Fundación de Estudios de Economía Aplicada

Multimarket Contact in Pharmaceutical Markets by Javier Coronado^* ** Sergi Jiménez-Martín *** Pedro L. Marín DOCUMENTO DE TRABAJO 2008-20 Serie Economía de la Salud y Hábitos de Vida CÁTEDRA Fedea – la Caixa

May 2008

* Universitat Pompeu Fabra. francisco.coronado@upf.edu ** Universitat Pompeu Fabra and FEDEA. sergi.jimenez@upf.edu *** Universidad Carlos III and CEPR. marin@eco.uc3m.es

Los Documentos de Trabajo se distribuyen gratuitamente a las Universidades e Instituciones de Investigación que lo solicitan. No obstante están disponibles en texto completo a través de Internet: http://www.fedea.es. These Working Paper are distributed free of charge to University Department and other Research Centres. They are also available through Internet: http://www.fedea.es.

ISSN:1696-750X

Multimarket Contact in Pharmaceutical Markets^∗

Javier Coronado Universitat Pompeu Fabra francisco.coronado@upf.edu

Sergi Jiménez-Martín Universitat Pompeu Fabra and FEDEA sergi.jimenez@upf.edu

Pedro L. Marín Universidad Carlos III and CEPR marin@eco.uc3m.es

First Version: April 20, 2007 This Version: April 15, 2008

Abstract

Multimarket rivalry theory predicts that firms engaged in price competition in several markets might find it optimal to redistribute market power from more collusive markets to more competitive instances. Price regulation is shown to afect this relation in a non-monotonic way. Mild or low price regulation may encourage further market power redistribution, whereas stronger price controls change the result to the point of making it irrelevant. We use data from the Pharmaceutical industry for nine OECD countries which are known to place diferent levels of price controls. We find evidence of the redistribution efect and the interaction with price regulations when considering contacts between chemically equivalent products; however, widening the contact dimension to consider interactions among substitute therapies make the result less transparent. We also find some evidence of the expected interaction between price controls and the redistribution efect driven by the multimarket structure of the industry. JEL Classification: L11, L13, L65, I18

Keywords: Pharmaceutical prices, Multimarket Contact, Regulation

^∗This study was supported by an unrestricted educational grant from the Merck Company Foundation, the philanthropic arm of Merck & Co. Inc., Whitehouse Station, NJ. Partial funding was also obtained from the Spanish Ministry of Science and Technology under grant SEJ2005-08783-C04-01. We thank MSD and IMS for providing the data and conference participants at IDEI/U.Toulouse, AES, Swiss IO Day conferences and seminar audiences at UPF and UC3M. The helpful suggestions of Antonio Cabrales, Félix Lobo, Guillem López, Vicente Ortún, Jaume Puig, Fran Ruiz-Aliseda, and Joel Shapiro are gratefully acknowledged. We are specially grateful to Patricia Danzon for her kind suggestions on critical issues in the data analysis

1 Introduction

1.1 Motivation

This paper study prices in pharmaceutical markets under a multimarket contact setting. Pharmaceutical markets are subject to price regulation (price ceilings) in some countries at diferent intensities. Price constrains according to applied Health Economics literature are responsible for reducing price competition and entry in this industry (c.f. Danzon and Chao (2000), Danzon and Epstein (2008) or Kyle (2007)), however there has been little research on plausible market mechanisms through which such distortions are realized. Therefore, the main contribution of our study is to show that price controls in some product markets may distort pricing in other markets through the multimarket contact structure of the industry in ways that can explain lack of entry in more regulated countries.

We perform a cross country analysis for nine OECD countries which are known to have diferent levels of price regulations for pharmaceutical markets. Within a country we study multimarket contact in terms of the simultaneous presence of corporations in various product markets; we adopt several of the contributions by Danzon and Chao (2000), Danzon and Furukawa (2003), Danzon et al. (2005), Kyle (2007) specially to classify countries in our sample according to their stringiness of price regulation.

In the old days, Edwards (1955) proposed that firms that meet each other in several markets may have incentives to relax competition because of mutual forbearance. In their seminal work Bernheim and Whinston (1990) presented a modern approach to formalize the expected efects of multimarket contacts and review the traditional approach using competition games with infinitely repeated interactions and trigger strategies. Multimarket contact may sustain higher prices if firms are able transfer slack of collective market power in one market to others where no slack is available. More interestingly under the presence of product diferentiation the multiple market contact setting may allow firms to optimally redistribute their market power across markets, rather than simply transfer excess. Firms create slack in more collusive markets, reducing prices, an allocate the slack to more competitive markets, increasing prices

Phillips and Mason (1996) study the case of multimarket contact under price regulation in which products are homogenous within markets and firms compete in quantities `a la Cournot. The main results suggest that binding mild price constrains in one market might increase further prices in other markets in comparison with the unregulated case. On the other hand, when price ceilings are further adjusted downwards then no slack might be available so prices in the second market will reflect only the price levels that would have been observed in the single market setting. We adapt this result to the case of an industry with product diferentiation and price competition to parallel the usual approach for competition in pharmaceutical markets.

¹Ease of collusion depends on a number of factors such as the number of firms operating in the market, product homogeneity, speed of interaction, cost asymmetries, demand stability, etc. Other theoretical works on the multimarket contact structure of industries are Spagnolo (1999) and Matsushima (2001).

Confirming the theory with pharmaceutical data has a number of important implications for policy making. Mild price restrictions are expected to further increase prices in markets that where not subject to price regulation or price controls are not binding, therefore undesirable welfare efects might be produced. Strong price regulations in some markets will reduce prices also in more competitive markets with respect to the unregulated case, therefore stringent price restrictions in one market may distort entry in more competitive markets, which might explain how price regulation afect competition dynamics.

1.2 Related empirical literature

After Bernheim and Whinston (1990) there have been several empirical works that successfully test some of the predictions of the multimarket contact theory. Evans and Kessides (1994) , Jans and Rosembaum (1996), Parker and Roller (1997) and Pillof (1999), provide evidence that multimarket contact is expected to increase prices for the airline, cement, mobile and banking US industries respectively. These studies basically assume that multimarket contact will have the same marginal efect across markets. Indeed they are able to show that firms are expected to place higher prices in markets where their rivals are also present in a larger set of independent contact markets. More recently, Fu (2003) has shown also for the newspaper industry that multimarket contact reduces substitutability among competing products.

Fernández and Marín (1998) performed an empirical strategy to test for the market power redistribution hypothesis in the Spanish Hotel industry. Their functional form for multimarket contact interacts a contact variable with a variable correlated with ease of collusion² with the main result that markets with lower ease of collusion are expected to have higher prices due to multimarket contacts.³

The most salient feature of our work is that we incorporate to Fernández and Marín (1998) approach price regulation and test whether the redistribution efect is afected in practice in a way compatible with the theory. Another important diference with respect to the existing literature is related to the way we define contacts. In all the studies reviewed contacts are counted in terms of geographical areas, in our case we define contacts across product markets much in the way is suggested in Bernheim and Whinston (1990, Sec. 7) for industries with various degrees in product diferentiation across markets. We maintain this approach is suitable for the pharmaceutical industry in light of the presence of several monopolistic product markets (due to patent protection) which provides a source of identification of the significance of multimarket contacts in explaining price variation.

²This functional form is taken from an unpublished paper by Gimeno and Woo (1994)

³This approach was also present in Jans and Rosembaum (1996), however they did not intend to test for the market power redistribution hypothesis. Nevertheless the latter found that prices are expected to be higher due to contacts across markets whenever concentration is higher in a market.

1.3 Main findings and results

We found evidence of the market power redistribution efect for the US, where markets for drugs are supposed to be free of price controls. When considering contacts among corporations with products having an equivalent composition prices are expected to increase for product markets below a concentration level of considering the Herfind¨ahl-Hirschman index and decrease above that level. For Canada, a medium or mild regulated country, data is also consistent with the redistribution efect, however the size of the efect is larger than for the US. That is, prices are expected to be higher for Canada than the US at low levels of concentration, and prices are expected to be lower for high levels of concentration. We interpret this result as strong evidence of the expected interaction between mild price regulation and the redistribution efect. For EU medium regulated countries there is some evidence of the redistribution power, however the size is lower compared to the US case. In this case, prices are predicted to be unambiguously lower for any level of ease of collusion with respect to the to first country cases. For strong price regulators, such as Japan and Spain, price variations across products and markets are not determined by multimarket contacts, which is taken as a result showing that price controls severely reduces market power slackness.

The paper develops the streamline of analysis in the following structure: In section 2 we ofer a brief description of relevant institutional aspects of markets for drugs as well as common issues related to competition analysis. Section 3 presents the theoretical implications of multimarket contacts; we also show how observed prices can be approached from this framework; section 4 describes the data set and the variables to be used; next, section 5 presents the empirical specification based on the discussion of section 3 and also present the relevant aspects of the econometric methods, identification problems and solutions; finally, section 6 describes the results and their interpretations as well as some robustness exercises and extensions.

2 Markets for pharmaceutical products

2.1 Agents and institutional issues

According to Puig-Junoy (2005) there are three diferent market types in pharmaceutical markets, on-patent markets for prescription drugs, of-patent markets for prescription drugs and markets where products are sold Over the Counter (OTC) without a prescription. Perhaps the most crucial complexity for the economic study of prescription drug markets is the number of agents involved in shaping demand and so the dificulty to approach market analysis by standard means. It is a doctor who is responsible for the choice given its advanced knowledge both of therapies available and patients condition. Hence, doctors might or might not be concerned about the economic standing of the patient, or even doctors might prescribe based on other type of incentives additional to the objective of successfully treating a patient. Hence, knowledge of the eficacy of a drug, brand reputation and other factors might induce small substitutability of available therapies in these markets. In addition there is usually a third party providing diferent levels of insurance coverage for diferent therapies, specially in OECD countries, which amounts to increase demand rigidity and may explain price variations across markets.

