The consequences of generic entry:
price, promotional e¤ort and market share
PRELIMINARY.
Micael Castanheira
Carmine Ornaghi
Maria-Angeles de Frutos
Georges Siotisy
July 5, 2016
Abstract
This paper identi…es a puzzling piece of evidence, speci…c to the pharmaceutical industry: typically, a molecule that loses patent protection (the originator drug plus its generic
competitors) experiences a steep drop in market share measured in quantity. This, despite being sold at a fraction of the original price. We combine theory and empirics to
investigate the economic forces that underpin this outcome. We …nd that, while intramolecular competition by generics leads to a steep fall in price, it also has a knock-on
e¤ect on promotion. This combined drop produces ambiguous e¤ects on inter-molecular
competition. Our models identi…es the condition under which a lower priced drug loses in
terms of quantity market share. Using a large dataset that covers all prescription drugs
sold in the US during the period 1994Q1 to 2003Q4, we test our theoretical predictions
and …nd strong support for them.
JEL Classi…cation: D22, I11, L13
Keywords: Pharmaceutical industry, generic entry
Micael Castanheira is “Maître de recherche” FRS-FNRS and gratefully acknowledges their …nancial support.
y
Siotis gratefully acknowledges the …nancial support from the Ministerio Economía y Competitividad
(Spain), grants, Beca I3 2006/04050/011, ECO2015-65204-P, MDM 2014-0431, and Comunidad de Madrid,
MadEco-CM (S2015/HUM-3444).
1
Introduction
The pharmaceutical industry brings together many characteristics that have been of interest
to Industrial Organization scholars. It is oligopolistic, regulated, and di¤erentiated – both
horizontally and vertically. It is also unique in combining high R&D and promotion intensity:
while …rms in other lines of business may have larger R&D or advertising budgets in absolute
value, R&D and promotion each represent 15% to 20% of Big Pharma’s sales.1 The industry’s
growth rate has also been impressive over the last decades, with worldwide sales nearly
totaling a trillion USD in 2013. Recent estimates regarding the size of the US market stand
at 374 billion USD in 2014.2 The cost of drugs is an important component of growing
health care costs; a concern for health insurers and public authorities alike. Encouraging
competition by generic bioequivalents is potentially a key to moderating these ballooning
costs. Throughout the paper,we use the terms “drug”, “molecule”, “chemical entity”, and
“active pharmaceutical ingredient” interchangeably.
This paper analyzes the e¤ects of generic competition from a theoretical and empirical
standpoint. Our motivation lies in a ba- ing piece of evidence: while generics appear to
be …erce competitors for the originally patent-protected drugs, they appear toothless against
other molecules. That is, while the generics gain volume at the expense of the originator,3 the
sales of the molecule (generics + originator) drop as a fraction of total market quantity. This
implies that few new patients are directed to the cheap genericized molecule while a number
of existing patients switch towards competing molecules at the time their initial treatment
becomes cheaper.4 This observation is derived from exploiting a dataset that encompasses
the entire set of prescription drugs sold in the US for the period 1994-2004. Cross-price
elasticities thus seem to have “the wrong sign”. That is, like Janus (the Roman god of
beginnings, transitions, time and endings), generics have two faces: they look like normal
goods with respect to their narrow bioequivalent, and Gi¤en goods in the market at large.
1
In the Oxford Hanbook of the Biopharmaceutical Industry, Harrington (2012) also estimates the R&D
to be at 17.9% of total net sales for the period 2001-2005, and Kenkel and Mathios (2012) report that the
advertising-to-sales ratio was 18% in 2005 in the U.S. As points of comparison, they highlight that advertising
is 4.5% of total net sales for General Motors (a car producer), 9.5% for Anheuser Busch (a beer producer)
and 10.8% for Kellogg (breakfast cereals). The …gures are typically smaller for most other R&D intensive
industries. For instance, in 2013, Apple spent 3% of its total net sales on R&D and 0.4% on advertisement
(Apple 2013: 10-K SEC submission).
2
Source:
http://www.statista.com/statistics/263102/pharmaceutical-market-worldwide-revenue-since2001/ and http://www.statista.com/statistics/238689/us-total-expenditure-on-medicine/
3
See also Gabrowski et al. (2014) who show that brands retain on average only 16% of the molecule’s
market share after 1 year.
4
Generic competition could be less than fully e¤ective because some patients and doctors fear that generics
are not an exact bio-equivalent (e.g. due to dosage problems) to the original drug. Yet, if this were the primary
reason for poor generic penetration, we should observe that they are also –perhaps mainly –toothless against
the original, branded, molecule. The opposite is observed in the data.
1
We develop a simple model to analyze which strategic interactions may produce this
outcome. Our initial conjecture is that promotional e¤ort, and change thereof, is a key driver
behind observed market outcomes.5 Concretely, when a drug faces generic entry, promotional
e¤ort also drops dramatically. Our model allows to identify the circumstances under which
this will more than o¤set the fall in prices.
We consider two pharmaceutical …rms that compete for patients. We compare the situation before and after the loss of exclusivity that triggers generic entry. The model builds on
three important features of the industry. First, patients di¤er in their therapeutic response
to each treatment, and they need doctors to identify the optimal treatment (see Inderst and
Ottaviani 2012 for a seminal model of inter-…rm competition through intermediaries). Second, some form of (private or public) health insurance coverage is common, thus attenuating
patients’price sensitivity. This is reinforced by the fact that doctors do not bear any direct
cost associated with their prescription choices –to the contrary, they may receive perks from
pharmaceutical …rms. Third, …rms compete through more than one instrument: prices matter but so do non-price instruments, such as “kickbacks” (which are typically unobserved),
direct-to-consumer-advertising and, most prominently, detailing e¤ort6 and free samples (for
which we have data).
Our analysis focuses on the post-R&D stage; i.e., we treat the extent of di¤erentiation as
exogenous. Before the loss of exclusivity (LoE), the two …rms choose the price and promotion
e¤ort that maximize their pro…ts. Then, …rm A loses exclusivity, and market competition
becomes lopsided: …rm A faces direct competition from generic bioequivalents (which we
assume to be perfect substitutes), whereas …rm B is still protected by its patent. Generic
competition on drug A drives down the price and promotion of that molecule. We …nd that
whether generic competition for market A weakens or strengthens the competitive pressure
faced by B depends on how close substitutes molecules A and B are.
Counter to standard intuition, B sees its demand increase, allowing it to gain market share
and further increase its price, when it is a close substitute to A. Conversely, competition is
only reinforced if A and B are su¢ ciently distant substitutes. The reason for these shifts in
demand is that, when generics enter, the price of A drops: this sti¤ens competition against
B. But a corollary is that pro…ts dwindle, which quickly drives the promotion of A to zero
(see also stylized facts in Section 2), thereby easing the pressure on B. These movements
create a permanent shift in the price/quantity demand locus for …rm B, with the promotion
5
According to …gures in Donohue et al. (2007), in year 2005, originator …rms spent on average 18% of
their revenues on promotion in its various forms: detailing, distribution of free samples, adverts in specialized
journals and, in the U.S., Direct-to-Consumer Advertising since 1997. The amounts spent on promotion are
slightly above R&D expenditure, indicating their strategic importance.
6
Detailing is the resources spent by the producer to inform doctors, hospitals, and/or patients about the
existence and the merits of their drugs.
2
e¤ect dominating the price competition e¤ect when A and B are close substitutes. This is
because, pre generic entry, …rm A promotes and prices its product more aggressively when it
is a closer substitute to B. In that case, generic entry a¤ects promotion more than it reduces
the price. This is when generics reveal their “Janus-like” nature.
Drawing on the theoretical results, we derive four testable implications that are taken
to the data. Our approach relies on the fact that there must be a causal link from LoE to
this shift in …rm behavior, since: (i) Losses of Exclusivity are evenly distributed over time,
(ii) there are no systematic technological shocks and/or new product launches concomitant
to LoE, (iii) there is no other competitive or legislative change that systematically happens
around the time of LoE. Hence, LoE is the only candidate event related to the systematic
changes in competitive interactions that we identify here.
In Section 2 we detail market evolutions around the time of loss of exclusivity, and show
that they require substantial shifts in the demand schedule to be rationalized. Empirically,
such changes in the demand schedule are a limiting factor to pursue empirical analyses that
encompass both the pre- and post-generic entry regimes. However, the high granularity of
our product-level dataset, which tracks the evolution of prices, promotion intensity, and sales
on a quarterly basis, allows us to overcome this constraint.
In Section 5, we estimate the parameters of the demand function, separately for hospitals
and pharmacies (the latter proves to be comparatively less price-sensitive). The promotion
and prices elasticities (own and cross) are large and precisely estimated. In addition, we
con…rm that generic entry does not a¤ect demand beyond the change in the price and promotion e¤ort for the drug experiencing LoE. Next, the di¤erence in the elasticities between
the pharmacy and hospital channels are consistent with the observed drop in the molecule’s
market share after patent expiration.
Turning to the supply side, the data con…rm our theoretical predictions that prices and
promotional e¤ort are strategic complements. Both are increasing in the molecule’s market
share. We also con…rm most of the theoretical predictions pertaining to …rms’ reaction to
generic entry. The competitors that remain under patent protection increase their market
shares when the initial level of promotion is high, and demand elasticity and product differentiation are low. Last, we note that, in most instances, the empirical results point to a
reduction in consumer welfare resulting from generic entry.
In our model, we assume that the entry of a bioequivalent generic leads to a Bertrand
outcome, driving the price of A down to the marginal cost of production. In the data, we
…nd that price elasticities for generics indeed increase up to 9, which is compatible with nearBertrand outcomes. However, we also observe some residual market segmentation within the
molecule. While a given generic’s cross-price elasticity is high for other generics producing
3
the same molecule, is it not signi…cantly di¤erent from zero for the originator drug.
Related literature.
The existing literature on competitive interactions in the pharma-
ceutical industry falls into two broad categories. The …rst analyses inter-brand competition
when drugs are still patent-protected (see, for instance, Brekke and Kuhn (2006), Dave and
Safer (2012) and de Frutos et al. (2013)). The second strand focuses on intramolecular competition following Loss of Exclusivity, i.e. when a generic bioequivalent drug can legally come
to market.7 It is in that context that the “generic entry paradox”, the price of the originator
branded drug going up following the launch of a bioequivalent generic version (c.f. footnote
9), has been unearthed (see Caves et al. (1991), Regan (2008) and Vandoros and Kanavos
(2013) among others). Our result that the demand for generics does not display any crossprice elasticity with respect to the originator drug is compatible with this paradox. To the
best of our knowledge, our paper is the …rst to encompass both the pre- and post-generic
entry situations.
While some scholars have analyzed the reaction of the originator producer to the entry of
generics (Scott Morton, 2000), there is little empirical evidence on the reaction of on-patent
branded producers experiencing generic entry into a competing drug (although Jena et al.
(2009) do provide suggestive evidence about prices and quantities for 5 events of generic
entry). One of our contributions is thus to combine the analysis of intra- and inter-molecule
competition, both theoretically and econometrically. This allows us to provide a unifying
framework for results that may appear contradictory at …rst sight. For instance, Stern (1996)
provides evidence of intense inter-molecular competition. By contrast, Ellison et al. (1997)
reports strong intra-molecular rivalry between the originator and the generic version of the
drug, and weak (or insigni…cant) inter-molecular competition. Our model and empirical
results reconcile these …ndings by endogenizing the joint promotion and price responses of
the di¤erent producers.
The remainder of the paper is organized as follows. Section 2 presents some surprising
facts that spur the paper’s central research question. Section 3 presents the model and derives
testable implications. Section 4 describes the data while Section 5 presents the empirical
results. Section 6 discusses some welfare considerations and concludes.
7
See the paper by Grabowski and Kyle (2007) for a description of generic entry in the US in the period
1995-2005 and Berndt and Dubois (2016) for an interesting comparison of generic penetration across OECD
countries for the period 2004-2010.
4
2
The main puzzle
2.1
Price, patent, and quantities
Once on the market, the life cycle of a patent-protected pharmaceutical drug can be broken
in two distinct stages. The …rst covers the period spanning market launch until the …rm
loses exclusivity, which typically stems from patent expiry. During that phase, the producing
…rm has exclusive rights over the production and distribution of the drug, and can thus
exercise market power. The second phase kicks o¤ after loss of exclusivity (LoE) when
generic equivalents can enter the market to compete with the incumbent …rm.
The introduction of a bioequivalent competitor produces a dramatic change in the nature
of competition. Generics are typically sold at half the price of their branded equivalent and
exert strong competitive pressure on the original branded product: Gabrowski et al. (2014)
show that, for branded drugs facing …rst generic entry in 2011-2012, brands retained on
average, only 16% of the molecule market after 1 year.
