Entry in Pharmaceutical Markets Margaret K. Kyle∗
December 2, 2002 Abstract The diffusion rate for new, patented technologies depends on the strategies implemented by innovators for entry into market segments. This paper examines the determinants of such entry, employing an international panel dataset on the pharmaceutical industry. A range of econometric models for a firm’s decision to launch a new pharmaceutical product in 21 OECD countries is estimated, with five main results. First, the likelihood of entry is increasing in the population of a country, but at a decreasing rate, and products are less likely to be launched in countries that regulate drug prices. Second, competition matters: one additional drug in a market reduces the probability of another’s entry by about 30%. Third, firm characteristics have a substantial impact on entry. The advantage to being a domestic firm is particularly strong. Fourth, drugs that compete in multiple therapeutic classes, that have been recently developed, and that are widely referenced in the scientific literature are more likely to enter a market. Finally, countryspecific effects remain. Certain firm and product characteristics have greater salience in some countries than in others, and the effect of competition also differs. The results suggest that the interaction between the innovating firm, drug, and target country is a critical component of profitability, and that regulations may not affect firms and products uniformly. In other words, even in an industry with the potential for ubiquitous licensing, the specific match quality between the innovating firm and idiosyncratic market conditions remains an important determinant of entry.
∗
Graduate School of Industrial Administration, Carnegie Mellon University,
[email protected]. This paper is based on Chapter 2 of my doctoral thesis. I thank my advisors, Scott Stern, Rebecca Henderson, Ernst Berndt, and Richard Schmalensee, as well as Iain Cockburn, Peter Davis, Jeffrey Furman, Judy Lewent, Robert Miglani, Stephen Propper, and participants at the 2001 CCC Conference, MIT IO Workshop, and the NBER 2002 Summer Institute for helpful suggestions. The Schulz fund at MIT provided funding for the data. I am responsible for all errors.
I.
Introduction The diffusion rate for new, patented technologies depends on the strategies implemented by innovators for entry into market segments. Economic theory contributes to our understanding of how many firms may compete in a market, stressing the role of market structure (e.g., Bresnahan and Reiss (1987, 1990, 1991) and Berry (1992)) and regulation (most recently, Djankov, La Porta, Lopez-de-Silanes, and Schleifer (2002)). In contrast, research in strategic management addresses which firms choose to enter and emphasizes firm characteristics, specifically differences in firm capabilities and resources (for example, Henderson and Clark (1990); Klepper and Simons (1997, 2000); and Teece (1987)). Joint testing of economic and strategic hypotheses is rare, largely because it requires a setting with a clear set of potential entrants and separate markets. One such setting is that of pharmaceutical markets, which are legally separated between countries and which apply to distinct medical conditions. This paper examines the influences of both market structure and of firm (and product) characteristics on the launch of new drugs around the world. Pharmaceutical markets provide an interesting empirical puzzle to explore.
Developed
nations differ from each other in the number of drugs that compete in a market as well as in the mix of available products. Over the past 20 years the US has had an average of three drugs (unique chemical entities) per therapeutic class, or medical condition for which a drug is prescribed. Italy, with a population of about 57 million, has an average of five drugs per therapeutic class. Switzerland has an average of four drugs per class for a population of just 7 million. Only one-third of the prescription pharmaceuticals marketed in one of the seven largest drug markets (the US, Japan, Germany, France, Italy, the UK, and Canada) are also marketed in the other six. This is a strikingly low figure given the size and wealth of these countries and the substantial trade between them, and since pharmaceutical firms should have incentive to spread the large sunk costs of drug development over as many markets as possible. In addition, some markets have no entrants at all, despite the availability of treatments in other countries. Economic theory suggests that entry is a function of market size, the level of competition, and the fixed costs associated with product launch. Research in strategic management suggests that firms are heterogeneous: some are better suited to a particular market than others. Disentangling these various effects is an empirical challenge, but one for which this setting is ideal. An identical product is launched (or not launched) in different markets, yielding three sources of variance to exploit: variation across countries, variation across therapeutic classes, and changes over time. 1
The entry patterns of pharmaceuticals are important to understand for several reasons. The cost of untreated conditions in markets with no entry may be substantial. In addition, there are many monopoly and duopoly markets. Competition usually results in lower prices, and given the widespread concern about the cost of pharmaceuticals, it is valuable to know what impedes further entry into a market. This study also contributes to the debate on the effect of regulations, particularly price controls, by examining their impact on the market structure of pharmaceutical markets within a country. Finally, understanding entry in this setting may provide insights into the diffusion of other new technologies, particularly those characterized by large development costs, relatively low marginal or transportation costs, and that are susceptible to creative destruction by subsequent innovators. There are five main results. First, the likelihood of entry is increasing in the population of a country, but at a decreasing rate, and products are less likely to be launched in countries with price controls.
Second, competition matters: one additional drug in a market reduces the
probability of another’s entry by about 30%. However, correct adjustment for market size is crucial. The importance of therapies varies substantially across countries, and failure to account for this heterogeneity in estimation leads to biased and misleading results.
Third, firm
characteristics have a substantial impact on entry. The home country effect is particularly strong, as is a firm’s local experience. Large firms are more likely to avoid price-controlled countries, and firms headquartered in different countries exhibit very different behaviors. This suggests that the match between the innovating firm and the target country is a critical component of profitability, and that regulations may not affect firms uniformly. Fourth, product characteristics affect entry in predictable ways. Drugs that compete in multiple therapeutic classes, that have been recently developed, and that are widely referenced in the scientific literature are more likely to enter a market.
Finally, country-specific effects remain.
Certain firm characteristics,
particularly domestic status, are more important in some countries than in others. Nations also differ from each other along dimensions of fixed costs that are hard to measure, such as regulatory stringency. These results are robust to a variety of assumptions about the error structure. In particular, estimating a very simple structural model of entry yields the same comparative statics as those obtained from reduced-form specifications. The following section gives a brief overview of the pharmaceutical industry and outlines regulatory regimes in the countries included in this study. Section III presents the rationale for examining market, firm, and product characteristics from the literature on empirical studies of entry and on the economics of pharmaceuticals. The empirical model is explained in Section IV.
2
Section V describes the data used in this research. Results are presented in Section VI, and Section VII concludes. II.
Description of Industry and Regulatory Regimes Expenditures on health care range from 5% of GDP in South Korea to over 13% in the US,
and the share of pharmaceutical sales in total health expenditures account for anywhere from 4% in the US to nearly 18% in France and Italy. The US is the largest single market at $97 billion of annual revenue; the five largest European markets amount to $51 billion, as does Japan.1 Table 1 provides revenues from the major markets and the distribution of revenues across broad therapeutic classifications. This table illustrates that the importance of certain therapies can vary substantially across countries. For example, nearly 22% of revenues in the US derive from drugs for the central nervous system, while in Japan this figure is only about 6%. Italian expenditures on anti-infectives are over twice those of the UK. The industry is highly fragmented: there are thousands of small firms around the world, only several hundred of which are research-based and have brought at least one drug to market. About forty multinational firms dominate the market. These firms, listed in Table 2, are responsible for half of all drugs available somewhere in the world and spent over $44 billion on research and development in 1999. Table 3 lists the number of firms in each major market, the number of drugs they have developed, and the average number of countries to which those drugs diffuse. The US is the origin of over a quarter of all drugs, and these products reach an average of about nine markets. Though many drugs are invented in Japan, they are launched in fewer foreign markets. Drugs with small domestic markets like Denmark, Switzerland, and the Netherlands spread to more foreign markets than drugs with large home markets. Pharmaceutical firms tend to specialize in certain therapeutic categories,2 and competition within therapies is relatively concentrated. A new drug is reported to require an average of 7.1 years to develop at a cost of $500-600 million.3 In 2000, pharmaceutical companies spent approximately $8 billion on sales and marketing and distributed samples worth an additional $7.95 billion in the US alone. 4 These markets differ on a number of dimensions, of which regulation is perhaps the most notable. The entry of pharmaceuticals is restricted and in most countries, so is the price. Each nation has an agency or ministry devoted to pharmaceutical evaluation, which have heterogeneous standards for establishing safety and efficacy and which vary in how quickly they 1
Figures are annual totals for 2000. Source: IMS Health. For a breakdown of the top twenty firms’ specializations, see DiMasi (2000). 3 Paraxel’s Pharmaceutical Statistical Sourcebook 1999, p. 49. 4 IMS Health Inc. 2
3
evaluate new drug applications. Some require that some clinical trials be performed on domestic patients and are less accepting of foreign data. Some European countries require proof of costeffectiveness. During the 1990s, mean approval times ranged from 1.3 years in France for 1990, to 4.8 years in Spain for 1991.5 In addition to differences in agency funding and bureaucratic efficiency, the number of drugs under review varies considerably across years and countries. There has been a gradual move towards harmonization of regulatory standards for all major markets, particularly within the European Union. Under the EU’s Mutual Recognition Procedure, enacted in 1995, a drug approved in one member state (the Reference Member State) must be granted marketing authorization in other member states (the Concerned Member States) within two months unless a Concerned Member State objects through a formal process. Another option is the Centralized Procedure, in which a drug is submitted to the European Medicines Evaluation Agency for marketing approval in all EU nations. However, the drug’s manufacturer must still negotiate with individual countries over price under either the Mutual Recognition Procedure or the Centralized Procedure. Price regulation has many variants. Most countries have adopted some form over the last thirty years or so. A few countries do not officially regulate prices, but may have considerable power in determining prices if the government, as the largest provider of health care, has monopsony power. For example., firms must negotiate price with the National Health Service in the UK. In countries with more explicit price controls, the government fixes the price for a drug based on some determination of therapeutic value, the cost of comparable treatments, the contribution of the drug’s manufacturer to the domestic economy, the prevailing price in other countries, and manufacturing cost; the weight given to each factor differs by country. In some cases, price controls apply only to “listed” drugs, those the government will reimburse through its public health insurance. Negotiations between pharmaceutical firms and national governments may be lengthy and tense, and drug companies often blame this process for delays in product launch. In a recent article in the Financial Times, Pfizer chairman Hank McKinnell stated “[w]e introduce our new products later and later on the French market, and if the government continues to put pressure on prices, there will be no more [new products].”6 Broadly speaking, northern European countries and the US have fewer or less intrusive price controls, while southern Europe has more extensive government intervention.7
5
Thomas et al. (1998), p. 790. “Drug companies hit out at French price controls,” Financial Times, June 10, 2001. 7 See Jacobzone (2000) for a detailed summary of regulations in each country. 6
4
An important consequence of price controls that relate the domestic price to the prices in foreign markets is that pharmaceutical firms now have incentive to launch their products first in countries where they have the freedom to set a higher price, since this will influence the price in markets with price controls. Price controls may have an additional effect in Europe through parallel imports, permitted between the 15 EU member states, which enable wholesalers to arbitrage price differences between EU countries. Launching a drug in a country with stringent price controls may depress global revenues if wholesalers in countries with higher prices purchase drugs in price-controlled markets for domestic resale. Several countries (South Korea, Mexico, Spain, and the UK) regulate the profits of pharmaceutical firms. The government negotiates with manufacturers and sets a rate of return according to complicated formulas accounting for operating costs, promotion expenditures, and R&D spending. During the 1990s, many countries also enacted price freezes or mandatory price cuts in response to the increasing cost of pharmaceuticals. Most Canadian provinces do not permit prices to increase by more than the rate of inflation, and the US Congress has threatened similar laws – and extracted non-binding commitments from major pharmaceutical firms to hold the rate of price increases.8 Three other sorts of regulation or government intervention are worthy of remark: policies on generic drugs, the use of demand-side controls, and restrictions on advertising. These are not explicitly considered here, but are described in Appendix A. Countries also differ in subtle non-regulatory aspects. The number and size of pharmacies are highly varied across countries, as are distribution and dispensing margins (see Figure 1). Physicians have diverse prescribing habits; in Japan, physicians both prescribe and dispense drugs, and they tend to prescribe lower doses than elsewhere in the world and combinations of drug therapies.
Consumer compliance and trust of doctors is multifarious.
Herbal and
“alternative” therapies are more widely used in Europe than in the US, though their popularity in the US is increasing. Finally, the practice of licensing products to one or more firms for launch is far more prevalent in some countries than others. It is particularly common in Italy, Spain, Japan, and South Korea. III.
Determinants of Entry Several general findings emerge from the literature on entry. Both market size and the degree
of competition influence the entry decision. The number of firms in equilibrium increases at a decreasing rate with the size of the market, and profit margins fall as the number of competitors increases (Bresnahan and Reiss (1987, 1990, 1991), Berry (1992), Scott Morton (1999)). Second, 8
Ellison and Wolfram (2000).