Although price regulation is easily justified for on-patent drugs (regulation aims at constraining the exercise of monopoly power), OECD countries, with the exception of US, also regulate price levels for of-patent markets whereby in principle competition is possible but afected by the aforementioned institutional factors. Given patients are insured, price regulation focuses in reducing the fiscal burden of public pharmaceutical expenditures. Given this simple approach to price regulation the kinds and size of distortions that these policies may produce are of course diverse and have been explored from many angles by a vast literature within the field of Health Economics.

Table A: Market Share by Type of Product (Standard deviation in parenthesis)

Danzon and Chao (2000), Danzon and Furukawa (2003), Danzon et al. (2005), Kyle (2007) have found evidence that price regulations reduce competition and delay entry of competitors in drug markets. In particular, some relevant findings suggest that price regulations reduce the incidence of competitive constrains for branded (originator) products by generic drugs, therefore the importance of generic products in of-patent markets difer across countries. Using these papers as references, OECD countries can be grouped by diferent levels of price constrains. In particular we investigate nine OECD countries from less regulated, US, medium regulated, Canada, Germany, UK and Netherlands to heavily regulated, France, Italy, Japan and Spain. Table A presents information for these countries on the average market share for of-patent drug markets for the period 1999-2003. Information comes from the IMS-Midas database which covers a large number of national drug markets for the aforementioned countries. The US markets have the highest average market share of generic products consistent with the claims of related health economics literature. The rest of the data is at least weakly consistent with the evidence provided by the literature, medium regulated countries have lower relative importance of generics in market shares (with the exception of Netherlands) but higher than the more regulated countries.

Regulatory mechanisms are diverse, and even if countries use the same mechanism differences can be identified in the implementation among national regulatory bodies. Most EU countries, and Canada, used Reference Prices (RP) to control price levels during our sample period. Reference Prices compare individual prices to an index (mean, median or delike) of prices of chemically equivalent products or therapeutically equivalent products and make the consumer/patient pay a fraction (or the total) of the diference of the targeted drug price with respect to the index of reference. The idea is to promote cost consciousness among patients and to provide incentives to firms to reduce prices by facing them with the risk of loosing market positions. Key alternative versions of RP are internal (local) and external based reference indexes. In EU countries, according to a very recent work by Danzon and Epstein (2008), international RP creates interdependence in strategic firms decisions across nations, a situation which is considered in our work in that it might place certain constrains for the identification of sources of price variations within a national market.

A similar source of interdependence is related to the so called parallel imports of drugs, that have risen serious concerns specially in the US. In the US and some EU counties, firms are afected by imports of their same products they ofer in the national market that come from countries whereby price regulation and specific market conditions allow distributors to commercialize them at much lower prices. US markets sufer from parallel imports from Canada, whereas UK, for example, sufer from parallel imports from other EU nations. The relevance of the interdependence produced either by regulations or parallel imports depends on the price diferentials of the countries involved in such interdependence. Table B presents average prices considering all types of drugs and also distinguishing the type of drugs within of-patent lines. In general prices in the US are much higher than in Canada, whereas prices in Germany, UK and Netherlands are higher than in the rest of European countries considered in our study. Although Japan is considered a heavy price regulator, prices are much higher than in the other highly regulated countries which can be explained by diferences in income levels.

Table B: Average prices in current USD

Source: worldbase.dta

2.2 Competition in pharmaceutical markets

Market competitive conditions are often considered in this industry as a function of market segmentation between branded products and generic copies, product diferentiation of distinguishable nature and the life cycle of drugs. On-patent drugs might compete with alternative drugs of close therapeutical efects but diferent composition, of-patent markets needs to consider the relevance of generic competition and competition in general is expected to place stricter constrains over pricing the more mature is the product market. A precise product market definition is therefore not straightforward.

In the light of these considerations price competition in pharmaceutical markets are thought to be imperfect due to product diferentiation coming from diferent sources. Products with equivalent composition might be vertically diferentiated due to advertisement outlays and firm reputation. Products with close therapeutical efects but diferent composition are diferentiated in their characteristics (recommended dosage, side efects, etc.) for a version of horizontal diferentiation. Finally two products can be diferentiated in both vertical and horizontal dimensions⁴. Theoretical works on pharmaceutical markets, however have placed more weight to the fact that goods are vertically diferentiated, basically because it is considered that the most important source of competition in a market comes from the entry of generic products. Anyhow, competition in pharmaceutical markets place two important questions for our study: First, market definition will have an impact on the source of product diferentiation. In simple words, a wide definition will introduce horizontal product diferentiation to our analysis, whereas a narrower definition will tend to omit horizontal product diferentiation in favor of quality diferentiation. Second, segmentation within product markets may introduce a specific structure of contacts among rivals.

3 A model of multimarket contact

3.1 Market power redistribution

We study an industry where there exist K product markets so as to approach what is observed in the pharmaceutical industry and N firms in each market. The detailed theoretical argumentation is reserved for Appendix A, here we present a synthesized version of the analytical results. Firms compete in prices and the one shot game equilibrium prices forma Nash Equilibrium, , ∀i. Assume firms compete in prices in a repeated fashion with infinite horizon. Firms may sustain in equilibrium higher than the stage game price equilibrium using trigger strategies in which any deviation of the collusive strategy is penalized by reverting forever to . If is the sustainable price for firm i in the repeated game in market it satisfies:

\[\frac {\delta}{1 - \delta} [ \pi_ {i k} ^ {c} - \pi_ {i k} ^ {n} ] \geq \pi_ {i k} (R _ {i k} (p _ {- i k} ^ {c}), p _ {- i k} ^ {c}) - \pi_ {i k} ^ {n}\tag{1}\]

where and are firm i’s profits when prices are and respectively, are firm i’s profits when all firms other that i set the collusive prices, and firm i chooses its best response to them, , and is the discount factor. Here we assume what is called the best collusive outcome so that at any given δ firms select the highest possible price. In the technical appendix we show that for certain usual assumptions about the profit function there is some above which the monopolistic price, is sustainable and the incentive constrain preserve some slack. Below this threshold, the equilibrium price saturate the incentive constrain. Assuming collusive prices are non-decreasing in the discount factor, from 1 a product market equilibrium price can be characterized by the stage game price and a function of the discount factor:

\[p _ {i k} ^ {c} = \Phi (\delta) p _ {i k} ^ {n}\]

⁴Some theoretical works on price competition in pharmaceutical markets are Cabrales (2003) K¨onigbauer (2006). Structural models of price competition with product diferentiation in this industry are scarce, Berndt et al. (1995) and Cleanthous (2004) are prominent examples.

where

\[\Phi (\delta) = \left\{ \begin{array}{l l} \frac {p _ {i k} ^ {m}}{p _ {i k} ^ {n}} & \text {if} \delta \geq \delta_ {k} ^ {0} \\ \frac {\tilde {p} _ {i k} (\delta)}{p _ {i k} ^ {n}} & \text {if} \delta < \delta_ {k} ^ {0} \end{array} \right.\]

where is the best collusive function whenever . Following Bernheim and Whinston (1990) a collusive strategy across markets can be sustained if firms realize that defection in one market will trigger retaliation of its rivals in the whole set of K markets. Accordingly, under multimarket contact a firm participation in the coalition will pool its K incentive constrains:

\[\sum_ {k = 1} ^ {K} \frac {\delta_ {k}}{1 - \delta_ {k}} [ \pi_ {i k} ^ {c} - \pi_ {i k} ^ {n} ] \geq \sum_ {k = 1} ^ {K} \left\{\pi_ {i k} (R _ {i k} (p _ {- i k} ^ {c}), p _ {- i k} ^ {c}) - \pi_ {i k} ^ {n} \right\}\tag{2}\]

Bernheim and Whinston (1990) analysis suggest that when markets difer in the number of firms, demand conditions or there are economies of scope, the pooled incentive constrain can be used to sustain higher prices in equilibrium. It is also claimed that asymmetries within markets, which are known to hinder collusion, might be softened through the multimarket contact setting. Absence of such diferences across markets makes contacts irrelevant for sustaining more collusive outcomes. However, it is not the mere existence of multimarket contact across difering markets which expands the set of collusive outcomes, but the ability of firms in the coalition to transference market power across markets. Imagine for instance that in some of the K markets firms are able to sustain in equilibrium so that the individual incentive constrains are verified in some of them with inequality (in our setting for some k). This slack can be used in markets where no collusive price is possible to sustain at the corresponding for example markets where is suficiently large. Indeed firms in the multimarket coalition can increase prices in more competitive markets violating the individual incentive constrains as long as 6 is verified. Under this result, average prices in the industry will be unambiguously higher than in a case without multimarket contacts.

More interestingly, whenever markets difer in their degrees of product diferentiation firms can find it optimal not only to transfer but to redistribute their market power. Differences in the degrees of product diferentiation across markets will also produce diferent levels of sustainable collusive prices. In the technical appendix we show that even if firms have no slack in any of the market specific incentive constrains, they may find it optimal to create slackness of one market’s incentive constrain by reducing the equilibrium price and use this slack to increase prices in markets where in isolation there are less favorable conditions to doing so. It is shown also that firms have an incentive to reduce prices in markets where it is easier to sustain collusion in isolation. The mechanism is somehow complex (see the Technical appendix) however the intuition is that for markets where prices close to are sustainable, a small reduction in price will produce more slack that can the be used to increase prices in markets where violating the corresponding incentive constrains is more profitable. It turns out that this is the case for markets where the marginal profitability of violating the incentive constrain is relatively high with respect to the marginal profitability of defection.

