`
`PHARMACEUTICAL INDUSTRY
`
`Fiona M. Scott Morton
`
`Working Paper 6190
`
`000001
`
`Exhibit 1 129
`
`Exhibit 1129
`IPR2017-00807
`ARGENTUM
`
`IPR2017-00807
`
`ARGENTUM
`
`000001
`
`
`
`NBER WORKING PAPER SERIES
`
`ENTRY DECISIONS IN THE GENERIC
`
`PHARMACEUTICAL INDUSTRY
`
`Fiona M. Scott Morton
`
`Working Paper 6190
`http://www.nber.org/papers/w6190
`
`NATIONAL BUREAU OF ECONOMIC RESEARCH
`
`1050 Massachusetts Avenue
`
`Cambridge, MA 02138
`September 1997
`
`Thanks are due to Peter Reiss, John Roberts, Garth Saloner, Andrea Shepard, Dimitri Vayanos,
`Matthew White, seminar participants at UBC, Northwestern, University of Chicago, University of
`Michigan, and the NBER 1.0. winter conference. I am also grateful to Merck and Co. for allowing
`me to gather data from their library. All errors are the responsibility of the author. This paper is part
`of NBER’s research program in Industrial Organization. Any opinions expressed are those of the
`author and not those of the National Bureau of Economic Research.
`
`© 1997 by Fiona M. Scott Morton. All rights reserved. Short sections of text, not to exceed two
`paragraphs, may be quoted without explicit permission provided that full credit, including © notice,
`is given to the source.
`
`000002
`
`000002
`
`
`
`Entry Decisions in the Generic Pharmaceutical
`Industry
`Fiona M. Scott Morton
`
`NBER Working Paper No. 6190
`September 1997
`
`JEL Nos. L65, L20, L21
`Industrial Organization
`
`ABSTRACT
`
`In this paper I use data on all generic drug approvals granted from 1984-1994 to examine
`
`whether heterogeneity among potential generic entrants can be used to predict which firms will
`
`choose to enter a particular market. The findings suggest that a firm’s portfolio characteristics,
`
`namely, its previous experience with a drug or therapy reduces the cost of preparing an ANDA and
`
`increases the probability of entry. A subsidiary’s parent ’s experience is not generally significant in
`
`predicting entry of the subsidiary. Firms also prefer entering markets that are similar, in terms of
`
`revenue and sales to hospitals, to markets already in their portfolios. On both scientific and
`
`marketing dimensions, the evidence shows that firms are specializing.
`
`I explore several different
`
`ways of constructing the set of potential entrants and find that the results are not affected by
`
`methodological variation. Standard IO theory suggests that profits per entrant will decline in the
`
`number of entrants. Previous research has found that generic prices depend on the number of generic
`
`entrants, and the results presented here show that the total number of entrants increases with the size
`
`of the market (revenue). These findings imply that generic firms face a negative competition
`
`extemality which makes their expectations about who else might be planning to enter any given
`
`market important in the entry decision. The limited evidence on entrant beliefs supports this
`
`conjecture as do several features of a regulatory upheaval when firms began entering different
`
`markets than they had in the past.
`
`Fiona M. Scott Morton
`
`Graduate School of Business
`
`University of Chicago
`1101 East 58th Street
`
`Chicago, IL 60637
`and NBER
`
`fionasm@gsb.uchicago.edu
`
`000003
`
`000003
`
`
`
`I. Introduction
`
`A firm’s decision to enter a particular market is one of the most important economic actions in a
`
`market economy. The number of firms in a market and distribution of market share have long been
`
`known to affect price levels and consumer welfare. Most research in this area uses the convenient
`
`assumption of symmetric firms, although this is of course not a good representation of reality. In
`
`contrast, this paper takes explicit account of heterogeneity among potential entrants to predict which
`
`firms are likely to enter which markets. In particular, it examines the entry choices of heterogeneous
`
`generic pharmaceutical firms and finds that they specialize along both scientific and marketing
`
`dimensions. The history and experience of a firm that lead it to enter particular markets can be thought of
`
`as firm ‘capabilities,’ in the sense that word is used in the business press. This industry provides a setting
`
`where a firm’s capabilities can be explicitly measured and the result of using existing capabilities or
`
`developing new ones can be observed.
