1 Introduction

Maintaining competition in the pharmaceutical industry has long been a challenge for regulators and competition authorities. The entry of an increasing number of generics should have “naturally” led to more competition and shrunk profit margins. Generic penetration is indeed high: the generics’ share of market volume has increased substantially, reaching 85% of drug prescriptions in 2016 (Bosworth et al., 2018). However, in terms of value, they commanded a market share of 19% in the US, 26% in the UK, 21% in Germany, and below 20% in France, Belgium, and Switzerland, with similar figures in other countries (Statista, 2020).Footnote 1

Before any potentially anti-competitive behaviour can be assessed, “a” relevant market must first be defined. Doing so can identify economically significant competitive constraints on a specific firm (or set of firms). Firms’ exercise of market power is constrained by demand- and supply-side substitution, as well as by potential entry. Assuming an initially competitive context, market definition involves identifying which of these constraints ought to be countered (e.g., through exclusion or a merger) for the firm (or the set of merging firms) to exercise significant market power.

In this paper, we gauge whether competitive constraints evolve as a consequence of the “genericization” of a particular molecule in a therapeutic market. We show that traditional (often implicit) assumptions about the shape of the relevant market must be re-assessed in the case of the pharmaceutical industry. In particular, we find that the entry of a generic alters market boundaries, decreasing the relevance of analyses based on pre-entry data for anticipating ex-post market power, and vice versa. For instance, a Hypothetical Monopolist with control of all generic versions of a single molecule (but not of the originator) would be able, on average, to bring about a Small but Significant Non-transitory Increase in Price (SSNIP) of at least 5%-10%, despite the presence of substitutes. That is, the market for generics of a given originator drug represents a relevant economic market (or relevant market for short) on its own. Concomitantly, we observe a (post entry) drop in the elasticity of substitution between the drugs that had been competing with each other prior to the entry of a generic. In other words, generic entry produces a softening of intermolecular competition.

Our analysis is, in part, inspired by a number of recent competition enforcement decisions where market definition was pivotal. While market definition is often a fairly routine exercise, in a number of cases it turned out to be central: Servier/PerindoprilFootnote 2 (EU), GSK/Paroxetine (UK),Footnote 3 and Vyera/Daraprim and Boehringer Ingelheim/Aggrenox (US), all reviewed in Sect. 4.

In the Servier case,Footnote 4 the EU’s General Court concluded that the EU Commission had “made a series of errors in defining the relevant market.”Footnote 5 At the time of writing, the appeal to the European Court of Justice (ECJ) brought by the Commission is pending resolution. In GSK/Paroxetine, the UK’s Competition Appeals Tribunal (CAT)Footnote 6 upheld the Competition and Markets Authority’s finding of abuse, but adopted a different line of reasoning to delineate the relevant market. In Daraprim, the Federal Trade Commission (FTC) observed that a sustained price hike was direct evidence pointing to a narrow market. Meanwhile, in the Aggrenox case, the Connecticut District CourtFootnote 7 reasoned that the post-generic entry price drop was direct evidence that the relevant market only included Aggrenox and its generic versions.Footnote 8 The econometric evidence reported below is consistent with the cogent economic reasoning of the CAT, FTC, and Connecticut District Court, while the ECJ has not yet pronounced itself.

To identify relevant markets in the pharmaceutical industry, we estimate residual demand functions under different market structures (Baker & Bresnahan, 1985; Scheffman & Spiller, 1987). Our empirical exercise is (conceptually) inspired by the logic of the Hypothetical Monopolist Test (HMT). While this approach rests on solid conceptual foundations, operationalizing the HMT is challenging.Footnote 9 In practice, competition authorities have had to approximate the HMT by exploiting historical data (customers’ reactions to a significant price change), estimates of diversion ratios (e.g., inferred from customer surveys) or simulations, to name a few. In some rare instances, authorities have relied upon econometric estimations of demand elasticities.Footnote 10 Although the residual demand approach and the HMT have their differences,Footnote 11 both focus on the ability to raise prices above some competitive benchmark.

In line with most panel econometrics approaches, we rely on the concept of an economic market (in the sense of Marshall 1920).Footnote 12,Footnote 13 More precisely, we approach market delineation by estimating residual demands that factor in competitors’ reactions. The estimation of own and cross-price elasticities allows for the identification of the source and strength of the competitive constraints faced by firms (Davis and Garcés 2010). This is in line with the European Commission’s emphasis on (demand) substitution to delineate markets.Footnote 14 Baker and Bresnahan (1985) and Scheffman and Spiller (1987) pioneered the implementation of the residual demand function approach in order to delineate antitrust markets for enforcement purposes.

Markets for prescription drugs are well suited to the residual demand approach. First, extremely rich product-level data are available. The dataset at our disposal comprises individual prices, quantities and promotional efforts for 125 molecules sold in the US over forty quarters.

Second, markets are exposed to frequent competitive shocks in the form of generic entry, which can be precisely identified and timed. Generic entry (GE) occurs frequently, and its timing is essentially exogenous to initial market conditions: most often, the trigger is the expiration of patents that were filed several decades earlier. In our sample, 64 molecules in 31 different candidate markets lost exclusivity during the period of analysis.

We exploit GE episodes as a competition shock that can be used to measure the initial level and the change in competitive pressure faced by originators that still benefit from exclusivity. A defining feature of our procedure is the identification of the set of drugs that exercise competitive pressure on one another. More precisely, we begin our analysis by estimating cross-price elasticities, distinguishing between generic and branded competitors. Our results suggest that, on average, intermolecular competition is vibrant prior to GE, pointing to broad relevant markets comprised of various drugs. We also find that GE generates market-wide shockwaves that reshape the nature and strength of competitive constraints. On average, the entry of generic competitors for one drug shrinks the initial relevant market: the genericized drug “drops out” in that it no longer constrains the pricing power of the drugs that still enjoy exclusivity. In other words, the intensification of intramolecular rivalry resulting from GE softens intermolecular competition.Footnote 15 This occurs because a previously significant competitive constraint fades into economic insignificance. We use this preliminary analysis as a guide for identifying potentially relevant markets.

Given the implication of the above, namely that market boundaries may be very narrow, we focus on generic producers of a given molecule—the narrowest possible market we can possibly identify with our data. Unsurprisingly, we find that own- and cross-price elasticities among generics are high, meaning they have essentially no market power and face very strong competition from other generic producers of the same molecule. With such estimates, we can proceed to more formal hypothetical monopoly tests (HMT), which we describe in Sect. 3.

Taken together, our empirical findings point to the existence of multiple candidate markets that are relevant for enforcement purposes.Footnote 16 Our empirical results, which identify the “average effects” on an “average product” in a large market, indicate that intermolecular competition is vibrant prior to GE, largely driven by non-price instruments. Post-GE, the molecule becomes a relevant market where price competition is rife. Indeed, the relevant market post-GE may be even narrower: generic versions (i.e., excluding the originator) may constitute an antitrust market. This serves to highlight that there is no “natural” or “unique” definition of the antitrust market: the delineation of the relevant market cannot be dissociated from the nature of the competitive concern, i.e., the theory of harm.

The consubstantial nature of market definition with the theory of harm can be illustrated with simple examples inspired by the results of Sect. 3. Imagine a market composed of three originator drugs, A, B, and C. C is the market leader with a 70% market share, while A and B each command a 15% share. These drugs initially benefit from exclusivity and hence significantly constrain each other, such that a firm controlling all three would be in a position to profitably and sustainably increase prices by 5–10% (or more). Thus, if the concern is coordinated behaviour, the candidate market should consist of A, B, and C.