Given the structure of each k market and as a consequence the structure of multimarket contacts for firm i, we can represent the firm’s equilibrium price of the repeated game in market k as a function of three separable components:

\[p _ {i k} ^ {*} = \Gamma (M M C _ {i k}) \Phi (\delta) p _ {i k} ^ {n}\tag{3}\]

where measures the efect of the multimarket contacts structure given by variable and the other two components come from the definition of the single market price equilibrium. The redistribution of market power can be tested in terms of the value of the multimarket function. will be expected for markets where a collusive price is easier to support (less toughness of price competition) in equilibrium and in markets with less favorable conditions to sustain collusion. Diferences in toughness of price competition among markets can be obtained by looking at the number of varieties available in a market, market concentration, product diferentiation and so on.

3.2 Multimarket contact and price regulation

The main point of this article is that price regulation, or price ceilings, may distort markets through the multimarket contact structure of the industry. Regulatory agencies for the pharmaceutical industry not only control on-patent product prices but also potentially competitive prices. Lack of entry, segmentation and price rigidities due to reputation, insurance coverage and other aspects are though to preclude the emergence of strong competition in product markets that are therefore subject to price regulation. We assume that price regulation is not eficient, because as noted in Puig-Junoy (2005), the objective of regulation is to reduce the public financing of national health systems.

In the technical appendix we show that price regulation in markets where prices close to the monopolistic prices are sustainable, will have an impact on the optimal price selected by firms under the multimarket contact setting. In particular, mild price regulations, that is binding price ceilings close to the monopoly price, are expected to further increase prices in other markets where price ceilings are not binding or are un-regulated. In this case, the size of the multimarket contact redistribution is expected to be larger with respect to the un-regulated case. For stronger price regulations, no slack will be available to be redistributed to the rest of the markets, therefore multimarket contact will have no efect over price variations.

By a continuity argument, it is expected that medium price ceilings will have at some point a negative efect over the sustainable prices in other markets.

In terms of observables, we require a very large set of prices and market conditions to analyze in practice whether it is possible to identify on one hand the un-regulated multimarket efect and on the other the predicted distortions incurred by mild, medium and strong price regulations. In the next section we present our data set and discuss the strategy followed to extract information that can be connected to the regulated multimarket contact analysis. In short we perform regressions for price equations of pharmaceutical markets for a set of OECD countries that are known to place diferent degrees of price controls among them. In each country there is a large number of product markets in which rival corporations meet, therefore albeit limitations due to diferences among countries, we are able to show evidence in support of out theoretical predictions.

4 The data

4.1 Data set description and relevant considerations

We use a multi-country and multi-product data set from the IMS MIDAS international dataset for the period 1998-2003⁵. This dataset encompasses a large number of countries including the top ten in terms of medicine expenditures, as well as medium size and small countries. We restrict the analysis to data from nine OECD countries, namely US, Canada, Germany, UK, Netherlands, France, Italy, Japan, and Spain. Following the comprehensive study by Danzon and Chao (2000) as well as the advice of recognized experts⁶ we group the countries considered in the sample in three regulatory categories. (I) US and Canada belong to the group of more market friendly policies, however Canada is considered to place relevant price controls and might also be included in the next category or in an intermediate category of mild regulated industry; (II) Germany, UK, Netherlands belong to the group of medium intensity of price regulation; and, finally, (III) France, Italy, Japan, and Spain belong to a family of countries that apply stricter price controls.

In our dataset products are classified in terms of their chemical, therapeutical and pharmacological characteristics using the standard Anatomical Therapeutic Chemical classification or ATC code. This classification is used regularly in defining and analyzing product markets in this industry. Notwithstanding studies analyzing specific product markets are able to confer more substance to the set of products considered, in this study it is crucial to consider all the information available which implies loosing precision at the time of shaping product markets. ⁷.

One drawback of the IMS Midas database is related to the units used to measure sales quantities. The IMS Midas uses a standard unit (SU) to measure a particular product sales based on a reference product or recommended daily dosage which may vary across countries. Unfortunately prices are also connected to these standard units introducing a measurement error in this variable. This will cause that some portion of price variation cannot be controlled in our model increasing the variance of the estimates, but leaving the consistency of our estimates unafected.

4^th

2^nd

⁵The dataset gather information from the quarter of each year, apart from 2003, for which the information is provided for the quarter.

⁶We are indebted to Guillem López (UPF) and Vicente Ortún (UPF) and Félix Lobo (UC3M) for helpful advice on this regard.

⁷The ATC classification is supported and maintained by the World Health Organization Collaborating Centre for Drug Statistics Methodology with a base in the Norwegian Institute of Public Health.

4.2 Market definitions

The pharmaceutical industry represents a complex exercise for market definition in practice for a number of reasons ⁸. Considering geographical boundaries within a country or clear mutually exclusive sets of substitute products is not straightforward. First, we disregard any delimitation of regional markets within a country, as it is reasonably to assume that value to cost transportation in this industry is high, therefore we study product markets at the country level. Second,to avoid subjective market definitions we adopt a common strategy in Health Economics by defining product markets in terms of the ATC classification. However, we first define a market to be the set of products belonging to the same molecule and second the set of products belonging to the same ATC-4 classification which groups medicines that have close therapeutical and pharmacological efects.

Both market definitions has disadvantages. Defining a market with respect to a single molecule will not capture possible competitive constrains that on-patent drugs may face from substitute therapies, which is equivalent to assume that on-patent drugs are pure monopolists. Although this limitations are expected to be (partially) solved by using the ATC-4 level classification as our reference market definition, yet we have to bear in mind that this broadened definition might be imposing irrelevant competitive constrains. Hence, individual product prices might appear to have lower covariance with the multimarket structure, compared to the case of the molecule definition.

Albeit some limitations of our market definitions, we can also persuade the reader that having this two somehow diferent exercises are beneficial at least in two aspects. First, using the molecule definition will still be valid to study the relevance of the multimarket contact structure derived from the interplay of firms with their closer rivals in of-patent markets. This is an original way to test whether the expected competitive climate among products with the same active ingredient are softened by their interactions across markets. Second, once we adopt the wider market definition comparisons with the previous molecule-based results can help us to discuss whether the new set of estimates are possibly due to the inclusion of too many irrelevant contacts. Note also that if multimarket rivalry appears not to explain price variation at the molecule level it is intuitive to expect that it will not explain it at the ATC-4 level either.

4.3 Variable definitions and expected marginal efects

The list of variables that we construct is the following. The variable price, called Price, corresponds to prices in SU terms. We convert the computed prices to current US dollars. As pointed out by several authors marginal costs are almost irrelevant in the industry [c.f. Stern (1996)]. This suggests the use of a hedonic approach.⁹ Accordingly, in our pricing regressions we incorporate variables that are considered to be correlated with some relevant characteristics of an individual product to proxy the stage game equilibrium price. These variables are: The firm’s size, Fsales, constructed as total corporation sales correcting it by excluding sales of the product under analysis. Generic is a dummy variable which takes one if the product is a generic (zero if its a branded or originator drug), and Composite is a dummy variable which takes the value one if the product is a combination of two or more molecules. Molecule age, Molage, is the time elapsed since the molecule was launched to December 31, 2003. Competition variables related to the mark-up are also computed. These variables are: Number of generics, Ngenerics, is the number of generic products in each market. The Hirschmand-Herfind¨ahl concentration index, HHI ^g , is constructed using corporation sales, with squared market shares of the corporation under analysis excluded from the index. We construct the market share of each variety in the market, Mshare, and the aggregate market share of all other varieties supplied by the same corporation in each market (if available), Cshare. For the regression analysis we use log transformations of Price, Fsales and Molage, so we value more the diferences in smaller than in larger values.

⁸Perhaps talk a bit of Coscellis view of practitioners’ problems

A number of dummy variables are also defined: New is a dummy variable equal to one if the product was launched in the previous year and zero otherwise, Censormol equals 1 if the molecule was launched before January 1, 1991 and zero otherwise, Censorlag equals one for products launched before January 1, 1991 and zero otherwise. These dummies are part of the quality variables used to capture price variation.

Table 1 in the Appendix B presents the mean and standard deviation of our control variables by country. Market share variables and the concentration index is presented for the two market definitions used in our sample. We have a total number of 67582 observations, Germany being the country with most data (17306 observations) and the Netherlands being the one with least observations (2614).

4.4 Alternative measures for multimarket contact

A contact of firm i with its rivals in the focal (or reference) market k in other markets should reflect the importance of these contact markets for the firm. This approach is not only motivated by a clear intuitive reasoning. Bernheim and Whinston (1990) has shown that multimarket rivalry can also be studied in terms of what Edwards (1955) called each firm’s ‘’spheres of influence‘’. Following this argument, the plain average contact statistics cannot reflect the relevant multimarket structure because market power transference needs to consider the quality of each contact market either as a source or destination of such market power. One firm will consider the contact with a rival present in the market of reference if the contact occurs in a market where the former has a specific interest or if both firms have suficient market power.

⁹See Berndt, Cockburn and Griliches, 1996, Berndt, Pindyck and Azoulay, 1999, Cockburn and Anis, 1998, and Suslow, 1996.

Keeping in mind the discussion of the treatment of a contact we define a generic multimarket contact variable in the following way: A contact occurs when a corporation i and its competitor l in the focal market k, are also rivals in an independent contact market m. If this is the case we define a binary variable , otherwise . The multimarket contact variable relative to this contact is defined as:

\[M M C _ {i l, k m} = C _ {i l, k m} w _ {m}\tag{4}\]

where is a weight aiming at measure the corporations’ interests in the contact market m or the both firms’ combined market power. We construct two alternative multimarket contact variables changing weight . First we use the Herfind¨ahl-Hirschman index of the contact market, second, we use the combined corporations’ operations in that market with respect to their respective total operations in sales. Finally, an individual firm is assumed to average its weighted contacts across markets relevant to the focal market in the following way:

\[A V M M C _ {i k} = \frac {1}{(N _ {k} - 1)} \sum_ {l \neq i} \sum_ {m \neq k} M M C _ {i l, k m}\tag{5}\]

where is the number of competitors in the focal market. Therefore, for each of the three alternative definitions the average multimarket contact variable will change for a firm across markets and across time due to entry/exit of firms as well as each firm’s commercial evolution. In this way we will extract the most of our Panel dataset as price variation will have many sources for identification.