`
`The entry decision is complex because the number of firms in a market affects the payoff to any
`
`one of them from entering that market; each entrant creates a negative extemality for the others that can
`
`be severe. Profits earned by an entrant firm therefore depend on entry decisions of other firms. Entrants
`
`sink entry costs simultaneously because firms do not typically announce their entry plans and the FDA
`
`does not reveal whose application it has received. The timing of the game, combined with research
`
`showing generic prices (and presumably profits) depend on the number of generic entrants, implies that
`
`generic firms face the difficult problem of how to form expectations about where others will enter. Those
`
`expectations will affect its own entry decisions.
`
`The question of which firms are expected to enter -- as well as do enter —- which markets in a
`
`simultaneous game is important. For a generic pharmaceutical manager making entry decisions for his or
`
`her firm, it is clearly a crucial problem. I discuss and examine how a generic pharmaceutical firm might
`
`form expectations of rivals’ actions and what firm equilibrium strategies might be. I argue that repeat
`
`players may use an entry strategy that provides stability of expectations: specialization. Specialization
`
`based on both scientific and marketing characteristics is natural because it reflects lower costs and
`
`provides a well-understood way to form conjectures about where competitors will enter.
`
`It is possible to conduct an empirical study of firm decision-making in the generic
`
`pharmaceutical industry because entry regulations create relatively good experiments and the regulatory
`
`agency, the Food and Drug Administration (FDA), generates data that are available to researchers. In this
`
`paper I use data on all generic drug entries from 1984 to 1994 to examine entry patterns and
`
`3
`
`000004
`
`000004
`
`
`
`specialization. In particular, I explore whether generic entrants are choosing markets based on past
`
`experience as measured by characteristics of their portfolios. I find that a firm’s previous experience with
`
`a drug or therapy increases the probability of entry into a similar market. The experience of a firm’s
`
`parent on various dimensions is generally not helpful in predicting entry, above and beyond the firm’s
`
`(subsidiary) experience. Marketing similarities between the entry opportunity and characteristics of the
`
`firm’s portfolio such as market revenue and hospital share are also important in explaining entry.
`
`Additionally, I show that larger markets, those that attract more entry, are markets with more sales to
`
`hospitals and those where the drug treats a chronic condition.
`
`In 1989 a major scandal erupted when various illegal practices were uncovered in the generic
`
`drug industry. I present results showing that the subsequent regulatory upheaval re-weighted the
`
`components of entry cost and disrupted established industry practices, including the pattern of
`
`specialization. Firms began to enter markets that looked different, rather than similar, to markets they
`
`were already in.
`
`II. Institutional Framework and Timing
`
`A firm that invents a new drug must get approval from the FDA by showing the drug is safe and
`
`effective. A New Drug Application (NDA) reports tests showing safety and efficacy and is typically
`
`expensive to construct and takes many years to be approved. A firm taking this route is called an
`
`innovator and the product is typically promoted under a proprietary brand name. In 1984 the
`
`pharmaceutical regulatory regime was significantly altered by the Waxman-Hatch Act. This legislation,
`
`among other things, allowed generic firms to submit Abbreviated New Drug Applications (ANDAs) for
`
`drugs approved since 1962. A flood of new ANDAs was filed in response to the law. The advantages of
`
`the ANDA process are summarized in the quotation below.
`
`“The benefit of the ANDA process to generic manufacturers is that it does not require these
`
`companies to repeat costly clinical and animal research on active ingredients or finished dosage
`
`forms already found to be safe and effective. A generic drug must contain the same active
`
`ingredients; be identical in strength, dosage form, and route; be bioequivalent; and be
`
`manufactured under the same strict standards as the brand-name drug to gain FDA approval.”
`
`(Frost and Sullivan report (1994))
`
`4
`
`000005
`
`000005
`
`
`
`Some firms submit both NDAs and ANDAs, but the vast majority specialize in one or the other.