In the context of generic foreclosure, the exercise involves evaluating the competitive constraints the infringing firm faces in both factual and counterfactual (i.e., absent the behaviour that is a source of concern) terms. For instance, in the case of generic foreclosure by B, the relevant competitive constraint is exercised by the generic version of B, and the relevant antitrust may well be molecular, even though the molecular market has yet to emerge, and despite the molecule’s comparatively low initial market share. The reason is that, following generic entry, a hypothetical monopolist limited to molecule B may be able to achieve a significant and profitable price increase over and above the competitive benchmark price.

With respect to mergers, our analysis indicates that, for a given theory of harm, market delineation may change in the event of generic entry. This is because vigorous intermolecular competition may vanish quickly if the drugs competing with those of the merged entities experience GE. A merger between A and B may be cleared under the assumption that the merged entity would face significant constraints from C. Our analysis indicates that if C is close to expiry, and if there are no product launches in the near future, the merger between A and B would create an entity that could rapidly gain significant market power. The reason underpinning C’s competitive constraint fading into irrelevance is the dramatic drop in promotional spending that follows GE (Castanheira et al., 2019; Lakdawalla & Philipson, 2012).Footnote 17 Such a situation would represent an instance of Type II error (an anticompetitive merger being waved through).

The suggestion that the relevant market should be made contingent on the infringement (actual or potential) has been discussed for some time (Salop, 2000, Rey et al. 2005, and Glasner & Sullivan, 2020 for an in-depth exposition and analysis), but enforcers and courts have so far been reluctant to endorse the approach, at least in the European Union.

The remainder of the paper is organized as follows. Section 2 describes the dynamics of competition in the pharmaceutical industry before and after patent expiration. In Sect. 3, we describe the data, implement our empirical analysis, and gauge how competitive constraints evolve over time. We also assess pricing power under different market structures. Section 4 discusses the potential implications of our results in light of past and ongoing cases. Section 5 concludes.

2 Competitive dynamics in the pharmaceutical industry

The Anatomical Therapeutic Chemical (ATC) classification groups drugs at different levels of aggregation, numbered 1–5. The ATC3 level encompasses the set of potential treatments for a given medical condition. As such, ATC3 groupings represent the potential substitute products that a given “customer”—here a patient-doctor pair—have at their disposal to address a medical condition. The EU Commission routinely uses ATC3 groupings as an initial step in defining a relevant market (Greenaway et al., 2009).

The definition of a “relevant market” can be complex when firms compete through price and non-price instruments (Sovinsky Goeree, 2008), and/or when other forms of public intervention dampen the role of price as a competitive tool. The combination of high R&D, high promotion intensity,Footnote 18 regulation, and information asymmetries generates particularly complex competition dynamics. In such a context, market shares and competitors’ price reactions, two standard indicators of market power, are less informative.Footnote 19

2.1 Market shares

Market shares are often used as a screening device to gauge market power (Motta, 2004). However, the mapping of market concentration onto consumer welfare is far from clear in the presence of product differentiation (e.g., quality differences) and non-price competition (e.g., promotion). For instance, in a model informed by the workings of the pharmaceutical industry, Lipatov et al. (2021) document the possibility that consumer welfare may be higher in concentrated markets. They also show that, somewhat surprisingly, the benefits of entry may be greater in less concentrated markets. These findings suggest that market shares (even derived from properly defined markets) may not be particularly informative for inferring potential harm (or its absence) in the context of the pharmaceutical industry.

In the scenarios described below, we propose to quantify market power directly. We compare competitive prices, quantities and profits in the factual and counter-factual scenarios.Footnote 20 Under such circumstances, large rents are direct evidence of dominance (Browdie et al., 2018), making the calculation of market shares less relevant.

2.2 Competitors’ and consumers’ reactions to price fluctuations

Patents offer 20 years of intellectual property rights (IPRs) to the firm that develops a new potential treatment.Footnote 21 Patent expiration triggers a unique competitive shock as generics, which are near perfect substitutes, legally compete with the originator. These generics are often sold at less than half the price of the original branded product and quickly erode the originator’s market share (Grabowski et al., 2014; Scott Morton & Kyle, 2012, Reiffen and Ward 2005). This is not surprising, as generics are bioequivalent products that have been explicitly recognized as such by health authorities.Footnote 22

A perplexing feature of generic entry is that the prices of the other on-patent molecules in the same ATC3 category are barely affected (Jena et al., 2009; Lakdawalla, 2018), and their volume market share can even increase (Castanheira et al., 2019; Grabowski et al., 2014; Lakdawalla & Philipson, 2012; Regan, 2008). The lack of price reaction has sometimes led competition authorities (and scholars) to conclude that markets are molecular. However, drawing such conclusions is highly contentious: prior to generic entry, the drug may have been actively competing against other originators, pointing to a broader antitrust market rife with intermolecular competition.

Hence, for a given set of drugs, evidence from different points in time may either indicate narrow (molecular) markets or a broad (multi-molecule) market, potentially leading to controversy (see Sect. 4 for concrete cases). In the next section, we show that these alternative market definitions need not be mutually exclusive. We also show that the “correct” market definition cannot be dissociated from the theory of harm.

3 Empirical analysis: delineating relevant markets in the pharmaceutical industry

This section proposes a three-pronged exercise. First, we perform an empirical estimation of a firm’s own- and cross-price and advertising elasticities. We exploit the existence of natural experiments—the entry of generics after a patent expires—to test whether the elasticity coefficients of a molecule i vary around the date of generic entry. This allows us to gauge whether competitive constraints evolve as a consequence of the “genericization” of a particular molecule in a particular ATC3 market. This first set of results indicates that on-patent drugs are only constrained by drugs that also benefit from exclusivity. Note that this exercise is preliminary: it identifies a change in the competitive landscape triggered by generic entry, but does not allow us to establish market boundaries.

Second, we perform a SSNIP test by assessing how demand elasticities vary across market structures. We start from the narrowest possible candidate market in our data: the generic versions of a given molecule. We then consider a more concentrated hypothetical market: one involving a single supplier controlling both the originator product and the generic versions. As detailed by Davis and Garcés (2010, pp. 204–17), this exercise evaluates whether a SSNIP would be profitable in the absence of a price reaction by other producers.

Third, as a robustness check, we propose an additional exercise that we apply to the same candidate markets. This complementary exercise is in line with the 2010 US Horizontal Merger Guidelines, which seek to identify whether the hypothetical monopolist would impose a SSNIP on at least one of the products under the control of the newly merged entity. Our implementation of these guidelines builds on Azar et al. (2018), and Azar and Vives (2020). Concretely, we use own- and cross-price elasticities to compute how a firm would separately modify the price of each of the varieties under its control, instead of holding all other prices constant. This exercise is also related to a Full Equilibrium Relevant Market (FERM) test, originally associated with the 1984 US Horizontal Merger Guidelines (see Davis and Garcés 2010, p162, and Ivaldi & Lorincz, 2011).

To summarize, we implement a SSNIP test using two approaches on two candidate markets. Implementing two methods in this way is similar to a sensitivity analysis and provides a robustness check of the market boundaries that we identify.

3.1 Data

Our starting dataset covers quarterly dollar revenues and physical quantities for all branded and generic prescription drugs sold in the US over the 40-quarter period 1994q1 to 2003q4. These have been obtained from the proprietary IMS-MIDAS database, which includes sales in both pharmacies and hospitals. The database is published by IMS-Health, one of the most important medical information providers (now known as IQVIA).