An important element of pharmaceutical markets is related to licensing. At any point of the life cycle of a drug, the originator firm may produce the drug by itself or grant licenses to one or more manufacturers to produce and commercialize it independently. This situation is reflected in the IMS Midas dataset. To avoid the introduction of irrelevant contacts to the multimarket contact structure, if a firm has given more than one license within a market to produce its original drug we count the presence of the originator corporation in that market rather than the presence of the manufacturers. Implicitly we assume that all corporations granting licenses do not allow for intrabrand competition whenever more than one license is given within a product market.

The mean and standard deviation of the two alternative definitions are presented in Table 2 in the Appendix B, figures are also presented by country and market definition. Note in this table that the mean levels are diferent which is not surprising given the diferent sizes in the weights.

5 Empirical specification and Econometric methods

In section 3 we have shown an expression for the observed price of a product considering the multimarket structure of the industry. Therefore the observed price for a product of firm i in market k, denoted , can be represented as a separable function of its equilibrium price in the stage game, , a mark-up on this price which depends on the discount factor, , and a function of the multimarket contact structure. We consider the following log-linear specification:

\[l o g (p _ {j i k t} ^ {*}) = \alpha + \Omega (A V M M C _ {i k t}) + \Phi (\delta_ {i}) + l o g (p _ {j i k t} ^ {n})\tag{6}\]

where t denotes time, , and is an intercept. The log of the stage game equilibrium price is specified in the following way:

\[\log (p _ {j i k t} ^ {n}) = X _ {j i k t} ^ {1 ^ {\prime}} \beta_ {1} + X _ {j i k} ^ {2 ^ {\prime}} \beta_ {2} + Z _ {k t} ^ {1 ^ {\prime}} \gamma_ {1} + Z _ {k} ^ {2 ^ {\prime}} \gamma_ {2} + \eta_ {1 i} + v _ {j i k t}\tag{7}\]

where the and are vectors of respectively time-variant and time-invariant variables concerning product of firm on one hand, and market on the other, that potentially afect the stage game equilibrium prices through diferent meaningful ways, and are the corresponding parameter vectors, is a corporation fixed efect and is a random variable with zero mean and finite variance. Given the product diferentiation nature of pharmaceutical markets we can interpret the pricing equation as a function of variables afecting marginal costs (which are usually thought to be negligible in this industry) and the product’s mark-up afected by attributes that are fixed or vary through time. From a structural point of view these attributes will afect the firm’s product market shares. At the same time, the fixed efect is included to control for elements of vertical (quality) product diferentiation which are one of the most highlighted peculiarities of this industry. The can be regarded as that information on attributes that are not observed by the econometrician but firm’s do take into account when taking their pricing decisions. To complete the specification we include time dummy variables for the last four time periods of the sample whose objective is to approach the function of the discount factor.

After replacing these expressions in the above equation we obtain:

\[l o g (p _ {j i k t} ^ {*}) = \alpha + \Omega (M M C _ {i k t}) + X _ {j i k t} ^ {1 ^ {\prime}} \beta_ {1} + X _ {j i k} ^ {2 ^ {\prime}} \beta_ {2} + Z _ {k t} ^ {1 ^ {\prime}} \gamma_ {1} + Z _ {k} ^ {2 ^ {\prime}} \gamma_ {2} + \eta_ {i} + \lambda_ {t} + v _ {j i k t}\tag{8}\]

where denotes the presence of the time dummy variables. In practice our specification is a version of the two-way error component model as described in Baltagi (2005). We do not consider product specific fixed efects first because we are considering already some product specific regressors that are invariant across time, and second because the usual consideration for this is to provide a control for vertical product diferentiation which we consider to be firm specific rather than product specific. Although this assumption may lead to some debate, we consider here that branded products gain reputation because they belong to certain corporations that are able to invest in quality diferentiation. ¹⁰

¹⁰In some of the related works reviewed [e.g. Evans and Kessides (1994)] market specific individual efects are also included in the specification because price variation across markets depends heavily in some exogenous characteristics of markets. We do not consider such fixed efects although some time invariant market characteristics are included in the list of control variables.

In section 2 we discussed the implications of diferent levels of insurance coverage over the price responsiveness to market conditions. National health systems usually ofers wider insurance over necessary therapies such as drugs controlling high blood pressure to other drugs that are designed mainly to increase the comfort of the consumer. Danzon and Chao (2000) proposed to study price competition controlling for diferent levels of insurance by including fixed efects at the first level of classification of the ATC code. The ATC-1 level groups drugs that act through the same organ of the body. We use a total of 12 ATC-1 dichotomous variables for the available groups in the data omitting the first group (drugs acting through the Alimentary Tract and Metabolism). We perform the same exercise in all our estimations following this line of argumentation so that finally the estimated regressions are versions of a third way error component model.

We estimate equation (8) country by country using a Within Groups panel data method where the firms’ specific heterogeneity efect wipes out of the estimation and considering the rest of the error components as dummy variables. Given the short length of the time dimension (five years) and relatively small number of ATC-1 groups with respect to the available observations by country (over 3,000 by country) this approach does not entail problems with our degrees of freedom.

5.1 Multimarket contact specification

Two specifications for are used. In the first place we follow a simple specification:

\[\Omega (A V M M C _ {i k t}) = \alpha_ {1} A V M M C _ {i k t}\tag{9}\]

which is independent of the characteristics of the focal market. This simple linear function will collect the sign and significance of the efect that the variable measuring multimarket contact has on prices on average. A second specification assumes that multimarket contact will have a distinguishable efect across markets depending on a variable correlated with ease of collusion, the HHI index:

\[\Omega (A V M M C _ {i k t}, H H I _ {k t}) = A V M M C _ {i k t} \times (\alpha_ {1} + \alpha_ {2} H H I _ {k t})\tag{10}\]

We use the variable , to measure ease of collusion adopting the result of most dynamic oligopoly models by which the higher the market concentration the more collusive the output of the repeated game. According to the market power redistribution efect described in the model 3, we expect to observe , which means that in markets with little capacity of collusion, i.e., low has a positive efect on prices. This efect has to decrease as the ease of collusion, measured by , increases, i.e., we expect and . By continuity the multimarket efect is expected to be equal to zero for a value of between the minimum and the maximum values in our set of observations. Summing up, the efect of multimarket contact is expected to be greater in absolute terms if the variable measuring the ease of collusion in the focal market, , is among either the largest or the smallest in the sample, being positive in markets with very low values for and negative in markets with very high values for

5.2 Identifying approach to potential endogeneity

Concerns on the endogeneity of some of the regressors related to the simultaneous nature of quantities used to construct shares and concentration indexes call for an identification strategy. The simplest way to avoid biased estimates is to assume independence of markets across time, much in the way it is implicitly assumed in the competition game presented in section 3. Therefore, we lag potential endogenous variables to minimize this risk while keeping the estimation procedure simple.

We also include an additional regressor in the specification to control for possible relevant omitted data. The argument is that in pharmaceutical markets, product diferentiation in terms of attributes is of particular relevance. However many important brand attributes are not observable from the econometrician’s point of view because are not measurable or as it is in this case are absent from our data set. The stage game price will be a reduced form equation of marginal costs and a mark-up term which depends on product attributes as well as its rival products’ characteristics within market k. As information on attributes such as strength of a drug or type of package under which the drug is commercialized are not available we approach the mark-up value with some available characteristics (as noted in the last section) and data on product and corporation market shares. These observed mark-up elements will be correlated with unobserved characteristics of product i as well as competitors attributes whose efects by definition will be located in the error term. Abusing the language of the Instrumental Variables approach to the problem, we put forward an identification assumption that the independence of markets across countries gives us the opportunity of using the price or an index of prices of other products in the same market definition in other countries to control for the unobserved attributes. At any point of time these prices will be correlated with unobserved attributes of a number of products that interact with product i in market k, information that is also relevant to the firm to determine its prices. However these attributes of other products are not correlated with product’s i own characteristics and as such helps us to control for some of the unknown price variability. The variable constructed is a global price, Gprice, which is the average price of the products belonging to the same molecule group of product i considering other countries in sample.

This independence assumption might be controversial for some products that, as noted in section 2, could be subject to parallel imports or external Reference Pricing regulations. In efect, this condition is not satisfied for a number of products in the UK which are subject to parallel imports from other EU markets. Likewise, in EU countries external reference pricing has been used to set launch prices for new drugs reducing the efectiveness of our strategy. For this reason we lag Gprice again resorting to the Panel configuration of our dataset to further provide for an independent correlate with unobserved product characteristics.

6 Results, interpretation and analysis

6.1 Results from baseline specifications

Tables 3 to 5 in Appendix E present the set of basic results for the molecule market definition. We run within groups estimations for the Log(Price) on the set of quality and competition characteristics as well as the multimarket contact variables. In these and other tables the results are shown with the countries grouped from the more market friendly ones to the more heavily regulated in prices. The regressions in all cases include time trends and fixed efects at the corporation level as well as ATC-1 level fixed efects to control for diferences in price variation due to diferent insurance coverage. Accordingly, the t-statistics shown in parenthesis are computed with robust standard errors. Table 3 does not include multimarket contact variables. Its purpose is to show to what extent the remaining variables explain prices in the diferent countries and the type of consequences that the omission of relevant structural variables entails. It can be seen that variables Fsales, New, Molage and Generic have the expected signs in all the cases, however not significant in very few of them. Firm size, F sales, is highly significant, indicating that large corporations enjoy higher prices either because its products are of higher quality or perceived as such. New has a negative and significant efect which means that on average products that enter an existing market bear a price discount. Molage has a neg ative impact showing that the prices fall with the life-cycle of the molecule. However, Censormol which is expected also to have a negative efect appear to be with the wrong sign but with weak significance in most cases. These variables proxy molecule eficiency since new molecules are expected to improve upon previously existing molecules. Generic has a negative impact which is fairly intuitive, with the exception of Canada for which the signs is reversed but the marginal efect is not significant.