5
firms tend to enter in markets that are similar to those they already compete in. Berry (1992) shows airlines that serve one or both of the cities in a city pair market are more likely to enter that market. Scott Morton (1999) demonstrates that generic drug firms in the US tend to enter product markets that match well to their existing products. Finally, the match between a product and a market is important. For example, Mazzeo (1998) finds that competing motels strategically differentiate themselves from each other in quality space to soften price competition. All these studies of entry have the distinct advantage of requiring little or no data on price and quantity, which is often expensive and difficult to obtain. However, these authors relied on a single crosssection of markets, which precludes simultaneous consideration of market, firm, and product characteristics.9 Many prior studies on the pharmaceutical industry identify factors that should be important in the decision to launch a new drug. Competition in pharmaceuticals exists both within a chemical (branded versus generic, prescription versus over-the-counter) and between different chemicals that treat the same condition. The generic segment garners significant market share within a few years of patent expiration when entry occurs, but not all therapeutic classes (and very few countries) attract such entry.10 While many have shown that generic competition has indisputable significance (at least in the US), there is substantial justification for focusing on competition between drugs. In a recent paper, Lichtenberg and Philipson (2000) estimate the loss in sales from entry by new drugs for the same therapeutic classification and find that entry by such drugs reduces the PDV of a drug by considerably more than generics. These results are broadly consistent with other studies that emphasize the importance of intermolecular competition, such as Stern (1996) and Berndt et al. (1997). In the context of a study on the diffusion of innovation, the creative destruction of intermolecular competition is more interesting than generic competition, which exists only for older drugs. In addition to competition, the regulatory environment has a significant bearing on prevailing prices (Danzon and Chao (2000a, 2000b)). Countries with stringent regulation of entry combined with relatively little price regulation, such as the US and the UK, have highly concentrated domestic industries whose products diffuse more extensively into foreign markets (Thomas (1994)).
The one study that explicitly addresses international entry (Parker (1984)) shows
9
Toivanen and Waterson (1999) observe entry decisions over time into fast-food markets in the UK, but like Berry (1992), assume all heterogeneity is at the firm level. 10 Generic competition in the US is the focus of Caves et al. (1991) and Grabowski and Vernon (1992), among others. Hanson (2000) looks at the determinants of generic entry in the US, the UK, Germany, and Japan. Ellison et al. (1997), who estimate demand for a class of antibiotics, and Berndt et al. (1997), who examine the antiulcer market, consider competition both within and between drugs.
6
regulation is related to large differences across countries in the number and mix of products introduced before 1978. Thus, there is much reason to expect regulation to influence entry. Regulation also affects drugs and firms differentially within a country, particularly in the costs of gaining regulatory approval (Dranove and Meltzer (1994), Carpenter and Turenne (2001)). Product characteristics, like therapeutic novelty or indication, and firm characteristics, such as experience with the FDA and domestic status, are related to the speed at which a new drug receives regulatory approval in the US. Data from three other large pharmaceutical markets (the UK, France, and Germany) displays a similar pattern in time-to-market of important drugs, and reveals a strong home country advantage: the drugs of domestic firms are approved earlier than those of foreign firms. Beyond the non-uniform effects of regulation, there is substantial evidence of significant firm and product heterogeneity in research productivity (Henderson and Cockburn (1996) and Cockburn and Henderson (1994, 1998)), and Scott Morton (1999) finds evidence of important firm-specific differences in the entry decisions of generic drug firms. Firm-specific costs are therefore likely to be important in drug launches.
IV.
Model The model used in this paper assumes that potential entrants for a market take existing market
structure as given and compete simultaneously in time t. Let i index drugs, j index firms, k index therapeutic classes, and l index countries. A market is thus a class-country-year triple. Define the reduced-form profit function as
Π ijklt = N klt δ + M klt θ + X klt β + Z jklt γ + Wikt α + ε ijklt where N is the number of competing drugs in the market, M is the number of potential entrants, X is a vector of market characteristics, Z is a vector of firm characteristics, and W is a vector of drug characteristics.11 Firms enter if their expected profits are at least zero, and any firm that elects not to enter must expect negative profits from entry. Several estimation approaches are possible: an ordered response model, a discrete choice model, a duration model, and the method of simulated moments. All are implemented to ensure that the results are robust to different assumptions about the error structure, and to demonstrate consistency with the findings of previous entry studies. A. Ordered response model
11
Product quality is considered exogenous. Once a drug has been developed and tested, its efficacy is fixed: a firm cannot re-position a low-quality drug as a high-quality product. In reality, some “tweaking” is possible, such as once-a-day dosing formulations, but such changes are second order.
7
The number of competing drugs may be modeled as pr(# new) = pr(Z jklt γ + ε klt < (N klt + # new)δ + X klt β)
and estimated as an ordered logit (probit) if we assume the cumulative distribution of ε is the logistic (normal). This specification requires several assumptions. The number of potential entrants must be very large, and all firms or drugs must be equally profitable after entry. In addition, there must be no unobserved firm heterogeneity and observable differences between firms must be small. Essentially, entry by any particular firm is a zero probability event; this specification is appropriate for answering questions related to the role of market characteristics, but not ideal for examining entry decisions by particular firms. B. Discrete choice model/Discrete-time hazard The probability that a firm expects positive profits, and therefore chooses to enter, is pr(enter) = pr( ε ijklt < N klt δ + M klt θ + X klt β + Z jklt γ + W ikt α)
which can be estimated as a linear probability model, probit, logit, etc. where the dependent variable is 1 if entry occurs, 0 otherwise. This model may also be thought of as a discrete-time hazard model. The probability that a drug is launched during a time interval t can be written as P(t) = a(t) + N klt δ + M klt θ + X klt β + Z jklt γ + W ikt α
where a(t) is a series of intercepts for each year of a drug’s age (or time at risk). A convenient transformation for estimation is the logit, i.e. P(t) log 1 - P(t)
= a(t) + N klt δ + M klt θ + X klt β + Z jklt γ + W ikt α
This method has the advantage of being quite flexible as well as accounting for rightcensored observations. Both the discrete choice and discrete-time logit also require several strong assumptions. To include N as an explanatory variable, we must assume that one drug’s entry does not induce another’s exit. The justification for such an assumption is provided in Section V. If M is excluded as an explanatory variable, then the estimation is nearly identical to that of Scott Morton. In this case, one is assuming that future entry has no affect on profits. If M is included and treated as an exogenous variable, then the threat of future competition is allowed to affect current entry decisions, but one must believe that firms do not behave strategically.
This
assumption is highly suspect. The number of drugs developed to treat a condition is almost certainly a function of the global profits associated with that disease. Firms in an oligopolistic setting (which most drug markets are) are very likely to react to the behavior of their competitors. 8
C. Continuous-time hazard model An alternative to the discrete-time logit is a continuous-time hazard model. Many are possible; here, the Cox Partial Maximum Likelihood Estimator, or proportional hazards model, is applied. The hazard rate is modeled as
λ(t) = λ 0 (t)exp(N
klt
δ + M klt θ + X klt β + Z jklt γ + W ikt α)
where λ0(t) is a drug’s baseline hazard rate, or individual heterogeneity. This heterogeneity is conditioned out so that the probability of entry for drug i, a potential entrant for therapeutic class k in country l is12
Pr(entry ikl | risk set kl ) =
exp(N klt δ + M klt θ + X klt β + Z jklt γ + Wikt α) M
∑ exp(N
klt δ +
M klt θ + X klt β + Z jklt γ + Wmkt α)
m =1
Since drug launches are observed at annual intervals in this dataset, a discrete-time model is probably most appropriate. As the interval of observation becomes small, the results from a discrete-time logit converge to those from a proportional hazard model. 13 The use of a duration model does not, however, address the issue of the endogeneity of M and strategic behavior by firms. D. Method of simulated moments Accounting for the endogeneity of potential competition requires structural estimation with the method of simulated moments. Following Berry (1992), assume profits take the form
Π ijklt = X klt β + Z jklt γ + Wikt α − δ ln( N klt ) + ρu kl + σu klj where ρ is the variance of unobserved market heterogeneity and σ is the variance of unobserved firm heterogeneity. Since ρ and σ are not separately identified, the variance of unobservables is normalized to one so that σ= 1− ρ 2 . Berry (1992) demonstrates that while multiple equilibria exist if firms play a simultaneous game, the number of firms is the same across all of them if firms differ only in the level of fixed costs and if variable profits are strictly decreasing in the number of competitors. If the game is played sequentially rather than simultaneously, then there is a unique subgame perfect equilibrium under the same conditions. Davis (1999) establishes that three assumptions are required to obtain a unique equilibrium level of market output in a more general model of quantity competition. In this setting only two apply: first, no firm that finds entry unattractive
12
This formula assumes only one drug enters each period; the likelihood function used in estimation must be adjusted for cases in which more than one drug enters. 13 See Amemiya (1985), pp. 433-455, or Allison (1984) for a more complete discussion of duration models.
9
with N drugs would enter if there were more than N drugs in the market; and second, firms are never indifferent between entering and staying out of the market. With numerous potential entrants, maximum likelihood estimation is impractical for two reasons. The region of integration is tedious to define; with as few as three firms, the probability of one firm in equilibrium is the sum of six disjoint terms. In addition, the usual discrete choice computation problems arise when solving high dimensional integrals. The method of simulated moments is usually employed to avoid such complexities. If the entry game is sequential, then imposing an order-of-entry rule allows one to use additional moment conditions in the estimation. Estimation is done as follows. For a given set of parameter values, R draws are taken from the error distribution. The last (mth) firm to move in a market takes the decisions of all other firms as given and enters if the simulated profits from entry are greater than zero. The (m-1)th firm again takes the decisions of prior movers as given and assumes the best response of the mth firm, and enters if its simulated profits are greater than zero. This process continues for all potential entrants. The prediction error is computed as: νˆ ijklt = E ijklt −
1 R
∑ pˆ
ijkltr
r
pˆ = predicted entry decision E = actual entry decision
The number of potential entrants varies across markets so an exogenous selection rule is needed to choose a set of C estimating equations for each market,
νˆ klt = (νˆ klt0 , νˆ kltc , c = 1...C) νˆ klt0 = prediction error for number of drugs in market νˆ kltc = prediction error for entry decision of drug c The following objective function is minimized using the downhill simplex algorithm: G(θ ) = g(θ)′Ag (θ ) 1 g (θ ) = νˆ klt ⊗ f(Wklt ) K * L * T k,l, t
∑
A = weighting matrix W = exogenous data
A is first set to be the identity matrix I to obtain an initial consistent estimate of θ. Then A is set to the sample variance of the individual moment conditions and θ re-estimated, with variance
(1 +
1 )((∇ θ g n (θˆ ))′A∇ θ g n (θˆ )) −1 R 10
Future work will examine other research questions using a different specification for profits. For example, the existence of price competition, firm heterogeneity, and endogenous sunk costs would each result in the number of competitors increasing at a decreasing rate in the size of the market. A structural model that allows variable profits to be a function of the number of competitors and fixed costs to be a function of market size, and which estimates a random coefficient for firm heterogeneity, could distinguish between these competing explanations. Despite the strong and sometimes uncomfortable assumptions necessary, estimating a static model, such as the discrete choice or duration models discussed above, can provide insights into the sources of unobserved heterogeneity that may inform future research. The linear models are numerically stable and robust enough to estimate a large number of coefficients and fixed effects. The estimation here offers several advantages over previous work. The set of potential entrants is clear, so the dependent variable is reliably defined. Unlike most previous studies, which use a single cross-section, the panel structure of this dataset permits a much richer set of controls. This is the first paper in which the separate impacts of firm and product characteristics are estimated. It is also the first entry study of this type to focus on highly R&D-intensive products, the management of which is likely to be quite different from single-outlet, local firms with relatively undifferentiated products. V.
Data Information on all drugs developed between 1980 and 2000 is obtained from the
Pharmaprojects database, which is maintained by the UK consulting firm PJB Publications. This dataset includes the drug’s chemical and brand names, the name and nationality of the firm that developed it, the identity of licensees, the country and year in which it was patented, its status (in clinical trials, registered, or launched) in the 28 largest pharmaceutical markets, and the year of launch where applicable. Each drug is assigned to up to six therapeutic classes. The system of classification used by Pharmaprojects is adapted from the European Pharmaceutical Market Research Association; there are 17 broad disease areas (for example, dermatological conditions) and 199 more specific classes (such as antipsoriasis treatments). The sample of drugs used in this research is restricted to those that are new chemical or molecular entities by dropping new formulations of existing products, OTC licensing opportunities, antidotes, and diagnostic agents. The OECD Health Data 2000 dataset provides population, GDP, data on access to health care, and other demographic information for OECD countries. Pharmaprojects, 21 are also OECD members.
Of the 28 countries in
The regulatory structure of each country is
classified as “price control regime” using the summary tables from Jacobzone’s “Pharmaceutical 11
Policies in OECD Countries: Reconciling Social and Industrial Goals.” Table 4 lists the countries included in this study, whether they have price controls, and the year such controls were enacted. While this is a crude indicator of policy, the variety of regulations in these nations is difficult to categorize neatly. A market is defined as a country-therapeutic class-year triple. This definition assumes that drugs with the same therapeutic classification are substitutes, and that there is no substitution between therapeutic classes. Of course, the latter assumption is a strong one. Different classes of products may be appropriate for the same condition. A patient with migraine headaches might be prescribed a treatment specifically for migraines, an NSAID, or a narcotic; these represent three distinct classes. Other therapeutic classes may be complements – drugs that have nausea as a side effect are often prescribed in conjunction with an anti-nausea treatment, for instance. In addition, this market definition requires that there be no trade in unapproved products across international borders: launching a drug in the US must not enable access to the Canadian market. While the move to a common market in Europe weakens the assumption of separate markets, negotiation with health ministries is still necessary for the drug to be reimbursed. Competition from drugs approved in nearby countries but without local insurance coverage is probably weak. The entry decision is fairly easy to observe, although the launch year is missing from the Pharmaprojects data for 2409 (~20%) launches. For these observations, the launch year is assumed to be the year of first global launch plus the median delay for that country (number of years between initial launch and launch into the country). A drug is “at risk” for entry into all markets beginning in the year of its first launch into any country. After launch in a market, it drops out of the risk set for that country. Any drug that has been approved somewhere in the world for a particular therapeutic class is a potential entrant into that therapeutic class in all other countries. This set excludes drugs currently under development for that therapeutic class. Future work will address this shortcoming, but the main focus here is on the fixed costs associated with product launch, not with development costs incurred by firms while still testing a chemical. Exit is more problematic. It is observed in only 59 markets, usually the consequence of a regulatory order for concerns about safety. In general, exit is rare, since a drug may continue to be an important therapy even after its patent expires, especially in nations without a significant generic segment.