In the US, pricing is free in the sense that publicly funded Social Security does not negotiate a price cap price for drugs that benefit from exclusivity.Footnote 23 Only the monopsony power of Third Party Payers (principally private insurers) mitigates the pharmaceutical firms’ pricing power.

All the drugs in IMS-Health are classified according to the Anatomical Therapeutic Chemical (ATC) classification system. In the IMS data, generics are listed under the name of the active ingredient.Footnote 24 We thus identified an initial list of ATC3 markets that feature at least one GE by selecting the markets where there are two or more different products for the same molecule name, and some of the drug names are the same as the molecule (e.g., Fluoxetine is the active ingredient of Prozac).Footnote 25,Footnote 26

For each of the drugs belonging to the selected markets, we computed deflated revenues (R) by dividing the nominal value of sales by the producer price index for the pharmaceutical industry as published by the Bureau of Labor Statistics. Quantities (Q) are reported in standard units that represent the number of dose units sold for each product; this corresponds to one capsule or tablet of the smallest dosage or five millilitres of a liquid (i.e., one teaspoon). Standard units allow a comparison across drug forms and dosages, since all presentations are subsumed into the same unit of observation. We then computed the average price of a molecule (P) by dividing R − i.e., the revenues for all the different packages − by total quantity Q. We applied the same method to construct an average price index of competing molecules in the same market.Footnote 27

We then purchased drug-level information on promotion expenditures in the most important ATC3 markets in terms of sales and promotional effort.Footnote 28 The promotion regressor used in the demand estimations is computed by applying the perpetual inventory method:

$$A_{it} = \, (1 - r) \, A_{it - 1} + \, I_{it} ,$$

where Iit is the quarterly expenditure on promotion for drug i retrieved from IMS, and ρ is the quarterly depreciation rate, assumed to be 0.1 − i.e., about 35% per year.Footnote 29

The final sample is determined by data availability: we include all the molecules for which we can compute all variables described above. The sample includes 31 different ATC3 markets encompassing 125 molecules, of which 64 experienced generic entry during our time window. The final sample includes the largest therapeutic markets by sales, such as proton pump inhibitors (used to suppress gastric acid), selective serotonin reuptake inhibitor (used to treat depression), statins (used to lower cholesterol), and all classes of anti-hypertensive drugs.

Table 1 provides descriptive statistics for these variables. We computed the “Promotion of Competing Molecules” as the total promotional spending on all other drugs in the same ATC3 market applying the perpetual inventory method described above. Since generic suppliers do not engage in promotion, this variable accounts for the promotional effort of originators.

Table 1 Descriptive statistics

With respect to price, we construct two indices: one pertaining to branded competitors and the other limited to generic suppliers. The “Price of competing branded (resp. generic) molecules in ATC3” is the ratio between total revenues and total quantities in the ATC3 market of branded drugs (resp. generics), after subtracting the revenues and quantities of drug i. The result represents the average price of all other branded molecules on the market as well as the average price of generics.

3.2 Intermolecular competition: specification and IV strategy

As summarized in the introduction to this section, the first set of regressions aims to assess whether price elasticities vary around GE. To this end, we estimate a log–log specification to obtain the elasticity of a given molecule’s volume market share with respect to its own price and promotion effort, as well as the corresponding cross-elasticities for competing drugs.Footnote 30 More precisely, prior to GE, the LHS variable pertains to the quantity market share of the originator drug. Post-GE, we use the quantity market share of the originator plus its generic versions. The case of the well-known antidepressant drug Prozac can serve as an illustration. The active ingredient in Prozac, which experienced GE in the third quarter of 2000, is fluoxetine. In this example, the dependent variable is the quantity market share of fluoxetine. The latter is equal to the quantity market share of Prozac prior to 2000q3, and to the combined market share of Prozac and its generic competitors (e.g., Teva fluoxetine, Barr fluoxetine etc.) from 2000q3 onwards.

Taking advantage of the fact that our dataset contains markets where generic entry occurred, we estimate (average) elasticities for a typical molecule depending on its exclusivity status. Concretely, we let the quantity market share of a molecule i at quarter t, msit, depend on own and competitors’ price p and advertising a (with all these variables expressed as logarithms):

$$\begin{aligned} ms_{it} = & \beta_{1} p_{it} + \beta_{2} a_{it} + \beta_{3}^{B} \left( {1 - E_{it} } \right)p_{ - it}^{B} + \beta_{3}^{G} \left( {1 - E_{it} } \right)p_{ - it}^{G} + \beta_{4}^{B} E_{it} p_{ - it}^{B} + \beta_{4}^{G} E_{it} p_{ - it}^{G} \\ + { }\beta_{5} \left( {1 - E_{it} } \right)a_{{ - { }it}} + \beta_{6} E_{it} a_{ - it} + \mu_{i} + \mu_{t} + \varepsilon_{it} \\ \end{aligned}$$

where \(p_{it}\) and \(a_{it}\) refer to the own price and advertising spend of drug i at time t, whereas the indices B and \(- i\) in \(p_{ - it}^{B}\) identifies the price of brand competitors and \(p_{ - it}^{G}\) is the price of generic competitors, both in the same ATC3 market. Eit is an indicator taking a value of 1 when the LHS molecule i experiences generic entry at time t, and zero otherwise.Footnote 31 Equation (1) includes a complete set of molecule/drug (\(\mu_{i} )\) and time (\(\mu_{t} )\) fixed effects. The fixed effect \(\mu_{i}\) captures persistent molecule-specific differences in market shares driven by unobserved factors such as the vintage of the drug, the quality of the sales force or the reputation of the pharmaceutical companies marketing the drugs. The fixed effect \(\mu_{t}\) controls for time-specific shocks that are common to all molecules. Finally, the error term \(\varepsilon_{it}\) captures molecule-specific demand shocks as well as measurement errors.

Since the variables are expressed in logarithms, the \(\beta\) coefficients measure the elasticity with respect to each of the associated variables, namely price (p) and advertising (a). This specification is similar to the one reported in Baker and Bresnahan (1985), who point out that, from an antitrust perspective, it is not necessary to determine whether it is competition from firm A or B that puts the most effective brake on C’s market power: it is the combined effect of all other firms that is of interest.Footnote 32

Thanks to the large set of GE events in our database, we can test whether competitive constraints are different before (\(\beta_{3}^{B}\) and \(\beta_{3}^{G}\)) and after (\(\beta_{4}^{B}\) and \(\beta_{4}^{G}\)) generic entry (\(E_{it}\)). If the point estimate of one of the \(\beta_{3}\) elasticities turns out to be statistically different from the associated \(\beta_{4}\), then we will infer that the constraints exercised by branded or generic competitors do change over the molecule’s lifecycle. Similarly, the coefficients \(\beta_{5}\) and \(\beta_{6}\) reveal the impact of competitors’ promotional effort before and after GE. Since generics generally do not advertise at all, it is not necessary to distinguish between promotional efforts by branded and generic competitors.

Own price and promotion are likely affected by endogeneity and measurement problems. Endogeneity can be due to feedback from market share shocks to subsequent price and promotional effort (reverse causality). Potential measurement errors of both price and promotion stem from the difficulty in observing and monetarily quantifying the effort of sales representatives when they visit physicians. Both of these issues can result in correlation between our regressors and the error term. To address them, we implement an IV strategy based on two sets of instruments that should be highly correlated with supply-side changes in promotion and prices, but not with the error term in Eq. (1).

