Journal of Health Economics 47 (2016) 5063

Contents lists available at ScienceDirect

Journal of Health Economics

journal homepage: www.elsevier.com/locate/econbase

Health insurance and diversity of treatment

David Bardey a,b , Bruno Jullien c , Jean-Marie Lozachmeur c,
a
University of Los Andes Cede (Bogota), Colombia
b
Toulouse School of Economics, France
c
Toulouse School of Economics, University of Toulouse 1 Capitole (CNRS), France

a r t i c l e i n f o a b s t r a c t

Article history: We determine the optimal health policy mix when the average utility of patients increases with the supply
Received 26 November 2014 of drugs available in a therapeutic class. Health risk coverage relies on two instruments, copayment and
Received in revised form 9 October 2015 reference pricing, both of which affect the risk associated with health expenses and diversity of treatment.
Accepted 4 January 2016
For a xed supply of drugs, the reference pricing policy aims at minimizing expenses, in which case the
Available online 17 February 2016
equilibrium price of drugs is independent of the copayment rate. However, with an endogenous supply of
drugs, diversity of treatment may susbtitute for insurance so that the reference pricing may depart from
Keywords:
maximal cost-containment in order to promote entry. We next analyze the determinants of the optimal
Health insurance
Drugs market
policy. While an increase in risk aversion, or in the side effect loss, increases diversity and decreases the
Price regulation copayment rate, an increase in entry cost decreases both diversity and the copayment rate.
2016 Published by Elsevier B.V.
JEL classication:
I18
L11
L15
L51

1. Introduction In terms of budgetary concerns, the key question that a health

A controversial issue has long been whether or not costly incre-
mental innovation leading to me-too or follow-on drugs (as
opposed to major innovation leading to breakthrough products)
has some value for society as a whole. As argued by Wertheimer
et al. (2001), me-too drugs have several advantages, among which
differing dose delivery systems and dosage forms that enable
extended uses with a variety of patient population; availability of
choice when patient response to and tolerance of a particular agent
is subject to great variation; [and] the ability to tailor therapy to
the needs and preferences of patients (pp 78). In line with these
comments, Di Masi and Paquette (2004) show that among 72 ther-
apeutic classes, one third of the me-too drugs received a priority
rating from the US FDA.1 Moreover, 57% of these classes include a
me-too drug that received such a priority rating. As such, diver-
sity of treatment itself has an insurance role in that it increases the
probability that a patient nds the best (or the least bad) treatment.
Corresponding author. Tel.: +33 5 61 12 87 37.
E-mail address: jean-marie.lozachmeur@tse-fr.eu (J.-M. Lozachmeur).
1
Drugs receiving priority review (as opposed to standard review) by FDA benet
from an accelerated procedure for approval (6 months as compared to 10 months
under standard review).
system faces is not only how much insurance coverage to provide
but also what level of diversity of treatments to offer for a given
pathology. In line with the literature on horizontal differentiation
with Bertrand competition (e.g. see Anderson et al., 1995), one may
object that the number of products at equilibrium in a unregulated
market is excessive from a social welfare point of view. This argu-
ment would be reinforced in the pharmaceutical market as entry is
subsidized via generous insurance coverage. This would ultimately
plead in favour of a strong price control in order to discourage entry
of me-too drugs (and possibly encourage major innovation). Our
contribution is to challenge this view, namely that strong price con-
trol should limit the entry of me too drugs. We show that in a
model where individuals are sufciently risk averse with respect to
poor health conditions, entry of me too drugs should be encour-
aged via soft price controls even when there is an (optimal or non
optimal) insurance plan in place.
We analyze this issue in a context where health services are
supplied in an imperfectly competitive market, as is the case for
drugs markets. In order to analyze the issue of the diversity of
treatments offered to policyholders, we focus on two policy instru-
ments that are used to maximize individuals expected utility. The
rst is a standard linear copayment rate, while the second is a ther-
apeutic reference pricing regulation. We use the latter instrument
as it provides a simple characterization of the degree of the price

http://dx.doi.org/10.1016/j.jhealeco.2016.01.003
0167-6296/ 2016 Published by Elsevier B.V.
 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063
control. Moreover, therapeutic reference pricing is used more and
more frequently in developed countries2 and its effectiveness has
been demonstrated (e.g. see Brekke et al., 2009, 2011). While terms
vary across countries, internal or therapeutic reference pricing (as
opposed to external reference pricing) consists of determining a
reference price as a weighted sum of drugs prices adopted in the
same therapeutic class. If the price of a drug is higher than this ref-
erence price, patients pay the full difference between the price of
the drug and the reference price. This regulatory scheme can be
perceived as a complement to copayment rates in order to encour-
age patients to consume low price drugs.
We build a model where there are several pharmaceutical rms
selling horizontally differentiated drugs that belong to the same
therapeutic class.3 All patients value the drugs differently because
of their different side effects.4 Risk averse consumers benet from
an insurance plan consisting of a premium and a linear copayment
rate subject to internal reference pricing. We consider a reference
price that is a linear combination of extreme value prices in the
market. By choosing the weight attached to the lowest price in
the reference pricing formula, the regulator determines the pres-
sure on the equilibrium price of drugs and thus on total health
expenses.5 Our results show that in the short run, the reference
pricing scheme aims at minimizing the price of drugs. Indeed, as
long as the number of drugs in the therapeutic class is xed, the
health insurer is only willing to lower drugs prices since he can-
not improve the drugs diversity. However, in the long run, there
may be room for a more lenient reference pricing policy accommo-
dating some increase in health expenses in order to improve the
diversity of treatments. The desirable level of the reference price
is the result of a trade-off between a diversity effect (where a new
drug decreases average side effects) and the xed cost generated
by additional drugs entries. We particularly emphasize the role of
policyholders risk aversion by showing that the higher the level of
risk aversion, the more likely it is that the regulator chooses a more
lenient reference pricing policy. While an increase in risk aversion
leads to a lower copayment rate and a higher diversity of treat-
ments, an increase in innovation costs implies a lower copayment
and a lower diversity of treatments.
1.1. Related literature
While mainly normative, our article borrows from the positive
literature dealing with drugs price regulation. To the best of our
knowledge, this is the rst paper that endogenizes both the ref-
erence pricing regulation and insurance scheme in a fully-edged
equilibrium model. Brekke et al. (2007) develop a general set-up
containing horizontal and vertical differentiations in drugs mar-
kets. They consider two reference pricing rules, namely internal and
external reference pricing as well as price cap regulation, with both
regulatory schemes being associated with an exogenous copay-
ment rate. Concerning the internal reference pricing, they consider
the laboratories best reply function, and provide comparative sta-
tics on the equilibrium allocation. They show that an increase in the
2
See Lopez-Casanovas and Puig-Junoy (2000), Danzon (2001) and Danzon and
Ketchman (2004) for more details on reference pricing and its applications.
3
This horizontal differentiation set-up can capture two types of competition in
drugs market. It accomodates competition between brand-name drugs, i.e. pioneer
drug and me-too drugs, which belong to the same therapeutic class. It also ts the
competition between generic drugs which reproduce the same molecule (horizontal
differentiation occurs because formulation, packaging as well as delivery systems
are allowed to vary). Conversely, an horizontal differentiation framework is not well
adapted to model competition between brand-name and generic drugs.
4 7
Side effects accommodate secondary effects per se, adverse drug reactions, drug-
drug interactions, dosing schedules as well as delivery systems.
5
In some European markets, such as Germany, the weight also depends on supply
conditions and market shares. We thank a referee for pointing that to us.
51
weight attached to the lowest price leads to a reduction in prices
and increases the market share of the cheapest drug (the generic
drug in their context).
In a model that distinguishes breakthrough and follow-on drugs,
Bardey et al. (2010) evaluate the long run impact of reference pri-
cing on pharmaceutical innovation, delays of introduction, patients
health and expenditures. They show that reference pricing reg-
ulation yields lower prices and therefore delays pioneer drugs
and me-too entries. Nevertheless, as me-too entries are more
delayed, it may favor costly breakthroughs and may increase health
expenditures in the long run. Miraldo (2007) compares the equi-
librium allocations under various alternative regimes: copayment
and reference pricing schemes.6 In contrast, our model is more nar-
row on the reference pricing modality as we focus our attention on
internal reference pricing, however the optimal copayment rate
and the drugs pricing regulations are jointly determined.
On the normative side, very few papers study health insurance
in the context of pharmaceutical markets. Lakdawalla and Sood
(2009) show that encouraging health insurance can be welfare
improving as it lowers static deadweight loss (i.e. it implies more
efcient utilization) without altering incentives for innovation.
They also show in another paper (2013) that a competitive health
insurance market can combine static and dynamic efciency in the
drugs market: a premium-nanced xed fee is offered to drugs
monopolists which ensures second best utilization and extracts
the full surplus so that incentives for innovation are optimal. Their
model however, does not provide for the optimal diversity of
treatment since individuals can only benet from at most one inno-
vation.
The next section presents the set-up. Section 3 is devoted to
the short-term analysis assuming a xed number of drugs. Sec-
tion 4 characterizes the optimal regulation in the long run, with an
endogenous number of drugs. Section 5 discusses some possible
extensions. Lastly, Section 6 concludes.
2. The set-up
Consider a collection of policyholders J (0, 1) who can be
unwell with probability  and healthy with probability 1 . We
normalize the size of the population to 1. In case of illness, each
patient chooses a drug among N ( 2) treatments, denoted by a
subscript i I (1, N). We refer to N as the diversity of treatment.
When choosing a drug i in case of illness, a patients net income is
wsi = w  pi ,7 where w is an exogenous income,  denotes the
premium paid to the health insurer and pi is the out-of-pocket price
paid for consuming drug i. When healthy, the net income is wh =
w . In the state of illness, policyholders are horizontally differ-
entiated: when consuming drug i, a policyholder j J is affected by
a side effect which depends on an individuals stochastic variable
xji X, observed by the policyholders before consumption takes
place. The shocks xji are identically and independently distributed
across policyholders and drugs over R+, and follow an exponential
distribution with distribution function F (x) = 1 exp (x) . This
assumption implies that the mass of individuals with large adverse
effects is always lower than the mass of individuals with small
adverse effects.8
6
A recent work by Brekke et al. (2015a) analyses a similar issue in a set-up
encompassing vertical and horizontal differentiation. The authors test the empirical
validation of their model in a companion paper (Brekke et al., 2015b).
There is no lump sum monetary compensation for being sick. This may be ruled
out by law or because the state of illness is not veriable by the regulator.
8
Our main result can be shown to hold if the distribution is such that 1 F (x) is
log-log concave (see Bardey et al., 2013)
52 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063

