|
Article Excerpt
1. Introduction
Understanding the determinants of the innovative performance of firms is central to innovation and entrepreneurship research. Yet, the development of innovations by firms is not simply a matter of technical ability. Incentives matter as well. A classical example is Arrow (1962), whereby larger firms have lower incentives to innovate for fear of cannibalizing existing products. Similarly, if a firm faces a larger potential market or has established downstream commercialization capabilities, it will have higher innovation incentives. Therefore, in studying the drivers of innovation performance, it is important to distinguish between incentives and capabilities, as one may confound lack of incentives with lower technical capabilities or vice versa.
A key contribution of this paper is to develop and estimate a structural model that distinguishes between these two dimensions. We use observations on more than 3,000 drug research and development (R&D) projects initiated between 1980 and 1994, and followed through 2005. The pharmaceutical industry is ideal for our analysis because the drug innovation process follows well-defined steps (see, e.g., FDA 1999, DiMasi et al. 2003). New compounds are generated in the laboratory (initial discovery), tested on animals (preclinical research), and if the firm considers them sufficiently promising, both technically and economically, they are advanced into clinical trials on humans. This process has two important regulatory gates: at the beginning of clinical trials, when an investigational new drug application is filed, and at the end of clinical trials, when a new drug application may be filed. These two main stages play an important role in our analysis.
We argue that firms do not advance all technically promising compounds to clinical trials; economic considerations also affect this decision, creating differences in selection behavior across firms. Thus, if one measures innovation capability by the share of preclinical compounds that eventually become new drugs, one may confound selection with capability (e.g., Shaver 1998).
The contribution of this paper to the literature goes beyond flagging a selection problem and developing a method to cope with it. We focus on two questions that can be studied particularly well in the context of the pharmaceutical industry. The first is the long-standing comparison between the innovation performance of large and small firms (e.g., Arrow 1983, Holmstrom 1989, Acs and Audretsch 1990, Henderson and Clark 1990, Chesbrough and Teece 1996, Levinthal and March 1993). One view holds that, whereas large firms have an advantage in commercialization, small firms have more flexible organizations that enhance creativity and attract more inventive minds, which makes them more productive in R&D. A different view holds that economies of scale and scope, and learning advantages, make larger firms better at innovation (e.g., Henderson and Cockburn 1996, Macher and Boerner 2006). In this view, the comparative advantage of small firms may lie in exploring technologies and markets that established firms are unwilling to explore (Christensen 1997, Klepper and Thompson 2006, Giarratana 2004). We cannot directly measure comparative advantage. However, because older, established firms and younger biotech firms each perform R&D, we can examine if older and younger firms differ in their ability to develop compounds successfully. This is our measure of innovation performance.
Our second question is whether markets for technology (Arora et al. 2001) are afflicted by a "lemons" problem (Akerlof 1970). Are compounds offered for license of lower quality than compounds developed in-house? The current evidence is mixed. Danzon et al. (2005) find that licensing has a positive effect on the probability of success of drug compounds, whereas Pisano (1997) and Guedj (2005) find that licensed drugs are less successful than those developed in-house. Our paper tests these contending views.
The next section reviews the main findings of the empirical literature on drug development in the pharmaceutical industry and places our contribution in context. Section 3 describes the model, and [section]4 develops our hypotheses. Section 5 describes our data and analysis. Section 6 discusses our estimates. Section 7 discusses the robustness of our results and concludes.
2. Innovation in Pharmaceuticals: Related Literature
One group of studies on drug innovation focuses on the internal economies of the firm (economies of scale, scope, spillovers, experience). Henderson and Cockburn (1996) study the determinants of research (drug discovery) performance, measured by the number of important drug patents. They use data at the level of the individual research program from the internal records of 10 major pharmaceutical firms, and find returns to scale at the level of both individual research programs and research expenditures of the firm as a whole, as well as evidence of economies of scope. Our paper investigates the stages after drug discovery, as do most of the studies described below.