Following the methodology proposed by Chaudhuri et al. (2006), our first set of instruments consists of the number of presentations, linear and squared.Footnote 33 The rationale for using this instrument is that the introduction of a new presentation is generally accompanied by increased promotional activities. Recall that our measure of promotional effort includes the distribution of free samples, which should increase when a new dosage or formulation is launched on the market. While the number of presentations is likely to be highly correlated with promotion expenditures, it is likely not correlated with the measurement error, since the number of presentations can be accurately measured in our data. At the same time, as explained in Chaudhuri et al. (2006), the number of presentations is related to a molecule’s average price p, since variations in p stem in part from variations in presentations available in each period.

As a second set of instruments, we also use the number of quarters before/after generic entry, linear and squared. These instruments capture the dramatic changes in pricing and promotion strategies by a brand manufacturer in the periods leading up to and following GE. As patent expiration is exogenous (unrelated to patients and doctors’ decisions) and can be accurately timed, these instruments can be reasonably considered unrelated to the error term,\(\varepsilon_{it}\). Before discussing our results, we note that this choice of instruments is validated by the Kleibergen-Paap rk-statistic (K-P) for under-identification, the Hansen J-test for over-identifying restrictions, and the C-statistic to test for the endogeneity of one or more instruments (regressors), as shown in the tables below.

Two last clarifications are in order. First, our reduced-form regression is similar to the standard logit demand model, although it has additional flexibility in estimating cross-price elasticities. The main difference with respect to a logit model is that cross-price elasticities are not assumed to be proportional to market shares and/or to the prices of competing products (see Werden & Froeb, 1994 and Nevo, 2000).Footnote 34 Second, in the context of this paper, a nested logit with a pre-determined set of nests would impose excessive restrictions: our purpose is precisely to assess whether the nest structure is affected by GE.

3.3 Intermolecular competition: results of the first exercise

Table 2 reports the estimates of price and promotion elasticities in our sample. Column (1) presents the results with molecule and time fixed effects, but before instrumenting for either price or promotion. All the coefficients are of the right sign and significant, save for competitors’ price post-GE. In column (2), we instrument for promotion but not for prices. The point estimates increase in absolute value, while the level of precision is maintained. In column (3), we augment the analysis by also instrumenting for price. The sign and precision of all the point estimates are maintained, and the magnitude of the estimated coefficients increases further.

Table 2 Price and advertising elasticities before and after loss of exclusivity

To be sure, a panel analysis can only reveal average effects, and is not a substitute for a detailed investigation by enforcement authorities, which must evaluate competitive constraints in the context of each particular case. However, we contend that the results reported below offer valuable background information. In particular, they reveal that, paradoxically, a drug may exercise less competitive pressure on substitutes when its price goes down as a consequence of generic entry.

Table 2 reveals a number of findings: first, both own price and own advertising are important drivers of a drug’s market share. In column (1), we do not instrument for prices or advertising. As a result, we would expect a price elasticity biased towards zero, since higher prices may be associated with more intense advertising. In column (2), where we instrument for promotion, this conjecture does appear to be confirmed, as both own price and promotion elasticities increase. In column (3), where we also instrument for prices, both coefficients increase further. Note that they are precisely estimated across all specifications.

Turning to the competitors’ promotion, we see a similar stability in statistical significance and the increasing magnitude of the point estimates. Interestingly, drug i’s market share is as sensitive to its competitors’ advertising after generic entry as it was before.

The large point estimates of \(\beta_{2} , \beta_{5}\), and \(\beta_{6}\) in column (3) confirm that promotion is a central driver of competitive interactions in the pharmaceutical industry. While own promotion may be a match for robust marketing by branded competitors, the dramatic drop in promotion activities after GE means that generic producers will only compete for a significantly smaller market share.

Cross-price elasticities offer insights on how each category of competitors can constrain a firm’s market share through price competition. We focus our comments on column (3), where we instrument for both price and promotion. Prior to GE, we find that only brand competitors exercise a competitive pressure on molecule i (\(\beta_{3}^{B} = 0.647\), significant at the 5% level).Footnote 35 By contrast, genericized molecules do not exercise a constraint (\(\beta_{3}^{G} = 0.034\) and statistically non-significant). In other words, while a drug is on patent, it is only price constrained by other on-patent drugs.Footnote 36 When one of its competitors loses exclusivity, the latter’s price ceases to influence the drug’s market share (\(\chi^{2}\) test for \(\beta_{3}^{B} = \beta_{3}^{G}\): 3.12; p-value: 0.077). Insofar as cross-price elasticities reveal the boundaries of an economic market, this implies that a genericized drug “exits” its former ATC3 market.

The picture is different for genericized drugs (coefficients \(\beta_{4}^{B}\) and \(\beta_{4}^{G}\)). After drug i experiences generic entry, its market share becomes dependent on the prices of branded as well as generic competitors: the respective cross-price elasticities are 0.387 and 0.488, both significant at 5%. For competing genericized molecules, the cross-price elasticity grows 14-fold in comparison to the pre-GE period. This points to intermolecular rivalry among (low-priced) generics. Thus, the number of relevant competitors increases for genericized molecules, in contrast with the drugs that still benefit from exclusivity.

An important corollary of this contrast is that competitive constraints appear to be asymmetric: on-patent drugs do restrain genericized molecules, but the reverse is not true. The asymmetry is even more notable when we include non-price competition in the picture: post-GE, molecules continue to be constrained by the promotion of other molecules (\(\beta_{6} = - 0.489,\) statistically significant at the 1% level).

These empirical findings are consistent with Grabowski et al. (2009) and Castanheira et al. (2019) who noted that generic entry generally allows other on-patent drugs to expand their quantity market shares. Hence, Table 2 identifies a marked change in the competitive landscape over a relatively short period of time, caused by the exogenous event of a GE, which results in the intensification of competition for a single pre-existing drug.

3.4 Generics: intra- and intermolecular competition

While the above analysis reveals information about the evolving boundaries of the relevant market around GE, it falls short of evaluating whether and when a SSNIP may be profitable. We must accordingly start from the narrowest market, and from there explore how extending a firm’s control over products (say, through the acquisition of competitors) would modify its pricing power.

In our data, the narrowest candidate market is that of a given producer of a generic drug. In our sample, there are on average 12.45 generic producers of a single drug after GE (median 11). In line with the previous exercise, we proceed by estimating the following equation:

$$\begin{aligned} ms_{{it}}^{G} = \gamma _{1} p_{{it}}^{G} + \gamma _{2} p_{{it}}^{{ - G}} + \gamma _{3} p_{{it}}^{B} + \gamma _{4} p_{{ - it}}^{G} + \gamma _{5} p_{{ - it}}^{B} + \gamma _{6} a_{{it}}^{B} + \gamma _{7} a_{{ - it}}^{B} \\ & + \mu _{i} + \mu _{t} + \varepsilon _{{it}}^{G} \\ \end{aligned}$$

where \(ms_{it}^{G}\) and \(p_{it}^{G}\) are, respectively, the ATC3 quantity market share and the price of a molecule i produced by generic firm G at time t (e.g. Teva fluoxetine, one of the generic versions of Prozac). We denote with “\(- G\)” the other generic producers of the same molecule (e.g. Mylan fluoxetine). Accordingly, the variable \(p_{it}^{ - G}\) refers to the price set by these other generic companies for that same molecule i, whereas \(p_{it}^{B}\) is the price of the originator company (Eli Lilly’s Prozac).

As in Eq. (1), the variables denoted by a subscript “\(- i\)” refer to the other molecules/drugs in the same ATC3 market: \(p_{ - it}^{G}\) is the price index of other drugs that had already lost patent protection before molecule i, and \(p_{ - it}^{B}\) is the price index of other branded products still covered by patent protection (e.g. Merck’s Zoloft). Finally, the specification includes the advertising effort made by the originator of molecule i (to account for the few instances where advertising on molecule i is sustained after patent expiration)Footnote 37 and the total advertising effort of other brand producers of molecules still covered by patent protection in the same ATC3.