While a firm may reduce its advertising efforts for a particular drug, it
generally does not withdraw the product from a market. It is therefore assumed that there is no
12
exit for economic reasons.14 Possible obsolescence is controlled for in the estimation by allowing older drugs to have a different impact on entry than newer therapies. Drug quality, or the therapeutic advance a treatment represents, is likely an important factor in both the fixed costs of entry (if regulators accelerate approval of breakthrough therapies, or if regulatory approval is more difficult to obtain for a novel type of therapy with which regulators are unfamiliar) and in variable profits. Unfortunately, objective measures of quality are difficult to obtain. Previous studies have used the ratings of therapeutic novelty assigned by the FDA upon application for approval, but these are unavailable for drugs that did not seek entry into the US. Pharmaprojects also ranks drugs according to their novelty, but this ranking is retrospective, so a drug that represented a therapeutic advance at its initial launch ten years ago may be rated an established therapy in the current database. The “Essential Drug List” of the World Health Organization is another possibility, but it is updated infrequently and most of the drugs on the list are more than twenty years old. Therefore, this research follows Dranove and Meltzer (1994) in using Medline citations; the construction of variables using citations is described in Appendix B. Other aspects of drug quality are the number and severity of adverse interactions and side effects, dosage form, and dosage frequency. Systematic data on these characteristics is unavailable, particularly for drugs not marketed in the US. The inclusion of a drug fixed effect should mitigate the bias from omitting better measures of quality, and the results presented later are unaffected by adding such fixed effects. Two additional assumptions are that all therapeutic uses for a drug are known at the time of its initial launch and that a drug is approved for all its therapeutic uses in each country. Neither is strictly true. Additional therapeutic uses for a drug are occasionally discovered after a drug’s introduction. For example, doctors noticed that a side effect of minoxidil, a drug used to treat high blood pressure, is a reduction in hair loss in men; it is now a treatment for baldness. Other drugs have well-known therapeutic uses for which they lack regulatory approval, since obtaining such approval would often require an additional battery of tests for safety and efficacy or perhaps because such uses are politically sensitive.15 They may be prescribed for off-label uses in practice, however.
In estimation, the dependent variable is entry into the drug’s primary
therapeutic class, although the drug is considered a competitor in all of the therapeutic classes to which it is assigned.
14
However, Lichtenberg and Philipson present evidence that 18% of the drugs approved between 1970 and 1979 in the US are no longer marketed in 1999. 15 For example, Cytotec (misoprostol) may act both as an antiulcer treatment and as an abortifacient, but is approved only as an antiulcer treatment in the US.
13
Quantifying the regulatory barrier to entry, as well as the severity of price regulation, is nearly impossible. One indication is the time between application and approval of a drug. However, not only is this unavailable in all markets, but is also likely to be a function of drug quality, firm characteristics, the number of other drugs under review, and perhaps the decisions of regulators in other countries, and is therefore an imperfect measure. The existence of price regulation in a country is captured by a dummy variable, which obscures differences in the implementation of such policies, as described in Section II.
All regulatory variables are
vulnerable to endogeneity problems, as such policies may be reactions to (the perception of) high profits earned by pharmaceutical companies.
Only four countries (Canada, Mexico, the
Netherlands, and Sweden) enacted price controls during the sample period.
Other omitted
variables include the importance of generic competition within a country (or therapeutic class), the degree to which marketing of pharmaceuticals is regulated, the cost of marketing in each country, heterogeneity in prescribing behavior, and other subtle but important distinctions between countries. These effects are subsumed in the country fixed effects included in each regression, with the unfortunate implication that the estimated fixed effect for each country is the net impact of many variables. Table 5 presents summary statistics for data used in estimation. The sample contains 1577 unique molecules produced by 310 firms in 147 therapeutic classifications, for a total of 58,624 country-class-year markets. There were 466,333 entry opportunities, only 12,505 (2.7%) of which had a product launch. The mean number of drugs competing in markets with entry opportunities is 2.8. The distribution of the number of competitors over all markets is shown in Figure 2, both for the entire time period and as of 2000; Figure 3 shows the distribution across therapeutic classes within several countries over 1980-2000.
Most markets are highly
concentrated, and over one-fourth have no entry at all. Over 28% of all potential markets are empty in the US, even though it accounts for twice the revenues of Japan and Europe. The large fraction of “0” markets reflects both that some drugs are never launched in a country and that some drugs are only introduced years after they first become available elsewhere. However, even as of 2000, 15% of markets are empty. Variables measured at the drug-year level include age, the number of therapeutic classes in which it competes, the number of countries in which the drug has been introduced, and its share of the stock of domestic and foreign Medline citations for its therapeutic class. Figure 4 shows the distribution of the number of countries in which a drug has been launched as of 2000. Most drugs enter only one country, usually the domestic market. There may be economies of scale in global production, as clinical trial data is accumulated and used in subsequent applications, or if 14
regulators are exposed to less political risk in approving a drug that has already been accepted by their counterparts in other countries. The probability of entry is thus expected to be concave in the number of launch countries. A drug’s value should decline with age, due to the limited period of patent protection and competition from newer therapies, so entry is predicted to be convex in age. Drugs that compete in multiple therapies and important drugs that are the subject of many scientific studies should be more profitable; positive coefficients on these variables are expected. Several firm-level variables are included. International experience is the count of the number of countries in which the firm markets any drug. A firm with a presence in many markets may have more resources to draw on, which would make entry more likely. However, such firms may also be less dependent on any single market and therefore be more selective in the timing of their launches. Thus there is no clear prediction for the effect of international experience. A firm’s experience in a country is defined as the count of drugs it markets in that country, and its experience in a therapeutic class is the count of other drugs it produces for that class. These capture economies of scope: experience with the regulator and the presence of a detailing force and distribution channels may be spread across all a firm’s products within a country, and there may be benefits to specialization within a therapeutic class. The number of drugs a firm has within a country-class market measures expertise in the local market. All firm variables apply to the innovating firm, which may license a drug to another firm for marketing in particular countries. If licensing were completely efficient, then the characteristics of the innovating firm would be irrelevant; only the market and product characteristics should matter. To the extent that licensing markets work well, this specification is biased against finding any significance on firm-level variables. Finally, country-level demographics provide rough measures of market size and demand. Ideally, incidence rates at the level of country-class would be included, but these are difficult to obtain and may also be endogenous if pharmaceuticals reduce the occurrence of disease. Instead the stock of Medline citations is computed for each therapeutic class authored by foreigners (as a measure of the global importance of the therapeutic class) and authored by domestic scientists (as a measure of the local importance). In general, additional country-level variables such as the number of doctors per capita, pharmaceutical spending, and life expectancy proved insignificant16 and so only a parsimonious set of variables is presented here.
16
This is likely because what these variables measure is unclear. A long life expectancy may indicate good health, but does this reflect low demand (healthy people don’t need drugs, so little entry) or is it the result of available treatments (lots of entry)? In addition, once demeaned by country and year, these measures have little variation.
15
VI.
Results Results from two ordered logits, four linear probabililty models, and two Cox proportional
hazard models are presented in Tables C1-C3 in Appendix C. In each case, the full set of market, firm, and product characteristics are included; the specifications vary in whether country and firm fixed effects are estimated. A probit or logit is generally preferable to the linear probability model with a discrete dependent variable. However, the large number of observations and fixed effects included in this application make maximum likelihood procedures difficult to implement. To ensure the LPM’s incorrect specification of the error structure is not driving the results, the model is estimated as a logit on smaller datasets for several random samples of therapeutic classes and the G7 markets and using Sequential Least Squares17 on the full sample. These robustness checks are presented in Appendix C; the results are little changed by assumptions about the error structure. The discrete-time logit yields similar results, though many country-therapeutic class fixed effects cannot be estimated; these results are available from the author on request (these will be included in future versions). All results from the LPM specifications are robust to the inclusion of a drug fixed effect and potential competition as an explanatory variable (which is suspected of endogeneity). 18 The effect of country characteristics Table 6 presents parameter estimates for country characteristics from an ordered logit (OL 1),
two linear probability models (LPM 1 and LPM 2), and a duration model (Hazard 1). Models without country fixed effects are discussed so that we may focus on the between-country variation. LPM 1 constrains price controls to have the same effect on all drugs or firms. LPM 2 includes interactions between price controls and dummy variables for the country in which the drug’s innovator is headquartered. The results from all four specifications indicate that drugs are more likely to be launched in large, rich countries. Consistent with other studies of entry, the number of drugs in a market is increasing in population, but at a decreasing rate, demonstrating that profits are concave in market
17
See Horrace and Oaxaca (2001) for details of this estimation technique, including conditions under which SLS is consistent and may outperform probit or logit models. 18 The standard errors of the LPM should be corrected for the heteroskedasticity that is a consequence of the discreteness of the dependent variable. Doing so requires dropping all observations for which the predicted probability is outside (0, 1), which is over 30% of the sample (the mean of the dependent variable is only .025). Using White standard errors to correct for heterogeneity is inefficient since it does not incorporate all information about the LPM’s error structure.
16
size.
Wealth is also positively related to entry.
The measures of the importance of the
therapeutic class are rather noisy and their net impact difficult to assess. Price controls appear to discourage entry. The negative coefficient on the price control dummy in OL 1 indicates that countries with price controls are more likely to have no drugs in a therapeutic class than are countries with free pricing. Similarly, the estimated hazard rate for price controls from Hazard 1 is significantly less than one, meaning that entry into price controlled markets is delayed relative to countries with free pricing. The effect of price controls differs substantially depending on a drug’s origin.
Italy and Japan have large domestic
pharmaceutical industries that tend to launch products only in their home countries (a point that will be discussed further below). Both countries also control drug prices, so the effect of price controls is masked if constrained to be the same for all firms, as in LPM 1. It is clear from LPM 2, however, that price controls have a significantly negative effect on entry by Swiss (the omitted category), British, and American firms: the implied change in probability ranges from 17% to 35% from the mean for drugs from those countries. The effect of competition Parameter estimates for variables measuring competition are displayed in Table 7. LPM 3
demonstrates the consequences of mismeasuring market size.
There is much unobserved
heterogeneity in the importance of therapies within countries. The probability of entry appears to be increasing in the number of current competitors in specifications that control only for year and cross-sectional effects. This specification overlooks the fact that the number of drugs in a market reflects underlying country-specific demand for a therapy, which is inadequately captured by population and wealth. As an example, consider lafutidine, a new antiulcer medication developed by a Belgian company, which was launched in Japan but not in the US. While quite a large market in both countries, Japan already had 18 antiulcer treatments, compared to 9 in the US. However, the Japanese have a much higher rate of stomach cancers and other gastrointestinal disorders,19 so demand for antiulcer drugs is especially large relative to other countries. Without accounting for the difference in demand for antiulcer treatments between these two countries, one would erroneously conclude that entry is more likely in markets with more competition. If country-therapeutic class interactions are included, as in LPM 4, the coefficient on the number of competitors – both recent entrants and older drugs – is negative and significant. One additional competitor that has entered the market within the previous five years reduces the probability of entry by about 30% from the mean of .025. Competition from older drugs appears 19
Source: Merck Manual.
17
to have an even greater impact, perhaps because brand-name capital takes time to develop or because doctors have “sticky” prescribing habits. The coefficients on the squared terms of the number of competitors are positive, indicating that the decline in expected profits is steeper when the number of competing drugs is low. This is to be expected: if each drug takes 1/N of the market, where N is the number of drugs, then moving from monopoly to duopoly is a greater loss of profits than the difference between nine and ten competitors. Results from Hazard 2 are similar. To make estimation feasible, it is estimated on a random subset of therapeutic classes, which reduces the number of country-therapeutic class interactions. The coefficients display the same pattern as those from LPM 4 in terms of directions of effects and significance. Estimating a logit on a subsample yields the same qualitative conclusions (see Appendix C). These results are robust to many assumptions about the structure of the error term. Markets with many potential competitors experience more entry, which is consistent with the hypothesis that firms have heterogeneous costs. Such markets get more idiosyncratic draws from the distribution of firm fixed costs and therefore experience more actual entry in expectation. As discussed earlier, this variable is endogenous if pharmaceutical companies behave strategically, so that the actions of one firm influence those of its competitors. It is also endogenous if pharmaceutical firms invest in the development of treatments for diseases with great profit potential. The measure used here includes only drugs that have been launched in at least one country, so that the bulk of their development costs have already been sunk; perhaps this mitigates the latter source of endogeneity. Excluding potential competition from the regression does not affect the estimates for other variables, so its inclusion is relatively harmless at least in terms of introducing bias. One solution to this endogeneity is the use of the method of simulated moments. Structural estimation is generally feasible only with a reasonably small dataset and a limited set of parameters.