The dashed light-grey curve in Figure 1 below illustrates the dynamics of mean price,
quantity and volume market shares (left) and median values (right) for 123 molecules that
lost patent protection in the U.S. during the period 1994-2004. We can also account for the
evolution of each of these molecules’ prices, including the e¤ect of the generic competitors
that enter the market after the loss of exclusivity (see Section 4 for a detailed description
of our dataset). Time is expressed in quarters and we denote as “date 0” the quarter when
…rms lose exclusivity. We normalize values to 1 at quarter -8.8 Before patent expiration
(quarters -8 to 0), the price of the original molecule is slightly increasing. Then, within a
year of the Loss of Exclusivity (LoE), mean (respectively, median) prices drop by about 30%
(20%). Within three years, the drop reaches about 60% (40%).9
8
We control as follows for the fact that not all molecules are observed in all quarters (for instance, if
a patent expires 4 quarters before the end of the covered data, we have only four data points after patent
expiration): i) we compute price change over two consecutive periods for all available molecules; ii) we compute
the average of these variations for each quarter before and after patent expiration, and iii) …nally, we construct
an index that starts at 1 and varies over time with the average variations computed at (ii). We followed the
same approach to compute the median price and all the other statistics in this graph.
9
Although the average price of the molecule (displayed) falls following generic entry, this is not always the
case for the price of the branded drug (not separately depicted in Figure 1). Sometimes, the latter remains
constant or even increases; see among others Regan (2008) for the U.S. and Vandaros and Kanavos (2013) for
the EU. This price setting behaviour of branded producers is compatible with a number of (non-competing)
conjectures. The …rst is that some patients may be unaware of the availability of a cheaper alternative. The
second is that a subset of patients insists on purchasing the brand, even at a higher price. Third, in view
of the inertia in prescription behaviour, possibly coupled with a low frequency of visits to doctors and slow
information di¤usion, the branded producers prefer to keep extracting rents on a (shrinking) subset of patients
rather than vigorously compete on price. These conjectures would also be compatible with a situation where
the branded producer decides on a “midway path”, namely to set a price lower than the pre-generic entry
level, but does not match that of the generics.
5
PRICE, QUANTITY and MARKET SHARE DYNAMICS
.6
.4
.6
.8
.8
1
1
PRICE, QUANTITY and MARKET SHARE DYNAMICS
-8
-4
0
Mean Quantity
Mean Price
4
8
12
Mean MS
-8
-4
0
Median Quantity
Median Price
4
8
Median Ms
Figure 1: Evolution of average (left) and median (right) price, quantity and market share
around the time of LoE
While the price drop is to be expected, Figure 1 presents a ba- ing piece of evidence:
despite being sold at a fraction of the original price, the genericized molecule actually experiences a steep drop in volumes after LoE (black curves): average and median quantity drop
by 20% three years after patent expiry. In other words, after LoE, the combined volumes of
the branded and generic producers are substantially below the volume of the single branded
drug when it was sold at a price embodying monopoly power. The dashed dark-grey lines
show that these molecules experience an even more dramatic decrease in their volume market
shares, thus suggesting that generic competition decreases volumes despite a growing market.
To the best of our knowledge, we are the …rst to document this fact.10
Another perplexing feature surrounding generic entry is that the price of the other molecules –those still patent-protected –does not decrease (see in Section 5 and Jena et al. 2009).
And yet, in many instances, they experience a gain in market share. Put di¤erently, few new
patients are directed to the cheap genericized molecule and a number of existing patients
switch towards competing molecules at the time their initial treatment becomes cheaper (see
European Commission, 2009). Neither the rationales for the so-called generic entry paradox
(cf. footnote 9), nor co-payment by health insurances may explain why cross-price elasticities
suddenly seem to take the wrong sign.
10
Jena et al. (2009) …nds that “For …ve major classes of drugs, we …nd that large reductions in the price
of pioneer molecules after patent expiration — which would typically lead to decreased consumption of strong
substitutes— have no e¤ ect on the trend in demand for follow-on drugs.”Our factual …ndings, based on almost
100 molecules losing patent protection, are not incompatible with theirs. We actually …nd substantial heterogeneity across molecules: some see their demand increase, some are constant, but most of them see their
demand dwindle. We exploit this heterogeneity in our econometric analysis.
6
12
Taken together, these facts show that, like Janus, generics display two di¤erent faces:
while they are …erce competitors for the branded drug that lost exclusivity, they are toothless
challengers with respect to the remaining patent-protected drugs. So much so that they could
be mistaken for Gi¤en goods.
This Janus-like feature is an open puzzle. It led to conjectures that, in our view, do not
accurately represent competitive interactions in the pharmaceutical industry. For instance,
Jena et al. (2009) tracks price and quantity evolutions in 5 therapeutic classes. The authors
interpret the absence of price reaction by competitors and/or demand shifting towards generics as evidence of single molecule markets, without actual substitutability. The implication
would be that patented molecules – including “me-too drugs” – are “monopolies”. This
conjecture contradicts vast evidence pointing to …erce intermolecular competition while all
molecules are patent-protected. In the same line, the EU Commission has de…ned a single
molecule market for Perindopril (an ACE inhibitor controlling blood pressure) in the context of the Servier case (CASE AT.39612 - Perindopril (Servier), EU Commission Decision,
09/07/2014).
2.2
Loss of Exclusivity and Promotion Intensity
Competitive interactions in the pharmaceutical industry are richer than the changes in prices
and quantities documented above. LoE also triggers major shifts in the …rms’promotional
e¤orts. Using data from IMS-Health, we measure the …rms’ drug-speci…c spending on personalized visits to general practitioners and hospital specialists, free samples dispensed to
physicians, and advertising in professional journals. All these instruments a¤ect the doctors’ incentives to prescribe one rather than another molecule to their patients. In this
paper, we thus focus on the doctors’ role as intermediaries between patients and pharmaceutical companies (see Inderst and Ottaviani (2012) for a seminal model of competition
through commissions and kickbacks, and Iizuaka (2012) for evidence that doctors respond
to economic incentives in their prescription decisions). The association between payments to
doctors and their prescription behaviour can be assessed on the basis of publicly available
data (Grochowski, Jones and Ornstein, 2016)
Figure 2 below shows how total spending on these three channels evolves around the time
of LoE for the molecules that are actively promoted two years before their patent expires.
Our data reveal that promotion falls continuously over the 8-quarter period before patent
expiration, with a sharp drop in the median spending at the time of LoE. At time 0, promotion
e¤ort has dropped by 50%. Within 3 years, the median spending is zero. The average lies
above the median because some molecules keep being promoted.11
11
For instance, high level of promotions are observed for Prozac (‡uoxetine) because Eli Lilly & Co. in-
7
0
.5
1
PRICE and PROMOTION D Y N AMIC S
-8
-4
0
4
Mean Price
Mean Promotion
8
12
Median Price
Median Promotion
Figure 2: Evolution of price and promotion around the time of LoE
While we believe we are the …rst to propose systematic evidence of this behavior, note that
Caves, Whinston and Hurwitz (1991) already observed that “generic entry brings with it two
o¤ setting e¤ ects: …rst, generic entrants o¤ ers signi…cantly lower prices, which tend to expand
overall sales of the drug, but their arrival also leads to a signi…cant reduction in the level of
advertising for the drug, which acts to counterbalance this price e¤ ect”. Building on this
intuition and on the stylized facts presented so far, the next section proposes a simple model
to analyze how …rms compete through price and non-price instruments, when consumption
is mediated through an intermediary (see also Inderst and Ottaviani 2012).
3
The model
In this section, we present a simple theoretical model in which …rms compete in both price
and promotional e¤ort. This allows us to understand how generic entry a¤ects both inter
and intra molecular competition. Our results identify the conditions under which the stylised
facts decribed above obtain as an equilibrium outcome. In turn, this allows us to directly
derive testable implications that will guide our econometric analysis.
Consider a market in which two molecules, developed by two pharmaceutical …rms J =
A; B compete for prescriptions by doctors for their patients. We want to capture the …rms’
detailing and pricing strategy given the quality
J
of their molecule, the degree of treatment
troduced weekly delayed release capsules of the drug just before LoE in an attempt to stem the post-patent
decrease in sales of their daily dosage. Similarly, we observe positive spending for Zantac (ranitidine) and
Tagamet (cimetidine) probably because some of their lower-dosage tablets are available over-the-counter.
8
response heterogeneity across patients, e,12 as well as the slope
of the patients’ demand
(which includes the in‡uence of co-payment by health insurance). Consider the case of a
patient who goes to her doctor with a condition that requires treatment. The patient/doctor
pair (PDP) i’s intrinsic utility from using treatment A or B is respectively:
UAi =
A
pA + "i ;
(1)
UBi =
B
pB
"i ;
(2)
where pJ is the price of molecule J 2 fA; Bg and "i
U
e e
2; 2
is the relative e¢ ciency of
drug A as opposed to B to treat patient i’s condition, and U denotes the uniform distribution.
A larger value of e implies that patients are more heterogeneous in their treatment responses
to molecules A and B, and hence that the two are more distant substitutes. Larger copayment by the patient’s health insurance implies that the PDP is less sensitive to prices:
is lower. We introduce detailing below –bear with us for a moment.
For the sake of simplicity, we focus on the case in which all patients who require a
treatment will buy one. This requires that the quality
J
of both molecules is su¢ ciently
high relative to equilibrium prices (see the exact condition below). Therefore, patients with
a su¢ ciently good …t with drug A (i.e. with "i su¢ ciently large) buy treatment A, and all
the others buy drug B. For lower drug qualities, prices would also a¤ect market size, which
complicates the algebra without bringing valuable new insights.13 Letting
denote the size
of the market (or the number of a- icted patients), we identify the patient who is indi¤erent
between A and B to determine quantities in the absence of detailing, and …nd that:
QA =
1
where F represents the CDF of "i ,
2
pB
B
QB = F
pB
B
F
;
2
B
B
A
and
pB
pB
pA . We associate
drug A with the oldest molecule, and drug B with the more recent one. Hence, …rm A loses
exclusivity before B does. We also let
0, which means that drug A is less e¤ective
B
than B. Finally, we normalize production costs to 0. Thus, when they cannot promote their
12
These parameters directly relate to vertical and horizontal product di¤erentiation in classical Industrial
Organization analyses. However, product di¤erentiation is not a choice variable in our model.
13
The main issue here is that we want to keep this illustrative model simple, since it is mainly a guide for
the empirical sections below. Generalizing it to include the possible switches and corner solutions in which all
patients buy treatment, or some don’t, because of the price of A or that of B, or both makes the model a lot
less tractable and muddies the message, without adding any interesting insight.
9
drugs (superscript N D for the “no detailing” case), …rms’respective pro…ts are:
ND
A
= pA
ND
B
= pB
1
2
1
+
2
pB
B
2e
pB
B
2e
;
(3)
;
(4)
where the terms between brackets are respectively the equilibrium markets shares of A and
B when we substitute F for its value with a uniform distribution.14
Generic Entry. Prior to A’s loss of exclusivity (LoE), patent protection ensures that each
…rm is the sole producer of its molecule. In the absence of detailing, an equilibrium would
be characterized by a pair of prices in which …rms maximize their pro…ts in (3
4). Loss of
exclusivity (typically associated with patent expiry) implies that generic bioequivalents can
compete directly for A consumers. Based on the evidence (see a.o. Gabrowski et al., 2014) we
let competition among generic producers reduce the price of A down to marginal costs, which
we normalized to zero without loss of generality – we empirically test that this assumption
is valid in Section 5.5. Instead, …rm B retains its patent protection and monopoly power.
In the absence of detailing, a post-generic-entry equilibrium would thus be characterized
by the price that maximizes B’s pro…ts when pA = 0. Unsurprisingly, generics competition
in the A market drives down the price of A, which can only reduce the price and the market
share of drug B. In Appendix 1, we show that, in an interior equilibrium, price sensitivity
determines the magnitude of the price reduction, but does not in‡uence market shares.
Detailing. As documented in the introduction, the pharmaceutical industry stands out for
its high promotional intensity. One of their main instruments is detailing: pharmaceutical
companies devote substantial resources to inform doctors and provide them with a number
of perquisites sometimes contingent on their prescription behavior. Often, they invest in
scienti…c studies to “objectify” the merits of their drugs. As we documented, this non-price
competition component is also dramatically modi…ed upon generic entry: price competition
amongst generic producers typically brings detailing down to 0.
