses were supported (H1, H3 and H4). New firms are more successful at commercializing inventions in industries with a large number of firms, but a relatively small market. We also find that the lower the manufacturing intensity of the industry, the more successful new firms are at commercializing inventions. --- Insert Tables 2 and 3 about here --- In Table 2, First sale is the dependent variable as described above. The first model reports the baseline where technical field (Chemical, Drug, Electrical, Mechanical) and time period dummies were included as control variables. In Model 2 the New Firm variable is added. Models

22 introduce the independent effects of Number of Firms, Venture Capital, Manufacturing Value- Added and Market Size variables, and their interactions with New Firm, respectively. In Table 3, Licensing Abandonment is the dependent variable. Again, the first model reports the baseline where technical field (Chemical, Drug, Electrical, Mechanical) and time period dummies were included as controls. In Model 2 the New Firm variable is added. Models 3-10 introduce the independent effects of Number of Firms, Venture Capital, Manufacturing Value- Added and Market Size variables, and their interactions with New Firm, respectively. Below, the results are discussed based on full models from both the First Sale and Licensing Abandonment models (Model 11 in Tables 2 and 3). Hypothesis 1 proposed that new firms would be more successful than established firms at commercialization in industries with a large number of firms. The interaction of New Firm and Number of Firms is positive and significant in Model 11 in Table 2, and negative and significant in Table 3, with significant main effects, supporting this hypothesis. New firms are more likely to reach first sale and less likely to abandon a licensed invention if the industry has a large number of firms. Hypothesis 2 proposed that the availability of venture capital would enhance the new firm success at commercialization. This hypothesis is not supported, however, since the main effect of Venture Capital in Model 11 in Table 2 is not significant, making the interpretation of the interaction between New Firm and Venture Capital difficult. Moreover, this interaction has a positive sign in Table 3, indicating that the amount of venture capital in the industry encourages rather than discourages license termination, contrary to our expectations. We return to the possible explanations for this unexpected result in the discussion section. 21

23 Hypothesis 3 proposed a negative relationship between the manufacturing intensity of the industry and new firm success in commercializing inventions. In Table 2 the interaction between the New Firm and Manufacturing Value-Added variables is negative, and in Table 3 is positive and significant, thus confirming this hypothesis. Hypothesis 4 predicted that new firms would be more successful in commercialization in smaller markets. The estimated negative interaction between New Firm and Market Size in Table 2, and the positive interaction between New Firm and Market Size in Table 3, provide support for this hypothesis. Model 12 in both Tables 2 and 3 provides a sensitivity test where selection corrections for license termination and self-selection are included. Even after controlling for these selection effects, the original results are consistently supported. We also tested for the possibility that the effect of the number of firms in the industry proposed in Hypothesis 1 was an artifact of industry concentration rather than firm density. To test the validity of this alternative hypothesis, we examined the effect of the Industry Concentration, measured as the annual market share of the four biggest firms in each industry, using data from the Census of Manufacturers. When we used Industry Concentration in place of the Number of Firms variable in interaction with New Firm, we failed to achieve significant results in either the regressions to predict First Sale or License Abandonment. In addition, when we included industry concentration as an additional control variable in our regressions that measured the effect of the number of firms, our results remained qualitatively the same as in our reported regressions. These results suggest that the number of firms in the industry captures the effect of a different construct than does industry concentration. 22

24 We also used an alternative measure for the New Firm variable to insure that the results were not affected by particular operationalizations. In place of our original dichotomous variable, we used a time-dependent Firm Age variable that records the age of each sample company since its founding. Unfortunately the firm age data were not available for all sample firms, reducing the final sample for this sensitivity test to 2844 observations. Despite this smaller sample, the results supported the original findings. In addition, we tested the robustness of the results to unobservable differences across firms. Since the lack of variation in the dependent variable over time for a large number of licensing attempts prevents the use of fixed effects estimation (many attempts never reach first sale, for example, and would thus be excluded from the analysis), we ran the models with two alternative methods that control for unobserved heterogeneity (see also Beck and Katz 2000): a Weibull event history model with gamma heterogeneity (Greene 2000), and a random effects complementary log-log estimation. Again, the pattern of the original results was supported. Finally, it is possible that technologically novel inventions are more difficult to commercialize, or take a longer time to develop. To ensure that our results are not affected by the effect of technological novelty, in unreported regressions we added this variable to the full model. Similar to Rosenkopf and Nerkar (2001), Technological Novelty was constructed as the number of technological sub-classes the invention is assigned to, but does not cite prior art from. Our original results were consistently supported when this variable was included, suggesting that the technological novelty of the inventions does not affect the industry contingencies we study. DISCUSSION This study demonstrates that industry conditions influence the relative advantages of new and established firms as innovators of new technologies. In industries composed of more firms, new 23

25 firms will be less likely than established firms to abandon efforts to commercialize technological inventions and more likely than established firms to achieve first sale. In industries that are more manufacturing intensive, new firms will be more likely than established firms to abandon efforts to commercialize technological inventions and less likely than established firms to achieve first sale. In larger markets, new firms will be more likely than established firms to abandon efforts to commercialize technological inventions and less likely than established firms to achieve first sale. However, the amount of venture capital flowing to the industry does not have a statistically significant effect on the relative advantage of new firms at commercializing technological inventions. Our empirical methodology strengthens our confidence in the findings. We explore the relative advantages of new firms at commercialization by identifying all efforts to commercialize MIT inventions between 1980 and As a result, we avoid the problem of sampling on the dependent variable, which plagues many efforts to explore technology commercialization efforts by new firms. Moreover, by examining university patents, we compare inventions that are simultaneously at risk of commercialization by new and established firms, thereby mitigating the problem of selection bias that can occur if patents eventually exploited by new firms were first not selected for commercialization by established firms. We are also able to include all firms that commercialized MIT inventions, avoiding the generalizability problem that many prior studies, which have been able to include only large firms, have faced. Finally, we conducted a selection correction to account for the fact that some MIT inventions were more likely than others to lead to the founding of new firms, perhaps because independent entrepreneurs foresee the advantages of new firms at commercializing some technologies and not others. 24

26 Despite these efforts, we did not find support for all of our hypotheses. The lack of a significant effect for amount of venture capital available in the industry on new firm advantage at technology commercialization begs further discussion. One explanation could be that venture capitalists are not good at, or interested in, identifying industries that will generate revenues quickly. As a result, sometimes they invest in hot industries like biotechnology, where reaching the first sale turns out to be very difficult and takes a lot of time. Another explanation could be that venture capital funding and achieving first sales are alternative ways to access capital. If venture capital is widely available, new firms do not need to reach first sale quickly to finance the development of technology. A third explanation is that formal venture capital is less important as a source of capital for the commercialization of new technologies than is popularly believed. Angel investing and government programs, such as the Small Business Investment Research grants, may be more important sources of capital for firms at early stages of technological development. As a result, differential flows of these types of capital across industries may drive the relative innovation advantages of new firms, while differential flows of venture capital may not. Whatever alternative explanation accounts for the null finding for venture capital flows, future research should investigate this question. Implications for Theory Our findings have implications for the development of theory about innovation and entrepreneurship. The dynamics of technological leadership depend heavily on whether new or established firms are better innovators. If established firms are better innovators, then established firms will come to dominate the innovation process, leading to a process of creative accumulation. In contrast, if new firms are better innovators, then technological leadership will be short-lived, and a process of creative destruction will prevail (Sørensen and Stuart 2000). 25