Excluding some parameters from estimation introduces additional sources of
variance, which may swamp the explanatory power of the model. It is therefore important to select a set of markets that are relatively homogeneous on the dimensions captured by excluded parameters. The results from estimating entry in a subset of five therapeutic classes over four five-year time periods are shown in Table 7b. This specification is estimated for anti-ulcer drugs, osteoporosis drugs, anti-epilepsy drugs, anti-migraine drugs, and anti-asthma drugs and all 21 countries. The first column presents results from estimating a linear probability model on this subset of data. These results are comparable to those obtained from estimating the linear probability on the full dataset, except that country-specific demand heterogeneity is smaller for this particular set of 18
therapeutic classes, so that fewer fixed effects need to be included. The second column shows the parameter estimates resulting from estimating entry using the method of simulated moments assuming that firms play a sequential game and move in order of experience within a country. The signs of the estimated parameters are in the expected direction and consistent with those from the reduced form models. The expected number of new entrants does significantly affect a firm’s entry decision. Failing to account for this endogeneity apparently does not introduce a large bias in estimating the relative importance of other factors. The effect of product characteristics Table 8 displays the coefficients of variables capturing product characteristics. As would be
expected given the limited period of patent protection, results from LPM 4 show that the probability of entry is convex in the age of a drug, implying large benefits to quick launch that are rapidly eroded with time. Both the LPM and the duration model indicate that profits are concave in the number of countries in which the drug has been launched. This result is consistent with firms introducing their products first in the most profitable countries, and with economies of scope from clinical trials or other data required for regulatory approval common to many countries. Drugs that compete in multiple therapeutic markets are more likely to be launched. Because the local and global importance measures are very highly correlated (ρ=.9), the local measure was first regressed on the global measure and the residuals used in the full regression. The probability of entry is increasing in both measures: the larger a drug’s share of the scientific work in its therapeutic class, the more likely its launch. However, the coefficients on the interactions of the local importance variable with the existence of price controls are negative (for LPM 4) or less than one (for Hazard 1), though imprecisely estimated. This result suggests that important drugs avoid price-controlled markets, where they may be unable to command high prices for their therapeutic value. The effect of firm characteristics The effects of variables measured at the firm level are presented in Table 9. While this table
includes only two specifications, the estimates are quite similar across all models estimated (see Appendix C). The diffusion of a new drug depends largely on the characteristics of its originator. The probability of launch in the home country of the firm is nearly five times greater than the average. A firm may have relatively low fixed costs in its domestic market for many reasons; perhaps it receives some favoritism by the regulator, is allowed more generous pricing in the interest of keeping the domestic industry strong, or enjoys superior marketing ability in its native environment. Alternatively, domestic firms may be most familiar with the therapeutic needs of 19
their home country, and therefore concentrate their drug development in those areas. Japanese firms have developed many of the antiulcer treatments available today in response to the local demand for such products, to continue with the example used earlier. Experience in a country also increases the likelihood of entry. For each additional drug a firm markets in a given country, the results from LPM 4 imply that the probability of launching another of its products there increases by about 9% from the mean. This suggests economies of scope in local distribution through familiarity with the regulator or the establishment of marketing and distribution forces. However, it is impossible in this model to distinguish whether the firm has exogenously low fixed costs in a given country, and therefore introduces more products, or whether it achieves lower costs through economies of scope. The coefficient on international experience is positive across all specifications, indicating that firms realize some economies of scope from marketing in additional countries.
However, this effect is rather small, since
marketing in one additional country translates into only a 1% increase from the mean probability of entry. Expertise in a therapeutic area is negatively related to entry, and entry is less likely if the firm has a larger number of drugs in its portfolio; these effects are also of little economic significance. Since all these firm measures correspond to the originating firm, rather than the firm that ultimately markets a drug in any particular country, they should have no effect if licensing markets work efficiently. That is, if firms enjoy large cost advantages in their local markets, foreign companies should license their products to domestic firms. An interesting topic for future research is why more licensing does not take place. As mentioned earlier in the discussion of country characteristics, price controls affect firms differently depending on where they are headquartered. Table 10 presents linear probability models that allow different coefficients for all variables by country of origin. The strong negative effect of price controls is again apparent for British and American firms. In contrast, the home country advantage is largest for Italian and Japanese firms. These results demonstrate that price regulations and industrial policies that favor local firms have important non-uniform effects on firms and on the mix of products available in a market. The previous discussion has assumed that every firm is a potential entrant into every country. However, about half of the firms in this sample compete in three or fewer countries. LPM 4 was estimated separately for multinationals (defined as a firm that competed in at least four countries) and for the remaining set of “local” firms.
The estimated home country advantage for
multinationals is about 15% smaller than for the entire sample, but still quite large. Local firms, since they compete primarily in their home market only, are less affected by competition than are 20
multinationals. Because these smaller firms are responsible for less than 14% of all drugs, excluding them from the analysis has little effect on the parameter estimates for the entire sample.20 Relative importance of market, firm, and product characteristics In order to assess the relative impact of market, firm, and product characteristics on the
probability of entry, the elasticities at the mean values of all variables are presented in Table 11. The elasticities for the country characteristics are computed using the results of LPM 2, which includes only year and therapeutic class fixed effects, because the differences between countries (rather than within countries over time) are more appropriate in this context. For all the other variables, the elasticities are computed using the results from LPM 4, with the full set of countrytherapeutic class interactions. The effects of the quadratic terms and interactions are included in the calculations for the corresponding variables. The variable with the largest impact on entry is clearly domestic status. Drugs originated by local firms are almost five times as likely to be launched in their home markets than in foreign countries. Drugs are about 17% less likely to enter a price-controlled market. The effect of competition is quite strong; an increase of only one recent competitor reduces the probability of another’s entry by almost 30%, and one additional older incumbent reduces the probability of entry by almost 40%. Drug age has a quantitatively similar effect, as does the number of countries launched in. Most of the other variables have small elasticities at their means. Country-specific effects Table 12 presents the results of entry models for five of the largest markets. In these
regressions, coefficients have country specific values. Competition appears to be more important in the US, the UK, Italy, and Germany, whose coefficients on the number of drugs in the market are negative and significant, than it does in France and Japan, whose coefficients are smaller in absolute value and less significant. The effect of domestic status is even more dramatic. Rapid entry into Japan and Italy is nearly certain if the drug is of domestic origin. French drugs are also far more likely to be launched in France than are foreign drugs. This could be due either to a protectionist industrial policy or to specialization by these firms in the treatments most desired by their domestic markets. It is not clear why the latter should hold true for Japanese, Italian, or French firms but not for British, German, or American firms, which realize far less from domestic status.
20
Results are available from the author on request.
21
The impact of some drug characteristics also differs from country to country. The effect of the number of countries previously launched in is much larger in the European markets than in Japan and the US. There are at least two explanations for this result. European countries are supposed to have fairly uniform standards for regulatory approval, so firms should be able to use the same clinical trial data in each European application. The FDA in the US is reputed to have more stringent requirements, and Japanese regulators usually require clinical trial data from domestic patients. For these two countries, the gains from launch in other markets are therefore smaller. Alternatively, European countries – especially those with price controls – may be the last markets firms elect to enter. Many of these countries reference each other’s prices when setting the domestic price or setting the level of reimbursement. There is therefore a strong incentive to enter the low-price markets only after launching in most other countries. VII. Conclusion This paper integrates predictions of economic theory with the views of strategic management in considering the relative impacts of firm, product, and country characteristics on the entry patterns of pharmaceuticals. It finds that entry is positively associated with market size and wealth and that expected profits decline in the number of competitors.
Thus, results are
consistent with predictions of industrial organization oligopoly models and the findings of previous studies of entry. Price controls are estimated to have a negative effect on entry, and drug characteristics are related to profits in expected ways. In addition, there is evidence of economies of scale in global production and economies of scope within a market.
Firm
characteristics, such as experience in a country and domestic status, are found to have an enormous bearing on the diffusion of a new drug. This provides support for ideas about the role of complementary assets advanced by the strategy field. While both market structure and firm/product characteristics have substantial effects on the entry pattern of a new drug, this research demonstrates that the interaction between them is crucial. The relative salience of domestic status and competition varies across countries, and price controls have differential impacts on firms headquartered in different countries. This is an important dimension of entry that most previous research has been unable to address. There are two implications for public policy from this research. Price controls appear to reduce the probability of a new drug’s entry, and disproportionately affect Swiss, British, and American firms. These companies are responsible for over 40% of all drugs developed between 1980 and 2000, and are generally considered the most innovative. The costs of deterring their products, over and above the possible long-run effects on incentives to invest in costly R&D and the development of future products, should be balanced against any short-run savings from lower 22
prices. Second, these results demonstrate that domestic firms are able to access their local markets at a lower cost than are foreign firms and that the advantage of domestic status varies across countries. While it is possible that local firms develop treatments for local needs more efficiently than foreign firms, an industrial policy that favors the drugs of domestic firms may result in crowding out of superior foreign products. For management, this research demonstrates the importance of understanding local pharmaceutical markets. The match between an innovating firm, a drug, and the local market appears to be a critical aspect of profitability. Demand is quite heterogeneous, and competitive effects vary substantially from country to country. The results suggest that there are gains from licensing to domestic firms or to firms with a large presence in a market. That regulations do not affect all firms and products uniformly has implications for selecting markets in which to launch products and in what order. These results indicate that there are important sources of competitive advantage that merit additional exploration. More information about pharmaceutical firms, such as their financial health, their patenting activities, and their licensing practices, would be very valuable. Future work should also incorporate better measures of country-specific demand and costs associated with product launch, such as indicators of regulatory stringency and advertising. Lastly, a structural approach that addresses the problem of endogenous entry by competitors and examines the nature of competition in these markets may be appropriate.
23
References Allison, P. (1984), Event History Analysis: Regression for Longitudinal Event Data, Newbury Park, CA: Sage Publications.