Here, we focus on the case in which detailing operates like persuasive advertisement: it
stimulates the demand of the patient-doctor pair without a¤ecting patient intrinsic utility
(1
2). Hurwitz and Caves (1988) and Rizzo (1999) provide provide extensive evidence
14
We thus have:
8
< 0; 8" < e=2
"+e=2
F (") =
= 12 + e" ; 8" 2 [ e=2; e=2]
e
:
1; 8" > e=2:
and the PDF is f (") = 1=e; 8" 2 [ e=2; e=2] :
10
(5)
pertaining to the persuasive nature of pharmaceutical detailing (see Bagwell (2007) for a
review of the advertising litterature). Formally, when …rm a2J =2 on detailing (or “promotion”:
we use the two terms interchangeably), it modi…es the patient-doctor pair’s perceived quality
of molecule J up to:
0
J
=
+ aJ :
J
This produces an autonomous increase in the demand for drug J. Given an action pro…le
faA ; aB ; pA ; pB g ; the resulting demands are then:
QD
A =
1
QD
B = F
B
F
B
+
3.1
= pA
D
B
= pB
1
2
1
+
2
B
aB
2
aB
2
where superscript D denotes Detailing and
D
A
+
+
pB
pB
aB
aB
aB
;
;
aA . The …rms’pro…ts become:
(pB
pA )
(pB
pA )
2e
B
+
aB
2e
a2A
;
2
a2B
:
2
Equilibrium analysis
Before generic entry, each PDP has a choice between two branded molecules. Each …rm
chooses its price and promotion intensity to maximize its pro…ts. Taking …rst-order conditions
yields the implicit solutions:
pD
=
J
2e QJ
aD
=
J
1
;
QJ ;
(6)
(7)
which in turn imply:
Testable implication 1 Both price pJ and promotion intensity aJ are increasing in total sales QJ and decreasing in the PDP’s price sensitivity .15 Moreover, prices (but not
promotion) are increasing with the degree of “horizontal di¤ erentiation” e.
15
This result is in line with e.g. de Frutos et al. (2013) or Brekke and Khun (2006) who write that detailing,
DTCA and price (if not regulated) are complementary strategies for the …rms. Thus, allowing DTCA induces
more detailing and higher prices. Similarly, Grossman and Shapiro (1984) …nd that more competitive markets
reduce both prices and advertisement in a model in which the purpose of advertising is to inform consumers
of the existence of the product.
11
In Appendix 2, we solve for the …rms’ equilibrium market shares. Among other results,
we …nd that when the two molecules are closer substitutes (smaller e), the market share of
the superior drug B decreases towards 50%, in part because it invests less in promotion,
whereas A invests more. Conversely, a larger market size ( ) magni…es the gap between B
and A: potential pro…ts being higher, B invests in promotion more aggressively and increases
its market share.
3.2
The e¤ects of generic entry
After LoE, generic entry drives the price of molecule A to pA = 0, and the dissipation of
pro…ts produces a drop in detailing: aA = 0. As a consequence, …rm B’s post-entry pro…t
function becomes:
G
B
= pB
1
+
2
B
+ aB
2e
a2B
:
2
pB
Where superscript G stands for generics: …rm B, which still bene…ts from market power,
faces sti¤er price competition but looser non-price competition from molecule A. We …nd
that:
Proposition 1 For qA > 0 both before and after generic entry, the loss of exclusivity on
molecule A allows B to increase its market share, promotional e¤ ort and price i¤ 2 e < :
D
Proof. From Propositions 2 and 3 in Appendix 2, after some tedious algebra one …nds that QG
B QB ,
G
aD
B , and pB
aG
B
negative whenever
pD
B are all a multiple of (2 e
QD
A
)( (
B
3e) + ), where the second factor is
is positive (see proof of 2).
It is striking that it is precisely when A and B are closer substitutes (e small enough)
that sti¤er price competition by the generic versions of A allows B to increase its market
share. Conversely, only if the two molecules are su¢ ciently distant substitutes or if market
size
is small, price competition will have the a priori intended e¤ect of boosting the sales
of molecule A.
The rationale for this result stems directly from …rm A’s behavior before LoE. When
market size is large and/or the two …rms sell close substitutes, competition becomes intense:
A invests a lot in advertising, and cuts down its price. As a consequence, generic entry will
substantially loosen non-price competition (aA dropping from a high level to 0) and have
a comparatively smaller e¤ect on price competition (pA dropping from a low level to 0).
Conversely, when market size is small and A and B are distant substitutes, competition is
lax pre-generic entry (i.e. prices are higher and promotion lower). In that case, generic entry
produces a comparatively stronger price drop, which dents the demand for B, and forces the
latter to react by adopting a more aggressive pricing strategy:
12
Testable implication 2 Generic entry should produce lower gains in market share for B
in markets where price sensitivity
is higher.
Testable implication 3 Generic entry should produce lower gains in market share for B
in markets where horizontal di¤ erentiation e is higher.
Testable implication 4 Generic entry should produce lower gains in market share for B
in smaller markets.
In Appendix 3, we extend the model to study welfare. Our …rst main result relates to
allocative e¢ ciency: we …nd that generic entry increases the market share of A when it is
already higher than in the …rst best before entry, and decreases its market share when it is
initially too low. In other words, it always distorts the market allocation farther away from
the …rst best. Banning promotion overall would not solve this issue: the results are identical
in the absence of promotion. The reason is that the highest quality drug tends to be sold at a
higher markup than the other one, and generic entry reduces the price of only one molecule,
generating ine¢ cient, lopsided, competition.
The second result aims at evaluating when generic entry increases or decreases consumer
(patient) surplus. We calculate it using the utility functions (1) and (2), evaluated at the
equilibrium levels of price and promotion. The welfare e¤ects of generic entry vary on a
case-by-case basis: even though allocative e¢ ciency worsens, patients bene…t from the lower
prices of A and, in some cases, of B. We identify that a drop in QB is a su¢ cient condition
for patients’utility to increase. Conversely, for
B
= 0, welfare only decreases if the market
share of B increases strongly, together with a signi…cant increase in its price and promotion
intensity. We return to these points in the conclusions, once we have assessed the actual
market responses in the data.
4
The Data
This section …rst describes our dataset and then additional stylized facts about the changes
in competitive conditions that occur upon generic entry. The empirical results are presented
in Section 5.
We have data on quarterly dollar revenues and physical quantities for hundreds of branded
and generic drugs sold in the U.S. in several therapeutic areas over the 40-quarter period
1994q1 to 2003q4. These have been obtained from the proprietary database IMS-Health, one
of the most important medical-information companies (IMS-MIDAS). We compute de‡ated
13
revenues (R) by dividing nominal value of sales by the producer price index for the pharmaceutical industry published by the Bureau of Labor Statistics. Quantities (Q) are measured
as mg of active ingredient.
For each drug, we observe R and Q for di¤erent packages, where a package refers to
di¤erent strengths (e.g. 20mg or 40mg of active ingredient) or di¤erent forms of administration
(e.g. tablets or vials). We can thus compute the average price of the molecule (P ) by dividing
R, i.e. the revenues for all the di¤erent packages, by total Q.16 An important feature of our
data is that we observe R and Q for two di¤erent distribution channels: hospital (HO) and
pharmacies (PH).
Variables
MarketShares
Channel
Hosp
Phar
Mean
0.117
0.134
SD
0.173
0.183
Min
0.01
0.01
Max
1
1
NumberofCompetitors
Hosp
13.2
8.17
0
46
(#ofotherMolecules)
Phar
12.1
6.62
0
41
OwnPrice
Hosp
16.73
70.31
0.016
902.8
Phar
16.78
71.23
0.05
910.1
Hosp
3268.77
8100.9
0
59469
Phar
11165.38
24591.7
0
198027
Competitors'Price
Hosp
8.86
28.32
0.02
197.29
(averagepriceinATC3)
Phar
4.15
15.19
0.02
122.13
Competitors'Promotion
Hosp
16314.7
37402.7
0
231144
(totalpromotioninATC3)
Phar
55871.2
124653.7
0
891515
OwnPromotion
Table 1. Summary Statistics
For a subsample of therapeutic clases, we also observe product-level expenditure for promoting drugs to doctors over the entire period. These promotional data, also obtained from
IMS-Health, include three main components: visits to o¢ ce-based practitioners and hospital specialists (this component is generally known as “detailing”), free samples dispensed to
physicians (their cost being estimated at the sales price of the drug), and advertising in professional journals. IMS Health data on detailing are computed using a representative panel
of physicians who track their contacts with sales representatives. The amount spent on free
samples is based on a panel of approximately 1200 o¢ ce sta¤ members in medical practices
16
The price we compute is the price per mg of active ingredient. We downloaded information on de…ned
daily dosage (DDD) from the website of the World Health Organization but the matching with molecules
in our dateset resulted in a substantial reduction of the sample size. Note that our empirical speci…cations
control for unobserved di¤erences, such as quality and DDD, across molecules.
14
while advertising expenditure in professional journals is computed by tracking advertisements placed in approximately 400 medical journals and then adding the publisher’s charge
for those advertisements. We will use the terms promotion and advertising as synonymous,
encompassing all forms of promotional e¤ort listed above. The empirical analysis assumes
that promotion to o¢ ce-based practitioners a¤ects sales in pharmacies while promotion to
hospital physicians a¤ects the use of drugs in hospitals. Table 1 reports descriptive statistics
of market shares, prices and promotion for hospitals and pharmacies, separately.
MarketDescriptiveStatistics
MoleculeDescriptiveStatistics
NmbofATC3
53
NmbofATC4
75
NmbofMolecules
NmbMolecules
losingPatent
NmbofATC4in
ATC3
1
2
3
4
5
6
Distribution
ObsinSample
56.8%
20.2%
13.3%
4.3%
3.7%
1.6%
NmbofGeneric
Entry
1
2
3
4
5
6
227
95
Nmbof
Markets
22
8
8
3
3
1
Table 2. Therapeutical Markets, Molecules and Generic Entry
For the main econometric analysis (whose focus is on pre-LoE competition), the sample
we use includes 227 molecules initially covered by patent protection, 95 of which lose patent
protection between 1994 and 2003.17 All the drugs are classi…ed according to the Anatomical
Therapeutic Chemical (ATC) classi…cation system. The “ATC3 level” corresponds to a
market: it groups the drugs that target a given condition. As shown in Table 2 below, drugs
in our sample are divided in 53 ATC3 markets, some of which are further subdivided in
ATC4 sub-classes. 46 of the ATC3 markets see the entry of at least one new generic molecule
during the observed time window, with one market experiencing LoE for six of its branded
drugs. Compared to previous studies, our sample includes a much larger number of drug
classes, including lipid-regulators, anti-depressants, anti-ulcerants and hypertension drugs.18
The complete list of therapeutic markets can be found in Appendix 5.
For illustrative purposes, Table 3 below reports the list of (plain) lipid regulators for the
main two ATC4 subclasses of anti-cholesterol drugs: statins and bill acid sequestrants.19 The
17
We also report the result of a complementary exercise (whose focus is on post-LoE competition) with a
larger sample made up of all drugs that have experienced LoE – see Section 5.5.
18
For instance, Regan (2008) investigates the e¤ect of generic entry on prices using a sample of eighteen
drugs, divided in 6 therapeutical markets. Half of the drugs in her sample are from the cardiovascular market
while …ve others treat diseases/conditions a¤ecting the central nervous system.
19
Statins (C10A1) lower cholesterol levels by inhibiting the enzyme HMG-CoA reductase, which plays
15
former has 6 di¤erent competing molecules; the latter has 4. In total, there are thus up to
10 di¤erent prescription possibilities for a doctor who has a patient with excess cholesterol.
The last column identi…es either the quarter when the drug loses its patent protection (4th
quarter of 2001 in the case of Mevacor, for instance) or the date when the company decided to withdraw the molecule from the market (lethal secondary e¤ects triggered the early
withdrawal of Lipobay in 2001).
ATC3
class
C10A
ATC4
class
C10A1
ATC4description
Statins
Molecule
Name
Atorvastatin
Lipidor
Cerivastatin
Lipobay
Fluvastatin
Lescol
Lovastatin
Mevacor
Pravastatin
Pravachol
Simvastatin
C10A
C10A3
BileAcidSequestrant Colestipol
Patent
Expiration
BrandName
withdrawn2001
2001q4
Zocor
Colestid
Cholestyramine Questran
Aspartame+cole
Prevalide
styramine
Colesevelam
Welchol
1996q3
Table 3. Classi…cation of Anti-cholesterol Drugs
To get an initial grasp on the di¤erences between hospitals and pharmacies, we explore
how quantities tend to vary between these two distribution channels, and across therapeutic
areas with di¤erent levels of promotion. Figure 3 shows the evolution of volume market
shares of molecules losing exclusivity for pharmacies (light-grey lines) and hospitals (black
lines).20 Two di¤erent statistics are presented: i) the simple average (solid line) and ii) the
average where other branded drugs in the same ATC3 have been weighted by their share in
promotional e¤ort in that particular ATC3 (dashed line). This is done to illustrate that the
shift from generics to branded competitors is more prominent in those ATC markets where
advertising (of branded competitors) is higher. Figure 3 illustrates that the quantity drop
observed after LoE originates mostly in pharmacies, and is less pronounced in hospitals. At
the same time, the di¤erence between the solid and dash line provides compelling evidence
a central role in the production of cholesterol in the liver. Bile Acid Sequestrants (C10A3) increase the
elimination of bile acids which the liver can replace only by converting cholesterol, thus reducing its level in
the blood. Note that there are other ATC4 groups in the therapeutic market C10A, for instance Fibrates
(C10A2). Although we observe quantities and prices for Fibrates, we have promotional data only for the most
important ATC4 markets. In fact, quarterly sales of Fibrates in year 2000 were around ten millions dollars
compared to quarterly sales of more than a billion dollars for Statins.