Consistently, Censorlag is positive in most cases, showing that products launched in the market before January 1991 maintain higher prices than those launched later within the same market. This efect however is not significant for most countries and appear with the wrong sign for Italy, Japan and Spain. Regarding Generic we find that the price of generics are, with the exception of Canada, significantly lower than other prices. Results for the number of Ngenerics, the number of generics in a market, are counter intuitive in that prices are predicted to be higher the greater the number of generic products available. As briefly mentioned in section 2, the presence of generics on a market does not mean that brand name products will reduce their prices. The evidence presented by the specialized literature is mixed. In some cases, the presence of generics will have the impact of concentrating brand name products over the inelastic portion of the demand which will then increase the price of these products. Hence, the expected sign of the number of generics will be positive. In our results this is the case of US, Germany, Netherlands, UK –the strongest efect, and France. On the other hand, the number of generics or generic competition, will reduce prices for everyone whenever, for instance, the quality of the existing products is not necessarily perceived to be high enough. In our results this seems to be the case of Canada alone. Note finally, that the efect of the number of generics is also positive for the heavy regulators (Italy, Japan, and Spain). Except for the case of US and Canada, this result is in line with the previous comprehensive analysis of regulation and competition performed in Danzon and Chao (2000).

The HHI concentration index, , is not significant and in some cases appear with the wrong sign for less regulates countries. For the market share of the product, Mshare, and of other corporation’s products in the same market, Cshare, the expected signs are observed except for the particular cases of Canada, and Spain. Also, for the majority of less regulated countries, Mshare is significant while Cshare is not.

From these first set of results interesting preliminary conclusions can be drawn. First, it appears that most attributes and quality characteristics explains a reasonable portion of price variations which is robust across countries. This suggest that diferent degrees of regulation does not distorts the efects of these attributes. The only attribute that seems to have a diferent efect with respect to the level of regulation is Censorlag, although the significance of the variable is in general poor. With respect to variables controlling competition, apart from the number of generics, although not significant in many cases at least the signs appear correct for most less regulated countries, excluding Canada.

Table 4 presents the results of the same regressions after including the average multimarket contact variable in its first version. The estimate for the parameter capture the average efect of the multimarket structure weighted by the HHI of the contact markets. All other coeficients remain fairly stable. This variable appears not significant for the US, UK and most of the highly regulated countries.

Table 5 allows for a diferentiated efect of the multimarket contact variable on prices depending on the concentration of the reference market, a variable that proxies ease of collusion. The efect is collected in parameter . According to the theory outlined in section 3, in presence of multimarket contact, prices are expected to fall in markets where it is easier to reach collusive outcomes whilst they are expected to increase where it is more dificult to collude. This means that the coeficient , is expected to be positive and the coeficient , is expected to be negative, with the latter larger in absolute value than the former. Notice that with respect to Table 3, in general the competition variables, either turn to be positive for some of the cases they where incorrectly negative or increase its value above zero. This suggest that omitting the multimarket structure in explaining price variations is relevant enough so as to bias the efect of competitive variables.

Turning back to the multimarket contact efect, the results for the US where pharmaceutical markets are free, is consistent with the theory. An increase on one weighted contact for a firm is expected to increase prices whenever that is for markets where concentration is below . Canada in its turn is also consistent and the size of the efect is greater than that of US countries showing evidence that mild price controls provides additional slackness in less competitive markets then redistributed to more competitive ones. Prices are expected to be higher in Canada with respect to the US for low levels of the HHI and lower for higher values of the HHI. In this case, all markets where HHI is below ≈ 0.73 are expected to have higher prices. Germany is also consistent with the theory although the size of the efect is smaller than that of the US and in particular show that the multimarket contact efect is relatively flat for diferent levels of concentration. For the remaining of the medium regulated countries, Netherland and UK, the corresponding coeficients are individually not significant nevertheless the sign of the coeficients is correct. France is strongly consistent with the theory even though it is classified as highly regulated. For Italy the coeficients are significant but contradicts the predictions of the market power redistribution hypothesis. Finally, Spain and Japan show that multimarket contact does not explain price variations which is also consistent with the theory; strong price regulation is expected to hinder the conditions for the redistribution efect. For all the estimated equations in Table 5 we report the results for the linear hypothesis test of joint significance of the contact variables. Given the number of observations in each regression the corresponding critical value for the F distribution function is at the 0.05 level of significance. The null is rejected for the UK, Japan and Spain.

As highlighted in the previous section, for the countries where the theory is supported it is possible to show a threshold for the concentration index below which the equilibrium price is afected positively and above which it is reduced through the multimarket contact mechanism. We show these thresholds graphically in figure 1 at the end of the document. It can be seen that for the less regulated countries, this level of concentration is above 0.8. The result then suggest that the multimarket contact mechanism of market power redistribution is predicted to function at very high levels of concentration, which in turns suggest that the creation of some market power slack is possible only at substantially low levels of competition.

The first set of results are completed by Table 6 showing the fixed efects at the ATC-1 level and Table 7 reporting comparative results of the multimarket contact efect for the two alternative definitions for AVMMC. Apart from obvious scale efects in the estimates for the second definition there are no specially interesting diferences. The only relatively important diference refers to Germany for which multimarket contact appears to deliver higher prices irrespective of the level of concentration of the market of reference with the second definition. This result is interpreted as evidence of market power transference.

We have found that the theory is consistent in the US and appears to be statistically significant. The same is obtained for Canada, however, as predicted by the theory, mild price regulations enlarges the efect of market power redistribution. For EU mild or medium regulated countries in general the signs of the coeficients and are in line with the theory but the size relation is dificult to find or not statistically significant. For these countries, multimarket contact is expected to increase prices no matter the level of concentration. This is in contrast with the market power redistribution hypothesis but, as tested with our specification, but can be reconciled with the market power transference hypothesis. France is highly consistent with the theory, although it is considered a heavy price regulator. Anyhow, in all two alternative definitions, multimarket contact is expected to deliver lower prices compared to the US for a wide range of values of the HHI. This is suggesting that France is relatively consistent with the theory because constrained prices may leave little room for market power redistribution. Japan and Spain have no multimarket contact efect showing that the theory also correctly predicts that for highly regulated markets stringent price constrains hinders the multimarket contact efect, possibly precluding some prices to increase and others to be reduced.

The next set of basic results are shown in Tables 8 to 10. We have repeated the estimation strategy for the ATC-4 market definition, however we only show the estimation output for the complete multimarket specification (comparable to Table 5). No significant changes are identified regarding the control variables with respect to the molecule market definition results. Nonetheless, the variable Nmols, number of molecules available at the ATC-4, has negative and significant efects for US, Canada, Italy and Spain, whereas it is not significant for Netherlands and UK. Substitution among available molecules within an ATC-4 definition are either significant or absent for this countries. For the rest this variable appears with a positive signs contradicting what was expected in the first place. Tables 9 and 10 contain the fixed efects at the ATC-1 level and the comparative results for alternative multimarket contact respectively. For space reasons it is more relevant to concentrate on the last table. Again, the multimarket contact prediction of market power redistribution can be found for the US for definition 1, however now it is predicted that prices are expected to increase due to this efect for markets with concentration indexes below, 0.89 Extending the market definition increases the set of markets for which prices are expected to be positively correlated with multimarket contact compared to the molecule definition results. For definition 2, however we find a threshold of 0.65 similar to what was found at the molecule level.

With the ATC-4 definition it is not possible to find the redistribution efect for Canada and Germany, in these cases now either multimarket contact has no significant efect over prices or prices are predicted to increase for any level of concentration. Same as before, Netherlands and UK show a positive efect of multimarket contact irrespective of the level of concentration. For the highly regulated countries, France is still consistent with the theory, now for Italy the theory has no significance as predicted for a strong regulator. Spain remains unafected by the multimarket structure and Japan now appears to suggest a negative efect of the contact variable which is inconsistent with the theory.

In sum, for the ATC-4 definition, US is consistent with the redistribution hypothesis as expected for an un-regulated case. Now for medium regulated countries the redistribution hypothesis is not verified, however multimarket contact is expected to increase prices for all levels of concentration in some cases. The model outlined in section 3 predicts this result both for product diferentiation and homogenous goods and some slackness is available to be transferred to other markets. However, the ATC-4 definition was expected to absorb precisely more dimensions of product diferentiation given that we consider competitive constrains coming from alternative varieties with close therapeutical efects. Therefore, our results does not support the underlying theory.

6.2 A restricted model for regulatory efects

To study further whether the theoretical predictions can be supported in our dataset we perform a series of restricted versions of our approach. Our aim is to test if the marginal efects for diferent groups of countries are statistically diferent and meaningful from the point of view of the theory of the multimarket contact redistribution efect. We first run three restricted models to compare three groups of countries with respect to the US benchmark. First we compare the US with Canada, then the US with the medium regulated countries in Europe, and finally the US with Japan and Spain which are the highly price regulators for which the results seems to be in accordance of the theory. Finally we pool Canada with medium EU regulators.