The policyholder js utility when unwell and consuming drug i where

is:
1
  Q =    < 1.
Uji = u wsi  xji , E exp mini I xji

where  > 0 is a xed effect of illness on the VNM utility function

when unwell.9  measures the side effects marginal loss (which
corresponds to the differentiation parameter in traditional models
with horizontal differentiation) so that  xji represents the mon-
etary equivalent loss incurred by a patient j consuming drug i.
There is also an outside alternative which is to forgo any treat-
ment: 0 xj0 denotes the monetary equivalent loss incurred by an
unwell individual when not consuming any drug. xj0 is assumed
to be exponentially distributed along with the xji . Throughout the
paper (except in Section 5.1), we consider that  0 is innite so that
The parameter Q is a measure of the average side effects of available
treatments which impacts the average utility when unwell. Given
that u is negative (due to the CARA specication), the expected util-
ity increases when Q increases. To ensure niteness of the expected
utility for all N we state the following:12
Assumption 1. The market is such that N >  .
Since x = minxji is exponentially distributed with mean 1/N,
iI
Assumption 1 ensures that:

the market is fully covered. Roughly speaking, the cost of not being 1


treated is so large compared to prices and adverse effects that all
policyholders buy a drug when unwell.10 Throughout the paper, we
use an exponential VNM utility function of the CARA form:
u () = exp () ,
where  represents the absolute risk aversion parameter. This spec-
ication allows us to avoid any income effect on the demand for
drugs which simplies the exposition of our results and allows us
to perform simple comparative statics with respect to the risk aver-
sion parameter. We show in Section 5.2 that our main results hold
when the adverse effect enters separately in the utility function.
Throughout the paper, we consider a sequential game where the
regulator moves rst and sets a reimbursement policy consisting a
copayment rate and a reference pricing scheme (described in the
next section). The rms decide to enter or not the market and then
compete in prices. To sum up, the precise timing of the game is as
follows:
1. The regulator chooses the reimbursement policy (copayment a
and reference pricing r dened in the section) and the insurance
premium .
2. Drug producers enter or not.
3. The N active drug producers simultaneously set prices.
4. A fraction  of consumers become ill. Each unwell consumer
privately observes the vector of individual effects xji , and decides
which drug to buy (or not to buy).11
At a symmetric equilibrium where pi = p for every i (1, N),
patients minimize the loss so that the aggregate expected utility
function writes:
  
EU = E u ws minxji + (1 )u(wh ), (1)
iI
where E is the expectation operator over J XN .
For a CARA utility function, Eq. (1) yields:
..., PN). A patient buying drug i is thus reimbursed
 but obtains no reimbursement for the excess payment above the
EU =  u (ws ) + (1 )u(wh ), (2)
Q reference price.
9 reference price is a weighted sum of the minimal and the maximal
The value of  can be lower than one as suggested by Finkelstein et al. (2013). As
we show later, this parameter only plays a role in the detemination of the insurance
coverage.
10
This assumption is quite realistic in countries characterized by high levels of
coverage, i.e. low levels of out-of-pocket prices.
11
An equivalent assumption is to assume that the doctor perfectly observes the
state of the patient and delivers good information on possible adverse effects of any
drug to ensure the best match between each drug and each patient.
= exp x N exp [Nx] dx,
Q (N) 0
is well dened so that:

Q (N) = 1 . (3)
N
Our model relates the average side effects to the diversity of avail-
able drugs in the therapeutic class. We thus refer Q (N) as the
welfare index of diversity (WID hereafter). Ceteris paribus, the higher
the number of drugs available in the therapeutic class, the higher
the perceived utility of the health service as patients benets from
a decrease in the average side effects. This marginal benet from
a reduction of side-effects decreases in the number of drugs. On
the contrary, the negative consequences of side effects on patients
welfare increase both in the side effects marginal loss  and in the
risk aversion parameter .
3. Insurance policy and the short-run equilibrium of the
pharmaceutical market
In this section, we rst determine the market equilibrium in
the therapeutic class and then analyze the short-run insurance
trade-off when diversity is given (there is no entry at stage 2). We
endogenize the diversity of treatment in the next section.
3.1. Insurance and market equilibrium
We consider one specic therapeutic class in which pharma-
ceutical rms each own one drug and compete in the market for
drugs by setting prices. While prices are freely chosen by rms,
the equilibrium is shaped by the insurance policy which is based
on two instruments: a copayment rate a [0, 1] and reference
pricing.13 We model reference pricing as a rule that associates
any vector of manufacturers prices (P1 , . . . PN ) with a reference
price R (P1 ,
(1 a) min Pi , R , that is a share 1 a of his expense up to R,
Several formulas for the reference price can be used. We con-
sider an internal reference pricing regulation by assuming that the
12
In the long-run analysis, the assumption constrains an endogenous variable N.
However, Assumption 2 below ensures that it holds in any equilibrium.
13
Alternative mechanisms to reference pricing could be price-cap or direct price
negotiation with producers.
 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063
prices of the drugs in the market.14 Formally, the reference price is
given by:

 (N)
R = rmin Pi + (1 r)max Pi ,
i i
where r [0, 1] is the policy variable, referred to as the strength of
the drug price regulation. When r = 1, there is maximal reference
pricing. This is the most stringent policy from the perspective of
pharmaceutical rms. Conversely, when r = 0, there is no reference
pricing. In such a case, all the treatments are reimbursed at a rate
1 a.
We rst discuss the last competition stage of the model (stage 3
and 4 mentioned above). The number of drugs N is given as well as
policy parameters (a, r, ). As already mentioned, in the main body
of the paper, we assume that the market is fully covered and relax
this assumption in Section 5.1. The demand for one drug is thus the
mass of individuals who prefer this drug to any other. We focus on
symmetric equilibria where rms set the same prices.
Consider any drug i (1, N) and suppose that the net out-
of-pocket price paid by patients for all other drugs is p. Our
assumptions on the utility functions (in particular, the fact that the
utility of unwell patients are symmetric and quasi-linear in prices)
(N) = . (5)
imply that the demand Di , when the out-of-pocket price paid for
consuming drug i is pi , can be expressed as follows at any symmetric
equilibrium:
Di = D (pi p, N) ,
where D is a function independent of i with D/pi < 0 for any pi ,
p and N. At a symmetric equilibrium, each pharmaceutical rms
market share is D (0, N) = /N.15
By denition, when all other rms charge P and rm i charges
Pi , the out-of-pocket price paid by patients is dened by:
pi = Pi (1 a) min [Pi , R]
so that,
pi p = Pi P (1 a) (min [Pi , R] min [P, R]) .
Using the denition of R in Eq. (4), this yields:
pi p = (Pi P),
where = a + (1 a)r measures the distortion introduced by the
insurance system on the market for drugs. We refer to as the
pass-through rate resulting from the joint regulation of insurance
coverage and reference pricing. This rate determines the fraction
of the price differential which is actually paid by individuals and is
increasing in the copayment rate and the reference pricing strength.
Note that our reference pricing specication does not generate any
kink in the demand function, as opposed to Danzon (2001). This is
due to the fact that we focus on symmetric equilibria in fully cov-
ered markets. In Section 5.1, which considers non-fully covered
markets, such a kink emerges.
For-prot pharmaceutical rms compete in prices, and marginal
costs are normalized to 0. Therefore, we have:
Pi arg max Pi D ( (Pi P) , N) .
As the symmetric equilibrium yields Di = /N, we obtain:
14
Even though each country may have some specicities regarding its applica-
tion, 20 of the 27 EU Member States had implemented an internal reference price
 and
regulation (see Ruggeri  Nolte, 2013).
15
More formally 1/ D (pi p, N) is the probability that the total loss  xji +
pi be less than minxjl + p.
l=
/ i
53
Proposition 1. For any N and policy (a, r), there exists a unique
symmetric price equilibrium with a drug price
(4) P= ,
where (N) = / (N 1) is the inverse semi-elasticity
  residual
of the
demand evaluated at pi = p(i.e. D (0, N) / D (0, N) /pi ).
Proof. See Appendix A.
The resulting price equilibrium is identical to the one in a
standard general model of horizontal differentiation in which the
differentiation parameter would be /. Here, the pass-through rate
increases in the strength of reference pricing and in the copay-
ment rate. In other words, increasing the insurance coverage lowers
the competition intensity. This feature of the model illustrates what
Danzon (2012) refers to as supplier moral hazard. Equivalently,
relaxing the strength of reference price regulation lowers the com-
petition intensity.
At a symmetric equilibrium, the pharmaceutical rms prot is
then:
 (N)
N
The closed form solution is useful in order to show in a more obvious
way the different effects at work. Ceteris paribus, the side effects
marginal loss  increases the drugs prices and the prot obtained by
laboratories. However, the overall differentiation (N) decreases in
the diversity of treatments available in the therapeutic class.
3.2. Fixed drugs diversity
Here we determine the optimal policy in the short-run, consid-
ering the market structure as given. Since the diversity N of
available drugs in the therapeutic class is xed, we drop the argu-
ment N in the functions of this section. The resource constraint
of the regulator states that the premium should cover reimburse-
ment cost that is  = (1 a)(1 + )P where  is to be interpreted as
a loading factor or as a shadow cost of public funds. Thereafter, we
consider two alternative scenarios. First, we study the case where
the social objective is to maximize the policyholders expected util-
ity. Second, as a robustness check, we integrate industry prots in
the regulators objective.
3.2.1. Social objective without laboratory prots
When the social objective is to maximize the policyholders
expected utility as dened in (2).16 For a xed diversity of treat-
ments, the optimal policy is the solution of:

max EU = (1 )u(w ) +  u (w  aP)
0a,r1 Q
s.t  = (1 + )(1 a)P, P= .
a + (1 a)r
For any r < 1, a decrease in a exacerbates the extent of supplier moral
hazard by increasing the price paid by patients. Nevertheless, this
problem can be dealt with by adjusting the strength of reference
pricing. For a xed supply of drugs, reducing the price P can only
benet consumers because the WID is xed and the price is always
above marginal cost. As increasing the weight r in the reference
price decreases the equilibrium price, it follows that it is always
optimal to choose maximal reference pricing r = 1. The price is then
at , the level that would prevail without any insurance scheme. In
16
This alllows us to compare our results with the ones in the following section
where prots will be nil.
54 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063

and 1, so that there is no need for insurance for  (1 ) / and

lambda 2.5 there is full insurance for  0 .
2.0
no insurance 3.2.2. Social objective with laboratory prots
1.5 partial Let us now assume that the policyholder holds a fraction  [0,
1] of laboratories prots. Since prots are equal to P, the policy-
1.0
holders expected utility becomes:
0.5 full insurance 
EU = (1 )u(w  + P) +  u (w  aP + P) .
Q
0.0
0.0 0.1 0.2 0.3 0.4 0.5 0.6 0.7 0.8 0.9 1.0 The only change with respect to the previous section is the change
WID
of net income due to prot redistribution (the equilibrium demand
Fig. 1. Optimal insurance scheme with xed supply. for drugs is unchanged so that the price P and the premium  remain
the same as well as the WID). We now show that maximal reference
other words, full reference pricing allows the supplier moral hazard pricing still applies. To see this, note that r only matters when there
effect to be cancelled out. is some insurance, i . e . when a < 1 . Then, using  = (1 a)(1 + )P,
Given that the price is xed at P = , the problem is reduced to we can write the net income in the two states as:
a standard insurance problem. The slope of the objective function wh = w (1 +  )P + (1 + )aP,
with respect to a is:
ws = w (1 +  )P + ((1 + ) 1) aP.
EU 
= EU (1 + )P u (ws ) P, (6) Now suppose that r < 1 and a < 1 . For given level of aP, the regula-
a Q
  tor can reduce P by increasing r and raise a. This change increases
where EU = (1 ) u (wh ) + /Q u (ws ) is the policyholders the net income in both states of nature and thus unambiguously
expected marginal utility of income. Rearranging (6) for a solution increases social welfare. It follows that at the optimal policy, either
with positive insurance coverage gives: r = 1 or a = 1 (in which case r = 1 is also optimal). Given that the
u (ws ) optimal strength of reference pricing is r = 1, the price is at its com-
EU (= 0 if a > 0). (7) petitive level P = . Thus the choice of copayment does not affect the
Q (1 + )
price P but only the level of insurance. As there is no wealth effect
The optimal policy is then characterized in the following proposi- in this model, redistribution of prots  amounts to change the
tion: wealth w to w +  that has no effect on the optimal copayment
Proposition 2. For a xed diversity of treatments N, the policy. More formally, we have
optimal strength of reference pricing is r = 1 and the copay-
EU EU
ment rate is non-decreasing in the loading factor . If  0 =
( 1) (1 ) / (1  + ), full insurance is optimal for all N.17 If
 > 0 , there exists N such that no insurance is optimal if diversity N
is such that N N, while the optimal copayment rate a is implicitly
determined by (7) if N < N.
Proof. See Appendix B.
For a small loading factor , the increase of the premium is not
sufcient to require a positive copayment rate. This is due to the fact
that the side effects render the marginal utility of income depend-
ent on the state, and in particular the marginal utility larger when
unwell.18 For  larger than 0 , the risk when N is large is not suf-
cient to grant insurance. Notice that when N is large, the price is
close to zero and the average side effect is also close to zero, but
there may be some residual risk due to the direct consequence of
illness on utility (as measured by ). Thus when  is small, the opti-
mal policy is to provide no insurance at all. Partial insurance occurs
when the supply of drugs is insufcient and in this case, the opti-
mal copayment rate, determined by condition (7), is the result of a
trade-off between risk pooling (ex-ante efciency) and an increase
in the size of the risk (ex-post efciency).19
The proposition is illustrated in Fig. 1 which represents the
choice in the space (Q (N) , ).20 There are two curves such that
full insurance is optimal below the lower curve while no insurance
is optimal above the upper one. The two curves coincide at zero
17
Of course, if  < 1, this cases never applies.
18
In this case a policy that would provide over-insurance would be optimal but
this option is ruled out.
19
See Geoffard (2006) for a complete analysis dealing with this trade-off in a
context where ex-post moral hazard is due to a quantity effect.
20
It is drawn for  = 1,  = 1.5,  = 1/3 and = 1.2 . Thus the diversity index is Q = 0.5
for N = 3.
|>0 = u () |=0 .
a a
We thus conclude that the optimal policy is the same as with no
prot redistribution. While our conclusion for reference pricing
is very robust, the conclusion for copayment relies on the utility
function.
4. Endogenous drugs diversity
In this section, we endogenize the market structure in the thera-
peutic class, and we focus on the free-entry equilibrium that occurs
in the long run (stage 2 of the timing presented in Section 2).
Laboratories enter into the market as long as the prots earned
in the drug market outweigh a uniform sunk cost k generated by
research and development activities.21 For conciseness we ignore
the fact that N is an integer and treat N as a continuous variable (see
Footnote 23 for more details).
Equalizing the sunk cost to equilibrium prots given by Eq. (5),
the diversity for a free-entry equilibrium is implicitly given by (up
to the integer):
 (N)
k= . (8)
N
Accounting for endogeneity of supply changes the nature of the
analysis compared to the case with xed supply. Indeed, when N
21
This assumption ts well with reality, as incremental innovation has often been
generated by unintentional research aimed towards producing breakthrough prod-
ucts (see Wertheimer et al., 2001). One can however argue that, starting from a
certain number of drugs existing in a therapeutic class, the xed cost of innovation is
decreasing since me-too drugs enter the market with less clinical trial experiments
(see Frothingham, 2004). A decreasing cost function would ultimately reinforce our
result stating that less than maximal reference pricing would be optimal.
 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063
is xed, the only scope for price policy is cost containment. Endo-
genizing the supply then creates a linkage between total expenses
and diversity. The free-entry condition implies an increasing rela-
tionship between diversity of treatment and expenses:
P = kN.
While the health policy may raise diversity, it has to account for
this cost. This suggests that we view the choice of health policy as
the joint choice of insurance, and diversity of treatment with a cost
function that is determined by the free-entry condition.
Given that = a + (1 a)r 1, the minimal number of labora-
tories entering the market is N given implicitly by the free entry
condition (8) with = 1 or:

N (N 1)
, (9)
k where EU is dened as in Eq. (6). Using Eq. (7) and u = u, this
where we used (N) = / (N 1). Recall that Q is dened only for
N > . To ensure this condition we assume (dening N = + if
 0 ):
Assumption 2. N 2 and  < N < N.
Assumption 2 guarantees that there are at least two rms in
the market and the WID is well dened (Q > 0). Moreover, the
condition N < N ensures that there will be some insurance in
equilibrium, i.e. the copayment will be less than 1.22
In order to provide more intuitions about optimal reference pri-
cing, we proceed in two steps. First, for any given copayment rate
a, we determine the optimal strength of reference pricing. In a sec-
ond step, we determine the joint optimal choice of copayment and
reference pricing.
4.1. Reference pricing and diversity for xed copayment
Even though we consider this step in order to more easily decon-
struct the different effects at work, it is worth stressing that such a
situation may occur when the copayment rate is set by another reg-
ulator, or at a high level of aggregation of pathologies/therapeutic
class. For a given copayment rate, the strength of reference pricing
determines the equilibrium price for any level of entry. Increasing r
results in lower equilibrium prices and consequently in less entry.
Hence, the optimal policy reects a trade-off between the total cost
of treatment and the WID.
Notice that the free-entry equilibrium (8) induces a decreasing
relationship between r and N. Hence, for a given a, choosing r is
equivalent to choosing the diversity N under the relations:
 (N) /kN a
0r= 1 and P = kN.
1a
Thus, the problem reduces to the choice of diversity within a fea-
sible range, and the optimal policy is the solution of the following
program:23

 kN
 of providing insurance helps to reduce the gap between marginal
maxEU = (1 )u (wh ) + u wh a , utilities. On the other hand, for low levels of differentiation and risk
N Q 
s.t. wh = w (1 + )(1 a)kN,
(10)
N N,
akN  (N) 0.
22
To see this, note that N = N if a = 1. The condition implies that a < 1 is optimal
for N = N, and therefore it is impossible that a = 1 is an optimal choice.
23
When N is treated as a discrete variable, several values of r generate the same
N . Clearly it is optimal to choose the maximal strength compatible with N, hence r
is such that P = kN/ .The problem is thus identical but restricted to integer values
of N . The comparative statics are the same.
55
By denition of N, we have that for a < 1 : kN =  (N) > akN.
Therefore the set of admissible N is a non-empty compact inter-
val so that an optimal diversity exists. Denote by ra the strength
of reference pricing as a function of a. When r = 1, the level of entry
is always equal to N. Hence, at any copayment level, it is possible
to implement the minimal WID at the minimal cost. One may raise
diversity above this level by reducing the strength r of reference pri-
cing. However, this would imply some more entry costs. The rst
question is thus whether or not it is optimal to minimize cost and to
rely only on insurance, or to accommodate some cost increase. The
gain of raising diversity is captured by the slope of EU with respect
to N which is equal to:
EU  Q 
= 2 u(ws ) k (1 + ) (1 a) EU aku (ws ) ,
N Q N Q
slope can be rewritten as:
    EU
EU  Q  1a
= k u (ws ) |N , (11)
N Q N Q N a
where EU/a|N is given by the LHS of Eq. (6). The rst term in
the RHS of Eq. (11) captures the trade-off between a higher WID
(associated with more entry) and the cost of entry. The last term
captures the extent to which reference pricing can correct for insuf-
cient or excessive copayment to provide some insurance. Indeed,
when there is insufcient coverage (i.e. EU/a|N < 0), additional
entry acts as a substitute to insurance since it reduces the difference
of marginal utilities between healthy and unwell individuals. Thus,
entry is more valuable when the copayment is too large. Whether
it is optimal to reduce the strength of reference pricing to induce
some entry depends on the level of the copayment rate.
Proposition 3. Suppose the copayment rate a is xed.
If  1, the reference pricing strength ra < 1 and the optimal
diversity N is above N for a (a, 1) where a < 1. Moreover ra < 1
for all a if full insurance is optimal in the short-run at N = N.
If  < 1, either diversity N is minimal for all a (i.e. ra = 1), or ra < 1
for a (a, a) with 0 a < a < 1.
Proof. See Appendix C.
When risk aversion and the side effects marginal losses are not
too small, it is optimal to induce entry by relaxing the intensity of
competition at least for large levels of coverage. This conclusion
contrasts with the welfare analysis of entry in standard models
of horizontal differentiation, which conclude that there is always
excessive diversity at an unregulated free-entry equilibrium (e.g.
see Anderson et al., 1992). The key difference between our setup
and these models is that diversity has an additional benet in a
pharmaceutical sector: improving the welfare of unwell individ-
uals improves the overall insurance performance. This is the case
in particular when full insurance is optimal, as alternative means
aversion, the insurance benet is not so important and the tradi-
tional result of excessive entry may apply. In this case, relaxing the
strength of reference pricing may be desirable only for intermediate
levels of coverage, or not at all.
We show in Appendix D that the objective is quasi-concave
which allows us to perform simple comparative statics.
Proposition 4. For a xed copayment a, the optimal diversity N:
i) is non-decreasing in the copayment rate a if there is excessive cov-
erage, i.e. if EU/a 0;
ii) is non-increasing in the loading factor  and non-decreasing in the
risk aversion  and the adverse effect risk .
56 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063
Proof. See Appendix D.
Increasing the copayment has two effects. It reduces the level
of insurance and the extent of the suppliers moral hazard. Since
entry induces some indirect form of insurance through a higher
WID, the two effects make entry more attractive if there is not
enough insurance. In such a case, the optimal diversity of treat-
ment increases in the copayment. However, if there is excessive
insurance, the two effects work in opposite directions and no con-
clusion can be derived. The other results outlined in Proposition 4
are rather more intuitive, and state that the policy becomes more
oriented toward cost containment when the cost of public fund
increases, risk aversion decreases or the differentiation parameter
decreases.
4.2. Optimal policy mix
We now consider the complete set of instruments and discuss
the optimal choice of coverage and strength of reference pricing,
(a , r ). At the optimum, r = ra which exists for all a . The value of
the program (10) is continuous in a so that an optimal copayment a*
exists. As already mentioned, Assumption 2 rules out the possibility
of no insurance. Hence, we only need to consider full insurance or
partial insurance. Before determining which regime prevails, we
discuss the strength of reference pricing in each regime.
Proposition 3 implies that for  larger than 1, a full insurance
policy is necessarily associated with some incentive to entry and
r* < 1 . To see this, note that if a* = 0 and r* = 1, then N = N. But this
would contradict Proposition 3 which implies that r0 < 1 (as full
insurance must be optimal for a* = 0).
Let us now consider the case where partial insurance is optimal.
The rst-order conditions with respect to N give (using Eqs. (6) and
(11) for EU/a = 0):
u (ws )
EU = ; (12)
Q (1 + )
 Q
k (= if N > N). (13)
Q N
Eq. (12) describes the same trade-off between risk pooling and size
of the risk as in the benchmark section, but obviously taken at a
different value of N. The second equation captures the trade-off
between the social benets and costs generated by the entry of a
new drug, respectively described by the LHS and the RHS of Eq. (13).
The social marginal cost is simply the welfare effect of an additional
sunk cost, accounting for the optimal allocation of this cost across
health states. The social marginal benet captures the increase in
the WID through the marginal reduction of average adverse effects
generated by a new drug entry. To follow, we analyze whether or
not the optimal diversity of treatment is interior, i.e. whether or not
there is less than maximal reference pricing.
Recall that the minimal diversity N is given implicitly by the
free-entry condition k =  (N) and is independent of the level
of absolute risk aversion . Substituting (13) in (12), a necessary
condition for an interior solution for r is:
 Q (N)
k > 0. (14)
Q (N) N
We can now state our main proposition:
Proposition 5. The optimal copayment rate a* is non-decreasing in
the loading factor  and diversity N* is non-increasing in .
i) If  1, then the reference pricing strength r* < 1 and diversity is
above N for all ;
ii) If  < 1, then the reference pricing strength is maximal, i.e. r* = 1
when the copayment rate a* > 0.
Proof. See Appendix E.
Increasing  always leads to less coverage and less diversity as
both are costly. Still, whenever  is large enough, it is optimal to
raise diversity above the minimum by relaxing the strength of refer-
ence pricing and inducing some entry. Conversely, if  is smaller
than 1, cost minimization tends to prevail and minimal diversity
will be chosen for  not too small.
As the copayment rate is non-decreasing in , the proposition
also reveals that there exists a threshold  such that there is par-
tial insurance if  > , and in this case the policy departs from
cost minimization when the combined value of risk aversion and
disutility from side effects is large enough.
When  < , the optimal policy coincides with the policy ana-
lyzed in the previous section for a = 0. On the range  > , we
can analyze how the optimal level of copayment and the optimal
diversity of treatment offered in the therapeutic class varies with
preferences and the xed cost.
Proposition 6. For  >  (a > 0) and  > 1 :
i) An increase in the risk aversion  or the adverse effect risk 
increases diversity N and decreases the copayment rate a;
ii) An increase in the xed cost k decreases both diversity N and the
copayment a;
iii) An increase in the marginal utility of income when unwell 
decreases the copayment rate a but does not affect diversity N.
Proof. See Appendix F.
 The results
 rst state that more risk aversion () or more risk
 and  imply more coverage. This is in line with standard insur-
ance analysis. The result on k follows the same logic: as providing a
given level of WID becomes more costly, the treatment expense
increases and the optimal policy compensates for the risk by a
reduction in the copayment rate. The result i) shows that insurance
and the diversity of treatment are complementary instruments to
accommodate an increase in risk aversion or in the heterogeneity of
side-effects. However, result ii) states that they evolve as a substi-
tute when the cost of entry increases. The last result is interesting as
it states that the optimal diversity is independent of . What mat-
ters is the marginal increase in utility that diversity of treatment
induces, and not the absolute value of disutility.
4.3. Numerical illustration
In order to illustrate the results of this section we run different
simulations. Our goal is to illustrate the trade-off between costs
and social benets from an additional entry of a drug. Remember
that the social cost is the xed cost k while the social benet comes
from the increase in the WID. We thus run simulations for different
values of k, and for each value of k we run three simulations depend-
ing upon the value taken by , the latter being positively related to
the WID. The other parameters are xed to  = 0.8,  = 0.5,  = 0.3,
w = 1 and  = 0.05. For each scenario, we report the long run val-
ues of n* , a* and r* of Section 4.2. We also provide the value a above
which it is optimal to introduce one additional entry as described
in Proposition 3. Recall that it is optimal to induce an additional
entry when the exogenous level of copayment is above a.
 Table 1 presents the results. In a rst scenario, we have k = k1 =
  /2. As described in (9), this is the highest possible xed cost
that implies a minimum number of rms exactly equal to N = 2
when reference pricing is at its maximum  strength.
 The value of
k is then successively decreased to k2 =   /2.5 and to k3 =
 