`
`The other main type of firm in the industry is a “generic firm” that sells generic products. Figuring out
`
`how to manufacture the drug and performing the bioequivalency studies required for the ANDA can take
`
`from several months to a couple of years depending on the formulation of the drug, the effort and skill of
`
`the firm, and the availability of good suppliers. The application process requires factory inspections
`
`from the FDA and independent laboratory tests of several preliminary batches of the product. A firm
`
`must therefore be ready to make the product -- new equipment must be purchased and operational, for
`
`example -- months before the firm is legally permitted to begin selling it.
`
`The time between submission of the ANDA and granting of approval from the FDA averaged
`
`about seventeen months over the 1984-94 period, although annual averages have varied greatly over the
`
`last decade (Scott Morton (1996)). Therefore, a firm must start applying for permission to manufacture a
`
`specific drug two to three years before its patent expires if the entrant wishes to be active in the market
`
`immediately following patent expiration. Historically, many firms have not applied in a timely fashion
`
`despite the well-known fact that a firm that is first into a market and the only generic for even a few
`
`months earns large profits. A useful rule of thumb is that generic markets for most drugs do not last very
`
`long compared to the product life of a patented and branded drug. By the time all the patents on a drug
`
`have expired, it is relatively old technology and superior therapeutic substitutes are very likely to have
`
`been invented. Demand for the drug in patient days and revenue begins to fall after patent expiration;
`
`not many drugs remain major markets five years after patent expiration. (See Caves et a1. (1991) and
`
`Stern (1995).) Thus, we should not see profit-maximizing generic firms wasting any time in entering a
`
`market —- once the relevant patents expire -- and the model will not build in an option for delay.1
`
`The FDA does not reveal what applications it has received from which firms; no one can receive
`
`information on pending applications from the FDA. However, firms may announce their own intentions
`
`or actions. An (admittedly incomplete) examination of annual reports and interviews suggests that
`
`announcing is rare. Industry participants claim they do not want to reveal that they think a particular
`
`market is a good one to enter. Since all firms have access to essentially the same information, this claim
`
`does not seem to be very convincing. However, another firm’s opinion could function as an additional
`
`signal of value for what could be an unknown, but common, value. Additionally, although a firm might
`
`announce it has applied for a particular drug, there is no guarantee approval will be granted in the time
`
`frame anticipated by the applicant. The announcement is not credible. The firm could still be missing
`
`some tests or have a sloppy application. At the last minute, the FDA could decide a trial did not meet
`
`1 Bolton and Farrell (1990) model an option for delay in their entry game.
`
`5
`
`000006
`
`000006
`
`
`
`FDA standards and must be redone; or the factory might fail a late inspection. A year’s delay (which
`
`could easily happen due to regulatory uncertainty) can cause an application that was submitted and
`
`announced to be approved after that of a competitor responding to the announcement. A firm therefore
`
`cannot precommit to a market with an application announcement. The result of this lack of announcing is
`
`that generic firms are effectively sinking entry costs simultaneously.
`
`The pharmaceutical industry is characterized by high fixed costs (invention) and low marginal
`
`costs (production). Although an ANDA and the research involved are much less costly than basic
`
`research and NDAs, entry cost is nevertheless significant for a generic drug project where there is likely
`
`to be vigorous price competition. The average size of brand markets in the sample that later attract one
`
`generic entrant is $22 million. Generic products usually capture about half of molecule volume, although
`
`at prices 30-50% lower than the brand price. Thus per firm generic revenues are likely to be less than $10
`
`million per year, perhaps as low as $5 million for a product (perhaps with multiple concentrations). In
`
`interviews I have been told that filing an ANDA costs one quarter of a million dollars, twenty million
`
`dollars and various figures in between. This order of magnitude matches a one-entrant market size,
`
`although ANDA costs clearly must vary across drugs and firms. Generic drug industry participants
`
`exhibit great concern over excess entry, falling prices, and the effect on profits. Stories of unexpected
`
`failures due to the appearance of extra entrants are common. If fixed and sunk costs were zero, marginal
`
`cost pricing would provide an acceptable rate of return, no one would be willing to charge less than
`
`marginal cost, and manufacturers would not worry about making a drug on which they are losing money
`
`overall.