To perform this restricted version we estimate the following specification for the multimarket contact efect:

\[\Omega^ {R} (\dots) = A V M M C S _ {i k} \left(\beta_ {1 0} + \beta_ {1 1} D r e g + \left(\beta_ {2 0} + \beta_ {2 1} D r e g\right) \times H H I _ {i k}\right)\tag{11}\]

All the control variables used in the baseline regressions are kept and the specification is completed by allowing a diferentiated efect of the number of generics as well as a dummy variable to allow for a diferent intercept for each group in a regression. Results are shown in the following table:

Table C: Estimates by group of countries ^a

* p<0.05, ** p<0.01,*** p<0.001 (1) The estimates correspond to the third definition of the multimarket contact variable (2) t-stats in parenthesis are computed with robust standard errors (3) The null hypothesis’s statistic has a critical value of 5.99 at 0.05

A test of the joint significance of the interacted variables with the dummy are also presented. The first column confirms that Canada, the mild regulator with respect to the rest of countries in our sample, is expected to deliver a more profound market power redistribution with respect to the benchmark case. Significance of the interactions of the redistribution efect specification with the dummy cannot be rejected. The second column suggests that prices are lower for low concentration values for medium regulated countries in EU with respect to the US, although appears to be not significant. The third column shows that in general an extra average contact will have a flatter efect over price with respect to the benchmark, however the profile estimated in this regression is increasing with respect to the concentration index. In the baseline results Spain and Japan showed no significant efect for the multimarket contact specification which we interpreted as the extreme case of regulation in which prices will follow only market specific conditions. This results cannot be confirmed in this restricted model. The fourth column show that medium regulated countries will experience lower prices with respect to a mild regulator, Canada, for low levels of concentration which is also consistent with the theory. Note that the diferent sub-samples used in each column does not suggest instability of the estimated marginal efects for US and Canada.

6.3 Analyzing segmented multimarket rivalry

In this subsection we are interested in asking the question whether segmentation of product markets between branded and generic products can either modify our baseline results or give additional insights concerning the relevant strategic efects of the multimarket structure for the special case of the pharmaceutical industry. Introducing segmentation in our analysis means that the relevant contacts for a corporation producing (or commercializing by a third party under license) a branded product are those with rivals producing branded drugs and likewise for generic producers.

We interact the multimarket contact variable with dichotomous variables, and indicating whether the observation belongs to the branded segment or the generic one respectively. The contacts are re-set to formulate multimarket rivalry in terms of the relevant contacts in the corresponding segment of the market. Therefore, we redefine the contact variable in (4) following:

\[M M C S _ {i l, k m} = C _ {i l, k m} w _ {m} \prod_ {k ^ {\prime} = k, m} (D _ {b i k ^ {\prime}} \times D _ {b l k ^ {\prime}} + D _ {g i k ^ {\prime}} \times D _ {g l k ^ {\prime}})\tag{12}\]

In this way, the contact variable will be strictly positive whenever firm i and l belong to the same segment in market k as well as in the contact market m. If we separate the number of corporations in the focal market, , between those producing branded, and generics, the average multimarket contact variable for corporation i in market k is re-defined in the following manner:

\[A V M M C S _ {i k} = \sum_ {s = b, g} \frac {D _ {s i k}}{(N _ {s k} - 1)} \sum_ {l \neq i} \sum_ {m \neq k} M M C S _ {i l, k m}\tag{13}\]

Where the added in the variable’s name stands for the segmentation exercise. The market power redistribution hypothesis is tested by handling a modified version of our original specification:

\[\Omega^ {S} (\dots) = A V M M C S _ {i k} \left(\left(\gamma_ {1} D _ {b i k} + \gamma_ {2} D _ {g i k}\right) + \left(\gamma_ {3} D _ {b i k} + \gamma_ {4} D _ {g i k}\right) \times H H I _ {k}\right)\tag{14}\]

Coeficient estimates for the molecule market definition and the third multimarket contact definition are reported in Table D. It also shows the result of a test for the null hypothesis that the multimarket contact redistribution efect is equal for both segments. The third definition seemed more appealing for this specification because it weights each contact by firm specific weights, the other two definitions consider information of the whole contact market which is dificult to make compatible with a segmentation analysis. For the US it appears that the redistribution efect is present in both segments and the corresponding marginal efects have statistically the same size. Interestingly, estimates for Canada and UK seem to support evidence of the redistribution efect only for branded products, γ₂ and are individually not significant. This is weakly observed also for the Netherlands, although the coeficients do not preserve the size relation. For France the result indicates that the redistribution efect appears stronger in the generic segment.

Table C: Segmented analysis of the redistribution efect

p<0.05, ** p<0.01,*** p<0.001 (1) The estimates correspond to the third definition of the multimarket contact variable (2) t-stats in parenthesis are computed with robust standard errors (3) The null hypothesis’s statistic has a critical value of 5.99 at 0.05

6.4 Lag of entry as a source of instability

We have argued that corporation fixed efects helps us to control for vertical product diferentiation, considered to be an important source for price variation across products. However we have been suggested that quality diferentiation across markets might change the results of the redistribution efect. The argument goes in the direction of a result by H¨ackner (1994) who showed that quality product diferentiation might hinder collusion. Therefore, as market concentration is a result of this source of product diferentiation, our results might be collecting these efect.

Consider the molecule market definition where vertical diferentiation may be a more influential feature as opposed to the ATC-4 definition where possible horizontal product diferentiation is more relevant. In this section we aim at evaluating whether our results are changed by including a variable that is expected to be correlated to quality diferentiation within the group of drugs belonging to the same active ingredient.

We assume that branded (originator) products are those regarded of high quality within the group of chemically equivalent products. Using originator drugs as a quality benchmark we calculate the average time in years elapsed until entry of other drugs within a molecule group. This variable is called and varies only across markets and time in our sample. The idea is that molecules where the originator has been more time alone in the market are thought to be more vertically diferentiated because the originator has been able to accumulate more reputation.

To control for the variable vertical product diferentiation we perform the following specification for the multimarket efect:

\[\Omega (M M C _ {i k t}, H H I _ {k t}, E n t r y l a g _ {k t}) = (\lambda_ {0} + \lambda_ {1} H H I _ {i k t} + \lambda_ {2} E n t r y l a g _ {i k t}) \times A V M M C _ {i k t}\]

Table D presents the results for the first contact variable definition. The results are comparable to those in Table 7 in the appendix. For the US, our benchmark case, introducing Entrylag within a molecule market changes the profile of the multimarket contact efect. In all three definitions, the higher the average lag of entry of competitors increases prices due to multimarket contacts. Still, prices are expected to be negatively afected for more concentrated markets, however for suficiently high average years of lag of entry, this last efect dominates. In particular for definitions 1 and an average entry lag of around ≈ 4 years ofsets the expected redistribution efect for a highly concentrated market. For definition 3, average entry lag that ofsets the redistribution efect for highly concentrated markets is estimated to be around 7 years. This result suggest that whenever the originator is allowed to accumulated greater reputation within a market, concentration does not necessarily commands the redistribution efect. In Canada and Germany where the multimarket contact efect is present entry lag of rivals seems not to place any sizeable efect, and in any case the sign of the efect is negative. The same applies to France. For UK, Japan and Spain the multimarket contact efect is not jointly significant.

Table E: Average entry lag and market power redistribution

* p<0.05, ** p<0.01, *** p<0.001 (1) t-stats in parenthesis are computed with robust standard errors (2) The null hypothesis’s statistic has a critical value of 7.82 at 0.05

For the regulated countries it might be the case that entry lag is not necessarily a good measure for vertical diferentiation because price regulation may distort the profitability of investing in publicity and reputation. It is also important to highlight that this exercise, absent an observable measure of vertical product diferentiation, implies inducing measurement error bias in our estimates, hence this results are to be taken with caution.

7 Concluding remarks and discussion

7.1 Reconciling theory with observations

We have focused to study whether a model of multimarket contact, whose source of asymmetry among markets is product heterogeneity, can be reconciled with observed data. The model predicts that if slack of market power in the sense of dynamic competition games is available, firms can transfer this slack to market with less favorable environments for collusion, hence prices will be higher in general whenever multimarket contacts increases. If firms cannot monopolize any market (or the industry), they may find it optimal to create slack by reducing prices in more collusive markets and apply this to sustain higher prices in more competitive markets. Firms are then expected to redistribute market power.

The main point of the paper is that price regulation can have quite diferent efects over firms’ pricing which can be characterized by diferences with respect to what is predicted for the unregulated case. The pharmaceutical industry is a paramount case of an industry where products are diferentiated within markets and price regulation is a common issue. Our strategy is to estimate a specification for pricing equations that seeks to capture the redistribution efect for market power with information from diferent countries which are known to place diferent intensities of drug price regulation among them.

Our results suggests that for the US, our un-regulated benchmark, the redistribution efect is present and is statistically significant for various contact definitions and market definitions. We have also obtained that Canada, a mild regulator, will produce the redistribution efect, however the size of the efect is significantly larger with respect to the US, which can be interpreted in terms of the multimarket contact theory. For medium regulated countries, belonging from the EU, it is not clear the presence of the redistribution efect, however in general it can be inferred that prices will be lower than the un-regulated case for more competitive product markets, a result that is at least weakly consistent with the theory. Finally, results for some countries that are known to place stricter price ceilings suggests as expected that firms are not able to use the existing multimarket structure as a collusive device.

In terms of significance of the size efect of the multimarket contact structure over prices, we have calculated predicted average elasticities for the diferent contact definitions and market definitions. In general, prices are inelastic to changes in the multirmaket structure of a market. For the US we calculate a very small but positive average elasticity (standard deviations in parenthesis) of 0.006(0.031) considering the molecule market definition, and 0.122(0.132) considering the ATC-4. Therefore, market definitions matters for the size of the average price efect in this case. For Canada, the average elasticity is around 0.10(0.18) for both markets definitions, for Germany the case is similar to the US.

7.2 Policy implications

From the point of view of policy evaluation, considering the multimarket contact structure of the industry is helpful to highlight interesting implications for regulators. The redistribution of market power creates price increases and price decreases in diferent product markets for the unregulated case. Hence consumer welfare might be improved with respect to a case in which corporations consider product markets in isolation. Mild price regulations can reduce this welfare efects because, as in Canada, they are expected to produce higher relative prices in more competitive product markets with respect to the

un-regulated benchmark.

