European Journal of Operational Research 249 (2016) 1131–1138

Contents lists available at ScienceDirect

European Journal of Operational Research

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

Decision Support

R&D for green technologies in a dynamic oligopoly: Schumpeter, arrow

and inverted-U’s✩
Gustav Feichtinger a, Luca Lambertini b,∗, George Leitmann c, Stefan Wrzaczek d
a
Institute of Statistics and Mathematical Methods in Economics, Vienna University of Technology, Austria
b
Department of Economics, University of Bologna, Strada Maggiore 45, 40125 Bologna, Italy
c
College of Engineering, University of California at Berkeley, California
d
Department of Business Administration, University of Vienna, Austria

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

Article history: We extend a well-known differential oligopoly game to encompass the possibility for production to generate
Received 10 March 2014 a negative environmental externality, regulated through Pigouvian taxation and price caps. We show that,
Accepted 14 September 2015
if the price cap is set so as to fix the tolerable maximum amount of emissions, the resulting equilibrium
Available online 9 October 2015
investment in green R&D is indeed concave in the structure of the industry. Our analysis appears to indicate
Keywords: that inverted-U-shaped investment curves are generated by regulatory measures instead of being a ‘natural’
Dynamic games feature of firms’ decisions.
Oligopoly
© 2015 Elsevier B.V. and Association of European Operational Research Societies (EURO) within the
Environmental externality
R&D
International Federation of Operational Research Societies (IFORS). All rights reserved.

1. Introduction
The departure point of the analysis illustrated in this paper lies
at the intersection between two different debates, one being cen-
tered upon the relation between competition and innovation, whose
most recent development is known as the Schumpeterian growth the-
ory initiated by Aghion and Howitt (1998), the other belonging to en-
vironmental economics and focussing on the optimal design of pol-
icy instruments, such as environmental standards, pollution rights
and Pigouvian taxation, to stimulate firms’ investments in abate-
ment and/or replacement technologies (for an updated survey, see
Lambertini, 2013).
The acquired industrial organisation approach to the bearings of
market power on the size and pace of technical progress can be
traced back to the indirect debate between Schumpeter (1934, 1942)
and Arrow and Nelson (1962) on the so-called Schumpeterian hy-
pothesis, which, in a nutshell, says that one should expect to see an
inverse relationship between innovation and the intensity of com-
petition or market structure. Irrespective of the nature of innova-
tion (either for cost reductions or for the introduction of new prod-
ucts), a large theoretical literature attains either Schumpeterian or
Arrovian conclusion (for exhaustive accounts, see Tirole, 1988; and
Reinganum, Schmalensee, & Willig, 1989).1 That is, partial equilib-
rium theoretical IO models systematically predict a monotone rela-
tionship, in either direction.
The picture drastically changes as soon as one takes instead the
standpoint of modern growth theory. In particular, Aghion, Bloom,
Blundell, Griffith, and Howitt (2005) stress that empirical evidence
shows a non-monotone relationship between industry concentration
(or, the intensity of market competition) and aggregate R&D efforts:
this takes the form of an inverted-U curve, at odds with all existing
theoretical IO models; in the same paper, the authors provide a model
yielding indeed such a concave result, and fitting the data. A thorough
discussion, accompanied by an exhaustive review of the related lively
debate, can be found in Aghion, Akcigit, and Howitt (2013).
One could say that the inverted-U emerging from data says that
Arrow is right for small numbers, while Schumpeter is right there-
after. Alternatively, on the same basis one could also say that neither
Arrow nor Schumpeter can match reality, if our interpretation of their
respective views is that “competition (resp., monopoly) outperforms
monopoly (resp., competition) along the R&D dimension” . Be that
as it may, there arises the need of constructing models delivering a
non-monotone relationship between some form of R&D (for process,
product or environmental-friendly innovations) and the number of
firms in the industry.

✩
We thank Lorenzo Peccati (Editor) and three anonymous referees for fruitful com-
ments and suggestions. The usual disclaimer applies. 1
See also Gilbert, Lerner, and Stern (2006), Vives (2008) and Schmutzler (2010) for
∗
Corresponding author. Tel.: +39 051 2092623; fax: +39 051 2092664. add-on’s on this discussion, where still the Schumpeter vs Arrow argument is unre-
E-mail address: luca.lambertini@unibo.it (L. Lambertini). solved.

http://dx.doi.org/10.1016/j.ejor.2015.09.025
0377-2217/© 2015 Elsevier B.V. and Association of European Operational Research Societies (EURO) within the International Federation of Operational Research Societies (IFORS).
All rights reserved.
1132 G. Feichtinger et al. / European Journal of Operational Research 249 (2016) 1131–1138