`
`I therefore make the assumption that fixed sunk costs are an important component in the
`
`project’s budget.
`
`The FDA imposes no requirement to produce once an ANDA is granted. Thus an unused but
`
`approved ANDA is an option, available to be used if prices rise, rivals exit, or the market becomes more
`
`attractive for any reason. Exit -- formal withdrawal of the ANDA -- is not an attractive alternative unless
`
`price is expected to be below marginal cost for the life of the drug (where marginal cost includes the
`
`opportunity cost of using equipment to make other drugs). There are several categories of ANDA
`
`withdrawal: the firm may discontinue the product, the product can be withdrawn by mutual consent, the
`
`firm might violate an FDA standard so that the FDA withdraws the ANDA, or the product may be found
`
`not to be efficacious (all NDA and ANDAs for that product are withdrawn). The last two (and maybe
`
`three) reasons are cases where the FDA is forcing the firm to exit, rather than situations where the firm
`
`chooses to leave the market because it is unsatisfactory. An ANDA can be sold or transferred easily with
`
`the physical production site, or if the original manufacturer becomes a contract manufacturer for the new
`
`owner. Buying an ANDA without its factory means new tests and inspections must be carried out to the
`
`6
`
`000007
`
`000007
`
`
`
`FDA’s satisfaction; this is not necessarily faster or cheaper than starting from scratch. Thus entry by a
`
`particular firm is a fairly irreversible decision; the costs that can be recovered upon exit from one drug
`
`(and not a factory) are close to zero.
`
`In 1989 what became known as the “generic scandal” broke out. Investigations by the US
`
`Attorney’s office in Baltimore began in 1988, and by the end of 1989 had uncovered several cases of
`
`bribery in the generic drug approval process. Four reviewers at the FDA were found to have been taking
`
`bribes in return for speeding approval of the bribing firms’ ANDAs.2 Additionally, some firms were
`
`found to have submitted the original branded product as their own in tests designed to compare a
`
`potential generic to the brand. The “generic” drug would pass the test with flying colors.3 In one case a
`
`generic firm was actually selling the recoated branded product.4 Later in the investigation, violations of
`
`manufacturing rules were also uncovered. The fallout from these discoveries greatly affected approvals
`
`and manufacturing in the following several years. The FDA’s Generic Drug Division fired many
`
`reviewers and the remaining reviewers proceeded very cautiously and slowly. New reviewers were
`
`hired, ethics standards promulgated, and the division was reorganized. ANDAs granted from 1989-1993
`
`took years longer than usual to be approved.
`
`Additionally, although this was not a concern of the FDA investigation, it became known that
`
`industry participants had been bribing employees of rival firms for information on approval activity.
`
`Many firms were engaging in gossip and inquiry into the entry plans of rival generic firms. The people
`
`in the industry included in my informal poll claim much of this activity came to a halt at the time of the
`
`FDA scandal and investigation because managers were worried that their behavior would be interpreted
`
`as unethical or illegal. This left firms with much less information on the entry plans of competitors.
`
`The discovery of irregularities in the approval process led to a review of what are known as
`
`“good manufacturing practices.” Many more, and stricter, inspections of manufacturing facilities were
`
`carried out than had gone on previously. Many firms were forced to withdraw products from the market
`
`for a period of months -- years, in some cases -- until their manufacturing satisfied the FDA. During the
`
`time the products were off the market the firms were upgrading their manufacturing process, suppliers,
`
`or procedures and training. The regulatory changes took place over several years and involved political
`
`actors from Congress, Health and Human Services, the FDA, and law enforcement. The criminal
`
`2 The investigation was launched because Mylan became suspicious that its drugs were not getting approval as
`quickly as competitors and found incriminating evidence in the garbage can of their FDA reviewer. (Chicago
`Tribune, Nov. 22, 1992)
`
`3 The independent labs doing the testing noticed when the coating chipped off, the brand logo was visible!
`4 This hardly seems like a profit-maximizing strategy.
`
`7
`
`000008
`
`000008
`
`
`
`investigation was still going strong in 1992, although most convictions had been secured by then.5
`
`Overall, the period from 1988 to at least 1992 was a time of great upheaval and uncertainty in the generic
`
`drug industry.