Amemiya, T. (1985), Advanced Econometrics, Cambridge, MA: Harvard University Press. Berndt, E. L. Bui, D. Lucking-Reily and G. Urban (1997), “The Roles of Marketing, Product Quality and Price Competition in the Growth and Composition of the US Anti-Ulcer Drug Industry," chapter 7 in Timothy F. Bresnahan and Robert J. Gordon, eds., The Economics of New Goods, Studies in Income and Wealth, Volume 58, Chicago: University of Chicago Press for the National Bureau of Economic Research, 277-322. Berry, S. (1992), “Estimation of a Model of Entry in the Airline Industry,” Econometrica, 60(4), 889-917. Bresnahan, T. and P. Reiss (1987), “Do Entry Conditions Vary Across Markets?” Brookings Papers on Economic Activity: Microeconomics, 833-71. Bresnahan, T. and P. Reiss (1990), “Entry in Monopoly Markets,” Review of Economic Studies, 57(4), 531-553. Bresnahan, T. and P. Reiss (1991), “Entry and Competition in Concentrated Markets,” Journal of Political Economy, 99(5), 977-1009. Carpenter, D. and M. Turenne (2000), “Why Do Bigger Firms Receive Faster Drug Approvals?” Mimeo, University of Michigan, Ann Arbor, MI. Caves, R., M. Whinston and M. Hurwitz (1991), “Patent Expiration, Entry, and Competition in the US Pharmaceutical Industry,” Brookings Papers on Economic Activity: Microeconomics, 1-48. Cockburn, I. and Henderson, R, (1994), “Racing to Invest? The Dynamics of Competition in Ethical Drug Discovery,” Journal of Economics & Management Strategy, 3(3), 481-519. Cockburn, I. and Henderson, R. (1998), “Absorptive Capacity, Coauthoring Behavior, and the Organization of Research in Drug Discovery,” Journal of Industrial Economics, 46(2), 157-82. Danzon, P. and L. Chao (2000), “Cross-National Price Differences for Pharmaceuticals: How Large, and Why?” Journal of Health Economics, 19, 159-195. Danzon, P. and L. Chao (2000), “Does Regulation Drive Out Competition in Pharmaceutical Markets?” Journal of Law and Economics, 43, 311-357. DiMasi, J. (2000), “New Drug Innovation and Pharmaceutical Industry Structure: Trends in the Output of Pharmaceutical Firms,” Drug Information Journal 34, 1169-1194. Djankov, S., R. La Porta, F. Lopez-de-Silanes, and A. Shleifer (2002), “The Regulation of Entry,” Quarterly Journal of Economics 117(1): 1-38. 24
Dranove, D. and D. Meltzer (1994), “Do Important Drugs Reach the Market Sooner?” RAND Journal of Economics 25(3), 402-423. Ellison, S., I. Cockburn, Z. Griliches and J. Hausman, “Characteristics of Demand for Pharmaceutical Products: An Examination of Four Cephalosporins,” RAND Journal of Economics, 28(3), 426-46. Ellison, S. and C. Wolfram (2000), “Pharmaceutical Prices and Political Activity,” Mimeo, MIT. Grabowski, H. G., J. Vernon and L.G. Thomas (1978), “Estimating the Effects of Regulation on Innovation: An International Comparative Analysis of the Pharmaceutical Industry,” Journal of Law and Economics, 21(1), 133-163. Grabowski, H. and J. Vernon (1992), “Brand Loyalty, Entry and Price Competition in Pharmaceuticals After the 1984 Drug Act,” Journal of Law and Economics 35, 331-350. Henderson, R. and K. Clark (1990). “Architectural Innovation: The Reconfiguration of Existing Product Technologies and the Failure of Established Firms,” Administrative Science Quarterly 35(1): 9-30. Henderson, R. and I. Cockburn, (1996), “Scale, Scope, and Spillovers: The Determinants of Research Productivity in Drug Discovery,” RAND Journal of Economics, 27(1), 32-59. Hudson, J. (2000), “Generic Take-up in the Pharmaceutical Market Following Patent Expiry: a Multi-country Study,” International Review of Law and Economics 20, 205-221. Horrace, W. and R. Oaxaca, “New Wine in Old Bottles: A Sequential Estimation Technique for the LPM,” Mimeo, University of Arizona, 2001. Jacobzone, S. (2000), “Pharmaceutical Policies in OECD Countries: Reconciling Social and Industrial Goals,” OECD Labour Market and Social Policy Occasional Paper #40. Klepper, S. and K. Simons (1997). “Technological Extinctions of Industrial Firms: An Inquiry into Their Nature and Causes,” Industrial and Corporate Change 6(2): 379-460. Klepper, S. and K. Simons (2000). “The Making of an Oligopoly: Firm Survival and Technological Change in the Evolution of the U.S. Tire Industry,” Journal of Political Economy 108(4): 728-760. Lichtenberg, F. and Philipson, T. (2000), “Creative vs. Uncreative Destruction of Innovative Returns: An Empirical Examination of the US Pharmaceuticals Market,” Mimeo, Columbia University, New York, NY. Mazzeo, M. (1998), “Product Choice and Oligopoly Market Structure,” Mimeo, Northwestern University, Evanston, IL. Parker, J. (1984), The International Diffusion of Pharmaceuticals, London: Macmillan Press. Scott Morton, F. (1999), “Entry Decisions in the Generic Pharmaceutical Industry,” RAND Journal of Economics, 30(3), 421-440. 25
Stern, S. (1996), “Market Definition and the Returns to Innovation: Substitution Patterns in Pharmaceutical Markets,” MIT POPI Working Paper. Teece, David (1987). “Profiting from Technological Innovation: Implications for Integration, Collaboration, Licensing and Public Policy.” The Competitive Challenge, ed. D. Teece, Ballinger Publishing, Cambridge (MA): 185-219. Thomas, K., N. McAuslane, C. Parkinson, S. Walker, and D. Luscombe (1998), “A Study of Trends in Pharmaceutical Regulatory Approval Times for Nine Major Markets in the 1990s,” Drug Information Journal 32, 787-801. Thomas, L.G. (1994), “Implicit Industrial Policy: The Triumph of Britain and the Failure of France in Global Pharmaceuticals,” Industrial and Corporate Change, 2(3), 451-489. Toivanen, O. and M. Waterson (1999), “Market Structure and Entry: Where’s the Beef?” Mimeo, University of Warwick, Coventry, UK.
26
Argentina
5524
5290
5153
4905
3422
2849
19.19%
23.45%
24.95%
24.26%
22.94%
23.77%
22.74%
14.63%
8.03%
16.42%
23.73%
UK
Aust/NZ
8888
Mexico
9035
Brazil
13283
Spain
14424
Canada
France
51434
Italy
Germany
97385 Pct of revenues by class Cardiovascular 17.51%
Japan
Millions of $US
US
Table 1: Revenues from major pharmaceutical markets and distribution across broad therapeutic classes, 2000
CNS
21.76%
6.05%
12.94%
15.23%
11.67%
18.13%
19.01%
17.56%
13.82%
11.76%
15.28%
16.74%
Alimentary
14.71%
15.69%
16.13%
14.96%
14.45%
16.09%
14.54%
15.05%
16.55%
18.94%
17.59%
16.04%
9.62%
11.50%
8.58%
10.25%
11.70%
4.71%
6.28%
7.88%
8.62%
17.49%
9.94%
6.18%
10.13%
6.93%
8.81%
9.15%
8.48%
13.09%
8.20%
10.74%
10.13%
11.17%
7.63%
11.65%
Musculo-skeletal
5.50%
6.72%
4.62%
4.73%
5.80%
5.23%
6.12%
5.29%
8.34%
7.54%
7.83%
5.05%
Genito-urinary
7.02%
2.06%
6.06%
6.05%
6.00%
5.99%
5.70%
4.99%
10.75%
6.97%
7.54%
4.49%
Cytostatics
2.68%
6.53%
5.12%
2.65%
4.53%
3.05%
3.51%
4.18%
0.45%
0.53%
1.49%
3.16%
Dermatologicals
3.60%
2.73%
3.72%
3.42%
3.27%
4.04%
4.54%
3.67%
7.63%
5.97%
6.28%
5.41%
Blood agents
1.61%
7.13%
2.93%
2.63%
4.06%
1.35%
1.88%
2.93%
1.36%
1.47%
1.69%
1.37%
Sensory organs
1.83%
3.17%
1.52%
1.89%
2.17%
1.79%
2.23%
2.06%
2.81%
2.14%
3.16%
2.42%
Diagnostic agents
1.30%
3.59%
2.24%
1.48%
1.26%
1.34%
1.81%
0.04%
0.12%
0.14%
0.64%
0.81%
Hormones
1.18%
2.26%
2.14%
1.72%
1.79%
1.34%
0.76%
2.74%
2.25%
1.75%
2.51%
0.49%
Miscellaneous
1.39%
2.54%
1.27%
0.58%
0.33%
0.42%
1.45%
0.06%
1.09%
4.87%
1.46%
1.97%
Hospital solutions
0.00%
3.91%
0.33%
0.09%
0.17%
0.11%
0.02%
0.04%
0.12%
0.29%
0.06%
0.00%
Parisitology
0.15%
0.01%
0.15%
0.22%
0.07%
0.39%
0.18%
0.04%
1.34%
0.92%
0.47%
0.49%
Anti-infective Respiratory
Source: IMS Health, “World-wide Pharmaceutical Market” Feb. 2001. Figures are revenues from retail pharmacies, in millions of $US at current exchange rates. Figures for Japan include both pharmacy and hospital sales.
27
Table 2: Top 40 (by R&D spending) pharmaceutical firms
Firm Pfizer Glaxo SmithKline Johnson & Johnson Aventis Roche Holding AstraZeneca Novartis Pharmacia Corporation Merck & Company Bristol-Myers Squibb Company Eli Lilly & Company American Home Products Corporation Bayer Group Abbott Laboratories Schering-Plough Corporation Sanofi-Synthelabo Boehringer Ingelheim Amgen Takeda Chemical Industries Schering AG BASF Group (Knoll) Sankyo Company Yamanouchi Pharmaceutical Company Merck KGaA E.I. du Pont de Nemours & Company Eisai Company Fujisawa Pharmaceutical Company Akzo Nobel Novo Nordisk Chugai Pharmaceutical Company Genentech Baxter International Daiichi Pharmaceutical Company Shionogi & Company Solvay Taisho Pharmaceutical Company Nycomed Amersham Kyowa Hakko Kogyo Company Ono Pharmaceutical Company
Nationality USA UK USA France Switzerland UK Switzerland USA USA USA USA USA Germany USA USA France Germany USA Japan Germany Germany Japan Japan Germany USA Japan Japan Netherlands Denmark Japan USA USA Japan Japan Belgium Japan UK Japan Japan
R&D Spending $4,035.0 $3,704.9 $2,600.0 $2,592.9 $2,462.7 $2,454.0 $2,233.3 $2,123.6 $2,068.3 $1,802.9 $1,783.6 $1,513.8 $1,270.9 $1,194.0 $1,191.0 $970.5 $880.4 $822.8 $728.9 $728.7 $707.4 $607.5 $517.2 $477.0 $442.0 $440.6 $429.9 $426.1 $393.1 $377.3 $367.3 $332.0 $322.2 $255.0 $244.0 $219.2 $203.8 $199.9 $189.6
Number of Drugs 43 78 43 79 46 28 40 54 33 27 17 30 25 8 9 54 27 4 27 16 23 16 15 11 6 12 11 22 6 6 10 8 9 11 6 3 8 5 8
Source: PharmaBusiness: 24, Nov. 2000. Figures are millions of 1999 dollars spent on healthcare research and development.
28
Table 3: Origin and diffusion of pharmaceuticals
Country
Number of firms Number of drugs 83 71 14 21 17 11 33 13 5 5 3 6 1 2 2 1 6 3 2 2 2 1 1 1 1 1 1
USA Japan France Germany UK Switzerland Italy Spain Netherlands South Korea Denmark Canada Norway Belgium Hungary Finland Sweden Argentina Australia Czech Republic Austria Israel Brazil Croatia Cuba Ireland New Zealand
420 301 195 147 128 110 100 37 36 18 17 8 8 7 7 6 6 5 5 3 2 2 1 1 1 1 1
Avg countries in which launched 8.9 4.4 7.3 6.9 9.2 9.5 4.5 2.7 8.1 1.2 13.3 6.0 9.0 8.3 5.7 6.0 6.3 2.2 3.0 9.0 1.0 5.5 1.0 15.0 2.0 1.0 1.0
Table 4: Countries in sample
Country Australia Austria Belgium Canada Denmark France Germany Greece Ireland Italy Japan
Price Controls Y Y Y Y N Y N Y N Y Y
Year 1951 1976 1963 1987 1945 1978 1978 1950
Country Mexico Netherlands Portugal South Korea Spain Sweden Switzerland Turkey UK USA
29
Price Controls Year Y 1993 Y 1996 N Y 1977 Y unknown Y 1993 Y 1962 Y 1928 N N
Table 5: Summary Statistics Number of drugs Number of firms Number of therapeutic classes Years covered Number of markets (country-class-year observations) Number of entry opportunities (drug-country-class-year observations) Number of entry events Frequency
Variable
Drugcountryclass-year
Drug global importance
Firmcountry-year
Country experience
Drug local importance
Country-class experience Countryclass-year
Number of new drugs in market Number of old drugs in market Number of potential competitors Class global importance Class local importance
1577 310 147 1980-1999 58,624 466,333 12,505
Definition Drug's share of stock of Medline citations for class by foreign authors Drug's share of stock of Medline citations for class by domestic authors Count of firm's other drugs launched in country Count of firm's drugs in country-class market Count of drugs in market launched less than 5 years ago Count of drugs in market launched more than 5 years ago Count of drugs launched in class elsewhere in the world Class share of the stock of Medline citations by foreign authors Class share of the stock of Medline citations by domestic authors
30
Obs
Mean
Std Dev
Min
Max
466333
0.0095
0.0650
0
1
466333
0.0118
0.0783
0
1
84337
1.1714
3.3205
0
51
84337
0.0525
0.2908
0
10
58624
1.0371
1.4892
0
15
58624
1.8669
2.9921
0
34
58624
8.5686
8.4244
1
81
58624
0.0084
0.0301
0
0.30
58624
0.0101
0.0407
0
1.00
Table 5 cont. Frequency Drug-year
Variable Drug age
Definition Number of years since drug's first launch anywhere
Number of countries launched in
Firm-country
Number of countries researched in Number of countries published in Number of languages published in Home country
Class experience
Number of countries in which Medline research authored Number of countries in which Medline research published Number of languages in which Medline research published Dummy = 1 if firm is headquartered in country Count of countries in which firm has launched any drugs Count of firm's drugs in therapeutic class
Firm-year
International experience
Portfolio
Total number of firm's drugs
Drug
Number of therapeutic classes
Country-year
Population
Population in 10s of millions
GDP per capita
GDP per capita in US$1000s, PPP
Price controls
Dummy = 1 if country uses price controls
31
Obs
Mean
Std Dev
Min
Max
20733
9.3124
6.9041
0
4
20733
6.1986
6.7298
0
27
20733
2.6034
5.7309
0
67
20733
2.0930
3.9817
0
42
20733
0.9927
1.6593
0
14
6470
0.0439
0.2049
0
1
4372
9.3882
9.3348
0
28
4372
1.1901
0.8085
1
17
4372
4.7884
8.5156
1
79
1577
1.5124
0.8069
1
6
420
4.5382
5.5274
0.34
27.29
420
14.2652
6.2790
2.25
31.94
420
0.5079
0.5005
0
1
Table 6: Effect of Country Characteristics on Entry
Variable
Population (10s of millions) Population squared GDP per capita ($1000s) Price controls in effect Class global importance Class local importance
OL 1
LPM 1
LPM 2
Hazard 1
Coef. (Std Err)
Coef. (Std Err)
Coef. (Std Err)
Hazard Rate (Std Err)
0.121**
0.00212**
0.00234**
1.041**
(0.007)
(0.00017)
(0.00017)
(0.008)
**
**
**
-0.005
-0.00009
-0.00008
(0.000)
(0.00001)
(0.00001)
0.077
**
**
**
0.00090
0.00101
(0.003)
(0.00007)
(0.00007)
-0.205**
0.00043
-0.00435**
(0.025)
(0.00081)
(0.00152)
*
*
**
0.998** (0.000) 1.042** (0.005) 0.838** (0.033)
-1.589
0.05689
0.05695
5.850
(0.559)
(0.02877)
(0.02891)
(10.916)
0.840
-0.00016
0.00352
(0.430)
(0.01392)
(0.01398)
0.264* (0.173)
*
French firm*price control
0.00581
(0.00232) German firm*price control
0.00323 (0.00223) 0.00918**
Italian firm*price control
(0.00237) 0.01001**
Japanese firm*price control
(0.00189) -0.00794**
UK firm*price control
(0.00276) -0.00509**
US firm*price control
(0.00193) Included Fixed Effects
Year Class
Year Class
Year Class
Year Class
N
55617
290344
290344
267391
Log L Pseudo-Rsq/Adj Rsq
-52249.624
-28518.857 0.1212
0.057
0.049
*= 5% significance, ** = 1 %. Standard errors have not been adjusted for heteroskedasticity. All specifications include the full set of country, firm, and product characteristics.