Table 3 reports summary statistics for the variables used in specification (2). Note that the number of observations is substantially larger. This is because the unit of observation is no longer a molecule, but a molecule and generic producer pair. For instance, our dataset contains 21 different generic producers of fluoxetine.

Table 3 Summary statistics of generic producers

Columns (1) and (2) in Table 4 contain the regression results of specification (2) above. We comment on columns (3) and (4) in the next sub-section. As in Table 2, column (1) is the specification without instrumental variables. In column (2), we instrument for both prices and promotion. Ceteris paribus, more generic entries reveal higher profit margins prior to GE, leading to steeper price drops (Reiffen & Ward, 2005; Scott Morton & Kyle, 2012). Accordingly, in column (2), we instrument for price and promotion using the number of generic producers of the same molecule and the headcount of producers of other drugs, on top of the number of presentations (linear and squared) that we used in specification (1). Comparing the coefficients in columns (1) and (2), we observe an almost tenfold increase in the size of the own-price elasticity and a 20-fold increase in the cross-price elasticity of other generic competitors. The Kleibergen-Paap rk-statistic (K–P) confirms the presence of endogeneity and the Hansen J-test validates the hypothesis that these instruments are both relevant and exogenous.

Table 4 Competition among generic producers

Focusing on column (2), the own- and cross-price elasticities for given generic producers of a single molecule (e.g., Barr and Teva fluoxetine) are large (resp. −8.693 and 6.643) and precisely estimated. This indicates that intramolecular competition is best described as undifferentiated Bertrand with no capacity constraints. By contrast, despite being bioequivalent, the originator drug behaves as a differentiated product, with a cross-price elasticity of 0.730.

As in the previous set of results, we observe a pattern of asymmetric competition: the price and promotion of other patent-protected molecules do exercise a significant constraint on genericized molecules. Hence, generic producers are “between a rock and a hard place”: they face fierce competition from other generic producers of the same molecule, and the overall market share of the genericized molecule is constrained by the price and promotion of other on-patent molecules.

Lastly, \(\gamma_{6}\), the coefficient for the promotion of the originator’s brand, has a positive sign and is significant. In our example, this would be the promotion for Prozac when the dependent variable is Teva fluoxetine or Barr fluoxetine. What this reflects is the well-known result that, absent significant differentiation, promotional effort also benefits direct competitors—promotion for Prozac, for example, benefits the generic producers of fluoxetine.

Since our dependent variable is the quantity market share, the cross-price results indicate that, following an increase in the price of one generic, the quantities would be principally diverted to other generics of the same molecule, while the leakage to the originator, other generics, or other drugs under exclusivity would be (very) limited. From Table 4, a 1% price increase by one generic producer would increase the combined market share of the other producers of the same molecule by 6.6%.

3.5 Simple (and direct) market delineation

With these results in hand, we are now in a position to perform our second and third exercises, each of which involves a different methodology. However, both address the same question: could a firm profitably increase its price if it controlled a wider proportion of the market?

To this end, we build on the results of Table 4. When there are multiple generic producers, a firm that only controls one version of the generic G would choose the price \(p_{i}^{G}\) that maximizes (static) profits:

$$\pi_{i} = \left( {p_{i}^{G} - c} \right)q_{i} = \left( {p_{i}^{G} - c} \right){ }A_{i} { }\left( {p_{i}^{G} } \right)^{{\upvarepsilon }} { },$$

where \(A_{i}\) is a demand shifter and \(\varepsilon ( < 0)\) is the own-price elasticity. In the logic of residual demand analysis, \(A_{i}\) depends on the competitors’ prices and promotion, as well as intrinsic characteristics that we captured through time- and product-specific fixed effects. The resulting pricing power obtained from the Lerner index is \(\frac{p - c}{p} = - \frac{1}{\varepsilon }\) or, equivalently, a price-to-cost ratio \(p/c = \frac{\varepsilon }{\varepsilon + 1}\). From column (2) of Table 4, the own-price elasticity of a given generic producer is estimated to be \(\gamma_{1} = - 8.693\). This implies a price-to-cost ratio equal to 1.13, which we use as our competitive market benchmark.

We build two counterfactual (hypothetical) scenarios, starting from different initial market structures. The aim of each is to delineate the relevant market, conditional on a prior degree of competitive rivalry.

3.5.1 Scenario 1: merger between generic suppliers

The first scenario we contemplate is that of a merger between generic suppliers of a particular molecule i. Under these circumstances, the narrowest candidate market is that of a single molecule and limited to non-originator producers. Hence, the test involves assessing whether a single firm gaining control of all the generic versions of molecule i could achieve a 5% price increase.

In the spirit of the HMT, our second exercise supposes that all other prices, and hence \(A_{i}\), would remain unchanged. The sole objective is to assess how the own-price elasticity of demand would vary if a single firm were to control the entire generic market of a given molecule (imposing that the price of all varieties would vary proportionally—we return to this issue in our third exercise below). This second exercise is thus a straightforward but direct test: if the elasticity drop justifies an increase in the price-to-cost ratio of more than 5%, then the boundaries of the relevant market will have been identified.

To assess the would-be elasticity associated with this hypothetical market structure, we estimate a variant of Eq. (3) where we aggregate the quantities of all the generic producers of a given molecule as a function of the average price of all other producer groups (e.g., generic suppliers of other molecules in the same ATC). The dependent variable then corresponds to the quantity market share of, say, all generic versions of fluoxetine (e.g. fluoxetine produced by Teva, Barr, Mylan, etc.). The results are presented in column (3) of Table 4. We find that the hypothetical entity with a monopoly on all generics of a given drug would face a price elasticity of − 4.5. Applying the same profit maximization logic, the hypothetical entity would set a price resulting in a price-to-cost ratio of 1.29.Footnote 38 In comparison with the benchmark price–cost ratio of 1.13 (and assuming constant marginal cost), this yields a price increase of 14%, meeting the SSNIP threshold. This is despite the fact the firm would still be constrained by the originator (cross-price elasticity: 2.9), other branded drugs (1.6), and generics of other molecules (0.61).

Our third exercise is less straightforward and allows the hypothetical entity to separately determine the price of each generic version of the drug. This is closely in line with the 2010 US Horizontal Merger Guidelines, article 4.1.1, which reads “Specifically, the test requires that a hypothetical profit-maximizing firm, not subject to price regulation […] likely would impose at least a [SSNIP] on at least one product […]. For the purpose of analysing this issue, the terms of sale of products outside the candidate market are held constant.” This third exercise is also close to a FERM test (Full Equilibrium Relevant Market, originally associated with the 1984 US Horizontal Merger Guidelines, Davis and Garcés 2010, p. 162)

To perform this exercise, we build on Azar et al. (2018), and Azar and Vives (2020). Their setting can be transposed to analyse the more complex problem of a firm separately choosing the prices of each version of the same genericized molecule. Taking the example of two generic drugs of the same molecule for simplicity, a firm that commands 100% of the profits of version i and a fraction \(\lambda\) of the profits of version j would maximize:

$$\left( {p_{i} - c} \right)q_{i} + \lambda (p_{j} - c)q_{j} = \left( {p_{i} - c} \right){ }A_{i} { }p_{i}^{\varepsilon } { }p_{j}^{\chi } + \lambda (p_{j} - c)A_{j} { }p_{i}^{\chi } { }p_{j}^{\varepsilon } { },$$

where \(\chi\) is the cross-price elasticity. We find that an interior solution to this optimization problem only exists for sufficiently small \(\chi\) and \(\lambda\). For \(\lambda = 1\) (full control of firm j), \(\chi\) would need to be smaller than 0.3 to warrant an interior solution. Instead, the estimate in column (2) of Table 4 is 6.6, implying a corner solution. In other words, the hypothetical entity would actually prefer to “choke” all but one version of the generics, to maintain one variant (e.g. Teva fluoxetine, and drive the production of Barr fluoxetine to zero).