  /4. These latter values still imply a minimum number of
rms equal to 2 (the maximum value  of kimplying a minimum
number of rms equal to 3 is equal to   /6). For each of these
 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063 57

Table 1
Examples with different values of k and .
     
k k1 =   /2 k2 =   /2.5 k3 =   /4

 1.4 1.8 1.9 1.4 1.8 1.9 1.4 1.8 1.9

N* 2 2 3 2 3 3 3 3 3
a* 0.13 0 0.12 0.17 0.18 0.14 0.60 0.29 0.23
r* 1 1 0.12 1 0.28 0.31 0.16 0.53 0.56
a 0.44 0.10 0 0.55 0 0 0.08 0 0

scenario, we report the values of interest variables corresponding in the extent to which this possibility implies more or less diversity
to three different values of risk aversion as measured by . In all at the optimum.
the scenarios, we have  > 1. Dene 0 = / 0 as the measure of the opportunity benet of
By Proposition 5, entry is thus desirable at the margin. However forgoing a treatment. The symmetric equilibrium policyholder js
N is discrete so that an additional entry may not be desirable for expected utility now becomes:
 low enough. When k is at its maximum value compatible with a    
minimum number of drugs equal to 2 (k = k1 ), an additional entry  i
EU = E u wh min minxji + p; x + (1 )u(wh ),
is desirable when  1.9. Below this level, the xed cost is too high iI 0 0
to induce one additional entry. As the xed cost decreases from
where the agent now has access to one more option.
k2 to k3 however, entry becomes desirable for lower values of risk
Suppose that all agents face an out-of-pocket price p except for
aversion.
drug i I. The demand function for drug i is given by the mass of
In all the scenarios, the long-run copayment rate a* is always
unwell individuals who prefer drug i over the others and whose
lower than 1. When an additional entry is desirable (N = 3), a*
severity of illness justies a treatment:
decreases with risk aversion as shown in Proposition 6. How-
 p p N1   p 
ever, when the desirable number of drugs jumps from 2 to 3, the i i
Di =  f (x) 1 F +x 1F 0 + 0 x dx.
copayment rate increases. This is due to the effect of entry of equi-  
0
librium prices. As N jumps from 2 to 3, the price (and the premium)
decreases so that insurance motives becomes less important. We then have:
 
Consider now the reference pricing mechanism implementing  (N 1) (pi p) + 0 pi
Di = exp .
the long-run solution. For each values of k, the value of r jumps N + 0 
from 1 to lower values when the desirable number of rms passes
Using the denition of the reference price mechanism and = a +
from 2 to 3. For example, when k = k2 , the value of r changes from
r (1 a), one thus has:
1 (the most stringent policy) to a value of approximately 1/3 of
 
the maximal price when  increases from 1.4 to 1.9. Alternatively,  (N 1) (Pi P) + 0 (Pi (1 a) min[Pi , R])
Di = exp ,
for a given value of risk aversion of 1.9, the reference price policy N + 0 
becomes less stringent as the xed cost decreases. It increases from
where R is given by (4). The demand function is thus dened by:
12% to 56% of the maximal price when k increases from k1 to k3  
since this is because it is easier to induce entry as k decreases. Now  (N 1) (Pi P) + 0 aP i
Di (Pi , P) = exp if Pi P;
consider the case where k = k3 in which an additional entry is always N + 0 
desirable. One can see that the value of r increases from 16% to  
 (N 1 + 0 ) (Pi P) + 0 aP
56% as  increases from 1.4 to 1.9. This is due to the fact that the Di (Pi , P) = exp if Pi P.
N + 0 
copayment rate decreases with the risk aversion parameter. For a
given value of r, it means that prices and thus prots are higher as As expected, with an elastic aggregate demand for drugs, the refer-
 increases. The value of r implementing a number of rms equal ence pricing system with r > 0 yields a kink in the residual demand
to 3 thus becomes lower. that a pharmaceutical rm faces, as pointed out in Danzon (2001).
Let us nally comment the value of a and remember that This kink implies that the demand is less elastic for a price Pi below
this variable reects the threshold value above which one addi- the market price than for a price Pi above. As a result there exist
tional entry is desirable when a is exogenously set. As shown by multiple equilibria.
Proposition 3, entry is always desirable only when a > a. In par- Proposition 7. Suppose that the opportunity benet of forgoing a
ticular, entry is always desirable when a is exogenously set to 0, i.e. treatment 0 >0. Then
the case when there is full insurance. It is worth to notice that the
 there exists a continuum of symmetric price
equilibria P P, P where
value a is non increasing with the risk aversion for each level of k.
 