32
Table 7: Effect of Competition on Entry
Variable
LPM 3 Coef. (Std Err)
LPM 4 Coef. (Std Err)
Number of new drugs in market
0.00081*
-0.00737**
(0.00040)
(0.00046)
Hazard 2 Hazard Rate (Robust Std Err)
0.787** (0.054)
Number of new drugs squared
0.00009 (0.00005)
0.00020 (0.00006)
1.016* (0.007)
Number of old drugs in market
-0.00019 (0.00024)
-0.00943** (0.00033)
0.726** (0.062)
Number of old drugs squared
0.00000 (0.00001)
0.00015** (0.00001)
1.010* (0.004)
Number of potential competitors
0.00002 (0.00007) Year Class Country 290344 0.058
0.00152** (0.00008) Year Country*Class
1.061* (0.029) Year Country*Class
290344 0.090
30284
Included Fixed Effects N Adjusted R2 Log likelihood
**
-5744.163
* = 5% significance, ** = 1 %. Standard errors have not been adjusted for heteroskedasticity. All specifications include the full set of country, firm, and product characteristics. Hazard 2 is estimated on a random subset of therapeutic classes, which reduces the number of observations.
33
Table 7b: Results from Method of Simulated Moments LPM Variable Coef. (Std. Err) Population 0.0134** (0.0026) Population squared -0.0005** (0.0001) Number of incumbents -0.0218** (0.0029) Number of incumbent squared 0.0006** (0.0001) Price controls -0.0412** (0.0098) Country experience 0.0069** (0.0008) Home country 0.3118** (0.02950) Anti-ulcer class 0.1995** (0.0185) Osteoporosis class -0.0448** (0.0175) Anti-epilepsy class -0.0492** (0.0189) Anti-migraine class -0.0472** (0.0209) Anti-asthma class -0.0049** (0.0113) Log(new entry)
Rho Number of markets Number of entry opportunities Number of entry events
420 4652 588
*= 5%, ** = 1 %.
34
MSM Coef. (Std. Err) 0.0426** 0.0125 -0.0009 0.0005 -0.0046 0.0051 0.0026** 0.0007 -0.0521 0.0315 0.0178** 0.0060 0.5201** 0.2203 0.4524** 0.1600 0.2372 0.1562 0.4136 0.2325 0.3019 0.1638 0.4152** 0.1624 5.0212** 0.3682 0.6439** 0.0992
Table 8: Effect of Product Characteristics on Entry
Variable
Drug age Drug age squared Number of countries launched in Number of countries launched in squared Number of therapeutic classes Global importance Local importance Global importance*price control Local importance*price control Included Fixed Effects
LPM 4 Coef. (Std Err)
Hazard 1 Hazard Rate. (Robust Std Err)
-0.00839** (0.00014) 0.00021** (0.00001) 0.00810** (0.00019) -0.00024** (0.00001) 0.00316** (0.00042) 0.01637* (0.00753) 0.17464** (0.01238) 0.00125 (0.00997) -0.01439 (0.01680) Year Country*Class
1.410** (0.021) 0.990** (0.001) 1.106** (0.032) 2.074** (0.549) 3.901** (0.994) 1.031 (0.335) 0.845 (0.279) Year Class
N Adjusted R2
290344 0.090
Log likelihood
267391 -52249.624
* = 5% significance, ** = 1 %. Standard errors for LPM have not been adjusted for heteroskedasticity. All specifications include the full set of country, firm, and product characteristics.
35
Table 9: Effect of Firm Characteristics on Entry
Variable
Country experience Country experience squared Home country International experience Class experience Country-class experience Portfolio Price control*portfolio Included Fixed Effects N Adjusted R2 Log likelihood
LPM 4 Coef. (Std Err)
0.00228** (0.00017) -0.00005** (0.00000) 0.12204** (0.00232) 0.00023** (0.00005) -0.00141** (0.00028) -0.00053 (0.00074) -0.00024** (0.00004) -0.00012** (0.00003) Year Country*Class
Hazard 1 Hazard Rate (Robust Std Err)
1.079** (0.011) 0.999** (0.000) 1.541** (0.119) 1.015** (0.004) 0.955 (0.027) 0.984 (0.046) 0.982** (0.003) 0.997* (0.001) Year Class 267391
290344 0.090
-52249.624
* = 5% significance, ** = 1 %. Standard errors for LPM have not been adjusted for heteroskedasticity. All specifications include the full set of country, firm, and product characteristics.
36
Table 10: LPMs by Country of Origin
Variable Number of new drugs in market N new drugs squared Number of old drugs in market N old drugs squared
France
Germany
Italy
Japan
UK
Coef. (Std Err) 0.00204*
Coef. (Std Err) 0.00190
Coef. (Std Err) 0.00007
Coef. (Std Err) 0.00165*
Coef. (Std Err) 0.00406*
Coef. (Std Err) 0.00325**
(0.00096)
(0.00112)
(0.00086)
(0.00064)
(0.00181)
(0.00111)
0.00002
-0.00023
-0.00006
(0.00011)
(0.00009)
(0.00022)
(0.00015)
0.00031
-0.00028
-0.00122
0.00089
(0.00049)
(0.00036)
(0.00108)
(0.00066)
0.00005
-0.00002
0.00002
0.00022
(0.00012)
(0.00014)
Class global importance Class local importance
0.00287
(0.00057)
(0.00066)
0.00000
-0.00012
-0.00034
*
(0.00016)
Population squared
(0.00040) -0.00005
**
0.00042
*
(0.00017) Price controls
0.00541
*
(0.00274) Country experience
0.00329
**
(0.00031) Country experience squared Home country
-0.00008
**
(0.00001) 0.07080
**
(0.00538) International experience Class experience
(0.00002)
(0.00005)
(0.00003)
0.00008
0.00004
0.00023
0.00000
-0.00030
(0.00022)
(0.00013)
(0.00036)
(0.00018)
(0.00028) **
-0.08496
0.43992
(0.07438)
(0.04402)
(0.07071)
(0.13704)
(0.08402)
-0.00331
-0.02167
0.00421
-0.06960
-0.00145
(0.03885)
(0.02886)
(0.02543)
(0.06371)
(0.03755)
-0.00040
0.00186
(0.00049)
(0.00034)
(0.00029)
-0.00002
0.00001
(0.00002)
(0.00001)
(0.00020)
(0.00014)
-0.00171
0.00287
(0.00305) 0.00303
0.00486
(0.00167)
-0.00006
-0.00137
(0.00008) 0.06148
**
**
(0.00034)
**
0.33061
(0.00589)
**
(0.00754)
**
-0.00023 0.00929
(0.00162)
0.00037
*
(0.00389)
**
(0.00012) **
(0.00150) 0.00606
**
(0.00073) -0.00092
**
(0.00007) 0.77303
**
(0.00698) -0.00032
(0.00023)
**
**
(0.00001)
-0.00409
(0.00224)
**
(0.00092)
-0.00512
(0.00138)
**
-0.00068
(0.00022)
-0.00293
-0.00007
0.00038
(0.00021)
*
**
0.00051
0.00088
0.00038
**
-0.04409
-0.00027
(0.00053) Country-class experience
(0.00002)
-0.01700
(0.00002) GDP per capita ($1000s)
(0.00003)
(0.07652)
0.00104
*
0.00003
0.26130
**
**
0.00000
0.10485
(0.02954) Population (10s of millions)
**
-0.00035
(0.00003) N of potential competitors
0.00036
**
USA
**
(0.00008) -0.00650
**
(0.00112)
0.00238**
-0.00020 (0.00080)
(0.00066)
0.00003
-0.00011*
(0.00003) 0.00100
**
(0.00037) -0.01395
**
(0.00524) 0.00395
**
(0.00071) -0.00004
**
(0.00001) 0.06261
**
(0.00005) 0.00165** (0.00021) -0.00936** (0.00276) 0.00277** (0.00057) -0.00003 (0.00002) 0.05733**
(0.00831)
(0.02061)
0.00006
0.00048*
(0.00042)
(0.00019)
0.00058
0.00052
(0.00206) *
0.00170
0.00005
0.00412
-0.00741
(0.00326)
(0.00431)
(0.00234)
(0.00306)
37
0.03968
(0.00113) -0.00565** (0.00180)
Variable Portfolio Price control*portfolio Drug age
France
Germany
Italy
Japan
Coef. (Std Err) -0.00024**
Coef. (Std Err) -0.00086**
Coef. (Std Err) 0.00100
Coef. (Std Err) 0.00036*
Coef. (Std Err) -0.00094**
Coef. (Std Err) -0.00078**
(0.00008)
(0.00028)
(0.00075)
(0.00016)
(0.00018)
(0.00016)
-0.00024**
0.00029
-0.00141**
0.00008
0.00020
0.00019
(0.00006)
(0.00022)
(0.00042)
(0.00015)
-0.00599
**
(0.00038) Drug age squared
0.00016
**
(0.00001) N of countries launched in N of countries launched in squared Number of therapeutic classes Global importance
0.00453
**
(0.00052) -0.00017
**
(0.00003) 0.00441
**
(0.00120) 0.02690 0.08858
**
(0.03375) Global importance*price control
0.00012 0.00557
0.00005
**
(0.00001)
**
0.00381
(0.00065) -0.00024
**
(0.00033)
**
(0.00002)
**
(0.00068)
**
-0.00015
(0.00003)
**
(0.00004)
0.00050 (0.00149) *
(0.01078)
0.14358
(0.04447)
N Adjusted Rsq
*
0.00959
**
(0.00047) -0.00029
**
(0.00003)
0.15273 **
(0.02543)
(0.06300)
(0.00001)
**
(0.00081)
(0.04407)
(0.02505)
0.00006
**
(0.00132)
0.07557
-0.06847
(0.00025)
0.00221
0.01036
0.02868
-0.00365
**
0.00027 -0.02112
(0.03550)
0.04789
Local importance*price control
-0.00238
(0.00045)
0.07092
(0.01858) Local importance
-0.00558
**
UK
**
(0.02628) 0.17890
**
(0.02807) **
0.02541
-0.10624
(0.01365)
(0.03339)
USA
(0.00011) -0.01307
**
(0.00083) 0.00039
**
(0.00003) 0.00580
**
(0.00101) -0.00015
**
(0.00005)
-0.01268** (0.00043) 0.00034** (0.00002) 0.00738** (0.00051) -0.00025** (0.00002) 0.00490**
0.00524 (0.00270)
(0.00143) 0.04980**
0.02030 (0.02464) 0.31262
(0.00011)
**
(0.01844) 0.14038**
(0.04409)
(0.02728)
0.00449
0.00299
(0.02924) **
(0.02281)
0.01586
-0.03294
-0.26831
-0.00896
(0.06717)
(0.04159)
(0.03667)
(0.06113)
(0.03612)
44253
31715
26524
62101
20976
58235
0.060
0.061
0.101
0.215
0.092
0.065
0.021 0.020 0.010 0.016 0.040 0.038 Mean of entry * = 5% significance, ** = 1 %. All specifications include year and class fixed effects. Standard errors have not been adjusted for heteroskedasticity.