Given that this firm would maintain all but one variant, intramolecular substitution would be limited to the originator. Hence, the own-price elasticity faced by this hypothetical firm can be proxied by \(\varepsilon + \chi = - 8.693 + 6.643 = - 2.05\). A firm facing that elasticity would select a price–cost ratio of 1.95, implying a price increase of 73% above the initial level (1.13). The higher predicted price increase (as compared to the elementary SSNIP test) aligns with theory. The Azar et al. inspired test provides a direct quantification of the diversion effects that a hypothetical monopolist would internalize. In the same spirit, a FERM test should point to narrower markets than the elementary SSNIP test.

While the two approaches produce different predicted price increases (14% and 73%), both clearly identify a narrow relevant market made up of a single molecule and limited to generic producers. Both estimates are below those obtained from comparisons of drugs that have attracted a differing number of generic entrants. For instance, Scott Morton and Kyle (2012, Fig. 12.6) provide “mirror image” evidence: price drops as a function of the number of entrants. They report that for molecules with 6 to 13 generic suppliers, prices are around 23% of the pre-GE brand price.Footnote 39

3.5.2 Scenario 2: from duopoly to monopoly on a molecular market

Our results indicate that, for an average market (panel analyses only yield average effects), the relevant market of a generic producer is the market of generics itself. This finding is obtained under a competitive benchmark of vigorous competition between generic suppliers of the same molecule. Market delineation may thus vary if the competitive benchmark is different.

We thus apply our second and third thought exercises to a different initial benchmark: a molecule duopoly consisting of the originator and a single generic supplier. Under this market structure, we assess whether a hypothetical monopolist encompassing the originator and the generic versions of a particular molecule would lead to a SSNIP. In contrast to the initial situation in column (2) of Table 4, the answer is immediate: by transitivity, moving from the crowded market (12.45 generic producers) to a monopoly at the molecular level must necessarily result in a SSNIP, since a hypothetical monopolist limited to generic versions of the drug would already impose a SSNIP.

However, the answer is less obvious if the starting point is a duopoly made up of the originator and a single generic producer. To answer that question, we re-estimate Eq. (2) with the total quantity market share (i.e. generic versions plus the originator) as the dependent variable. The results are displayed in column (4) of Table 4. The point estimate of the own-price elasticity that this hypothetical producer would face stands at − 2.7, yielding a price-to-cost ratio of 1.59. Hence, moving from a molecular duopoly to a single seller would result in a 23% price increase (price–cost ratio of 1.59 vs. 1.29), again surpassing the 5% threshold.

Applying the third test, the sum of own- and cross-price elasticities is: \(- 8.69 + 6.64 + 0.73 = - 1.32\), yielding a price–cost ratio of 4.125, which may appear high. Still, this magnitude is not dissimilar to the point estimates implied by Table 2, column (3), and are compatible with the price differences depicted in Scott Morton and Kyle (2012, Fig. 12.6).

The findings of our “elementary SSNIP” and “Azar et al.” tests are thus twofold. First, the two proposed methods yield price increases that range from conservative to high price hikes, but both approximations point to the same relevant market, given the competitive benchmark. In that sense, the resulting relevant market delineation should be considered to be robust.

Second, this evidence indicates the appropriate market definition is intrinsically linked to the competitive concern. To illustrate, it is possible to think of different merger scenarios. If the merger involves a subset of generic versions of a particular molecule, our results indicate that the relevant market should only encompass generic suppliers of that particular molecule, leaving out the originator. If the proposed merger encompasses the generic suppliers and the originator, then the adequate market should be limited to the molecule itself, i.e. excluding other originators or genericized molecules in the same ATC3. Finally, if the proposed transaction involves originators that still benefit from exclusivity, the relevant market should not include genericized molecules in the same ATC3 class. A similar reasoning applies to the case of foreclosure or abusive prices (cf. next section). These examples only serve to illustrate that the definition of the relevant market cannot be dissociated from the theory of harm.

4 Relevance for enforcement

This section discusses the relevance of our findings for competition enforcement in the EU and the US. We review instances of foreclosure (Sect. 4.1 and 4.2), pay for delay (Sect.4.3) and excessive (or unfair) pricing (Sect. 4.3).Footnote 40

In the EU, current practice in abuse of dominance proceedings requires that the market share of the firm under investigation be above a certain threshold (typically, 40%-50%) as a pre-requisite to establish “dominance.”Footnote 41 In addition, according to our reading of case law, a finding of abuse requires that the firm be dominant at the time of the allegedly illicit behaviour (i.e., prior to generic entry in our example). This is often referred to as the “concomitance requirement.”

4.1 Foreclosure: EU cases

Imagine an originator company selling a blockbuster drug in a crowded market (a situation akin to our sample). In terms of revenue, the firm is the largest seller, but only commands a market share of 20–30%. That firm successfully manages to temporarily block generic entry. According to our estimates in Sect. 3, large-scale generic entry would lead to an expected drop in prices of about 30–55%–note that most observed price dynamics after generic entry tend to fall in that range. This magnitude provides a clear indication of the rents that accrued to the incumbent because of (temporary) foreclosure and the resulting consumer harm.

In the hypothetical case described above, a narrow application of the market share threshold criteria combined with concomitance would not allow for the opening of proceedings (market share below 40%-50% at the time of the abuse). This despite direct evidence that the foreclosure led to (i) significant consumer harm in the form of higher prices and (ii) additional rents for the incumbent.

Three avenues provide a means of addressing this conundrum. We first reflect on two of them, highlighting their limitations. We then argue that a third avenue, that of the theory of harm, should guide market delineation as it meets the concomitance and dominance requirements; the novelty lies in identifying antitrust markets that have not yet emerged at the time of the infringement.

4.1.1 First avenue: under-enforcement

The first avenue consists of refraining from pursuing the infringement on the basis that the concomitance condition (high market share at the time of the abuse) is not met. This entails the risk of significant Type-II errors: an abuse of dominance left unprosecuted. Type-II errors are a likely concern in the pharmaceutical industry since large markets (with one or more blockbuster drugs) attract entry in the form of me-too drugs, resulting in low market shares. However, margins (measured as the price to cost ratio) and sales tend to remain high pre-GE, even in crowded markets.Footnote 42

4.1.2 Second avenue: narrow market definition pre-Generic Entry

The second avenue involves defining narrow markets prior to generic entry. Under this approach, market share thresholds and concomitance requirements are –formally– met. Two sophisticated competition authorities, the EU Commission and the UK’s Competition and Markets Authority (CMA), have taken this route, defining molecular markets pre-GE. Presumably, this was driven by the case law pertaining to the application of Art. 102 or Sect. 2 of the UK’s Competition Act. We briefly describe two cases of generic foreclosure and how they unfolded.