20
The volume ATC3 market share is de…ned as (Quantity of Molecule/Total Quantity at the corresponding
ATC3 level). Note that one mg of one molecule is not necessarily equivalent to one mg of another molecule in
the same ATC3. Our empirical speci…cations will control for di¤erences in de…ned daily dose (DDD) of drugs.
16
that promotion plays a crucial role in shifting market shares from drugs with LoE to drugs
still covered by patent protection. We will exploit this inter-molecule and inter-channel
heterogeneity in the econometric analysis.
FIGURE 3:
evolution of market shares
Figure 4 highlights another puzzling feature of the pharmaceutical industry: as compared
to price, promotional e¤ort varies signi…cantly over time. Contrary to other industries, pricing
appears to be a medium-to-long-run decision variable, while advertising seems to be the key
short-term strategic variable. As illustrated in Figure 4, there is a tenfold di¤erence in the
standard deviations of the respective variables.
Descriptive statistics of the variables used in our main econometric speci…cation are reported in Table 1 above. Note that, for each molecule, the promotion level at time t is
computed with the perpetual inventory method, commonly used for physical capital:
ait = (1
) ait
1
+ Iit ;
where Iit is the quarterly expenditure in promotion retrieved from IMS and
depreciation rate, assumed to be 0.1, i:e: about 40% per
year.21
is the quarterly
Competitors’ price refers
to the average price of all the other molecules in the market, including generics and it is
computed as the ratio between total revenues and total quantities in the ATC3 market, after
21
All the results reported in Section 5 are robust to setting
17
= 0:25.
subtructing revenues and quantities of drug i. Competitors’ promotion refers to the sum
of the promotion of all other drugs in the ATC3 market, each computed according to the
equation above.
FIGURE 4:
5
variation in price and promotion
Empirical Results
In this section, we …rst estimate the demand equation prior to LoE to assess whether PDPs’
demand behavior matches the main hypotheses underlying the model (section 5.1). We …nd
that the elasticity of demand with respect to both own and competitors’promotion and prices
are large and signi…cant. Importantly, generic entry into a competitor’s molecule does not
appear to a¤ect the parameters of the demand function. In addition, the estimated elasticities
of demand with respect to promotional e¤ort and price before LoE are consistent with the
average observed drop in the molecule’s market share after LoE (section 5.2). In particular,
the empirical results correctly predict that, despite their substantially lower price, molecules
losing patent protection experience a decrease in quantity and (volume) market shares of the
molecule –i.e. the original brand, plus all generics taken together.
In section 5.3, we turn to the producers’equilibrium behaviour pre LoE. Testable implication #1 provides guidance on how to identify the …rms’pricing and advertising behavior
pre-LoE. In section 5.4 we assess how on-patent drugs’market shares react when a competitor faces generic entry. Testable implications #2,3,4 state that it will depend on the price
18
sensitivity of demand ( ), horizontal di¤erentiation (e), and market size ( ). Proposition 1
extends the predictions to prices and promotion.
Last, section 5.5 focuses on competitive interactions post LoE, i.e. when a molecule has
become generic. For each substance, our dataset distinguishes between the originator product
that used to bene…t from exclusivity, and generic producers. In the model, we assume that
generic entry leads to a Bertrand outcome, driving the price of A down to the marginal cost
of production. In the data, we …nd that price elasticities for generics increase up to 9, which
is compatible with near-Bertrand outcomes. We also observe some residual market segmentation within the molecule: while cross-price elasticities are high among generic producers of
the same molecule, it is not signi…cantly di¤erent from zero with respect to the originator
drug. We also …nd the intense pre-LoE inter-molecular rivalry becomes subdued post LoE.
The results reported below have been obtained with quantity market shares (relative
quantity in the ATC3 market) as the dependent variable. The results with (absolute) quantity
as the dependent variable are similar and available upon request. Our preference for relative
quantity is that most of these markets are growing in both value and volume.
5.1
Patients-Doctors’demand schedule
We start our empirical analysis by evaluating the parameters of the demand side of the
market to assess the relative importance of promotion. Exploiting the data, we conduct
the exercise separately for hospital and pharmacies: we shall see in the following section
that the measured di¤erences in prescription behavior can explain a signi…cant part of the
heterogeneity in market share responses upon generic entry.
We estimate the demand for each active ingredient while the innovator still enjoys exclusivity. Focusing on the pre-LoE period means that all …rms in this subsample only face
competition from imperfect substitutes. That is, we track how a …rm B’s quantity market
shares, price, and promotion e¤ort, are in‡uenced by the same variables for other drugs A;
some of which no longer bene…t from exclusivity. But when molecule B loses exclusivity, we
remove it from the left-hand side in (8). We turn to post-LoE competition from generics in
Sections 5.4 and 5.5.
De…ning a market j as a given ATC3 group, equation (8) describes the PDPs’ demand
for a particular molecule i at time t. It is estimated in …rst-di¤erences in order to remove all
19
time-invariant, drug-speci…c …xed e¤ects, such as quality di¤rences:22
yi;j;t =
0
+
::: +
where
yi;j;t ,
pi;j;t and
1
pi;j;t +
5 GEN i;j;t
2
p
i;j;t
+
3
ai;j;t +
4
a
i;j;t
(8)
+ T Ei;j;t + "i;j;t
ai;j;t refer to the quarterly evolution of, respectively, the quantity
market shares, price, and promotion e¤ort of molecule i. Similarly,
p
i;j;t
and
a
i;j;t
are the competing molecules’evolution of prices and promotion in the same ATC3 market.
All these variables are in logarithms, implying that the coe¢ cients can be interpreted as
elasticities. GEN
i;j;t
is the number of these competing molecules who lost exclusivity. We
do not expect it to be signi…cant: according to our hypotheses, generic entry a¤ects demand
through its e¤ect on prices and promotional e¤ort. It should not produce additional shifts
in demand. Finally, the variable T Ei;j;t (Time to Expiration) counts the number of quarters
until patent expiration to account for the impact of the drug’s life cycle on demand.23
The regressors are likely a¤ected by two di¤erent problems. The …rst one is that feedbacks
from shocks on market shares to future prices and advertising may produce endogeneity
issues (reverse causality). The second problem is measurement errors in the computation of
promotion e¤ort, which are likely to create an attenuation bias.24 To address these, we follow
previous empirical work by instrumenting prices and advertising with: the number of packages
(linear and squared),25 the average price of drugs sold by …rm i in quarter t in other ATC3
markets (“Hausman” instruments), and a dummy indicating whether a branded drug has
experienced the entry of a generic competitor before LoE following a successful “Paragraph
IV” challenge.26 This choice of instruments is validated by the Kleibergen-Paap rk-statistic
22
Results using FE are qualitatively similar but we could not identify a set of instruments that simultaneously pass the relevant tests for under-identi…cation/weak-identi…cation and the Hansen J test for the
exogeneity (orthogonality) of the instruments.
23
We use negative values so that the variable is increasing as we approach patent expiration. For instance,
-10 and -1 refer, respectively, to ten quarters and one quarter before LoE.
24
In the words of Griliches and Hausman (1986), “Errors of measurement will usually bias the …rst di¤erence
estimators downward (toward zero) by more than they will bias the within estimators.”This problem has been
largely discussed in the empirical literature on the estimation of production function (see Griliches and Mairesse
(1995) among others). It is not by coincidence that the estimation of the demand elasticity to promotion is
a¤ected by the same problem of the elasticity of production with respect to capital. In fact, in both cases,
econometricians need to construct a measure of “e¤ective stock” looking at account data on past and present
expenditure in physical capita or promotional e¤ort.
25
The number of packages has been used by Chaudhuri, Goldberg and Jia (2006) and Branstetter, Chatterjee and Higgins (2014) among others. The variable is related to the average price p as variations in p stem in
part from variations in the set of packages available in each period. First stage results show that this variable
is also highly correlated with advertising. This is because the introduction of a new posology or delivery mode
is often accompanied by resurgence in the promotional e¤ort. For instance, Ely Lilly started a huge marketing
campaign to promote Prozac Weekly, a once-weekly formulation of Prozac. Other examples include Paxil CR,
a controlled release version of Paxil.
26
Paragraph IV of the Hatch-Waxman Act allows generic manufacturers to attempt to enter the market
before patent expiration of the original branded drug, either by claiming non-infringement or invalidity of the
20
(K-P) for under-identi…cation, the Angrist-Pischke (A-P) F -test for weak instruments, the
Hansen J-test for the orthogonality conditions and also the C statistic to test the exogeneity
(endogeneity) of one or more instruments (regressors).27 .
We estimated the empirical model in (8) separately for hospitals and pharmacies in order to unearth possible channel-speci…c idiosyncrasies. The dependent variable pertains to
branded drugs until one quarter before they lose exclusivity (e.g. if a drug loses patent protection in 2000q3, we use the relevant data for the period 1994q1 to 2000q2). While the
dependent variable only pertains to drugs while they are on patent, total quantity (market
size) and competitors’indices (for price and promotion) also include generic drugs. Columns
(1a,b) of Table 4 report the estimates have been obtained without instrumenting for prices
and advertisement. Column (2a,b) report the results when we instrument for these variables.
Note that the C-statistic indicates that own promotion is endogenous, thus con…rming its
importance as a competition tool.
branded product’s patent. Successful Paragraph-IV challenge represents an exogenous shift in the promotional
e¤ort of branded drugs uncorrelated with demand shocks or measurement errors. Branstetter, Chatterjee and
Higgins (2011) investigate the welfare e¤ect of accelerated generic entry via Para-IV challenge challenges for
hypertensive drugs.
27
These tests lead us to use slightly di¤erent instruments for the two channels. Concretely, the average …rm
prices in other markets can be used only for hospitals, while in pharmacies we use the price of the molecule
in hospitals.
21
Specification:
FD
Coeff.
Hospital
FD-IV
FD
a
(1a)
(1b)
a
a
(2a)
OwnPrice
α1
-1.281*** -1.378*** -2.278*
(0.18)
(0.18)
(1.37)
Price_Comp
α2
0.841***
(0.15)
Price_Comp_Brand
α2.1
Price_Comp_Generic
α2.2
OwnPromotion
Pharmacies
FD
FD-IV
FD-IV
FD
a
(1a)
(2b)
-2.631**
(1.28)
0.932***
(0.27)
a
(1b)
a
-1.004*** -1.029*** -1.729***
(0.22)
(0.22)
(0.65)
0.249**
(0.13)
FD-IV
a
(2a)
(2b)
a
-1.787***
(0.67)
0.306*
(0.17)
0.751***
(0.14)
0.022
(0.02)
0.943***
(0.28)
0.002
(0.07)
α3
0.085*** 0.084*** 2.021***
(0.02)
(0.02)
(0.51)
2.080***
(0.51)
0.096*** 0.096*** 1.978***
(0.04)
(0.04)
(0.71)
1.996***
(0.72)
Promotion_Comp
α4
-0.061*** -0.061*** -1.442***
(0.02)
(0.02)
(0.38)
-1.485***
(0.38)
-0.067** -0.068** -1.511***
(0.03)
(0.03)
(0.55)
-1.527***
(0.55)
GenericEntry
α5
0.005***
(0.00)
-0.004***
(0.00)
5046
0.002
(0.00)
-0.003***
(0.00)
5046
.0052
.0015
.0269
.670(3)
<.0001
.6704
0.003*
(0.00)
-0.003***
(0.00)
5032
-0.002
(0.00)
-0.001*
(0.00)
5032
.0116
.0103
.0084
.526(3)
.0007
.5261
TimeTrend
Observations
b
K-PUnderidentification
c
AP_F-test-Promotion
c
AP_F-test-Price
d
HansenJtest(df)
e
C- test-Endogeneity
e
C test-Exogeneity
0.005***
(0.00)
-0.004***
(0.00)
5046
0.002
(0.00)
-0.003***
(0.00)
5046
.0058
.0017
.0386
.604(3)
<.0001
.4452
0.243
(0.26)
0.054
(0.04)
0.002*
(0.00)
-0.003***
(0.00)
5032
0.335
(0.35)
0.067
(0.09)
-0.002
(0.00)
-0.001*
(0.00)
5032
.0117
.0104
.0085
.522(3)
.0007
.3486
Robust standard errors clustered at market level in parentheses. * significant at 10% level; ** significant at 5%; *** significant at 1%. a All Specifications in First Differences. Endogenous
b
c
variables: Own Price and Promotion. P-value for the Kleibergen-Paap rk statistics testing the null hypothesis that the model is underidentified. P-value for the Angrist-Pischke F test for
d
excluded instruments testing for weak instruments in the first stage regressions of promotion and price. Hansen J test of overidentifying restrictions with degrees of freedom reported in
parentheses.
e
P-value of C (GMM distance) test of Endogeneity for own price and promotion and of Exogeneity for price and promotion of competitors.