On the other hand, medium regulators, for which prices are expected to be lower than for the un-regulated benchmark and the mild regulation case, might be in efect enjoying larger consumer welfare efects. However, this is expected to be a short-run efect. As price regulation in some product markets is translated into lower prices in more competitive markets, entry will be discourage as the return of investment will be lower than in a un-regulated case. On aggregate, lack of entry or postponing entry decisions can be explained in terms of the multimarket contact mechanism. The same idea applies to strong price regulators. Therefore, our evidence suggest efectively that a plausible source for observing less dynamic entry, in EU with respect to the US for example, is the way in which regulation afects the profile of multimarket contact pricing.

Although in the end welfare efects of price constrains cannot be determined unambiguously, our results provide evidence of a mechanism by which price intervention may distort firm’s behavior. Discussions on the expected efects of introducing price regulation for the US (Santere and Vernon (2005)), have focused basically in the disincentives to research and development with the crude conclusion that consumer welfare will sufer from the reductions of availability of innovative drugs. From our perspective, regulation might also reduce incentives to enter markets where barriers to entry are lower (more competitive markets) reducing also the availability of varieties or substitute treatments.

References

1. Berndt, E., Bui, L., Reiley, D. and Urban, G.: 1995, Information, marketing and princing in the u.s. antiulcer drug market, American Economic Review 85(2), 100–105.
2. Bernheim, D. and Whinston, M.: 1990, Multimarket contact and collusive behavior, Rand Journal of Economics 21, 1–26.
3. Cabrales, A.: 2003, Pharmaceutical generics, vertical product diferentiation and public policy. Working Paper 662, Universitat Pompeu Fabra.
4. Cleanthous, P.: 2004, Evaluating innovation in the pharmaceutical industry. MIMEO, New York Univeristy.
5. Danzon, P. and Chao, L.: 2000, Does regulation drive out competition in pharmaceutical markets?, Journal of Law and Economics 43, 311–357.
6. Danzon, P. and Epstein, A.: 2008, Efects of regulation on drug launch and pricing in interdependent markets. MIMEO.
7. Danzon, P. and Furukawa, M.: 2003, Prices and availability of pharmaceuticals: Evidence from nine countries, Health Afairs 22, 521–536.
8. Danzon, P., Wang, Y. R. and Wang, L.: 2005, The impact of price regulation on the launch delay of new drugs., Health Economics 14(3), 521–536.
9. Edwards, C.: 1955, Conglomerate bigness as a source of power, The National Bureau of Economic Research Conference Report, Business Concentration and Price Policy, Princeton University Press, pp. 331–359.
10. Evans, W. and Kessides, I.: 1994, Living by the “golden rule”: multimarket contact in the u.s. airline industry, Quarterly Journal of Economics 109, 341–366.
11. Fernández, N. and Marín, P.: 1998, Market power and multimarket contact. some evidence from the spanish hotel industry, Journal of Industrial Economics 46, 301–315.
12. Fu, W.: 2003, Multimarket contact of u.s. newspaper chains: Circulation competition and market coordination, Information Economics and Policy 15, 501–519.
13. Gimeno, J. and Woo, C. Y.: 1994, Multimarket contact as a mechanism of transfer of power across structurally diferent markets. Working Paper/Texas AM University.
14. H¨ackner, J.: 1994, Collusive pricing in markets for vertically diferentiated products, International Journal of Industrial Organization 12, 155–177.
15. Jans, I. and Rosembaum, D.: 1996, Multimarket contact and pricing: Evidence from the u.s. cement industry, International Journal of Industrial Organization 15, 391–412.
16. K¨onigbauer, I.: 2006, Advertising, generic competition and price regulation in pharmaceutical markets, CESifo Economic Studies 53(3).
17. Kyle, M.: 2007, Price controls and entry strategy, Review of Economics and Statistics 89, 88–99.
18. Matsushima, H.: 2001, Multimarket conctact, imperfect monitoring and implicit collusion, Journal of Economic Theory 98, 158–178.
19. Parker, P. and Roller, L.: 1997, Collusive conduct in duopolies: Multimarket contact and cross-ownership in the mobile telephone industry, Rand Journal of Economics 28, 304– 322.
20. Phillips, O. and Mason, C.: 1996, Marker regulation and market rivalry, The Rand Journal of Economics 27(3), 596–617.
21. Pillof, S. J.: 1999, Multimarket contact in banking, Review of Industrial Organization 14, 163–182.
22. Puig-Junoy, J.: 2005, Price regulation systems in the pharmacetical market, in J. Puig-Junoy (ed.), The Public Financing of Pharmaceuticals, Edgard Elward Publishing.
23. Santere, R. E. and Vernon, J. A.: 2005, Assesing consumer gains from a drug price control policy in the u.s., NBER Working Paper Series (11139).
24. Spagnolo, G.: 1999, On interdependent games: Multimarket contact, concavity and collusion, Journal of Economic Theory 89, 127–139.
25. Stern, S.: 1996, Market definition and the returns to innovation: Substitution patterns in pharmaceutical markets. MIMEO, MIT Sloan School.

Appendices

A Technical Appendix: Multimarket Contact

A.1 Product diferentiation case

We study an industry where there exist K product markets so as to approach what is observed in the pharmaceutical industry. Lets identify a single product market by k and denote by the number of firms producing diferentiated products with the same constant returns to scale technology. Assume that firms compete in prices under product diferentiation. Denote the individual profit function of a firm in market k as and assume it is concave in prices, where denotes the rival’s prices. Equilibrium prices in the one shot pricing game are denoted by and form a unique Nash equilibrium. ¹¹. Let denote the price that allow firms to achieve monopoly profits in the market. Assume firms compete in prices in a repeated fashion with infinite horizon. Firms may sustain in equilibrium higher than the stage game price equilibrium using trigger strategies in which any deviation of the collusive strategy is penalized by reverting forever to . Denote by the price for firm i in the repeated game in market concavity of the profit function implies is assumed to be selected such that it maximizes the present discounted value of the firm’s expected flow of profits subject to the incentive constraint that loses implied by deviations from the collusive path are greater than the implied gains ¹²:

\[\frac {\delta}{1 - \delta} [ \pi_ {i k} ^ {c} - \pi_ {i k} ^ {c} ] \geq \pi_ {i k} (R _ {i k} (p _ {- i k} ^ {c}), p _ {- i k} ^ {c}) - \pi_ {i k} ^ {c}\tag{15}\]

where and are firm i’s profits when prices are and respectively, are firm i’s profits when all firms other that i set their collusive prices, , and firm i chooses its best response to them, , and is the discount factor. Now, note that if is to be supported as a sub game perfect equilibrium, then it must be the case that there is no profitable deviation from it, in other words it satisfies:

\[\frac {\delta}{1 - \delta} [ \pi_ {i} ^ {m} - \pi_ {i} ^ {*} ] \geq \pi_ {i} (R _ {i} (p _ {- i} ^ {m}), p _ {- i} ^ {m}) - \pi_ {i} ^ {m}\tag{16}\]

Where is firm’s i profits from the joint profit maximization outcome. While the left hand side of this expression depends on δ, and increases monotonically in this argument, the right hand side is independent of the discount factor. If we denote the left hand side by the following condition is true:

\[F (0, \pi_ {i} ^ {m} - \pi_ {i} ^ {*}) < \pi_ {i} (R _ {i} (p _ {- i} ^ {m}), p _ {- i} ^ {m}) - \pi_ {i} ^ {m} < F (1, \pi_ {i} ^ {m} - \pi_ {i} ^ {*})\tag{17}\]

case, pⁿik = pⁿ_k , ∀i

pik

¹¹Given concavity and identical technologies we can look for the symmetric price equilibrium in which

¹²More precisely, in a symmetric product diferentiation set up p^cik is the price that maximizes joint profits under the constraint that loses from deviations are greater or equal than the gains from such an strategy.

This expression implies the existence of some threshold for the discount factor, call it above which the joint profit maximization outcome is a sub game perfect equilibrium. As it is evident, whenever the discount factor is above , higher than prices also conform with , hence if is the equilibrium it will be the case that there is some slackness in the incentive compatibility constrain. Below cannot be supported and we assume that firms will choose the maximum sustainable price defined as:

\[\tilde {p} _ {i k} (p _ {i k} ^ {*}, \delta) = \max \{p _ {i k} \in [ p _ {i k} ^ {c}, p _ {k i} ^ {m}) | F (\delta , \pi_ {i k} - \pi_ {i k} ^ {c}) \geq \pi_ {i k} (R _ {i k} (p _ {- i k}), p _ {- i k}) - \pi_ {i k}) \}\tag{18}\]

Note that the condition in this function when implies and when implies ; therefore whenever it should be the case that the condition holds with equality. This result holds also under the assumption that the optimal deviation profit is convex in prices ¹³. To proceed we place the following assumption:

Assumption: [Monotonicity] Function satisfies

It is not obvious how equilibrium prices under product diferentiation change as the discount factor increases, the above assumption says that whenever future profits are more valuable, short run benefits from defecting are accordingly less preferred. Therefore the collusive price, will be a non-decreasing function of

At any given function 18 shows that will depend on the same cost and demand conditions that determine . In fact, the best collusive price can be characterized based on the incentive compatibility constraint and Assumption 1 in the following way:

\[p _ {i k} ^ {*} = \Phi (\delta) p _ {i k} ^ {n}\]

where

\[\Phi (\delta) = \left\{ \begin{array}{l l} \frac {p _ {i k} ^ {m}}{p _ {i k} ^ {*}} & \text {if} \delta \geq \delta^ {0} \\ \frac {\tilde {p} _ {i k} (\delta)}{p _ {i k} ^ {*}} & \text {if} \delta < \delta^ {0} \end{array} \right.\]