In the Servier/Perindopril case, the EU Commission imposed a fine for violations of Art. 101 and Art. 102. Our discussion focuses on the concomitance requirement in market definition and hence only applies to the “abuse of dominance” (Art. 102) leg of the Decision.Footnote 43 The EU Commission deemed that Perindopril (an ACE inhibitor to treat hypertension) and its generic versions formed an antitrust market on their own.Footnote 44 This conclusion was largely motivated by the observation that the drop in price associated with generic entry into other anti-hypertensive drugs did not seem to dent the sales of Perindopril (in Sect. 2, we highlighted that this is a common outcome in the industry, even when two drugs are substitutes; in Sect. 3, we report that market boundaries are indeed narrow after generic entry). In a similar case, the UK’s CMA established that the relevant market was molecular in the context of Paroxetine, a Selective Serotonin Reuptake Inhibitor (SSRI, an antidepressant) initially produced by GSK.Footnote 45 The CMA also relied on the lack of price response following GE events among competitors to Paroxetine.

Our reading of these cases is that there is prima facie evidence of consumer harm resulting from foreclosure (monopoly prices beyond the –legitimate– period of exclusivity). The extent of consumer harm can be gauged by the observed –delayed– price drops once the GE materialized. At the same time, the “market share at the time of the abuse” requirement has apparently led competition authorities to define excessively narrow markets prior to generic entry, despite evidence pointing to the contrary.

Both Servier and GSK appealed, pointing to vigorous intermolecular competition among patent protected molecules. In Servier, the EU’s General Court (GC) issued its ruling on 12 December 2018.Footnote 46 The GC found that the Commission had committed a series of errors in its analysis of the relevant market (§ 1589) and thus annulled the Art. 102 leg of the Decision. In particular, it found that the European Commission had not properly evaluated the role played by non-price competition when establishing market boundaries. In the Paroxetine case, the UK’s Competition Appeals Tribunal (CAT) rendered its judgement on 8 March 2018.Footnote 47 Regarding the alleged violation of Sect. 2 of the UK’s Competition Act, the CAT was not convinced by the CMA’s analysis.Footnote 48

Thus, subsequent judicial review by the EU’s General Court and the UK’s Competition Appeals Tribunal (CAT) highlights the pitfalls associated with a narrow, time-invariant market definition. Both the CAT and GC rulings pointed to intermolecular competition prior to GE.

Based on our results in Sect. 3, we would label the Servier and GSK/Paroxetine cases as Type III errors: the correct decision (i.e., there was an abuse) based on an incorrect premise (with the null hypothesis defined as no competition infringement).Footnote 49 Both involved significant consumer harm, even though the infringers were not “dominant” at the time of the abuse in the sense that they did not command a sufficiently large market share on the observable market at the time.

4.1.3 Third avenue: market definition contingent on the competitive concern

The third avenue for resolving this apparent tension is to explicitly recognize that a “natural” and “unique” antitrust market may not exist. We argue that, as a consequence, the definition of the relevant antitrust market should be contingent on the (actual or potential) competitive concern. As indicated, this approach has been discussed for some time, but enforcers and courts have so far been reluctant to endorse it, at least in the European Union.

In the Servier/Perindopril and GSK/Paroxetine cases, we conjecture that, having correctly identified the relevant markets as being molecular post-GE, competition authorities felt compelled to define molecular markets pre-GE, even though the evidence pointed in the opposite direction. Authorities were looking for dominance (high market share) on a pre-GE market that was not the relevant one.

Given these considerations, the market definition exercise for enforcement purposes should be contingent on the nature of the infringement. This leads to the conclusion that (i) the boundaries of the relevant market may change at the time of the (potential) entry of generic producers and (ii) the relevant market can narrow down to the molecular level ex post.Footnote 50 The empirical evidence reported in the previous section indicates that this is compatible with a multi-molecular market definition pre-GE. The advantage of this contingent market definition approach is that it avoids the need to artificially define narrow markets prior to generic entry. Properly identifying the relevant market, even if it has not yet emerged at the time of infringement, would reduce the risk of Type III errors.

As mentioned above, both authorities had relied on the lack of a price response following the launch of cheaper (generic) versions of competitors’ molecules to define narrow markets. One notable difference between the two cases is that the CMA observed that a SSNIP test that compares pre- and post-entry prices would have indicated a narrow, molecular, market.

The CMA’s stance is compatible with a market definition that is contingent on the nature of the alleged infringement. In addition, the CAT clearly recognized that other on-patent drugs exercised less competitive pressure than did generics.Footnote 51Although the CAT was not convinced by the CMA’s market definition, it was attracted by the idea of a molecular market.Footnote 52 The CAT wondered whether the constraint by generics that are not yet on the market, but soon will be, ought to be taken into account when defining the relevant market. As the CAT noted (§ 403): “The definition sought is of the relevant market: this is not an absolute but should reflect relevance to the issue under consideration, and can vary accordingly” (emphasis in the original). The UK Court recognized “that this approach is novel” (Ibid). In order to clarify the matter, the CAT requested a preliminary ruling from the European Court of Justice (ECJ) on March 27, 2018.Footnote 53 In January 2020, the ECJ ruled that, as long as generic competitors have made preparations such that they are in a position to enter with sufficient strength (i.e., entry is “imminent”), generics exercise a competitive constraint that should be factored into the market definition exercise.Footnote 54

Moreover, the opinion of the ECJ’s Advocate General (AG) also took the momentous step of recognizing that market boundaries may evolve over a short period of time.Footnote 55 In its judgment,Footnote 56 the ECJ was more circumspect, but essentially endorsed the approach.Footnote 57

The ECJ’s recognition that market boundaries may change over time opened the door for the CAT to define a molecular market, albeit using a reasoning that departed from that initially applied by the CMA to reach the same conclusion. Consequently, the CAT confirmed the finding of abuse in a supplementary judgement.Footnote 58

GSK argued that, since the CAT had established that the CMA’s reasoning contained errors, it would be “entirely inappropriate” to “uphold the Decision on other grounds” (§90 CAT 2021). The CAT reminded GSK that it was “entitled to make any decision which the CMA could itself have made.” This would suggest that the CMA had reached the right conclusion but based on a flawed reasoning, an error that we christen as “Type III.”

4.2 Foreclosure: US

In the US Daraprim case, the FTC addressed the issue of market definition by directly applying a SSNIP test. The infringement essentially amounted to preventing the emergence of supply-side constraints on the pricing of Daraprim.

In the US, Daraprim (generic name pyrimethamine) is a drug treatment for toxoplasmosis, a parasitic infection that is normally fought off by the immune system but which can lead to fatal brain or lung infections for vulnerable patients. Daraprim was first approved by the FDA in 1953. Although it no longer benefits from any patent protection, it is the only FDA-approved pyrimethamine product (branded or generic) on the market in the United States. Daraprim faced little competition from imperfect substitutes to treat toxoplasmosis, such as (non-FDA approved) compounded pyrimethamine, or non-pyrimethamine drugs.

In 2015, Vyera acquired the right to market Daraprim in the US and increased its price by 4000%, from $17.5 to $750 per tablet. In order to sustain this price level, Vyera engaged in a three-pronged generic foreclosure strategy. First, it signed exclusivity agreements with the two Active Pharmaceutical Ingredient (API, i.e., pyrimethamine) suppliers that could potentially have supplied the input to other generic producers. Second, it prevented the sale of Daraprim samples to potential entrants (the FDA requires generic applicants to conduct bioequivalence testing comparing its product to samples of the original drug). Third, Vyera signed “data-blocking” agreements with distributors to prevent them from selling their Daraprim sales data to third-party data market intelligence providers, such as IQVIA. The purpose was to hide the profitability of the price hike to potential entrants.