Table 4. Regression results: molecules’volume market shares, demand equation.
A number of interesting …ndings emerge from the results in Columns 2a –the results pertaining to Columns 2b are discussed below. First, our estimate of the own-price elasticity of
demand is higher than those reported in most of the literature. For instance, Dave and Sa¤er
(2012) report estimates below unity; a …nding di¢ cult to reconcile with pro…t maximization
in the absence of a binding price cap. By contrast, our point estimates in Columns (2a) are
compatible with price-cost margins in the 40-50% range. This order of magnitude is in line
with the price reductions observed following generic entry.
Second, the elasticity of demand w.r.t. prices seems higher in hospitals than in pharmacies.28 This probably re‡ects the following di¤erence: private practice doctors do not directly
bene…t from the patient’s buying a cheaper alternative (possibly, they even lose perks o¤ered
by some pharmaceutical companies), while patients only pay a fraction of drug’s price, thanks
28
A Wald test of the null hypothesis that the coe¢ cients of own price and competitors’price in speci…cation
(1) are the same for the two channels is rejected at 1% signi…cance level.
22
to co-payment by the health insurance. In contrast, hospitals are residual claimants: their
margin depends one-for-one on procuring drugs at a discount, since they then charge the
patient a pre-determined reference price. There are thus two opposite biases that magnify
the di¤erences between hospital and pharmacies. How they a¤ect welfare, remains of course
a di¤erent question.
Third, the coe¢ cient pertaining to price of competitors (P rice_Comp) is of the right sign
and signi…cant for both channels. However, it is about three times as large in the hospital
channel, and more precisely estimated. This reinforces the message that hospitals are more
price sensitive than private practice doctors.
Fourth, in columns (2b), we decompose the rivals’ price index into a sub-index for
branded competitors (P rice_Comp_Brand) and another one for generic competitors
(P rice_Generic_Comp).29
That is, we test how the demand for an on-patent drug is
a¤ected by the price of other branded molecules (whether they are still on-patent or not)
and that of generics producers present in the same ATC3 market. We observe that the coe¢ cients on P rice_Comp_Brand are close to the coe¢ cients for P rice_Comp in columns
(2a), whereas the coe¢ cients on P rice_Comp_Generic are much smaller and nowhere close
to statistical signi…cance. This is direct evidence that, even though generics are …erce competitors for their branded bioequivalent, they are toothless challengers with respect to the
remaining patent-protected drugs.
Fifth, the coe¢ cients for own- and cross-promotion elasticity are of the right sign, large,
and precisely estimated in both channels. When instrumented, the point estimates increase
substantially, con…rming the extent to which this variable is a¤ected by endogeneity problems.
The point estimates barely change when we split the price index (2a vs. 2b). These results
strongly indicate that short term strategic interactions in the pharmaceutical industry cannot
be understood without properly accounting for promotional e¤ort.
Sixth, the coe¢ cient on the headcount of genericized molecules is statistically insigni…cant
once we properly control for endogeneity (and economically minute in columns (1a) and (1b)).
In other words, the regression con…rms our initial hypothesis that generic entry a¤ects demand
exclusively through the price of, and promotion for, a given molecule. Generic entry does
not modify the e¤ective number of products on the market or the level of the demand for the
other molecules.
To the best of our knowledge, these are the …rst results that identify the relative importance of price and non-price instruments in determining market shares separately for hospitals
and pharmacies for a large set of prescription drugs sold in the US. While the data samples are
29
Doing the same for promotional e¤ort would not make much sense, since generic drugs are not actively
promoted through detailing.
23
quite di¤erent, Anderson et al. (2015) also identify the importance of comparative advertising
for analgesics sold over the counter (OTC).
5.2
Out-of-sample prediction
To test our model’s predictive capacity, we carried out some "back-of-the enveloppe" calculations to gauge how well our pre-LoE point estimates are able to predict post-LoE (i.e. out
of sample) developments. The exercise runs as follows: …rst, we use the evidence in Section
2, which indicates that losing exclusivity produces an average price drop of about 60% , and
that the drop in promotion intensity is about 95%. Focusing on hospitals …rst, the results in
Table 4 indicate that the own-price elasticity is
2:631, and the own-advertising elasticity is
2:080. Piecing these elements together, the change in market share is predicted to be:30
2:631 ( 0:6) + 2:080 ( 0:95) =
which compares to an observed average drop of about
0:397;
0:25. Following the same procedure
for pharmacies, the predicted evolution is:
1:787 ( 0:6) + 1:996 ( 0:95) =
which compares to an observed average drop of about
0:824;
0:5. In other words, in line with what
is observed in the data, the predicted drop in the pharmacy segment is about twice as large
as that in hospitals (-40% vs. -80% in our computation, -25% vs. -50% in the data). Table 4
shows that this di¤erence in responses is driven by the higher price sensitivity of hospitals.
We note that both estimates overshoot the actual drop in market shares. More importantly though, once endogeneity is controlled for, both estimates (2a and 2b) point to a
substantial drop in market shares despite lower prices.31 . This is further evidence supporting
the conjecture that promotional e¤ort is a key strategic variable.
The fact that both estimates overshoot the actual drop in market shares may be due
various factors. First, the predictions do not factor in the change in reimbursement rules
that accompanies generic entry. Typically, the reimbursement rate of on-patent drugs is
reduced. This mitigates the drop in market shares and partly explains the gap between our
estimates and the actual outcome.32 Second, in anticipation of generic entry, promotion (but
not prices) starts falling before LoE. Hence, the market shares that a drug A will lose to
30
This ignores the reaction of the other …rms, but the latter is small in absolute value.
We experimented with di¤erent estimating techniniques and speci…cations. Once endogeneity is dealt
with, the …nding of a volume market share drop is consistent across speci…cations and/or estimation techniques.
32
Unfortunately, we do not have the appropriate data on reimbursement rules to factor in this e¤ect.
31
24
a branded drug B still covered by patent protection but facing imminent generic entry will
be lower than the one derived from our estimations. Third, as we will show in Section 5.5,
genericization leads to …erce price competition stemming from generics (which is not the case
pre LoE, cf. estimates reported in Table 4). The fact that the predicted drop in market share
is twice larger in hospitals than in pharmacies suggests that these three factors likely a¤ect
the two channels in a similar way.
5.3
Firm’s strategic behavior prior to LoE (testable implication 1)
The results presented in the previous section show the importance of both prices and promotion e¤ort as strategic tools to gain market shares. In this section, we focus on the …rms’
strategic behavior in the phase when they still bene…t from exclusivity. In the next section,
we analyze how strategic interactions are modi…ed by generic entry.
Based on Testable Implication 1 we expect that a producer with a higher-quality molecule
will (a) sell more, (b) charge a higher price, and (c) promote its molecule more intensely than
its lower-quality competitors. In addition, (d) both promotional e¤ort and price should be
decreasing in price sensitivity : Given the di¤erences in price elasticities that we identi…ed
in the previous section, we thus expect lower prices in hospitals than in pharmacies. Last,
(e) prices should be increasing in the degree of “horizontal” di¤erentiation, while promotion
weakly decreases with the degree of horizontal di¤erentiation.
To test prediction (e), we construct a measure of horizontal di¤erentiation for each ATC3
market, based on each molecule’s therapeutical categorization. As detailed in Section 4, the
ATC3 class is associated to a pathology, while each of the ATC4 therapeutic sub-groups
within the same ATC3 correspond to di¤erent modes of actions to treat that pathology. Our
conjecture is that the higher the number of ATC4 sub-groups in the same ATC3, the more
di¤erentiation there is in that market. We de…ne M oA (for Modes of Action) as the number
of ATC4 sub-groups in each ATC3 market, minus 1. That is, for each drug, it identi…es the
number of rival modes of action to treat the same pathology.
Empirically, the …rst set of predictions (a-c) compare di¤erent molecules/…rms in a particular market. However, we cannot directly observe the relative quality nor the de…ned daily
dose (see footnote 20, p16) of di¤erent molecules. Hence, we use …rst-di¤erence estimations
to isolate unobserved heterogeneity across molecules and markets. We then test implication
1 by checking whether a higher market share is indeed associated with higher promotion and
prices. To test (d), we exploit higher price elasticity in hospitals that we measured in Section
5.1: we expect
2
to be negative in equation (9) below.
25
Firm i’s behavior in market j at quarter t is described by the following equation:
yi;j;t =
0
+
1
zi;j;t +
2 Hosp
+
3 M oAj;t
+ Xi;j;t + "i;j;t ;
(9)
where yi;j;t ={price, promotion} and zi;j;t ={market share, price}. Hosp is a dummy taking
value 1 for hospital channel. M oAj;t takes value "0" if the ATC4 and ATC3 fully coincide,
"1" if there are two ATC4 subclasses in one ATC3, "2" if there are 3 and so on. Control
variables X includes a trend that identi…es the number of quarters before patent expiration
of brand i, T ime_to_expiration, and a complete set of time dummies. Note that in line
with equations (6) and (7) above, we use quantity (respectively, market shares) on the r.h.s.
when promotion (price) is used on the l.h.s. of equation (9).
Table 5 reports the results obtained again limiting the sample to branded drugs until
one quarter before LoE for the two channels. The two Columns (1) looks at the relationship
between quantity and promotion, columns (2) between prices and promotion and columns (3)
between market shares and prices. The latter su¤ers from the classic identi…cation problem
that both demand and supply in‡uence that relationship. To address this issue, we exploit
the results in Section 5.1, which showed that promotion intensity produces an outward shift
of demand. We can thus use promotion as an instrument to isolate supply-side e¤ects in
Columns (4).
In line with theoretical predictions, a molecule that sells more commands a higher price
(prediction (b) above) and is more heavily promoted (prediction (c)). Hence, quantity, price,
and promotion are indeed strategic complements.33 Next, promotion and prices appear to
be lower in hospitals (prediction (d)) although the e¤ect for advertising is not signi…cant. In
line with the theoretical model, e does not directly in‡uence promotion. However, in contrast
with our predictions, we do not …nd a signi…cant impact of e on price.
Identi…cation of parameters in equation (9) comes from comparing changes in the dependent and independent variables over time. Results in Table 5 shows that prices, promotion
and market shares positively co-move over time.34 This is in line with the idea that as more
evidence is gathered by patients and doctors on the true quality of a molecule, …rms tend to
increase prices and promotional e¤ort of better and, in turn, best selling drugs. Next, the
growth of promotion and prices appear to be lower in hospitals (prediction (d)) although the
e¤ect the growth of promotion is not signi…cant. In line with the theoretical model, e does
33
In Columns (3), we observe that the demand-side of the ledger dominates
prices and market shares. Once Market Shares is properly instrumented for in
changes sign and is precisely estimated.
34
In Columns (3), we observe that the demand-side of the ledger dominates
prices and market shares. Once Market Shares is properly instrumented for in
changes sign and is precisely estimated.
26
the raw correlation between
Columns (4), the coe¢ cient
the raw correlation between
Columns (4), the coe¢ cient
not directly in‡uence promotion. However, in contrast with our predictions, we do not …nd
a signi…cant impact of e on price. This is not too suprising, given that most of the data
variation for Modes of Action comes from di¤erences across markets and not over time, it
is not surprising that we cannot …nd evidence supporting prediction (4) of our theoretical
model.
35
Dep.Variable:
Coeff.
Promotion Promotion
FD(1)
FD(2)
Price
FD(3)
Price-IV
FD(4)
Quantity
β 1q
Market-Shares
β 1ms
Price
β 1p
Hospital
β2
-0.006
(0.005)
-0.001
(0.004)
-0.003*** -0.005***
(0.001) (0.002)
ModesofAction"e "
β3
-0.080
(0.059)
-0.001
(0.001)
10425
.232
-0.075
(0.056)
-0.002**
(0.001)
10425
.225
-0.007***
(0.002)
-0.001
(0.001)
10425
.065
TimetoExpiration
N
R-squared
a
Under-identification
0.253***
(0.054)
-0.039*** 0.049**
(0.008) (0.023)
0.411**
(0.184)
-0.003
(0.004)
0.001***
(0.000)
10425
<0.001
Robust standard errors clustered at ATC3-market level in parentheses. * significant at 10% level; ** significant at
5%; *** significant at 1%. Model (4) control for endogeneity of price using promotion in as shifter of the
demand. a P-values of F-test for excluded instruments in the first-stage.
Table 5. Market Shares, Price and Promotion under Patent Protection
5.4
The e¤ects of LoE (testable implications 2-4)
While the previous section studies the …rms’strategic interactions before the Loss of Exclusivity (LoE), this section turns to the e¤ects of competing molecules’losing exclusivity on drugs
that remain on patent. According to testable implications #2-4, the reaction of branded drug
B (the one that remains patent protected) when molecule A loses patent protection should
depend on three elements: (i) the price sensitivity of demand , (ii) the degree of horizontal
di¤erentiation e; and (iii) market size . The purpose of this section is to implement a test
of these predictions. Further, Proposition 1 also states that B’s price and promotion should
co-move with market share after generic entry. We also test that prediction.