38
Table 11: Relative Impacts of Competition, Firm, Drug, and Country Variables Variable
Competition
Firm Characteristics
Drug Characteristics
Country Characteristics
Number of new drugs in market Number of old drugs in market Number of potential competitors Country experience Home country International experience Class experience Country-class experience Portfolio Drug age Number of countries launched in Number of therapeutic classes Global importance Local importance Population (10s of millions) GDP per capita ($1000s) Price controls in effect Class global importance Class local importance
∆(x) 1 1 1 1 1 1 1 1 1 1 1 1 0.01 0.01 1 1 1 0.01 0.01
39
∆(y) -0.0074 -0.0094 0.0015 0.0023 0.1220 0.0002 -0.0014 -0.0005 -0.0002 -0.0084 0.0081 0.0032 0.0002 0.0000 0.0023 0.0010 -0.0044 -0.0135 0.0040
% change in x % change in y Elasticity at mean
61.8% 29.4% 6.0% 21.3% 5.4% 59.0% 569.0% 5.3% 11.9% 24.9% 66.4% 135.8% 134.1% 2.0% 190.4% 120.8% 113.9%
-28.7% -37.1% 6.1% 8.9% 488.2% 0.9% -5.7% -2.1% -1.0% -32.7% 31.4% 12.7% 0.7% 0.1% 9.1% 4.0% -17.6% -53.9% 16.0%
-0.464 -1.263 1.011 0.419 0.173 -0.096 -0.004 -0.181 -2.745 1.261 0.191 0.005 0.001 4.610 0.021 -0.446 0.141
Table 12: LPMs by Destination Country Variable
Marketspecific variables
Number of new drugs in market N new drugs squared Number of old drugs in market N old drugs squared
Firmspecific variables
Number of potential competitors Class global importance Class local importance Country experience Country experience squared Home country International experience Class experience Country-class experience Portfolio Drug age
France Coef. (Std Err) -0.00516 (0.00276) 0.00000 (0.00037) -0.01524** (0.00205) 0.00034** (0.00009) 0.00291** (0.00051) 0.29924 (0.19302) -0.13563 (0.40338) 0.00276** (0.00105) -0.00015** (0.00002) 0.11763** (0.01029) 0.00012 (0.00028) -0.00344 (0.00259) -0.00397 (0.00394) 0.00058 (0.00035) -0.01361**
Germany Coef. (Std Err) -0.00795** (0.00291) 0.00016 (0.00031) -0.01567** (0.00210) 0.00021** (0.00006) 0.00480** (0.00076) 0.37483 (0.24595) -0.10777 (0.19670) 0.00257* (0.00128) -0.00003 (0.00002) 0.05796** (0.00915) 0.00050 (0.00036) -0.00551* (0.00242) 0.01066* (0.00524) -0.00032 (0.00040) -0.01855**
Italy Coef. (Std Err) -0.01083** (0.00292) 0.00034 (0.00035) -0.00903** (0.00172) 0.00006 (0.00006) 0.00386** (0.00061) 0.36925 (0.23778) -0.31804 (0.17119) 0.00135 (0.00132) -0.00001 (0.00002) 0.32398** (0.01690) 0.00035 (0.00034) -0.00521** (0.00191) 0.00895 (0.00477) -0.00030 (0.00044) -0.01311**
40
Japan Coef. (Std Err) -0.00316 (0.00230) -0.00009 (0.00024) -0.00507** (0.00149) 0.00005 (0.00006) 0.00131** (0.00047) 0.11770 (0.15871) 0.00875 (0.05028) 0.00103 (0.00131) -0.00008 (0.00005) 0.72450** (0.01117) 0.00029 (0.00030) -0.00130 (0.00141) 0.00017 (0.00416) 0.00018 (0.00024) -0.00550**
UK Coef. (Std Err) -0.00876** (0.00302) 0.00032 (0.00040) -0.00901** (0.00206) 0.00017 (0.00010) 0.00339** (0.00050) 0.25228 (0.19279) -0.16154 (0.12247) 0.00348** (0.00085) -0.00006** (0.00002) 0.07725** (0.00877) 0.00090** (0.00025) 0.00046 (0.00179) -0.00454 (0.00397) -0.00056* (0.00022) -0.01585**
USA Coef. (Std Err) -0.00677** (0.00222) 0.00005 (0.00024) -0.00522* (0.00213) -0.00023* (0.00012) 0.00256** (0.00063) 0.05833 (0.16861) -0.10257 (0.13539) -0.00124 (0.00097) 0.00005* (0.00002) 0.05818** (0.00557) 0.00061** (0.00023) -0.00159 (0.00191) 0.00195 (0.00420) 0.00020 (0.00019) -0.01402**
Variable
Drugspecific variables
Drug age squared Number of countries launched in Number of countries launched in squared Number of therapeutic classes
N Adjusted Rsq Mean of entry
France Coef. (Std Err) -0.01361** (0.00080) 0.00032** (0.00003) 0.01673** (0.00131) -0.00050** (0.00008) 0.00312 (0.00256) 0.12610** (0.02619) 0.68352** (0.05051) 11893 0.141 0.039
Germany Coef. (Std Err) -0.01855** (0.00100) 0.00049** (0.00004) 0.02278** (0.00216) -0.00093** (0.00016) 0.00799** (0.00309) 0.04872 (0.02745) 0.18293** (0.05519) 10731 0.134 0.051
Italy Coef. (Std Err) -0.01311** (0.00104) 0.00038** (0.00005) 0.01751** (0.00166) -0.00051** (0.00011) 0.00903** (0.00310) 0.10807** (0.03361) 0.23089** (0.05422) 10680 0.132 0.046
Japan Coef. (Std Err) -0.00550** (0.00078) 0.00014** (0.00003) 0.00074 (0.00096) 0.00017** (0.00005) -0.00158 (0.00240) 0.04168 (0.02371) 0.15936** (0.03950) 11584 0.374 0.046
UK Coef. (Std Err) -0.01585** (0.00076) 0.00039** (0.00003) 0.00917** (0.00130) -0.00015 (0.00008) 0.00259 (0.00223) 0.00083 (0.02149) 0.27016** (0.04450) 13145 0.124 0.035
USA Coef. (Std Err) -0.01402** (0.00073) 0.00034** (0.00003) 0.00299** (0.00097) 0.00003 (0.00005) -0.00176 (0.00208) 0.07955** (0.02044) 0.40491** (0.03822) 13833 0.156 0.035
* = 5% significance, * = 1 %. All specifications include year and class fixed effects. Standard errors have not been adjusted for heteroskedasticity.
41
Figure 1: European price structure of pharmaceuticals
UK Sweden France Italy Spain Norway Netherlands Switzerland Belgium Germany Finland Austria 0%
10%
20%
30% Ex-Factory Price
40%
50% Wholesaler's Price
60%
70%
Pharmacy's Margin
80%
90%
100%
VAT
Source: European Federation of Pharmaceutical Industry Associations (EFPIA), 1998 in: Pharmaceutical Pricing and Reimbursement in Europe, 1999.
42
Figure 2: Distribution of the Number of Drugs in a Market
All markets, 1980-2000
All markets in 2000
30
25
Percent of Markets
20
15
10
5
0 0
1
2
3
4
5
6
7
8
9
10
11
12
Number of Drugs in Market
43
13
14
15
16
17
18
19
20
>20
Figure 3: Distribution of the Number of Drugs in a Market, Selected Countries Canada
France
Italy
Japan
USA
40
35
Percent of Markets
30
25
20
15
10
5
0 0
1
2
3
4
5
6
Number of Drugs in Market
44
7
8
9
10
>10
Figure 4: Distribution of the number of OECD countries entered and population reached
Percent of Drugs
Population
35.0%
100 90
30.0%
80 70 60
20.0%
50 15.0%
40 30
10.0%
20 5.0%
10
0.0%
0 1
2
3
4
5
6
7
8
9
10
11
12
13
Number of countries
45
14
15
16
17
18
19
20
21
10s of millions
Percent of drugs
25.0%
Appendix A: Additional regulatory information Generic Drugs Countries differ in the amount of testing required for approval of generic products, and
many have made policy changes over the past several decades. Even the generic drug industry in the US, where generic penetration is highest, achieved significance only after the Waxman-Hatch Act of 1984. European countries have only recently adapted their policies to encourage generic entry, such as providing incentives for pharmacists to fill prescriptions with generics, encouraging doctors to prescribe generics, or requiring patients covered by the government health plan to accept generics. In most countries, though, generics garner only a small market share. Demand-side Controls Reference pricing is a regime in which the government sets a price at which it will
reimburse a treatment for a condition. The patient must then pay the difference between that reference price and the price of the treatment he elects to take. The reference price is usually determined by a formula that accounts for the average cost of alternative therapies, the cheapest available treatment, etc. Reference pricing is relatively new, beginning first in Germany in 1989, and is in limited use in about six major markets. Most countries require a patient co-payment for a prescription covered under the government insurance plan, which varies across patients (by income or age), therapeutic class, type of drug, etc. and may be fixed or a percentage of total cost. Some governments, notably Britain and Germany, monitor or limit the prescribing behavior of physicians.
Many countries provide guidelines for prescribing and some impose financial
sanctions on doctors who deviate. Others also limit the volume a physician may prescribe of a particular drug or restrict him to a fixed budget. In the US, health maintenance organizations attempt similar controls on cost. For a detailed treatment of the many varieties of regulation, see Jacobzone (2000). These distinctions amount to variations in the price sensitivity of patients and doctors in different countries. Advertising Prescription drug advertising is highly regulated in all countries, in its content and in
some cases its quantity as well. Only three countries (the US, China, and New Zealand) permit direct-to-consumer advertising, and the US only recently relaxed its position on this. Italy restricts the number of minutes a firm may detail a drug. One consequence of this policy is extensive licensing of the same drug to several firms, so that the total number of detailing minutes
46
is increased. France and Spain set targets for limiting promotional expenditures to a percentage of revenues or selling price.
Appendix B: Construction of variables using MEDLINE citation data.
The description of MEDLINE from the National Library of Medicine website is as follows: MEDLINE is the NLM's premier bibliographic database covering the fields of medicine, nursing, dentistry, veterinary medicine, the health care system, and the preclinical sciences. MEDLINE contains bibliographic citations and author abstracts from more than 4,000 biomedical journals published in the United States and 70 other countries. The file contains over 11 million citations dating back to the mid-1960s. Coverage is worldwide, but most records are from Englishlanguage sources or have English abstracts. All MEDLINE citations that were classified as a clinical trial, meta-analysis, practice guideline, or randomized controlled trial, and pertained to humans, were downloaded from the National Library of Medicine website (http://www.ncbi.nlm.nih.gov/PubMed/) for 1965 through 2000, a total of 307,527 articles. To match the drugs in Pharmaprojects to the Medline data, the drug’s generic name, chemical name, and its synonyms (such as brand names in different countries) were located in the title and abstract of citations, resulting in 764,384 drug mentions. For 81% of drug mentions, there is a field for the affiliation of the lead author; the geographic locations of the lead authors were identified from this field for all but ~1% of these mentions. “Global drug importance” is defined as a drug’s share of the stock of drug mentions for its therapeutic class from articles by foreign authors. “Local drug importance” is analogously defined as a drug’s share of the stock of drug mentions for its therapeutic class from articles by domestic authors.
Similarly, “class global importance” and “class local importance” are
computed as the share of the stock of drug mentions for all drugs in a therapeutic class relative to all drug mentions authored by foreign or domestic scientists, respectively. The stocks were computed using 5, 10, and 15% rates of depreciation; the results from the regression analysis are robust to the assumed rate (15% is the rate used for the reported results). The most class citations pertain to anticancers, anti-infectives, and antidiabetics. Not surprisingly, the anti-AIDS/antiHIV therapeutic classes account for an increasing share of citations over time (about 10% in 2000). In using this data, one must assume that a drug’s importance is positively correlated with the number of studies and publications that refer to it, and that Medline’s coverage is not biased 47
towards a particular country.
There are several unavoidable weaknesses.
The measure of
importance might reflect not therapeutic value but safety concerns, if a potentially dangerous drug is the subject of more studies. Large pharmaceutical firms may have more resources to devote to the funding of clinical trials that are published in Medline journals, thus biasing “importance” towards larger firms. Although the Medline database includes publications from more than 100 countries, its coverage is most complete for English-language journals, which could lead to an upward bias for the importance of drugs from English-speaking nations.
Each country’s
contribution to the Medline database is summarized in Table B1. Finally, it is possible that the search algorithm misses mentions of drugs in abstracts that are not in English or finds fewer matches if abstracts from non-English articles are often unavailable. These shortcomings are acknowledged, but alternative objective measures of drug importance are few. Table B1: Fraction of Medline Drug Mentions, by Country Country
Freq.
Percent
USA Location missing Italy Germany UK Japan France Netherlands Sweden Canada Spain India Denmark Finland Australia Austria China Switzerland Belgium Other
171893 144883 73929 46732 44899 37180 33078 20078 17725 16689 16070 14865 11143 10539 9616 9389 8432 8177 7615 61452
22.49 18.95 9.67 6.11 5.87 4.86 4.33 2.63 2.32 2.18 2.10 1.94 1.46 1.38 1.26 1.23 1.10 1.07 1.00 8.05
Total
764384
100.00
48
Appendix C: Complete Specifications Table C1: Ordered Logits Variable
Number of new drugs in market N new drugs squared Number of old drugs in market N old drugs squared Number of potential competitors Class global importance Class local importance Population (10s of millions) Population squared GDP per capita ($1000s) Price controls Included Fixed Effects
N Log likelihood
OL 1 Coef. (Std Err) 0.092** (0.017) 0.005* (0.002) 0.005 (0.011) -0.001 (0.001) 0.051** (0.004) -1.589** (0.559) 0.840 (0.430) 0.121** (0.007) -0.005** (0.000) 0.077** (0.003) -0.205** (0.025) Year Class
OL 2 Coef. (Std Err) 0.014 (0.017) 0.011** (0.002) -0.062** (0.011) 0.001* (0.001) 0.065** (0.004) -1.490** (0.566) 0.966* (0.430) 0.082 (0.067) 0.006** (0.002) 0.022 (0.012) 0.024 (0.070) Year Class Country
55617 -28518.857
55617 -28166.191
*= 5% significance, ** = 1 %.