Nerkar and Roberts (2004) examine the determinants of commercial success of new pharmaceutical products. They find that proximal technological experience (patents in the same therapeutic class) has a positive and significant effect on first-year sales of a new product. Distal technological experience is positive and significant only when accompanied by a high level of distal product-market experience. Finally, they find that the interaction between distal and proximal technological experience is negative, which suggests that focused and diversified innovations may be alternative strategies. Nerkar (2003) also studies the relationship between experience and the probability that a drug receives U.S. Food and Drug Administration (FDA) approval. He hypothesizes that experience may not lead to better performance if the experience is of the wrong sort (e.g., oriented toward drug discovery rather than drug development), and if feedback is inadequate or delayed. With drug development cycles stretching to 10 years or more, researchers may be rewarded for discovery (measured by patenting) rather than commercialization (measured by FDA approval). As a result, scientists may continue to work in areas that provide them with clumps of patents but do not lead to commercially useful drugs. As we discuss below, even small firms may suffer from a misalignment of incentives, as reflected in their selection behavior.
A second group of studies looks specifically into the drug development process by examining projects in clinical trials. Adams and Brantner (2003) and Abrantes-Metz et al. (2006) analyze drugs in clinical trials around the world between 1989 and 2002. They find that success rates and durations can vary substantially across observable characteristics of the drugs, including primary indication, originating company, route of administration, and chemistry. Macher and Boerner (2006) also estimate time to complete clinical trials, but by looking at contract research organizations rather than integrated R&D projects in pharmaceutical firms. They find that scale and scope economies, and experience, are valuable.
Danzon et al, (2005) focus on the role of experience (both overall experience and experience in a particular therapeutic category) and alliances on the outcome of R&D projects. Their sample consists of R&D projects from more than 900 firms during 1988-2000. They too find evidence of large, positive returns to a firm's overall experience in the larger and more complex late-stage trials. They recognize, but do not explicitly model, differences across firms in the quality of their drug candidates.
Our paper shares the approach and goals of these papers. However, none of them explicitly address the concern that firms may systematically vary in how selective they are in advancing compounds into clinical trials. As is well known, clinical trials are much more expensive than preclinical trials, particularly late-stage clinical trials. Which compounds make it into clinical trials depends on whether managers think the compounds are promising. All else being equal, compounds with a higher likelihood of FDA approval are more likely to be advanced into clinical trials than compounds with a lower probability of success.
Commercial considerations are also very important. Acemoglu and Linn (2004) present evidence that larger pharmaceutical markets see more innovation. Furthermore, a firm with downstream marketing capabilities in a particular market (such as Bristol-Myers Squibb in anticancer or Merck in cardiovascular) may be able to extract greater value from a drug than a firm that lacks such capabilities, and may therefore be willing to take a chance on a compound with a lower likelihood of success. Other considerations may also be important. A drug championed by a high-status research team may be selected for clinical trials even if it is below threshold (Nerkar 2003). Guedj and Scharfstein (2005) point to the agency problem between managers and investors. In biotech startups, managers typically have one or two compounds to bet on, and therefore try to push them into clinical trials. By contrast, larger firms have many compounds from which to choose and are less likely to have a vested interest in any particular compound. As a result, biotech companies are more likely to advance products from phase I into phase II clinical trials, but these compounds are more likely to fail in later stages (Guedj and Scharfstein 2005). In our paper, we distinguish between preclinical and clinical trials, but do not separately analyze progress across the various clinical stages.
To appreciate further the contribution of our paper, it may be useful to compare it with Chandy et al. (2006), who examine the "conversion ability" of firms in translating drug-related patents into new drugs. They find that conversion ability is greatest in firms that develop an intermediate number of drug-related patents. Because firms patent compounds soon after discovery, Chandy et al. (2006), in essence, study the success rate of preclinical compounds. Our analysis "unpacks" this sequence into whether a compound is selected into clinical trials, and if selected, whether it successfully receives FDA approval, and models the drivers of the underlying processes. This enables us to move away from the assumption implicit in Chandy et al. (2006) that firms attempt to convert every compound. Instead, our model allows the stringency of selection to be driven by the expected profitability of the market, the scale and scope of the firm, and other technology, market, and firm characteristics, ensuring that the rate at which preclinical compounds succeed is determined not only by the factors that drive innovation performance but also by the factors that drive selection into clinical trials. In econometric terms, what allows us to identify selection is that the major resource commitments are made in clinical trials rather than in preclinical trials (DiMasi et al., 2003), so that we can assume that commercial considerations affect selection, but do not affect the outcome once a compound is into clinical trials.
Finally, the...
|