35
Only ten of the …fty-three ATC3 markets register an increase in the number of Modes of Action during
the time window we consider.
27
We estimate the following speci…cation:
yi;j;t =
6
yi;j;t
4
+
1 GEN i;j;t
::: +
4 small_ j;t
::: +
0
Xi;j;t +
i
+
GEN
2 Hosp
i;j;t
+
GEN
5 M oAj;t
i;j;t
+
3 M oAj;t
small_
j;t
GEN
GEN
i;j;t
i;j;t
(10)
+ "i;j;t
where yi;j;t is {market share, price, promotion} of the patent protected drug, and GEN
i;j;t
counts the number of molecules other than i that lost exclusivity in the ATC3 market. Hosp
is a dummy for the hospital channel, and M oA is our proxy for horizontal di¤erentiation,
de…ned in the previous section. The speci…cation includes yi;j;t
4,
the one-year lag of the
dependent variable (i.e. four quarters) to capture dynamic autoregressive processes. The set of
control variables X includes the number of competitors to proxy the intensity of competition,
T ime_to_expiration, and a complete set of time dummies. As in all previous cases, the data
for yi;j;t pertains to branded drugs until one quarter before patent expiration.
Because of the presence of the lagged dependent variable, the use of …xed e¤ects to remove
unobserved heterogeneity would lead to a downward bias (generally known as “Nickell bias”
in the point estimate of
6,
which can be transmitted to the other coe¢ cients. To tackle this
problem we use an IV approach based on lags of predetermined variables and the forward
orthogonal deviations (FOD) transformation, proposed by Arellano and Bover (1995). The
reason for preferring the FOD over the First Di¤erence estimator is that some of the regressors
do not vary (e.g. Hosp) or vary little over time (such as M oA and GEN ). Accordingly, taking
these regressors in First Di¤erence would not capture the transition from the pre-entry to
post-LoE equilibrium as de…ned in our theoretical model.
According to testable implication #2, the market share of a B-molecule should increase
less (or decrease more) if
is higher. Using again the results of Section 5.1, where we
found that the elasticity of demand is higher in hospitals, we thus expect
According to testable implication #3, we also expect
3
2
to be negative.
to be negative: the market share
of molecule B should increase less (or decrease more) if the ATC3 market features more
di¤erentiated molecules. According to testable implication #4, we expect that B will lose
market share if the market is deemed small by the …rm. To implement this test, we need
a measure of market size that re‡ects the importance of an ATC3 market from the vantage
point of the …rms. However, we do not observe pro…tability directly, and sheer quantities
would be a poor measure of market size. In addition, our econometric approach implies that
average di¤erences in market size across markets are absorbed as we control for …rm …xed
e¤ect. We thus apply a revealed-preference approach: we de…ne an ATC3 market as “small”
if, before generic entry, promotion intensity was already below the median (5% of sales) –
28
this is exactly the theoretical rationale for market reversals that we identi…ed in Section 3.
Hence, we expect
4
to be negative as well.
We also interacted these three dimensions to check if they have a stronger in‡uence when
combined with one another. Proposition 1 identi…ed that it was the combination of high
demand elasticity, high di¤erentiation, and small market size that is most likely to produce
a drop in the market share (as well as price and promotion) of the molecule that remains
on patent. Hence, we expect
5
to be negative as well. Our speci…cation in (10) omits the
interactions that did not prove signi…cant.
The regression results are presented in Table 6. Volume market share is the dependent
variable in columns FOD(1)-(3) while price and promotional e¤ort are the dependent variable respectively in FOD-IV(4) and FOD-IV(5). The …rst columns, FOD(1), do not control
for endogeneity. In FOD-IV (2)-(3) instead, we instrument the lagged market share with
the number of competitors in period t
4. Similarly, we instrument for lagged prices and
advertising using a price index of competitors in (4) and competitors’total promotion in (5).
Our preferred speci…cation is the one appearing in columns FOD-IV(3)-(5).
It is interesting to note that our IV strategy leads to an increase in the point estimate
of the lagged dependent variable with little changes in the estimates of the coe¢ cients we
are interested in. For market shares as the dependent variable (columns FOD-IV(2) and
(3)), all results are in line with the three testable implications. The …rst …nding is that we
mostly observe instances of branded drugs increasing their volume market share following
the genericization of a competitor. Indeed, the coe¢ cient on GEN is positive and precisely
estimated. As already discussed, the reason is that generic entry also drastically reduces
promotional e¤ort, which more than compensates the price drop. Importantly, and in line
with the higher elasticity of demand found in Section 5.1, this e¤ect is more limited in the
hospital channel: in columns FOD-IV(2) and (3), the interaction between the two dummies
(HOSP GEN ) is signi…cantly negative, reducing by two thirds the market share gain for the
drugs remaining on patent. The negative coe¢ cient
3
con…rms that markets characterized
by a higher degree of horizontal di¤erentiation experience a smaller increase in market.36 The
interaction between the low promotion indicator variable and the generic count (
4)
does not
prove signi…cant in FOD-IV(2) and is therefore dropped in FOD-IV(3). In the latter, we
added the interaction between GEN , the degree of di¤erentiation, and our indicator variable
for small markets.37 The results are in line with testable implication #4: the market gains
for on patent drugs are much reduced in small markets characterised by a higher degree of
di¤erentiation. This is also what emerges form Graph 3: the drop in market share is larger
36
Remember that M odes_of _Action = 0; 1; 2::: when there are 1; 2; 3::: ATC4 subclasses in the same
ATC3.
37
Further interacting with HOSP did not provide additional information.
29
in promotion intensive markets for both distribution channels. Interestingly, the sum of
2;
3
and
5
1;
is signi…cant and negative (a Wald test rejects the null hypothesis that the sum
is zero at 5% signi…cance level), indicating that when the conditions for the three testable
implications are met together, on-patent molecules do lose market share to generics.
DepVbl:
Coeff.
MarketShare
Price
Promotion
FOD(1) FOD-IV(2) FOD-IV(3) FOD-IV(4) FOD-IV(5)
GEN
γ1
0.070*
(0.04)
0.068**
(0.03)
0.067***
(0.02)
0.001
(0.01)
0.025
(0.11)
Hosp*GEN
γ2
-0.026
(0.02)
-0.046**
(0.02)
-0.044**
(0.02)
0.008
(0.01)
-0.055**
(0.03)
Modes_of_Action*GEN
γ3
-0.022
(0.01)
-0.035**
(0.02)
-0.028***
(0.01)
-0.005
(0.00)
-0.029
(0.05)
LowPromotion*GEN
γ4
0.053
(0.05)
0.023
(0.04)
Mod_of_Act*LowProm*GEN
γ5
LaggedDependentVariable
γ6
-0.058***
(0.01)
0.555*** 0.797*** 0.778***
(0.05)
(0.13)
(0.14)
-0.017*** -0.019*** -0.019***
(0.01)
(0.00)
(0.00)
-0.015
-0.015
-0.015
(0.01)
(0.01)
(0.01)
8686
8686
8686
0.460(34) 0.424(34)
-0.001
(0.00)
0.671***
(0.23)
0.004*
(0.00)
0.003
(0.00)
8686
0.314(34)
-0.037
(0.06)
0.166***
(0.05)
-0.018
(0.01)
-0.045
(0.05)
8686
0.156(34)
TimetoExpiration
NumberofCompetitors
N.Observations
a
HansenJ-test(df)
Robust standard errors clustered at ATC3-market level in parentheses. * significant at 10% level; ** significant at 5%; *** significant at 1%.
All Specifications estimated using Forward Orthogonal Deviations (FOD) transformation. Endogenous variable: Lagged Dependent Vbl 1yr
before. Instrument: number of competitors in column (1)-(3), ATC3 weighted average price in column (4) and ATC3 promotion in column
a
(5). P-value of Hansen J-test of overidentifying restrictions with degrees of freedom reported in parentheses
Table 6. On-patent drugs’reaction to a competitor’s LoE.
Columns FOD-IV(4)-(5) report the results of the same exercise with price and promotion
as the dependent variable. Given the nature of the exercise, it is the ‡ow measure (instead
of the stock) that is used in FOD-IV(6). In both instances, the coe¢ cients are of the right
sign but, with exception of the lagged dependent variable, not signi…cant. From a technical
standpoint, the lack of signi…cance for price could be due to the fact that payers (health
management organisations and health insurance companies) succesfully cap signi…cant price
increases. For promotion, the reason could be that it is imperfectly measured and highly
variable over time (cf. Graph 4). From an economic perspective, what appears to be occuring
is that on-patent drugs do not need to proactively seek to gain volume from molecules that
have experienced entry: the drop in promotion by the originator facing generic competitors
is su¢ cient.
30
5.5
Competitive environment post genericization
The analysis above has been focused on drugs that are patent protected. The demand estimation con…rmed the central role played by promotion and the model’s predictions regarding
on-patent drugs have largely been con…rmed. In this sub-section, we analyze whether the
model’s predictions are also con…rmed in the post generic entry universe.
The model’s prediction is that price will converge to marginal cost as Bertrand competition kicks in following the entry of a bioequivalent substitute. Since the only di¤erence
between products is residual (di¤erent packages, colors) and the choice between varieties
is made by the doctor/pharmacist (these are prescription drugs), we expect the market to
converge to the Bertrand outcome in the absence of capacity constraints. During the initial
stages of generic entry, the stock of past promotion may still drive PDPs’choices. However,
the investment in fresh promotion should drop after generics enter the market, because of
strong spillover e¤ects (promotion of Prozac following LoE would bene…t generic producers
of ‡uoxetine). In addition, zero or low pro…t margins make promotion unpro…table.
In contrast to the previous estimations, the analysis is now based on all molecules that
have experienced generic entry. Our sample includes 409 such molecules with a mean of
around seven molecules per ATC3 market. The median and mean number of generic producers for the same molecule is respectively 7 and 12.
This sample is used to estimate the following speci…cation in FD:
msG;i;j;t =
1
pG;i;t +
::: +
6
2
aB;t +
pG;
7
i;t
+
3
pB;t +
4
pOG;t +
5
pOB;t
aOB;t + Xijt + "ijt
where msG;i is j-market share of a generic molecule G produced by …rm i (e.g. TEVA Fluoxetine), pG;i is the corresponding price of that version of the generic drug, pG;
i
is the average
price set by the other generic producers of the same molecule (e.g. MYLAN ‡uoxetine, BARR
‡uoxetine, etc..), pB is the price of the original branded drug (i.e. ELI LILLY ‡uoxetine sold
under the brand name Prozac), pOG and pOB refer to the price index of respectively, other
antidepressants that have experienced LoE (e.g. amytriptilyn), and other on-patent antidepressants (e.g. sertraline brand name Zoloft). Finally, aB and aOB pertains to the promotion
associated with the originator drug and other on-patent drugs in the ATC3 (i.e. Prozac and
Zoloft in our example). All variables are in logarithms.
31
Hospital
FD(1)
FD-IV(2)
Coeff.
OwnPrice
δ1
PriceGenerics
(samemolecule)
PriceBrand
(samemolecule)
PriceOtherGenerics
δ2
PriceOtherBrands
δ5
PromotionBrand
(samemolecule)
PromotionOtherBrands
δ6
Observations
a
K-PUnderidentification
b
AP_F_test-Promotion
b
AP_F_test-Price
c
HansenJtest(df)
d
C- test-Endogeneity
d
C test-Exogeneity
δ3
δ4
δ7
-0.448***
Pharmacies
FD(1)
FD-IV(2)
-9.234*** -0.539*** -8.706***
(0.04)
(1.97)
(0.03)
(2.17)
0.005
(0.04)
0.012
(0.03)
0.143***
(0.03)
0.308***
(0.06)
0.061*
(0.03)
0.149***
(0.03)
61443
1.566***
(0.43)
0.059
(0.08)
0.588***
(0.18)
0.275*
(0.16)
0.229*
(0.12)
-1.578**
(0.71)
61443
<.0001
<.0001
<.0001
.512(3)
<.0001
.3289
-0.054
(0.04)
-0.031
(0.03)
0.035
(0.02)
0.038
(0.10)
0.031
(0.03)
0.164***
(0.04)
61260
1.392***
(0.41)
0.139
(0.11)
0.212**
(0.09)
-0.359
(0.28)
0.304**
(0.14)
-1.930**
(0.83)
61260
.0025
.0003
<.0001
.755(3)
<.0001
0.8572
Robust standard errors clustered at molecule level in parentheses. *, **, *** significant at respectively 10%, 5% and 1%.
a
Specification (2) assumes own price and market promotion is endogenous. P-value for the Kleibergen-Paap rk statistics
b
testing the null hypothesis that the model is underidentified. P-value for the Angrist-Pischke F test for excluded instruments
testing for weak instruments in the first stage regressions of price and promotion. c Hansen J test of overidentifying
d
restrictions with degrees of freedom reported in parentheses. P-value of C (GMM distance) test of Endogeneity for own
price and promotion and of Exogeneity for price of generics and brands for the same molecule.