49
Table C2: Linear Probability Models Variable
Number of new drugs in market N new drugs squared Number of old drugs in market N old drugs squared Number of potential competitors Class global importance Class local importance Population (10s of millions) Population squared GDP per capita ($1000s) Price controls Country experience Country experience squared Home country International experience Class experience Country-class experience
LPM 1 Coef. (Std Err) 0.00178** (0.00040) 0.00006 (0.00005) 0.00057* (0.00023) -0.00001 (0.00001) -0.00013 (0.00007) 0.05689* (0.02877) -0.00016 (0.01392) 0.00212** (0.00017) -0.00009** (0.00001) 0.00090** (0.00007) 0.00043 (0.00081) 0.00287** (0.00016) -0.00006** (0.00000) 0.11541** (0.00217) 0.00023** (0.00005) -0.00093** (0.00028) -0.00063 (0.00071)
LPM 2 Coef. (Std Err) 0.00163** (0.00040) 0.00007 (0.00005) 0.00045* (0.00023) -0.00001 (0.00001) -0.00007 (0.00007) 0.05695* (0.02891) 0.00352 (0.01398) 0.00234** (0.00017) -0.00008** (0.00001) 0.00101** (0.00007) -0.00435** (0.00152) 0.00368** (0.00017) -0.00005** (0.00000)
0.00031** (0.00005) -0.00121** (0.00028) 0.00028 (0.00072) 50
LPM 3 Coef. (Std Err) 0.00081* (0.00040) 0.00009 (0.00005) -0.00019 (0.00024) 0.00000 (0.00001) 0.00002 (0.00007) 0.05440 (0.02903) 0.00784 (0.01396) -0.00216 (0.00184) 0.00024** (0.00005) -0.00093** (0.00027) 0.00523** (0.00157) 0.00239** (0.00017) -0.00006** (0.00000) 0.11716** (0.00219) 0.00025** (0.00005) -0.00119** (0.00028) 0.00033 (0.00072)
LPM 4 Coef. (Std Err) -0.00737** (0.00046) 0.00020** (0.00006) -0.00943** (0.00033) 0.00015** (0.00001) 0.00152** (0.00008) 0.14931** (0.03030) -0.02076 (0.01490) -0.00186 (0.00186) 0.00025** (0.00005) 0.00032 (0.00028) 0.00078 (0.00159) 0.00228** (0.00017) -0.00005** (0.00000) 0.12204** (0.00232) 0.00023** (0.00005) -0.00141** (0.00028) -0.00053 (0.00074)
LPM 5 Coef. (Std Err) -0.00716** (0.00046) 0.00019** (0.00006) -0.00936** (0.00034) 0.00015** (0.00001) 0.00168** (0.00008) 0.13894** (0.03055) -0.01938 (0.01486) -0.00205 (0.00185) 0.00026** (0.00005) 0.00034 (0.00028) 0.00050 (0.00158) 0.00223** (0.00021) -0.00005** (0.00001) 0.12259** (0.00234) -0.00083** (0.00012) -0.00275** (0.00037) 0.00006 (0.00077)
Variable
Portfolio Price control*portfolio Drug age Drug age squared Number of countries launched in Number of countries launched in squared Number of therapeutic classes Global importance Local importance Global importance*price control Local importance*price control French firm German firm Italian firm Japanese firm UK firm US firm French firm*price control
LPM 1 Coef. (Std Err) -0.00033** (0.00004) -0.00013** (0.00003) -0.00883** (0.00014) 0.00023** (0.00001) 0.00714** (0.00019) -0.00022** (0.00001) 0.00290** (0.00042) 0.02870** (0.00720) 0.17952** (0.01235) -0.00614 (0.00945)
LPM 2 Coef. (Std Err) -0.00076** (0.00004) -0.00001 (0.00004) -0.00893** (0.00014) 0.00023** (0.00001) 0.00683** (0.00020) -0.00021** (0.00001) 0.00300** (0.00042) 0.02725** (0.00724) 0.18707** (0.01240) -0.00464 (0.00951)
LPM 3 Coef. (Std Err) -0.00023** (0.00004) -0.00012** (0.00003) -0.00883** (0.00014) 0.00023** (0.00001) 0.00720** (0.00019) -0.00022** (0.00001) 0.00286** (0.00042) 0.02610** (0.00721) 0.18220** (0.01234) -0.00334 (0.00946)
LPM 4 Coef. (Std Err) -0.00024** (0.00004) -0.00012** (0.00003) -0.00839** (0.00014) 0.00021** (0.00001) 0.00810** (0.00019) -0.00024** (0.00001) 0.00316** (0.00042) 0.01637* (0.00753) 0.17464** (0.01238) 0.00125 (0.00997)
LPM 5 Coef. (Std Err) -0.00078** (0.00008) -0.00010** (0.00003) -0.00834** (0.00015) 0.00021** (0.00001) 0.00796** (0.00021) -0.00024** (0.00001) 0.00325** (0.00046) 0.01471 (0.00774) 0.17082** (0.01236) 0.00132 (0.00995)
-0.01780 (0.01678)
-0.01484 (0.01686)
-0.02118 (0.01677)
-0.01439 (0.01680)
-0.01330 (0.01676)
0.00231 (0.00179) -0.00308 (0.00169) -0.00654** (0.00180) -0.00875** (0.00146) 0.01556** (0.00211) 0.00048 (0.00149) 0.00581* (0.00232) 51
Variable
LPM 1 Coef. (Std Err)
German firm*price control Italian firm*price control Japanese firm*price control UK firm*price control US firm*price control
LPM 2 Coef. (Std Err) 0.00323 (0.00223) 0.00918** (0.00237) 0.01001** (0.00189) -0.00794** (0.00276) -0.00509** (0.00193)
LPM 3 Coef. (Std Err)
LPM 4 Coef. (Std Err)
LPM 5 Coef. (Std Err)
Included Fixed Effects
Year Class
Year Class
Year Class Country
Year Cntry*Class
Year Cntry*Class Firm
N Adjusted Rsq Mean of entry
290344 0.057 0.025
290344 0.049 0.025
290344 0.058 0.025
290344 0.090 0.025
290344 0.097 0.025
* = 5% significance, ** = 1 %. All specifications include year and class fixed effects. Standard errors have not been adjusted for heteroskedasticity.
52
Table C3: Duration Models
Variable
Number of new drugs in market N new drugs squared Number of old drugs in market N old drugs squared Number of potential competitors Class global importance Class local importance Population (10s of millions) Population squared GDP per capita ($1000s) Price controls Country experience Country experience squared Home country International experience Class experience Country-class experience
Hazard 1 Hazard Rate (Robust Std Err)
Hazard 2 Hazard Rate (Robust Std Err)
1.131** (0.024) 0.995* (0.002) 1.040* (0.016) 0.999 (0.001) 0.994 (0.006) 5.850 (10.916) 0.264* (0.173) 1.041** (0.008) 0.998** (0.000) 1.042** (0.005) 0.838** (0.033) 1.079** (0.011) 0.999** (0.000) 1.541** (0.119) 1.015** (0.004) 0.955 (0.027) 0.984 (0.046)
0.787** (0.054) 1.016* (0.007) 0.726** (0.062) 1.010* (0.004) 1.061* (0.029) 9.279 (65.540) 0.049 (0.076) 0.918 (0.172) 1.009 (0.005) 1.020 (0.038) 0.501** (0.118) 1.061* (0.028) 0.999 (0.001) 2.171** (0.435) 1.019* (0.010) 0.998 (0.077) 0.814 (0.102)
53
Variable
Portfolio Price control*portfolio Number of countries launched in Number of countries launched in squared Number of therapeutic classes Global importance Local importance Global importance*price control Local importance*price control Included Fixed Effects N Log likelihood
Hazard 1 Hazard Rate (Robust Std Err)
Hazard 2 Hazard Rate (Robust Std Err)
0.982** (0.003) 0.997* (0.001) 1.410** (0.021) 0.990** (0.001) 1.106** (0.032) 2.074** (0.549) 3.901** (0.994) 1.031 (0.335) 0.845 (0.279) Year Class
0.992 (0.006) 0.996 (0.004) 1.364** (0.055) 0.992** (0.002) 1.072 (0.084) 4.863* (3.170) 6.214** (3.866) 0.489 (0.344) 0.667 (0.454) Year Country*Class
267391 -52249.624
30284 -5744.1628
*= 5% significance, ** = 1 %. Hazard 2 is estimated on a random subset of therapeutic classes, which reduces the number of observations.
54
Table C4: Robustness to Adjustments for Standard Errors LPM Variable Number of new drugs in market N new drugs squared Number of old drugs in market N old drugs squared Number of potential competitors Class global importance Class local importance Price controls Population (10s of millions) Population squared GDP per capita ($1000s) Country experience Country experience squared Home country International experience Class experience Country-class experience Portfolio Price control*portfolio Drug age Drug age squared Number of countries launched in Number of countries launched in squared Number of therapeutic classes
Coef. (Std Err) -0.00737** (0.00046) 0.00020** (0.00006) -0.00943** (0.00033) 0.00015** (0.00001) 0.00152** (0.00008) 0.14931** (0.03030) -0.02076 (0.01490) 0.00078 (0.00159) -0.00186 (0.00186) 0.00025** (0.00005) 0.00032 (0.00028) 0.00228** (0.00017) -0.00005** (0.00000) 0.12204** (0.00232) 0.00023** (0.00005) -0.00141** (0.00028) -0.00053 (0.00074) -0.00024** (0.00004) -0.00012** (0.00003) -0.00839** (0.00014) 0.00021** (0.00001) 0.00810** (0.00019) -0.00024** (0.00001) 0.00316** (0.00042)
LPM, White SE Coef. (Std Err) -0.00737** (0.00053) 0.00020* (0.00008) -0.00943** (0.00040) 0.00015** (0.00002) 0.00152** (0.00008) 0.14931** (0.03321) -0.02076 (0.01613) 0.00078 (0.00136) -0.00186 (0.00226) 0.00025** (0.00006) 0.00032 (0.00028) 0.00228** (0.00020) -0.00005** (0.00001) 0.12204** (0.00525) 0.00023** (0.00004) -0.00141** (0.00024) -0.00053 (0.00088) -0.00024** (0.00004) -0.00012** (0.00003) -0.00839** (0.00015) 0.00021** (0.00000) 0.00810** (0.00026) -0.00024** (0.00001) 0.00316** (0.00047)
55
LPM, GLS Coef. (Std Err) -0.00943** (0.00048) 0.00035** (0.00007) -0.01144** (0.00036) 0.00020** (0.00002) 0.00195** (0.00009) 0.18561** (0.03166) -0.03322* (0.01613) -0.00080 (0.00164) -0.00109 (0.00199) 0.00030** (0.00005) 0.00061* (0.00029) 0.00282** (0.00018) -0.00005** (0.00001) 0.11425** (0.00319) 0.00031** (0.00005) -0.00181** (0.00028) -0.00014 (0.00076) -0.00038** (0.00004) -0.00012** (0.00003) -0.01101** (0.00016) 0.00026** (0.00001) 0.01149** (0.00021) -0.00035** (0.00001) 0.00428** (0.00042)
LPM, SLS
Logit
Coef. (Std Err) -0.01784** (0.00112) 0.00097** (0.00014) -0.02057** (0.00091) 0.00040** (0.00004) 0.00429** (0.00024) 0.26927** (0.08089) -0.06583 (0.04080) -0.00497 (0.00449) -0.00194 (0.00472) 0.00072** (0.00013) 0.00013 (0.00078) 0.00490** (0.00046) -0.00007** (0.00001) 0.18783** (0.00580) 0.00073** (0.00012) -0.00274** (0.00090) -0.00086 (0.00192) -0.00089** (0.00010) -0.00032** (0.00009) -0.03068** (0.00059) 0.00069** (0.00002) 0.02942** (0.00061) -0.00082** (0.00003) 0.00890** (0.00105)
Coef. (Std Err) -0.179* (0.084) 0.010 (0.009) -0.213** (0.076) 0.004 (0.004) 0.082** (0.030) 3.916 (6.246) -3.764 (3.908) -1.307** (0.438) -0.376 (0.233) 0.015* (0.006) -0.134 (0.098) 0.072* (0.030) -0.001* (0.001) 2.434** (0.202) -0.005 (0.009) -0.082 (0.079) -0.003 (0.135) 0.004 (0.009) -0.020** (0.006) -0.569** (0.038) 0.015** (0.002) 0.412** (0.037) -0.009** (0.002) 0.098 (0.072)
LPM Variable Global importance Local importance Global importance*price controls Local importance*price controls Included Fixed Effects
Coef. (Std Err) 0.01637* (0.00753) 0.17464** (0.01238) 0.00125 (0.00997) -0.01439 (0.01680)
N 290344 * = 5% significance, ** = 1 %.
LPM, White SE Coef. (Std Err) 0.01637 (0.00904) 0.17464** (0.02544) 0.00125 (0.01197) -0.01439 (0.03495) 290344
56
LPM, GLS Coef. (Std Err) 0.01846* (0.00774) 0.18284** (0.01626) 0.00141 (0.01033) -0.03741 (0.02187) Year Country*Class 204250
LPM, SLS
Logit
Coef. (Std Err) 0.02347 (0.02028) 0.22447** (0.02827) 0.02431 (0.02667) -0.00246 (0.03880)
Coef. (Std Err) 1.667 (1.186) 1.965 (1.626) 0.058 (1.595) 1.706 (1.965)
88616
10879