`
`III. Model
`
`The entry game and the role of competition and fixed costs has been addressed by a number of
`
`researchers (Bresnahan and Reiss (1988, 1991b), for example). These models predict the total number of
`
`entrants in a market, partly because this piece of information is of great interest to 1.0. economists and
`
`policy makers. However, the issue of exactly which potential entrant ends up entering is undetermined in
`
`many game theoretic models of entry. Typically, entrants are modeled as drawn from a limitless pool of
`
`identical potential entrants, probably an unrealistic assumption for most markets. Often it is difficult to
`
`measure and model heterogeneity among potential entrants; additionally, heterogeneity has a strong
`
`impact on form of the entry game. Berry (1992) addresses the question of which entrants enter which
`
`markets, as well as how many entrants enter a market, in his study of the airline industry. However, the
`
`generic pharmaceutical industry is characterized by simultaneous entry, rather than sequential, as in the
`
`Berry model. The following simple model illustrates its main features.
`
`The game has two periods. In the first period firms choose whether or not to enter specific
`
`markets by sinking their fixed costs. In the second period production and sales determine net profits.
`
`Profits will depend on the number of entrants in two ways. First, equilibrium price will decline in the
`
`number of entrants. (Price will hereafter mean generic price.) Despite the fact that the product is very
`
`homogeneous and the industry looks like it should exhibit Bertrand pricing, several studies have
`
`documented that generic prices fall steadily with the number of generic firms (Caves et a1. (1991), Frank
`
`and Salkever (I995), Wiggins and Maness (1995)).
`
`Secondly, an increasing number of firms means smaller quantity sold for each, all else equal. I
`
`assume a molecule’s market size is fixed because products are relatively old by the time their patents
`
`expire; the brand has expanded the market as much as it can. The drop in price due to the introduction of
`
`generics does not increase total quantity sold of the molecule, probably due to the price-insensitivity of
`
`doctors (see CWH (1991)).6 However, lower generic prices will increase generic quantity at the expense
`
`of brand quantity. If total generic quantity sold is Q, and n is the number of generic firms, I assume Q/n
`
`falls with n, despite the expansion in Q due to lower prices. The important feature of the model is that
`
`5 Chicago Tribune, Nov. 22, 1992 p 13A
`
`8
`
`000009
`
`000009
`
`
`
`profits decline with an additional entrant. This will hold, and the coordination problem will be
`
`particularly important, when both price and quantity sold per firm decline with additional entry. The
`
`expression below represents total profits for firm i earned in market j.
`
`U"! = (Mn) —Ca')qij(n) - Fa
`
`where price (p) and quantity (q) depend on the number of firms (n) in the manner described. Cij and F,j
`
`denote the marginal cost and fixed entry cost respectively that are specific to firm i and market j.
`
`The idea here is that there is heterogeneity across firms and it shows up in firm costs. The model
`
`will produce the same entry rule whether firms differ in marginal cost, fixed cost, or both.7 For the
`
`purposes of discussion, for the rest of the paper I will assume that it is the fixed cost of entry that differs
`
`across firms and markets, while marginal costs are constant across firms.8 Firms differ in the skills of
`
`their research group, suppliers, equipment, etc. which will affect ANDA cost. Drug markets also vary in
`
`how difficult a drug is to formulate and test, and therefore how much an entrant will spend to enter. The
`
`basic implications of this sort of model are that in larger markets more firms will find it profitable to
`
`enter, each additional entrant will lower the net profit of all other entrants, and lower cost firms should
`
`enter where higher cost firms should not. The analysis in the paper will test these implications.
`
`In equilibrium firms choose markets to enter based on differences in past experience. Each firm
`
`has a fixed cost Fi that is observable and common knowledge. Strategies are enter if Fi<F* and stay out if
`
`Fi>F*. F* is defined such that firms with costs higher in the distribution than F* earn negative profit
`
`when all firms with Fi<F* enter. No mistakes occur under this scheme because each firm knows where it
`
`lies in the realized distribution of fixed costs for the market and will not enter if there is not “room” for
`
`it. The postulated strategies form a Nash equilibrium because the low cost entrants will earn non-
`
`negative profit when the high cost firms stay out and therefore prefer to enter, while the high cost firms
`
`6 The elasticity of demand is clearly not zero at all price levels. However, in the observed price range, elasticity of
`demand for a molecule is very low.