Table 7. Demand function of a generic drug (e.g. Mylan ‡uoxetine)
The thrust of these results is that generic entry leads to a near homogneous good Bertrand
outcome. We focus on columns (2), where the endogenous variables are instrumented for.
The …rst salient fact is the 4-5 fold increase in the own price elasticity of demand as compared
to the results reported in Table 4. By any standard, this is a quantum change. Second, the
cross price elasticity pertaining to other generic versions of the same molecule is signi…cant
and of the right sign. It is about 6 times smaller that the own price elasticity, which is in
line with the fact that the median number of generic competitors for a given molecule is 7.
The third striking …nding regards the cross price elasticity w.r.t. the originator brand: it has
the right sign, but it is small and statistically insigni…cant. This …nding is compatible with
instances of the generic entry paradox, whereby a previously unique market is segmented in
two parts. Fourth, the only signi…cant competitive price pressure is exercised by other generic
producers producing the same molecule. This is fully in line with a near Bertrand outcome
among bioequivalent drugs. Fifth, the cross price elasticity of other generics in the same
ATC3 is of the right sign and signi…cant, but its magnitude is small. Sixth, other branded
products barely have an in‡uence (the point estimates is very small and barely signi…cant,
only in the case of the hospital channel). Seventh, we observe that the relative importance of
detailing is greatly diminished. Promotion Brand (same molecule) has a positive, but small,
e¤ect on market share. More importantly, we observe that promotion by other on-patent
32
drugs within the same ATC3 does exercise a large negative in‡uence on a generic’s market
share. This is again in line with the previous results that the competing molecules acquire
additional market shares by sustaining promotional e¤ort.
Overall, our results strongly suggest that competition is intermolecular while drugs are
on patent, and shifts to prominently intramolecular when generic entry occurs. In the …rst
quarters after LoE, promotional e¤ort seems to be the key short-term strategic variable. Then,
as time passes, detailing fades into insigni…cance and is replaced by price as the strategic
weapon of choice. Accordingly, the market share of the originator drug progressively fades
into insigni…cance, while the generics drugs are engaged in Bertrand competition. At the
same time, other molecules reduce the overall market share of these generics through heavy
promotion.
6
Conclusions and welfare implications
Competition stemming from generic drugs is perceived as an e¤ective tool to keep a lid on
healthcare costs. With this objective in mind, legislators on both side of the Atlantic have
actively promoted generic penetration. Despite these proactive e¤orts, expenditures on drugs
have kept on rising, pouring cold water on these initial (high) hopes. This has led public
authorities (in Europe), health management organisations and insurances companies (in the
US) to use their buyer power to cap expenditures on drugs.
Generic competition was expected to reduce market-wide monopoly power by reducing the
cost of older medical treatments as well as having a knock-on e¤ect on other drugs launched
more recently. The fact that, by and large, this has not happened motivates our paper. We
developed a simple model that we bring to the data. In the model, two …rms compete for
prescriptions to patients. They compete both in prices and in promotion – we documented
that the pharmaceutical industry posts a 15%-20% promotion-to-revenue ratio, which is high
by any industry’s standard, even more so for an R&D intensive one – and we analyze how
competitive constraints changes when one of the two …rms is hit by a loss of exclusivity. Based
on data covering the universe of prescription drugs in the U.S. for the period 1994 to 2003,
we …nd that the originator essentially loses its market power. We observe price elasticities as
high as 9 for generics. The originator …rm’s market share for that molecule then plummets:
on the face of it, competition works.
The issue is that this intense intra-molecular competition proves ruinous for the molecule.:
Because of the drop in pro…tability, the originator …rm stops advertising its product, which
decreases doctors’incentive to prescribe this particular drug. This may more than compensate
the lower price of generics. In the data, we observed that, on average a drug that faces generic
33
competition loses 25% of its volume market share in the hospital channel, despite the fact
that hospitals can typically pro…t from procuring the molecule at a lower price. It loses
50% of its volume market share in the pharmacy channel, where prescribing doctors do not
directly bene…t from lower prices. In other words, generic fails to stimulate inter -molecular
competition. This, despite a price elasticity of demand that we estimated to be about 2.6
for hospitals and 1.8 in the pharmacy channel. These levels imply prices about twice as high
as production costs. These elasticities are also compatible with the observed market share
drops in each of the distribution channels (see Sections 5.1 and 5.2).
The model also identi…es the circumstances in which the molecule facing generic entry
will not experience a market share drop. This will happen in markets that are small, where
the price elasticity of demand is high, and – surprisingly – where there is a high degree of
di¤erentiation between molecules. Under such a con…guration, pre-LoE advertising intensity
is low, and the two prices are relatively high. Hence, when generic entry occurs, the procompetitive price e¤ect dominates the competition-reducing drop in promotion. Taking these
predictions to the data, we …nd that the market share of on-patent drugs increases less in
these cases. It even reverses in the hospital channel when the market is also characterised by
both low levels of promotion and a high degree of di¤erentiation prior to loss of exclusivity.
Going one step further, the model allows us to infer some welfare implications. In Appendix 3, we extend the analysis to patient/consumer surplus. Our modelling does not attribute
any direct bene…t to promotion: the model assumes its informative role away. We …nd that a
su¢ cient condition for patient welfare to increase after generic entry is that the market share
of molecules remaining on-patent decreases in equilibrium (the price of both molecules then
falls). As we saw, this condition is rarely met in practice.
Does that mean that generic entry typically decreases welfare? Proposition 5 in Appendix
3 identi…es that welfare should decrease in markets that are very large, in which case generic
entry triggers an increase in prices and advertising by the other originator …rms, that produce
competing molecules. Our econometric evidence instead points to a quasi absence of reaction
by these …rms (oddly enough, the only reaction we identify is a 1% price increase in hospitals,
not in pharmacies). Beyond welfare, we also use the model to assess when the change in
market shares produced by generic entry would bring the market allocation closer to the …rst
best. The answer is never, and a ban on promotion would not resolve the problem.
34
References
Apple (2013). UNITED STATES SECURITIES AND EXCHANGE COMMISSION,
Form 10-K, September 28. http://investor.apple.com/sec…ling.cfm?…lingid=1193125-13416534&cik=#D590790D10K_HTM_TOC590790_1
Anderson, S.P. Cilibertoz, F. Liaukonytex, J. and Renault R. (2015). Push-Me PullYou: Comparative Advertising in the OTC Analgesics Industry. Rand Journal of Economics,
forthcoming
Arellano, M. and O. Bover (1995). Another look at the instrumental-variable estimation
of error-components models. Journal of Econometrics, 68 (1995), pp. 29–52
Bagwell, K. (2007). The Economic Analysis of Advertising," Handbook of Industrial
Organization, vol. 3. Elsevier.
Berndt, E.R. and Dubois P. (2016). Impacts of Patent Expiry of Pharmaceutical Treatments in Eight OECD Countries, 2004-2010. International Journal of the Economics of
Business, vol. 23, n. 2, May 2016, pp. 125–147.
Branstetter, L.G., Chatterjee C. and Higgins M.J. (2011). Regulation and Welfare: Evidence from Paragraph IV Generic Entry in the Pharmaceutical Industry. NBER Working
Paper No. 17188
Branstetter L.G., Chatterjee, C. and Higgins M.J., (2014). Starving (or Fattening) the
Golden Goose?: Generic Entry and the Incentives for Early-Stage Pharmaceutical Innovation.
NBER Working Paper 20532. http://www.nber.org/papers/w20532
Brekke, K., Kuhn, M., (2006). Direct to Consumer Advertising in Pharmaceutical Markets. Journal of Health Economics 5 (1), 102–130.
Caves, R.E., Whinston, M.D., and Hurwitz, M.A., (1991). “Patent expiration, entry, and
competition in the U.S. Pharmaceutical Industry”, Brookings papers on Economic Activity.
Microeconomics, pp. 1–23
Chaudhuri, S. Goldberg, P. K. and Jia P. (2006). Estimating the E¤ects of Global Patent
Protection in Pharmaceuticals: A Case Study of Quinolones in India. The American Economic Review, Vol. 96, No. 5, pp. 1477-1514
Dave, D. and Sa¤er H. (2012). The Impact of Direct-to-Consumer Advertising on Pharmaceutical Prices and Demand. Southern Economic Journal, 79, 1. 97-126.
de Frutos, M.A., Ornaghi C. and Siotis G. (2013). Competition in the pharmaceutical industry: how do quality di¤erences shape advertising strategies?. Journal of Health Economics
–32, 268-285
Donohue J.M., Cevasco M. and Rosenthal M.B. (2007). A Decade of Direct-to-Consumer
Advertising of Prescription Drugs. The New England Journal of Medicine. 357:673-81
Elison, S., Cockburn I., Griliches, Z. and Hausman J. (1997). Characteristics of Demand
for Pharmaceutical Products: An Exploration of Four Cephalosporins. RAND Journal of
Economics, 28(3), 426-446.
European Commission (2009).
Pharmaceutical Inquiry Final Report.
35
Available at:
http://ec.europa.eu/competition/sectors/pharmaceuticals/inquiry/sta¤_working_paper_part1.pdf
Grabowski H., and Kyle M. (2007). Generic Competition and Market Exclusivity Periods
in Pharmaceuticals. Managerial and Decision Economics 28(4-5), 491-502.
Grabowski H., Long G and Mortimer R. (2014). Recent trends in brand-name and generic
drug competition. Journal of Medical Economics. 17(3):207-14.
Grabowski H. and Vernon J., (1996). Longer Patents for Increased Generic Competition:
The Waxman Hatch Act After One Decade. Pharmaco-Economics. 110–23.
Griliches, Z and Hausman A.J., (1986). Errors in Variables in Panel Data. Journal of
Econometrics 31: 93-118.
Griliches, Z. and Mairesse J. (1995). Production Function: The Search for Indenti…cation.
NBER Working Paper, No. 5067
Grochowski Jones, R. and Ornstein, C. (2016). Matching Industry Payments to Medicare
Prescribing Patterns: An Analysis. ProPublica. (The undelying data is avalaible at:
https://projects.propublica.org/docdollars/)
Harrington, S.E. (2012). Cost of Capital for Pharmaceutical, Biotechnology, and Medical
Device Firms. Oxford Hanbook of the Economics of the Biopharmaceutical Industry.
Hurwitz, M.A. and Caves, R. E (1988). Persuasion or Information? Promotion and the
Shares of Brand Name and Generic Pharmaceuticals. Journal of Law and Economics 31, No.
2, 299–320
Iizuaka, T., (2012). Physician Agency and Adoption of Generic Pharmaceuticals, American Economic Review 102(6), 2826–2858
Inderst, R., and Ottaviani, M., (2012). Competition through Commissions and Kickbacks.
American Economic Review, 102(2): 780-809.
Jena A.B, Calfee. J.E., Mansley E.C. and T.J. Philipson, (2009). Me-Too Innovation in
Pharmaceutical Markets. NBER Chapters, in: Frontiers in Health Policy Research, volume
12, NBER
Johnson, J. and Myatt, D., (2006). On the simple economics of advertising, marketing
and product design. American Economic Review 96, 756–784
Kenkel, D., and Mathios, A. (2012). Promotion to Physicians and Consumers. Oxford
Hanbook of the Economics of the Biopharmaceutical Industry.
Le- er, K. (1981). Persuasion or information? The economics of prescription drug advertising. Journal of Law Economics, 24(1) 45–74.
Regan, T.L., (2008). Generic entry, price competition, and market segmentation in the
prescription drug market. International Journal of Industrial Organization 26, 930–948.
Rizzo, J. (1999). Advertising and competition in the ethical pharmaceutical industry:
The case of antihypertensive drugs. Journal of Law and Economics 42(1) 89–116.
Scott Morton, F. M. (2000) “Barriers to entry, brand advertising, and generic entry
in the US pharmaceutical industry” International Journal of Industrial Organization 18(7)
1085-1104.
36
Stern, S. (1996). Market De…nition and the Returns to Innovation: Substitution Patterns
in Pharmaceutical Markets. mimeo.
Vandoros S. and P. Kanavos, (2013). The generics paradox revisited: empirical evidence
from regulated markets. Applied Economics, 45:22, 3230-3239
37
Appendix 1: Benchmark. The e¤ects of generic entry in the
absence of promotion
This section presents the case in which …rms cannot use detailing, and only compete in prices. One of the
main reasons for promoting generics is to make drugs more a¤ordable. The expectation is two pronged: …rst,
the entry of a generic version of molecule A should put substantial pressure on the price of molecule A. As we
saw in Section 2, this e¤ect is unquestionably present. We will thus assume that generic entry does produce
such a strong competition on molecule A that the price of drug A drops to the marginal cost of production,
which we normalized to 0. This is the direct e¤ect of generic entry.
The second, indirect, e¤ect is the price reaction of …rm B. We can check how each of these e¤ects operates
when …rms can only compete in price, that is when we impose that …rm cannot use promotion: aA = aB 0.