P= and P =
(N 1 + 0 ) (N 1) + 0 a
5. Extensions
with = a + (1 a) r.
In this section, we explore two alternative issues related to our Proof. See Appendix G.
set-up. First, we develop the model when the drug market may not
be fully covered. Second, we check whether or not our main results We can measure the size of the equilibrium set by the following
hold with preferences in which side effects enter separately in the ratio:
utility function. P P 0 r (1 a)
= .
P N 1 + 0 a + r (1 a)
5.1. Partially covered market This is increasing with 0 and decreasing with N. Hence, the range
increases when the treatments become less useful either because
This section extends the previous analysis to the case where the option of no treatment becomes less detrimental or because
individuals may forgo any treatment. We are particularly interested the diversity of drugs decreases. The range also increases with the
58 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063

The equilibrium number of rms is thus lower when the opportu-

nity benet of forgoing a treatment is high. Using this zero prot
free entry condition, the actuarial premium must just cover the
total cost of entry so that one has  = (1 + )Nk. It follows that the
socially optimal level of r is obtained by maximizing welfare (15)
subject to wh = w (1 + )kN and condition (18).

Corollary 1. Suppose that the copayment rate a = 0, then the optimal

diversity N is decreasing with the opportunity benet of forgoing a
treatment 0 .

Proof. See Appendix H.

As the opportunity cost of forgoing a treatment decreases, entry

becomes less desirable. Note that all the comparative statics in
Proposition 4 still hold.

Fig. 2. Price equilibria with elastic demand. 5.2. Separable utility function

strength of reference pricing r. Fig. 2 illustrates the price equilibria Until now, we have used a particular specication of the prefer-
in the (r, P) space. Every price between the upper and lower curves ences against side effects. It may be asked whether our main result,
are possible price equilibria in the market with an elastic demand. namely that less than maximal reference pricing is optimal, still
We now turn to the welfare analysis. In this model, we have holds under an alternative model. To show this, we now use a
to consider three situations for individuals. An individual may be model in which the side effect of a drug enters separately in the
healthy, unwell with treatment or unwell without treatment. This utility function. Formally, we now assume that the policyholder js
implies that we adjust our notion of index of diversity and that we expected utility when consuming drug i is:
introduce a WID that accounts for the probability of not using a     
U =  u wsi v xji + (1 )u(wh ),
treatment. Considering that policyholders preferences are repre-
sented by a CARA utility function and that Assumption 1 holds, one  
can rewrite the expected utility function as:
where v (x) = exp x and  represents the absolute risk aver-
1 1
 sion parameter with respect to adverse effects. Keeping the
EU =  u (ws ) + u (wh ) + (1 ) u (wh ) , (15) assumption that xji follows an i.i.d. exponential distribution, the
Q1 Q0
expected utility function now writes:24
where:
   
1 N 0 aP 1
= exp , (16) EU =  u (ws ) + (1 )u(wh ),
Q1 N + 0    (N)
Q
1 0
 NaP 
= exp . (17) at a symmetric equilibrium where
Q0 N + 0  
Q1 is the WID accounting both for the probability of consuming  (N) = 1 
Q .
a drug and the expected side effect conditional on the fact that N
individuals consume a drug. This WID depends on prices as this Following the same reasoning as in Section 3.1, we can state the
affects the decision to consume. In contrast, Q0 measures the value following proposition:
of the new option of not consuming drugs when the severity of ill-
ness is low. Contrary to Q1 , it is obviously decreasing in 0 since it Proposition 8. For a given copayment rate a, there exists a reference
decreases the disutility of not consuming any drug. As in the pre- pricing strength r inducing a free entry equilibrium with N drugs if
ceding section, both Q1 and Q0 are decreasing (resp. increasing) in
kN (N 1) N
 kN

the risk aversion parameter (the treatment diversity).   exp  ( (1 + ) (1 a) + a)

Notice that both Q1 and Q0 are functions of aP. This reects the  N + 
effect of the out-of-pocket payment on the probability of using
a treatment and on the conditional expected utility. Indeed as p
 exp 
(w) exp (w)
, ;
increases, it is less likely that the patient uses a treatment. This in  a
turn also implies that a patient using a treatment supports lower
side effects on average. the price is then P = kN/.
To conduct our policy analysis we need to make an assumption
Proof. See Appendix I.
on the relevant price that emerges given the policy mix (a, r). Given
that we are interested by, among other things, the conditions under The long run equilibrium is more complex than that described
which the optimal policy involves r < 1 under some insurance a < 1, in Eq. (8). This is due to the fact that an increase in the price paid by
for conciseness, we develop the analysis only for a = 0 and P = P, the patients now has an income effect: increasing the price decreases
largest possible equilibrium price. Then, we have P = / (N 1) r the disposable income when unwell which increases the marginal
and the average demand for a drug is D = / (N + 0 ). In the long utility of income. The semi-elasticity of residual demand is there-
run, the free-entry condition k = PD/N yields the diversity as an fore higher with this specication of preferences.
implicit function r as follows:


(N + 0 ) (N 1) r = . (18)
k 24
We assume that  < N, which is the counterpart of Assumption 1.
 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063 59

The problem of the insurer thus now amounts to solve: introducing sequential entry and delays in order to take into
  account the impact of health insurance and drug price regulation
1 on drugs introduction delay. Indeed, these delays of introduction
maxEU = (1 )u (wh ) +  u (ws ) ,
N,a  (N)
Q may have a huge impact in terms of welfare. All these aspects are
kN left for future research.
s.t. wh = w (1 + )(1 a)kN, ws = wh a ,

  Appendix A. Proof of Proposition 1
kN (N 1) kN exp (w)
exp  ( (1 + ) (1 a) + a) (r 1),
  
  The equilibrium condition is
kN (N 1) kN exp (w)
exp  ( (1 + ) (1 a) + a) (r 0).
    D
+ (a + (1 a)r) P = 0. (19)
N pi
Following the same reasoning as in Section 4, we can state the
following: Given prices pi and p, the probability that drug i is chosen by a
patient j is determined by
Proposition 9. Assuming an interior solution for a* , the optimal
policy mix veries r* < 1 if  1.
  
Pr pi + xji = min p +  xjl | xji
l
Proof. See Appendix J.  
l pi p
One can see that the main result still holds with this kind of = Pr min xj + xji | xji
l=
/ j 
preference. Everything works as if the risk aversion parameter with
 p p N1
respect to side effects risks  corresponds to the product of the i
= 1F + xji .
risk aversion parameter and the side effects marginal loss  in 
the previous model. A noticeable difference here is that  1 is a
sufcient (but not necessary) condition to have less than maximal Integrating over all possible realizations of xji , the expected demand
reference pricing. of drug i is thus:

 p p N1
i
6. Conclusive remarks Di =  f (x) 1 F +x dx
0


Our analysis sheds light on various mechanisms and underlines  N1

 
= exp (pi p) .
several results related to drug price regulation, such as reference N 
pricing. First, in the short run, it is always optimal for the health
It yields
insurer to implement the strongest policy in order to lower health
care expenditure. In the long run, the regulator takes into account D 
  N1
the drug entry process and the overall quality offered to patients. = .
pi N 
Diversity of treatment is an indirect form of insurance that may
substitute for or complement monetary insurance. In terms of pol- Substituting this expression in (19) thus yields:
icy making, our model points out the importance of having good 1 
estimations of the impact of prices on the supply and of the social P= .
a + (1 a)r N 1
value of drugs diversity (which depends positively on the level of
policyholders risk aversion) before deciding on a reference pricing
rule. Appendix B. Proof of Proposition 2
Our model focused on a specic price regulation mechanism,
namely reference pricing. Our main message would not change if For a xed price P = (N) the slope EU/a increases in 
we introduce other price mechanism controls such as price caps or implying that the optimal copayment is non-decreasing in  . The
exogenous reference pricing as it is done in Europe. The important rst order condition at a = 0 can be rewritten (using wh = ws =
aspect of the analysis is to show that a lenient price control can w (1 + )P)
be optimal in a context where drugs diversity is valued by risk
averse patients. One important and neglected aspect of our analysis  u (w )
(1 ) u (w ) +  u (w ) ,
is that we do not model vertical differentiation so that one can Q (N) Q (N) (1 + )
not see how the policy may affect breakthrough innovations. As or
suggested by Brekke et al. (2007), this would allow us to compare
 Q (N)
more explicitly the two modalities of reference pricing, i.e. generic  (1 ) . (20)
(1 ) Q (N) + 
versus therapeutic. This would also be useful to take into account
the situation where brand-name drugs and generic drugs compete Similarly, we have a = 1 if
in the same reference pricing class.25 We however expect that long-
 u (w P)
term entry considerations would still mitigate the benets of tough (1 ) u (w) +  u (w P) ,
price regulation as long as reference pricing hurts the prots of all
Q (N) Q (N) (1 + )
active producers in the market. or
Finally, we have determined the optimal public regulation in  exp ( (N)) Q (N)
short and long run equilibrium analysis. Nevertheless, as in Bardey  (1 ) . (21)
(1 ) Q (N) +  exp ( (N))
et al. (2010), it would be useful to complete this analysis by
The RHS of (20) and (21) decrease in N and take value 0 when Q = 1.
Hence, (20) holds for all Q if  0 . When (1 ) / >  > 0 on
25
This may occur when dates of patent rights expiration differ within a given
the contrary, the condition (21) is satised if N N where N solves
therapeutic class, or when the sole brand-name drug stays in the market after patent (21) with equality. Finally, when  (1 ) /, the condition holds
expiration. for all N.
60 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063

Appendix C. Proof of Proposition 3

  
Hence, we have 2/  N  + k (2a + (1 a) (1 + )) < 0
so that the rst term in brackets in the RHS of (23) is nega-
We rst prove that the objective is quasi-concave in Lemma 1 tive hence /N|=0 < 0 at any point where  = 0. This implies
and then prove Proposition 3. that 2 EU/N2 < 0 when EU/N = 0 so that the objective is quasi-
Lemma 1. The objective is quasi-concave in N. concave.

Proof. The slope By quasi-concavity, r < 1 if and only if the RHS of (22) is positive

EU
 i.e. if:
=  u(ws ) k (1 + ) (1 a)   Q (N) 
N N Q k
  
  Q 2 N Q (N)
(1 ) u (wh ) + u (ws ) aku (ws )  kN

Q Q
(1 ) exp a
can be written as u (ws )  where  
(1 a) k (1 + ) Q (N) > 0
 
  +
= k (a + (1 a) (1 + )) k(1 + ) Q (N)
 N Q Q (24)
kN
 
(1 a) (1 ) exp a . (22) where the LHS of (24) is concave in a. This implies that r < 1 on an

interval (a, a).
We have One has
 


= k (a + (1 a) (1 + )) +  Q  
N  N 2 Q N Q =   ,
Q 2 N N N  Q
k
 kN
  
a (1 a) k (1 ) exp a (1 + ) . so that
 
 
Hence:   Q (N)   k 
k = N +    .
 

  Q (N)2 N Q (N) kN N  Q (N)
|=0 = k (a + (1 a) (1 + )) +
N  N 2 Q N Q But the free-entry condition yields
  
k    
a k (a + (1 a) (1 + )) . kN = 1 = N.
  N Q Q (N 1) kN
Using Therefore, we obtain

     k
  Q (N)  
Q
=  2 =  ,  k = 1 +  .
N N  N N  Q  Q (N) 2 N Q (N) N  Q (N)

 2 2 
= Substituting this last expression in (24) it yields r < 1 if
 3 =  2 Q ,
N 2 Q N  N N     1 a  EU
 1 
k > |N , (25)
we obtain N  Q (N) Nu (ws ) a
  2  
|=0 =
  2 Q + k (a + (1 a) (1 + ))   + where EU |N is evaluated at r = 1.
N N N  N N  Q a
           
1a EU 1a (1 ) u (wh )  
k  |N = +  (1 + )  N
a     k (a + (1 a) (1 + )) Q , Nu (ws ) a N u (ws ) Q Q
  N N  Q           
1a  
= (1 ) exp a N +  (1 + )  N ,
so N Q Q
2

Q     
which is decreasing with a when positive.
| =  N  Suppose that  > 1. Then as 1a EU | decreases when
 N =0 Nu (ws ) a N
+k (2a + (1 a) (1 + )) positive, condition (25) holds on an interval (a, 1) . This interval is
 k non-empty as the RHS vanishes at a = 1. At a = 0, the condition is
  a k (a + (1 a) (1 + )) . (23) veried if EU/a|N,a=0 < 0 (i.e. full insurance is optimal), in which
N N  
case a = 0.
Notice that  = 0 implies that 
For  = 1, the same conclusion  holds because our assumption
 that N < N ensures that 1a EU | <0 for a close to 1.
  Nu (ws ) a N
k (a + (1 a) (1 + )) =
 N Q Q  If <1,EUthe condition is violated at a
1. The function
  1a
|N is convex. Thus, the condition holds on some inter-
Nu (ws )
  a
  k (a + (1 a) (1 + )) >0 val (a, a) which may be empty.
N N  Q

implying (because N > ): Appendix D. Proof of Proposition 4


k (a + (1 a) (1 + )) <  . At any point where Na is interior, the effect of a parameter on
 N 
Na is given by the sign of the derivative of  with respect to this
 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063 61

parameter. Differentiating the LHS of (22) with respect to a and  because a > 0 and N = 0. But this contradicts the assumption
respectively yields: made that a* decreases in . Thus, a* is non-decreasing with  and
a* = 0 for  . Notice also that at an interior solution

 1
 EU
= |N
dN
a Nu (ws ) a = 0.
kN kN
  d
+ (1 a) k (1 ) exp a (1 + ), Hence, N* is constant in  when it is larger than N on  > . Since
 
    N* decreases in  for   from Proposition 4, we conclude that
 kN  notice that N (N) =
= (1 a) k (1 ) exp a + < 0.  everywhere.
it is non-increasing   Moreover,

  Q kN (N 1) kN N  = kN  1 is positive if and only if
 > 1. Hence r* < 1 for all a* if  > 1. Finally, if  < 1, r* = 1 if a* > 0.
so that /a > 0 if EU/a|N > 0 and / < 0. Then, using

   Appendix F. Proof of Proposition 6
=  
N Q N N  Q
First, from Eq. (27), it is immediate that N* increases in ,  and
we have , and N* decreases in k . Applying the Cramers rule yields

Q  a aN
 =   (a + (1 a) (1 + )) k
| |
 N N  da N NN
  N  =
kN d det
(1 a) k (1 ) exp a (1 + )
 N where


which is increasing in  and . Thus / > 0 and / > 0 at kN 
a = (1 + ) (1 ) exp a (1 + a )
=0.  N


kN 
Appendix E. Proof of Proposition 5 aN = (1 + ) (1 ) exp a (1 a )
 N 2

Suppose a* decreases in  for some value. Then it must be the N = kN

case that a* > 0 . Moreover N > N because a* is non-decreasing in  
NN = k 2N 
 for a xed N from Proposition 2. At such a point, the rst-order
conditions are: so that
 

u (ws ) da
2 (1 + ) (1 ) k exp a kN

 
Q = N  + a N < 0.
EU (.) = 0, d N det
1+

 
Moreover,
 u(ws ) k u (ws ) = 0,
N Q Q a aN
| |

which can be rewritten (up to a proportional factor) as: da N NN
=
    d det
N  kN

a = (1 + ) (1 ) exp a where
N  

 kN
+  (1 + ) 1 = 0, (26) a = (1 + ) (1 ) exp a
N 

N =  + kN 

 
N =  kN N  = 0. (27) so that
 
  da
 (1 + ) (1 ) exp a kN

We have at a = N = 0 (using N  k/ = /N for aN ): =
d N det
   
  k kN
  
 
aa = (1 + ) (1 ) N   exp a < 0, k 2N  +  + kN  (1 a ) .
  N


 
kN  Using  = kN N  , we obtain
aN = (1 + ) (1 ) exp a (1 a ) > 0,
 N 2  
da
 (1 + ) (1 ) exp a kN