`7 Later in the paper, I use drug experience variables to proxy for fixed cost differences across firms. Experience
`variables will also pick up any experience-based differences in marginal cost. Other marginal cost differences come
`from the batch production nature of pharmaceutical manufacturing. The marginal cost of a batch (rather than a pill)
`includes the opportunity cost of capital. A firm with strong demand for more drugs than it can produce at one time
`has an attractive outside option, production of one of its other drugs, for any given piece of capital. Since the quality
`of the outside option will vary across firms and over time, the marginal cost of a batch will also vary. The problem
`of opportunity cost of a batch is an important factor in the firm’s decision to enter a market, and it is impossible to
`measure without detailed plant-level data. Finally, there will be some marginal cost differences are due to
`characteristics of the production process that I cannot measure or control for. These final two sources of marginal
`cost differences are unaccounted for in this study.
`
`9
`
`000010
`
`000010
`
`
`
`will only lose money if they enter and the low cost firms also enter, so they prefer to stay out. A firm’s
`
`fixed cost level will help predict entry in this game, as will the cutoff level for the market, F*.
`
`All firms can work out the cutoff point in the fixed cost distribution. The firm enters if its own
`
`fixed cost falls below that cutoff level.
`
`pr(entry y) = pr(Fg < Ff")
`
`where Fj* represents the cutoff fixed cost level for marketj.
`
`The model most closely related to this problem is that of Dixit and Shapiro (1986). They
`
`describe a case of N identical potential entrants vying for spots in a market that “holds” only M firms,
`
`and build a model based on mixed strategies.9 There is a probability of entry from which each firm can
`
`calculate its expected profits from entering or not (using a binomial probability density). The equilibrium
`
`probability of entry is derived from the indifference of a firm to choosing enter or not enter.10 Dixit and
`
`Shapiro point out that choosing a pure strategy equilibrium in which only M firms play “enter” is
`
`artificial unless the reason for choosing those M firms is incorporated into the game. The model and the
`
`empirics below do incorporate reasons for particular firms to enter, allowing the firms to do better than
`
`just flipping a coin, and allowing the researcher (and presumably competitors) to predict entry, although
`
`not perfectly.
`
`Firms bear considerable risk due to simultaneous entry in the standard model of competition
`
`described above. “Over-entry” results in a lower than expected market price, volume, and profits, and the
`
`ex post desire not to have entered the market in the first place. Price might stay above marginal cost, and
`
`yet below a level that yields zero profit on the project, without firms having the incentive to exit the
`
`industry. “Under-entry” of course, produces economic profits until additional firms are approved to make
`
`the drug (which normally would occur within two to three years). At any moment, each firm is trying to
`
`enter only “good” markets. Each is trying to be in a market with the equilibrium, or fewer than
`
`equilibrium, number of firms.
`
`————______________—__
`
`3 The correct assumption undoubtedly varies by market.
`9 Large markets cannot be served by only one firm because each firm has an upward sloping supply curve due to the
`opportunity cost of capital (using all its machines to increase output of one drug) and the risk associated with
`dedicating a large part of the firm’s capital to one drug.
`10 The probability of entry times the number of potential entrants can be greater or less than the number of fu'ms
`which can earn positive profits; which direction the inequality goes depends on the shape of the profit function.
`Dixit and Shapiro show that for convex profit functions (Coumot, for example) on average there will be over-entry.
`
`10
`
`000011
`
`000011
`
`
`
`This raises the difficult problem of how generic firms make entry decisions in conjunction with
`
`other firms when they cannot explicitly coordinate. Bolton and Farrell (1990) examine the classic
`
`problem of two firms deciding whether or not to enter a natural monopoly market. They are primarily
`
`interested in when a central planner can do better than the decentralized case of each firm deciding based
`
`on its own private information. The randomness inherent in not coordinating will result in “mistakes.” If
`
`firms could avoid those mistakes by using the services of a central planner, welfare would be increased.