The Federal Trade Commission (FTC) filed a complaint on 27 January 2020 for violation of Sect. 2 of the Sherman Act (analogous to abuse of dominance in the EU’s legal order).Footnote 59 A settlement involving compensation for the aggrieved parties was reached on 7 December 2021. The FTC defined the relevant market as comprising the territory of the US,Footnote 60 and limited to FDA-approved pyrimethamine products (§ 292). It pointed to the price increase, concluding that Vyera enjoyed durable monopoly power, the erosion of which was prevented by the successful foreclosure of generic entrants. The FTC furthermore applied a SSNIP test and decided that the relevant market was molecular and limited to FDA-approved drugs (§ 302).

4.3 Pay for delay: Boehringer/Aggrenox

“Pay for delay” refers to situations in which (potential) entrants challenge incumbent patents as being either invalid or not infringed, while the originator alleges an IPR infringement. “Pay for delay” involves the parties settling prior to the ruling, with the originator rewarding the generic producer to postpone entry. These agreements are also known as “reverse payment settlements” since the compensation flows from the (allegedly) aggrieved party (the originator, who claims infringements of its IPR), to the “infringer” (the generic challenger).Footnote 61

Since an agreement between two (or more) parties is involved, the “high market share” pre-condition does not apply (for instance, in the EU, pay for delay agreements are covered under Art. 101). However, the issue of market definition has been implicitly addressed in a US reverse patent settlement case: the AggrenoxFootnote 62 case, heard before the US Connecticut District Court. On the basis of the evidence, the Court concluded that the market should be defined narrowly (Aggrenox and its bioequivalent), hence refusing the defendants’ requests for discovery into other drugs that were therapeutic substitutes of Aggrenox. The Court concluded that such discovery and evidence would be “irrelevant.”Footnote 63

The Court’s reasoning was both clear and cogent:

Moreover, market power exists in degrees. […] The patented drug could face some competition from imperfectly interchangeable drugs (which may or may not themselves be supracompetitively priced) yet still have a meaningful degree of market power, enabling it to be sold profitably at supracompetitive prices

[…] (p. 5).

Moreover, in this case we have a history of actual market data because generics have already entered, and the effect of new competitors entering a market provides an additional direct basis to evaluate the question of supracompetitive pricing. We have data from a less competitive market of the molecule in question, and we have data from a more competitive market of that molecule.

[…] (pp. 7–8).

Sales and pricing data about other drugs would at best be redundant, because any substitution effect constraining the price of Aggrenox will already be “priced in” to this analysis.

[p. 8].

Essentially, the Court established a competitive benchmark in the absence of the restrictive practice (the reverse payment settlement), i.e., the counterfactual. The latter was directly observed, since generic versions of Aggrenox did eventually launch. This allowed the Court to argue that a single supplier of Aggrenox and its generic version would have been able to sustain supracompetitive pricing, consequently defining a narrow molecular market. The District Court’s reasoning is thus consonant with the conjecture that post-GE, a SSNIP would point to such a market. Our econometric results likewise indicate that this is, on average, the correct assessment.

4.4 Exploitative abuse (abusive prices)

Our results also have a direct bearing on potential abuses following generic entry: a situation in which the initial market has “shrunk” and rivalry becomes primarily intramolecular. In economically significant markets, large-scale GE ensures competitive conditions: with no capacity constraints, the own- and cross-price elasticities reported in Table 4 point to a near-Bertrand outcome. The converse implication is that a single supplier would have the opportunity to exercise significant market power, absent regulatory constraints. Since free pricing is the default rule in genericized markets, our results indicate that a supplier with a monopoly on generics would be in a position to extract significant rents.

A number of recent cases support our inference of narrow (molecular) antitrust markets post-GE. Aspen, a South African generic producer, purchased the rights to a series of anti-cancer treatments from GlaxoSmithKline (GSK). The patents for these five active ingredients (chlorambucil, melphalan, mercaptopurine, tioguanine, and busulfan) had long since lapsed. They are used in drugs sold in the EU under different formulations (tablets or injections) and brand names (e.g., Cosmos in Italy).Footnote 64 After purchasing the rights, Aspen imposed very significant price increases in a number of EU countries, among them Italy. When the Italian health authority indicated its reluctance to pay inflated prices, it was threatened with withdrawal of supply. Through these tactics, Aspen obtained high price increases ranging between 300 and 1500%. The Italian Competition Authority (ICA) opened proceedings and concluded that this practice was unfair. The ICA noted that the evolution of Aspen’s costs could not justify such price increases. In May 2017, the EU Commission opened proceedings against Aspen related to its practices in various Member States (except Italy). Footnote 65

Aspen would not have been able to impose such price increases had it been competing with other generic producers of the same molecules.Footnote 66 It was a de facto monopolist on the molecular market, and acted accordingly; this behaviour also appears to be at the core of the EU proceedings.

Aspen entered discussions with the EU Commission with a view to settle.Footnote 67 In exchange for the EU Commission ending the proceedings, Aspen offered to reduce its prices by an average of 73% across the European Economic Area.Footnote 68 Interestingly, this settlement offer is within the range implied by our estimates reported in Sect. 3. On 10 February 2021, the Commission deemed that Aspen’s commitments were satisfactory, and the matter was settled.Footnote 69

5 Conclusions

Market definition is built around the identification of competitive constraints. The Hypothetical Monopoly Test (HMT) is a thought exercise geared towards discerning the limitations to pricing power that must be neutered to achieve a profitable SSNIP. While the HMT rests on solid analytical foundations, it is rarely applied. Instead, competition authorities rely on approximations (e.g., diversion ratios) to delineate antitrust markets.

This paper was initially motivated by the controversies surrounding market definition in the context of unilateral behaviour in the pharmaceutical industry. The market delineation established by two sophisticated enforcement agencies, the EU Commission (Servier) and the UK’s CMA (GSK/Paroxetine), failed to convince the competent courts. Enforcers defined narrow markets that yielded the “required” market shares to allow for the cases to be prosecuted, despite abundant evidence of vibrant competition among drugs.

Our empirical results provide two important takeaways. The first is that the market definition exercise ought to be contingent on the nature of the alleged competitive infringement, i.e., the theory of harm. As noted by Glasner and Sullivan (2020, p. 330): “Because there is no economically meaningful natural market, relevant markets must be analytic devices. Because analytic devices are tied to the subject of analysis, relevant markets can be defined only by reference to specified theories of harm.” Our paper provides a concrete and quantified illustration of the ontological relationship between the theory of harm and a cogent market definition. Accepting this premise may contribute to reducing the risk of what we term Type III errors (a correct decision based on the wrong premises) on the part of enforcement agencies.

The second takeaway is that estimating a logit (or nested logit) demand system may not be adequate in certain circumstances (this holds for pharma, but similar issues could arise in other industries). This is because entry (here, by generics) may alter market structure, i.e., the underlying nests. Our results imply that, following GE, the market may split, giving rise to a different nest structure, despite the fact that the physical properties of the products remain unaltered. In terms of enforcement, a molecular market may be warranted in a GE foreclosure case. By contrast, in a merger of two producers of different drugs, the relevant market is likely the broader intermolecular market (with the proviso that a competing drug may lose patent protection soon, and then “drop out” of the relevant market).

The cases discussed in this paper may have a bearing in Servier (the Commission’s appeal of the EU General Court’s December 2018 ruling is pending before the European Court of Justice). In its supplementary judgement, the CAT noted that “neither the Advocate General nor the CJEU gives any support to the concept of the relevant market being assessed according to the conduct under scrutiny” (§ 86 CAT 2021). Our analysis indicates that conditioning market definition on the nature of the competitive concern would appear to be the only (cogent) path out of the Servier conundrum.