 
Na = 0, NN = 2kN + k < 0. = k N (1 + a )  < 0.
d N det
First, notice that the determinant of the modied Hessian is Finally, we have
det = aa NN > 0, hence the matrix is semi-denite negative. Dif-
ferentiation of a* with respect to  gives ak aN
| |

da Nk NN
a aN = < 0,
| | dk det
a N NN a NN as ak , NN and Nk are negative while aN is positive. Similarly,
= = > 0,
 det det a < 0 and N = 0 imply da* /d < 0.
62 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063

Appendix G. Proof of Proposition 7 Given prices pi and p, the probability that drug i is chosen by a
patient j is determined by
Since     
i max
Pr u wsi exp x u (ws ) exp  xjl | xji
j
l
Pi arg max Pi Di (Pi , P)
  
 
every price Pi that satisfy = Pr min exp  xjl i | xi
u (ws ) u wsi + exp xj j
l=
/ j

Di (Pi , P)     N1
Di (Pi , P) + Pi |Pi P 0for Pi = P i
ln u (ws ) u wsi + exp x
Pi j
= 1 F .
Di (Pi , P) 
Di (Pi , P) + Pi |Pi P 0for Pi = P
Pi
Integrating over all possible realizations of xji , the expected demand
are candidates for symmetric equilibria. Since of drug i is thus:

      N1
Di (P, P)
ln u (ws ) u wsi
+ exp x
Di + Pi |Pi P = 0 Di =  f (x) 1F dx
Pi 
0

leads to
N1

   
=  expx u (ws ) u wsi + exp x
 dx

Pi = 0
(N 1) + 0 a
Differentiating Di with respect to pi yields:
and Di  
=  (N 1) u wsi
Di (Pi , P) pi
Di + Pi |Pi P = 0
  N 1 1
Pi
 i
exp x
u (ws ) u ws + exp x  dx
leads to 0

 
 
x N+
 =  (N 1) u wsi exp dx
Pi =
(N 1) + 0 0
  1
  =  (N 1) u wsi  
and that Pi < Pi , any price belonging to the interval Pi , Pi = N + 
 
P, P are possible candidates for the symmetric price competition
game. so that the symmetric equilibrium price is given by:
 
N + 
Appendix H. Proof of Corollary 1 1
P exp [ ( (1 + ) (1 a) + a) P] exp (w) =
N (N 1)
The optimal diversity N solves
The long term equilibrium can thus be described by:
 N + 0
  
maxEU = u (wh ) + (1 ) u (wh ) N + 
N N + 0  1 exp (w)
P exp [ ( (1 + ) (1 a) + a) P] =
 (N 1) N 
s.t. wh = w (1 + )kN and (N + 0 ) (N 1)
k P = kN
The gain of raising diversity is now captured by: which yields:
   
kN kN (N 1) N exp (w)
EU  exp  ( (1 + ) (1 a) + a)   =
=   2 u (wh )   
N N+
N + 0 
 N + 0
 (28)
(1 + ) k (1 ) +  u (wh ) .
N + 0  It is easy to show that N is decreasing in [a, 1] which proves
Proposition 8.
Thus at any interior equilibrium one has
   Appendix J. Proof of Proposition 9
 (1 + ) k N + 0  N + 0 (1 )  = 0.

The LHS is decreasing with N and 0 , which shows that N is unique Assuming an interior solution for a one has:
and N/ 0 < 0 when N is interior. EU
= kN (1 + ) EU kNu (ws ) = 0. (29)
a
Appendix I. Proof of Proposition 8 Differentiating the objective function with respect to N yields:

The rst order condition for prot maximization of rm i is still EU Q

=  k (1 + ) (1 a) EU aku (ws ) . (30)
given by (19). N N
 D. Bardey et al. / Journal of Health Economics 47 (2016) 5063 63

Substituting (29) in (30), the gain from rising diversity above N is Brekke, K., Konigbauer, I., Straume, O.R., 2007. Reference pricing of pharmaceuticals.
given by: Journal of Health Economics 26, 613642.
Brekke, K., Grasdal, A.L., Holms, T.H., 2009. Regulation and pricing of pharmaceut-
EU Q icals: reference pricing or price cap regulation? European Economic Review 53,
|N =  ku (ws ) (31) 170185.
N N Brekke, K., Holms, T.H., Straume, O.R., 2011. Reference pricing, competition, and
pharmaceutical expenditures: theory and evidence from a natural experiment.
where N is implicitly given by (28) where = 1:
Journal of Public Economics 95, 624638.
N 2 (N 1)
 kN
 Danzon, P.M., 2001. Reference Pricing: Theory and Evidence. In: Lopez-Casasnovas,
  k exp  ( (1 + ) (1 a) + a) = exp (w) (32) G., Jonsson, B. (Eds.), Reference Pricing and Pharmaceutical Policy. Springer, New
 York.
 N + 
Danzon, P.M., 2012. Regulation of Price and Reimbursement for Pharmaceuticals,
The Oxford Handbook of The Economics of the Biopharmaceutical Industry.
Substituting (32) in (31) thus yields: Oxford University press, pp. 266301.
  Danzon, P., Ketchman, J., 2004. Reference pricing of pharmaceuticals for medicare:
evidence from Germany, the Netherlands and New Zealand. Frontiers in Health
N + 
EU  Policy Research 7 (2), bepress.
|N =   2 Di Masi, J.A., Paquette, C., 2004. The economics of follow-on drug research and
N N 2 (N 1) development. Pharmacoeconomics 22 (Supl 2), 114.
N  Finkelstein, A., Luttmer, E., Notowidigdo, M., 2013. What good is wealth without
   2   health? The effect of health on the marginal utility of consumption. Journal of
the European Economic Association 11, 221258.
N 3  1 +  N  Frothingham, R., 2004. Me-too products friend or foe? New England Journal of
=  2 Medicine 350 (20), 21002101.
Geoffard, P.-Y., 2006. Incentive and selection effects in health insurance. In: Jones,
N  N 2 (N 1)
A. (Ed.), Elgar Companion to Health Economics. Elgar.
Lakdawalla, D., Sood, N., 2009. Innovation and the welfare effects of public drug
which is positive if  1. insurance. Journal of Public Economics 93 (3-4), 541548.
Lakdawalla, D., Sood, N., 2013. Health insurance as a two-part pricing contract.
Journal of Public Economics 102, 112.
References Lopez-Casanovas, G., Puig-Junoy, J., 2000. Review of the literature on reference pri-
cing. Health Policy 54, 87123.
Anderson, S.P., de Palma, A., Nesterov, Y., 1995. Oligopolistic competition and the Miraldo, M., 2007. Reference Pricing Versus Co-Payment in the Pharmaceutical
optimal provision of products. Econometrica 63 (6), 12811301. Industry: Firms Pricing Strategies. Center for Health Economics, University of
Bardey, D., Bommier, A., Jullien, B., 2010. Retail price regulation and innovation: York, Research Paper 27.
reference pricing in the pharmaceutical industry. Journal of Health Economics Ruggeri, K., Nolte, E., 2013. Pharmaceutical Pricing: The Use of Reference Pricing.
29 (2), 303316. Rand Corporation Research Report Series.
Bardey, D., Jullien, B., Lozachmeur, J.-M., 2013. Health Insurance and Diversity of Wertheimer, A., Levy, R., OConnor, T., 2001. Too many drugs? The clinical and eco-
Treatment: A Policy Mix Perspective. In: CESifo Working Paper Series No. 4111. nomic value of incremental innovations. In: Irina Farquhar, Kent Summers, Alan
Brekke, K., Canta, C., Straume, O.R., 2015a. Reference pricing with endogenous Sorkin (Eds.), Investing in Health: The Social and Economic Benets of Health
generic entry. In: NHH Working Paper. Care Innovation (Research in Human Capital and Development, Volume 14).
Brekke, K., Canta, C., Straume, O.R., 2015b. Does Reference Pricing Drive Out Generic Emerald Group Publishing Limited, pp. 77118.
Competition in Pharmaceutical Markets? Evidence from a Policy Reform. In:
NIPE Working Paper.