`
`The increase in social welfare in their model comes from the reduction in “duplication” (both firms
`
`enter) and “delay” (both firms wait for the other to enter). These are analogous to the generic
`
`pharmaceutical industry where producer surplus is saved if the fixed cost of entry is not paid twice and
`
`consumer surplus is increased if the entry occurs quickly so that prices fall. However, Bolton and Farrell
`
`show that if private information, perhaps about the entry costs of the two firms, is sufficiently important,
`
`central planning will result in lower total surplus.
`
`It is difficult to isolate evidence of the negative externality problem from the ‘low-cost firms
`
`should enter’ story. If it is the case that an entrant’s profits depend on the number of other entrants in the
`
`market, then an entry decision cannot depend on a firm’s own characteristics alone. Because of oligopoly
`
`or competitive interaction, the entrant’s beliefs about the actions of other entrants will affect its entry
`
`decision.“ Each firm has an interest in not entering if it believes too many other firms will enter. Beliefs
`
`about other players’ strategies determines which equilibrium (of many) is selected. Any set of beliefs
`
`that coordinated firms’ actions would work: geographic, alphabetic, phases of the moon, etc.. However,
`
`beliefs are difficult to measure and I will be unable to use them in the empirical analysis.
`
`The cost story, on the other hand, reflects a situation where there are firms with sufficiently high
`
`fixed costs that they cannot earn positive profit even if given an entry slot; beliefs about who else will
`
`enter do not affect entry decisions. However, if any firm with costs below a certain threshold can earn
`
`positive profits in the market, then market size should not be a helpful predictor of entry. The reason
`
`more firms enter larger markets is because they can fit - despite the business stealing efforts of their
`
`rivals. Thus the significance of market size measures in predicting entry will provide some evidence for
`
`the existence of negative competitive externalities in the industry. I use observable characteristics of the
`
`firm and its portfolio to proxy for firm fixed costs of entry. Estimation relies on those firm characteristics
`
`as well as knowledge of actual and potential entry events and measures of market size such as revenue.
`
`11 Notice that beliefs matter only among firms that could make a profit if they were given an entry slot. For
`example, it is easy to construct an example of a natural monopoly where any firm in the cost distribution could earn
`non-negative profits as a monopolist, but none could earn non-negative profits in a duopoly.
`
`11
`
`000012
`
`000012
`
`
`
`IV. Data
`
`The data I use to examine entry behavior are ANDA approvals granted from 1984-1994. A firm
`
`becomes part of the dataset if it is granted an ANDA in the time period; a particular drug becomes part of
`
`the dataset if it is applied for in an ANDA approved during this time period. Each approved ANDA is an
`
`observation. The FDA provides the submission date, the approval date, the applicant name as well as
`
`characteristics of the drug such as ingredients, form, route (into the body), and strength.
`
`I recode the
`
`FDA’s detailed form and route descriptions into five basic categories according to the type of machinery
`
`needed to manufacture a drug with certain characteristics and the cleanliness standards required in the
`
`manufacturing facility. The first category is oral solid, which forms the bulk of the observations in the
`
`data. The second group is injectable drugs, third is topical preparations, followed by oral liquids, and
`
`then ocular drugs.12 Note that the ANDAS making up the dataset are not a sample, but the complete
`
`universe of ANDAS approved in the United States during the time period 1984-94.
`
`I have calculated or collected information on additional variables such as the firm's primary type
`
`of product (brand or generic) and the parent of the applicant (if it exists). A drug’s therapeutic class is
`
`assigned according to the description under “indications” in the 1993 volume of Physician’s GenRx, a
`
`comprehensive pharmaceutical reference. There are approximately thirty therapeutic categories in the
`
`dataset.
`
`I also include the date legal restrictions on entry into the drug expire. Simply including the last
`
`patent expiration date is not a good method because, first of all, there are usually several patents in force
`
`and it is not obvious to the uninformed observer which one is the binding patent. Also, the FDA may
`
`ha