A.2 Multimarket contact implications

We assume there is a number of rival firms present in the entire set of the K product markets. When a firm intends to deviate from the collusive equilibrium in any market k it will need to balance the gains of such deviation with the trigger of penalty reactions in the rest of the markets where it compete with its rivals. Therefore, firm i’s incentive constraint under the multimarket contact hypothesis becomes a pooling of the K individual market incentive constrains:

¹³Under this assumption, net benefits of collusion increases at a slower pace than the net gains of deviation up to the point a marginal increase in the collusive price will be in conflict with the incentive constrain

¹⁴Bernheim and Whinston (1990) nevertheless mentions that ‘’Product heterogeneity within each market adds considerable complexity since the maximum sustainable price typically increases continuously as the discount factor, δ, rises‘’.

\[\sum_ {k = 1} ^ {K} \frac {\delta_ {k}}{1 - \delta_ {k}} [ \pi_ {i k} ^ {'} - \pi_ {i k} ^ {*} ] \geq \sum_ {k = 1} ^ {K} \left\{\pi_ {i k} (R _ {i k} (p _ {- i k} ^ {'}), p _ {- i k} ^ {'}) - \pi_ {i k} ^ {'} \right\}\tag{19}\]

According to Bernheim and Whinston (1990) analysis, when markets difer in the number of firms, demand conditions or there are economies of scope, the pooled incentive constrain can be used to sustain higher prices in equilibrium. It is also claimed that asymmetries within markets, which are known to hinder collusion, might be softened through the multimarket contact setting. Absence of such diferences across markets makes contacts irrelevant for sustaining more collusive outcomes. Also, it is not the mere existence of multimarket contact across difering markets which expands the set of collusive outcomes, but the ability of firms in the coalition to transference market power across markets. Imagine for instance that in some of the K markets firms are able to sustain in equilibrium so that the individual incentive constrains are verified in some of them with inequality (in our setting for some . This slack can be used in markets where no collusive price is possible to sustain at the corresponding δ, for example markets where is suficiently large. Indeed firms in the multimarket coalition can increase prices in more competitive markets violating the individual incentive constrains as long as 19 is verified. Under this result, average prices in the industry will be unambiguously higher than in a case without multimarket contacts.

More interestingly, whenever markets difer in their degrees of product diferentiation firms can find it optimal not only to transfer but to redistribute their market power. To provide further intuition for this result we introduce the following assumption:

Assumption: [Product diferentiation and Collusion] For a given value of the discount factor δ, the sustainable best collusive price increases with the degree of product diferentiation.

Assumption above although not directly intuitive can be supported by the fact that the one shot pricing game delivers higher profits the higher the product diferentiation however under higher product diferentiation the net gains of deviation are lower. Under the above assumption we will expect that high prices, perhaps close to the monopoly price, are sustainable in equilibrium in some markets whereas in other markets sustainable prices are close to the stage game price equilibrium based only in variations of the degree of product diferentiation across markets.

Example: [Hotelling model] For the case of a duopolistic multimarket contact structure with two independent markets. Both the monotonicity assumption and the efect of product diferentiation over the best collusive price are rather observed properties.

To parallel Bernheim and Whinston (1990) lets assume that each of the K markets in our model is a duopoly and that the coalition cannot sustain monopoly prices in the whole industry which implies that at the equilibrium prices, the pooled incentive constrain must be binding. To simplify the analysis we can look for a symmetric price equilibrium in which each corporation i in market k set . The optimal pricing decision for a firm in the multimarket coalition solves:

\[\max _ {\{p _ {k} \} _ {1} ^ {K}} \sum_ {k = 1} ^ {K} \pi_ {k} (p _ {k}, p _ {k})\tag{20}\]

s.t

\[\frac {\delta}{1 - \delta} \sum_ {k = 1} ^ {K} [ \pi_ {k} (p _ {k}, p _ {k}) - \pi_ {i k} ^ {*} ] = \sum_ {k = 1} ^ {K} \{\pi_ {k} (R _ {k} (p _ {k}), p _ {k})) - \pi_ {k} (p _ {k}, p _ {k}) \}\]

The K first order (necessary) conditions together with the incentive compatibility constrain delivers the following optimal conditions for any :

\[\frac {\pi_ {k} ^ {'} (\hat {p} _ {k} , \hat {p} _ {k})}{\pi_ {k} ^ {'} (R (\hat {p} _ {k}) , \hat {p} _ {k})} = \sum_ {s, s \neq k} ^ {K} \frac {\pi_ {s} ^ {'} (\hat {p} _ {s} , \hat {p} _ {s})}{\pi_ {s} ^ {'} (R (\hat {p} _ {s}) , \hat {p} _ {s})} \forall k\tag{21}\]

This last condition relates equilibrium prices in one market, k to market conditions in the rest of the markets where a firm is in contact with its market k rivals providing interesting predictions. For instance, assume that for the given is sustainable in market k whenever it is considered in isolation and for the rest of the markets, product diferentiation is small enough so that no collusive price is sustainable. Concavity of the profit function implies that . Therefore, the left hand side of condition 21 indicates that it is optimal for the firm to set market’s k price below the monopolistic level, and at least for one contact market prices will have to be adjusted above the one shot game price. This example shows that multimarket contact may incentive optimal redistribution of collective market power.

Given the structure of each k market and as a consequence the structure of multimarket contacts for firm i, we can represent the firm’s equilibrium price of the repeated game in market k as a function of three separable components:

\[p _ {i k} ^ {*} = \Gamma (M M C _ {i k}) \Phi (\delta) p _ {i k} ^ {n}\tag{22}\]

where measures the efect of the multimarket contacts structure given by variable and the other two components come from the definition of the single market price equilibrium. The redistribution of market power can be tested in terms of the value of the multimarket function. will be expected for markets where a collusive price is easier to support (less toughness of price competition) in equilibrium and in markets with less favorable conditions to sustain collusion. Diferences in toughness of price competition among markets can be obtained by looking at the number of varieties available in a market, market concentration, product diferentiation and so on.

A.3 Multimarket contact and price regulation

This section poses an argument close to the one presented in Phillips and Mason (1996). Consider the example presented in the last sub-section. Consider now that there is a regulatory agency whose only objective is to reduce prices. In particular lets assume the agency focuses in product markets where market conditions allow firms to sustain monopolistic pricing, and the corresponding individual market constrains are not binding according to our model. Assume the agency does not consider the multimarket structure, that is the agency targets each market individually, considering the market specific circumstances. Recall condition 21 and assume the regulatory agency sets an exogenous binding price ceiling with respect to the un-regulated multimarket case for market k (where the monopolistic price was assumed to be just sustainable), . We highlight three points of interest:

1. Whenever the ceiling close to concavity of the profit function and convexity of the deviation profits:

\[\frac {\pi_ {k} ^ {'} (\hat {p} _ {k} , \hat {p} _ {k})}{\pi_ {k} ^ {'} (R (\hat {p} _ {k}) , \hat {p} _ {k}) [ \frac {\partial R (\hat {p} _ {k})}{\partial \hat {p} _ {k}} + 1 ]} < \frac {\pi_ {k} ^ {'} (\bar {p} _ {k} , \bar {p} _ {k})}{\pi_ {k} ^ {'} (R (\bar {p} _ {k}) , \bar {p} _ {k}) [ \frac {\partial R (\bar {p} _ {k})}{\partial \bar {p} _ {k}} + 1 ]}\]

This situation suggests that not too restrictive price ceilings will increase the slack produced in market k freeing additional market power to be distributed across the rest of the markets involved in the multimarket contact structure. Therefore, additional price increases might be observed in other un-regulated markets or where price ceilings are not binding.

2. Now imagine an extreme case in which , that is the price ceiling is equal to the stage game price equilibrium. Condition (A.1) implies that no slack will be available to re-distribute to the rest of the contact markets. Therefore, prices in these contact markets will be given by equation (18). In our example, equilibrium prices in the other markets will reflect only the stage game conditions.

3. By continuity of the profit functions, the reasoning in the last point can be extended to cases in which some of the other markets may sustain prices above the stage game equilibrium. Given function for the unregulated case, there must be a level of price ceiling for so that there are corresponding outcomes satisfying . That is price regulation in one market might reduce prices in others with respect to the level of the unregulated multimarket case.

B Control Variables Definitions

C Descriptive statistics

_n and S.D. _{(in parenthesis)} for variables in sam^pl

_n and S.D. _{(in parenthesis)} of mul^timarket conta

D Graphics

Figure 1: Efect of multimarket contact in selected markets. Market definition: molecule.

E Estimation results

ing regressions for molecule markets No Multima ions include corporation and _A^TC-1 fixed efects and ^t _n parenthesis are com^puted with robust stan _{0.05, ** p<0.01,} ^{*** p<}

ing regressions for molecule m^arkets: Average multi ions include corporation and _A^TC-1 fixed efects and ^t _n parenthesis are com^puted with robust stan

regressions for molecule marke^ts: Multimarket contact redis _{ypothesis’s} statistic _2×F(2,∞⁾ has ^a critical value ions include corporation and _A^TC-1 fixed efects and ^t _n parenthesis are com^puted with robust stan

_n parenthesis are com^puted with robust stan ₍⁽ _{0.05, ** p<0.01,} ^{*** p<} cing regressions for molecule markets: ^ATC-1

ation C^luste Table7:ComparativeestimatesbyMultimarketContactVariableDefinition(t-statsbasedonRobustStandardErrorsbyCor

Pricing regressions for _ATC-4 m _{ypothesis’s} statistic _2×F(2,∞⁾ has ^a critical value ^o ions include corporation and _A^TC-1 fixed efects and ^t _n parenthesis are com^puted with robust stan _{5, ** p<0.01,} ^{*** p<}

_n parenthesis are com^puted with robust stan _n of _ATC-1 fixed efects _(t-stats based on R^obust Standard Errors by Corp