Derivating the …rms’respective reaction functions is straightforward:
pA
=
pB
=
1
(e
2
1
(e +
2
B
+
pB )
(11)
B
+
pA ) :
(12)
Hence, in the pre-entry equilibrium:
Pre-entry benchmark: Before generic entry and in the absence of detailing, for
j B =ej 3 the equilibrium is such that:
pA
=
pB
=
1
e
B
e
+
and QA =
3
B
3
and QB =
1+
B
1
3e
2
B
3e
2
A
+
B
> 2e, and
;
:
The drug with highest quality J , sells more and at a higher price. Moreover, both prices increase in patient
heterogeneity e and decrease in price elasticity, :
In that benchmark, the condition A + B > 2e is necessary and su¢ cient to ensure that all patients
obtain their treatment in equilibrium. For lower values of J , some patients would …nd both treatments
una¤ordable. Note that this condition does not depend on price sensitivity because, in the absence of price
constraints, …rms vary their price exactly to compensate a variation in price sensitivity. In other words, if
health insurances double their intervention, total prices will double as well, and the patient pays the same
…nal price. Second, the condition j B =ej
3 ensures that both …rms sell positive quantities for the price
levels derived in benchmark 1.
Upon generic entry, the price of A falls to 0. As a consequence, by (12) ; the equilibrium becomes:
Post-entry benchmark: After generic entry and in the absence of detailing, an interior equilibrium is such
that:
pA
=
0 and QA =
pB
=
e+
2
B
3
4
and QB =
B
4e
;
1
B
+
4
4e
:
Note that for any interior solution (i.e. for 0 <
B =e < 3), the price of B must decrease and the market
share of A must increase in comparison to the pre-entry benchmark. Hence, the only question is the magnitude
of this change: the more vertically superior is molecule B, the less generic entry in‡uences market shares and
equilibrium prices.
38
Appendix 2: equilibrium before and after generic entry in the
presence of promotion
We derive the explicit solutions for the equilibrium levels of prices, advertisement, and quantities before and
after generic entry. We work under the following condition, which ensures that solutions are interior (i.e. the
market share of A before entry is strictly positive):
Condition 1
< (3e
B)
Equilibrium before generic entry
Letting KJ
1+
J
J
3 e
and solving explicitly yields:
Proposition 2 Under Condition 1, the equilibrium prior to generic entry is unique and such that:
pD
J
=
aD
J
=
QD
J
=
e
2
2
(13)
KJ
KJ
(14)
KJ ;
(15)
for J 2 fA; Bg. Hence, the most advanced molecule, B, has a higher price, advertisement level and market
share than A.
Proof. Everything follows directly from the FOCs:
@ A
@aA
@ B
@aB
pA
2e
pB
2e
=
=
aA = 0
aB = 0:
Hence:
aB =
2e
pB :
Next:
@ A
@pA
@ B
@pB
=
1
2
=
1
+
2
B
+
aB
(pB
"
e 1
= 0 , pA =
2
"
e 1
= 0 , pB =
+
2
2pA )
2e
B
+
aB
(2pB
pA )
2e
Solving jointly for pA and pB yields:
pA
=
pB
=
e
e
1
1+
39
B
3 e
B
3 e
B
+
pB pA
2e
pB
2e
B
+
pB pA
2e
2e
+ pA
#
#
Finally, Condition 1 follows from the fact that qA
+
B
1
aB
(pB
0 i¤ :
pA )
e
e
+
B
e
1+
B
2e
2
e
B
2
3 e
B
3 e
3 e
+
=
3 e
2 e
3 e
3 e
e
2e
=
e
3 e
3 e
B
Equilibrium after generic entry
Pro…t maximization yields:
Proposition 3 Post generic entry, the unique equilibrium is given by:
pG
A
=
aG
A = 0
pG
B
=
2e
aG
B
=
QG
B
=
2e
2
B
+e
4 e
pG
B
1+
2
(2 e
B
)
4 e
(16)
Proof. Perfect competition amongst generics implies pA = aA = 0: Taking …rst order conditions of
the maximization of
B
with respect to pB and aB yields the Proposition.
Appendix 3: Outcome E¢ ciency and Consumer Welfare
The results in Section 3.2 can be exploited to analyze how generic entry in‡uences static allocative e¢ ciency
and patient welfare. The fact that our results focus on static e¢ ciency is important to note: our analysis
considers …rm behavior for pre-existing molecules, around the time of patent expiration. As we know, patents
are essential to to provide the incentives to pharmaceutical …rms to develop these molecules in the …rst place.
Yet, extending the analysis to dynamic allocative e¢ ciency and welfare would require accounting for the
endogenous R&D investments by di¤erent …rms, which is beyond the scope of this paper.
Keeping this in mind, static e¢ ciency is reached when both …rm A and B face perfect competition. In
B
B
this …rst-best case (F B), the price of both molecules is driven down to marginal costs, pF
= pF
= 0, and
A
B
B
FB
promotion consequently drops to zero: aF
=
a
=
0.
The
resulting
quantities
are:
A
B
B
QF
=
J
2
1+
J
e
;
(17)
and consumer welfare is maximized.
Quite trivially, we would thus reach …rst best as soon as all molecules have lost exclusivity; just a matter
of time. In reality, new vintages of old treatments (so-called “me too” drugs) and new treatments appear
constantly. In our quarterly data, for instance, there is at least one molecule still under patent protection for
all “anatomic therapeutic classes” and periods. That is, reality is best described either by the situation D or
the situation G that we analyzed above. Our purpose here will be to identify when the movement from D
(only patent-protected molecules) to G (some patent-protected molecules) improves market e¢ ciency and/or
consumer welfare. We …nd that:
40
Proposition 4 Whether pharmaceutical …rm can (D), or cannot (N D), use detailing, generic entry never
brings the market equilibrium closer to the …rst-best allocation (17).
Proof. First, let us concentrate on the case in which …rms can use detailing and we focus on the
D
FB
quantities sold by B since A serves the rest of the market. Consider …rst the case in which qB
> qB
;
i.e. B sells too much prior to generic entry. By (15), and (17) this only happens if:
>
3 e
1
,2 e< ;
e
G
D
which is the exact condition in Proposition 1 for qB
> qB
. Hence, generic entry produces an increase
in the quantities sold by B precisely when convergence to the …rst best requires a drop. By the same
token, the reverse is true for 2 e > :
Second, let us turn to the case in which …rms cannot use detailing. As shown in Appendix 1, the
quantities before and after generic entry are respectively:
QD
B
=
QG
B
=
2
4
1+
1+
B
3e
B
e
=
B
QF
B
;
2
G
and the condition for QD
B < 3e. It is straightB and QB to be less than the whole market ( ) is:
FB
D
forward to check that, for such parameter values, we always have QG
B < QB < QB :
The main reason for this result is the same as for Proposition 1: generic entry produces a very lopsided
market, in which one molecule becomes cheaper by also loses all detailing e¤ort. As we saw, when molecules
are distant substitutes and market size is small (2 e > ), the genericization of A intensi…es competition for
B, which loses market share, and cuts down prices and advertisement as a result. This is exactly the intended
purpose of generic competition. The problem is that this happens precisely when B’s market share is initially
lower than in the …rst best.
Conversely, when the two molecules are close substitutes or market size is large (2 e < ), the genericization of A actually relaxes the competitive pressure on B. This allows …rm B to actually gain market power
and market share. In this case, …rm B also grabs the opportunity to increase its prices and its detailing e¤ort.
This only happens when B’s market share was already initially too high.
Interestingly, prohibiting detailing altogether would not be a panacea. In that case, the market share of
the more advanced molecule B is always too low (except in the corner solutions in which it grabs 100% of the
market), and the genericization of A further reduces it.
From a policy perspective, being confronted with explosive expenditure on healthcare, authorities have
actively promoted the use of generics partly in an e¤ort to contain the costs borne (directly or indirectly) by
patients. In addition, competition authorities have, de facto, adopted a consumer welfare standard. We de…ne
consumer surplus as their utility minus the price they pay for the molecule they actually buy. Importantly,
their utility depends on the actual quality of the molecule, J , whereas which molecule they actually buy
depends on perceived quality, 0J
J + aJ . We thus calculate welfare as the integral of the patient/doctor
G
38
pairs’utilities (1-2) when they actually consume the quantities QD
We
J before, and QJ after, generic entry.
…nd that:
Proposition 5 Generic entry necessarily increases consumer welfare when B’s market share (weakly) decreases after generic
B = 0, consumer welfare decreases upon generic entry
p entry (i.e. for < 2 e). For
10 ' 2:56 e, in which case B gains substantial market share, and increases prices and
when > 23e 7
promotion intensity.
38
The derivations of consumer welfare are available upon request. We do not present them here because
they are tedious but rather straightforward.
41
Proof. After tedious algebra (see supplementary material), we …nd that:
2 e=
) CW G
CW D =
2
(e
1)
2
0;
where CW D is consumer welfare when both A and B bene…t from patent protection, and CW G is
consumer welfare when only B still bene…ts from patent protection. When is smaller, prices must
be decreasing for all consumers, and their welfare thus increases.
Imposing
B = 0 for the sake of tractability, one can also identify
p the value of for which this
10 . For any values of below
di¤erence in consumer welfare is zero. The threshold is: = 23e 7
that threshold, consumer welfare increases upon generic entry.
42
Appendix 4: Supplementary tables
Table A1: list of ATC3 markets
ATC3_code ATC4_code
A10B
A10B
A10B
A10B
A10B
A10B
A2B
A2B
A2B
B1C
B1C
C10A
C10A
C10A
C1B
C1F
C1X
C2A
C2A
C3A
C3A
C4A
C7A
C8A
C9A
C9B
C9B
D10A
D11A
D1A
D6D
D6D
D7A
G4A
G4B
G4B
G4B
J1C
J1D
J1G
J2A
J4A
J5B
L1A
L1B
L1C
L1D
L1X
L1X
L1X
L2B
L2B
L2B
L4A
M1A
M1A
M1C
M5B
N1A
N2A
N2B
N3A
A10B1
A10B2
A10B3
A10B4
A10B5
A10B9
A2B1
A2B2
A2B9
B1C2
B1C4
C10A1
C10A3
C10A9
C1B0
C1F0
C1X0
C2A1
C2A2
C3A2
C3A3
C4A1
C7A0
C8A0
C9A0
C9B1
C9B3
D10A0
D11A0
D1A1
D6D1
D6D9
D7A0
G4A1
G4B3
G4B4
G4B9
J1C1
J1D1
J1G1
J2A0
J4A1
J5B0
L1A0
L1B0
L1C0
L1D0
L1X1
L1X2
L1X9
L2B1
L2B2
L2B3
L4A0
M1A1
M1A3
M1C0
M5B1
N1A2
N2A0
N2B0
N3A0
ATC4_name
SulphonylureaAntidiabetics
BiguanideAntidiabetics
CombSulph+BiguanAntidiabetics
ThiazolinedioneAntidiabetics
Alpha-Gluc.Inhib.Antidiabetics
OtherOralAntidiabetics
H2Antagonists
AcidPumpInhibitors
AllOtherAntiulcerants
AdpRecepAntagPlatInhibitors
PlCampEnhPlatAgInhibitors
Statins-Hmg-CoaReductaseInhibitors
BileAcidSequestrant
AllOthChol/TriglycRed
Antiarrhythmics
PositiveInotropicAgent
AllOtherCardiacPreps
Antihyper.PlMainlyCent
Antihyper.PlMainlyPeri
LoopDiureticsPlain
Thiazide+AnaloguePlain
Cereb/PeriphVasotheraps
B-BlockingAgents,Plain
CalciumAntagonistPlain
AceInhibitorsPlain
AceInhibitorsComb+A-Hyp/Diur
AceInhibitorsComb+CalcAntag
TopicalAnti-AcnePreps
OtherDermatologicalPrd
TopicalDermatAntifung
TopicalAntivirals
OthTopPrdsViralInf
Top.CorticosteroidPlain
UrinaryAntibiot/Sulphon
ErectileDysfunctionPrd
UrinaryIncontinencePrd
AllOthUrologicalProds
Brd.Spect.Penicill.Oral
Cephalosporins,Oral
OralFluoroquinolones
SystAntifungalAgents
Anti-Tb,SingleIngred
AntiviralsExclAnti-Hiv
AlkylatingAgents
Antimetabolites
VincaAlkaloids
Antineoplas.Antibiotics
AdjPrepForCancerTher
PlatinumCompounds
AllOth.Antineoplastics
CytoAnti-Oestrogens
CytoAnti-Androgens
CytostatAromataseInhibitors
ImmunosuppressiveAgents
AntirheumaticsNon-SPln
Coxibs
SpecAntirheumaticAgent
OralBisphBoneCalcReg
InjectGenAnaesthetics
NarcoticAnalgesics
Non-NarcoticAnalgesics
Anti-Epileptics
43
44
45