With this purpose in mind, here we extend a noncooperative Each firm has two control variables, investment in capacity ui (t)
differential game model dating back to Leitmann and Schmitendorf and investment in green R&D ki (t). The policy maker has two instru-
(1978) and Feichtinger (1983) to describe an industry in which firms ments, the Pigouvian tax rate τ (which may be usefully thought of as
sell a homogeneous good and accumulate capacity over time through incorporating the price of emission rights) and the regulated mark-
costly investments; firms’ activities entail polluting emissions hin- up p. To avoid time inconsistency issues, we consider the policy menu
dering welfare, and the government adopts a Pigouvian taxation applied onto the steady state only. The structure of the model iden-
policy aimed at providing them with an incentive to internalise the tifies a linear state game (it wouldn’t be so if either the policy were
environmental externality and therefore undertake R&D projects for function of the state or the demand function were endogenously de-
pollution abatement. As in the original model, the mark-up is exoge- termined). Therefore, the open-loop solution is subgame perfect, re-
nously fixed, and here is thought of as an additional regulatory tool specting the original LSF formulation.
in the hands of the public authority.
Our main results can be outlined as follows. First, we show that 3. Equilibrium analysis
there exists a unique open-loop equilibrium which is subgame per-
fect and saddle-point stable, for any pair of policy instruments. Then, Firm i’s (i = 1, . . . , N) current-value Hamiltonian (from now on we
taking again the mark-up and tax rate as given, we prove that the ag- suppress the time argument)3
gregate green R&D effort is monotonically increasing in the number  
Hi (s, x, k, u) = πi + λii ẋi + λi j ẋ j + μii ṡi + μi j ṡ j (5)
of firms, which is a definitely Arrovian result. Subsequently, we en-
j=i j=i
dogenise the regulatory toolkit, allowing first the policy-maker to set
that Pigouvian tax rate so as to maximise steady state social welfare; generates the following first order conditions (inner solution) for firm
in such a case, the aggregate R&D effort is strictly convex in the num- i’s (i = 1, . . . , N) controls
ber of firms. If optimal taxation is accompanied by a mark-up tailored ∂ Hi
on industry structure so as to limit the overall volume of emissions, = λii − ui = 0 (6)
∂ ui
then there emerges a general condition on the shape of the price reg-
ulation scheme whereby the industry investment is indeed concave 
∂ Hi
w.r.t. the number of firms. = −γ − ki − zμii − h μi j = 0 (7)
The remainder of the paper is organised as follows. The setup is
∂ ki j=i
laid out in Section 2, while the equilibrium analysis is in Section 3. Thus we obtain the following optimal controls of firm i (i = 1, . . . , N)
Section 4 illustrates the design of policy tools and its consequences
on aggregate R&D efforts. Concluding remarks are in Section 5. u∗i (t ) = λii (t )

2. The game
k∗i (t ) = −γ − zμii (t ) − h μi j (t ). (8)
j=i

As anticipated in the introduction, here we extend the model in- Furthermore, each firm i obtains the following dynamic equations
troduced by Leitmann and Schmitendorf (1978) and further investi- for the costates (i, j = 1, . . . , N, i = j)
gated by Feichtinger (1983), to allow for the presence of an environ- 
λ̇ii = (ρ + δ)λii − p − μii − μi j v
mental externality and green R&D investments. In the remainder, we
j=i
will label this framework as the ‘LSF model’ for brevity. The market 
exists over t ∈ [0, ∞), and, as in Dragone, Lambertini, and Palestini λ̇i j = (ρ + δ)λi j − μi j
(2010), it is served by N ≥ 1 a priori symmetric firms with individual j=i
capacity xi (t) ≥ 0.2 Given a fixed profit margin p ≥ 0, the instanta- μ̇ii = (ρ + η)μii + τ
neous profit of firm i is
μ̇i j = (ρ + η)μi j . (9)
u2i (t ) k2 (t )
πi (t ) = pxi (t ) − − γ ki (t ) − i − τ si , (1) In order to characterize the optimal long run solution of the sys-
2 2
tem we have to derive the equilibria of the above defined system of
where γ > 0 is a parameter. Capacity xi (t) changes according to
differential equations (i.e. state and costate equations of all firms). In
ẋi (t ) = ui (t ) − δ xi (t ), (2) this model the equilibrium is unique. For the adjoint variables we ob-
where ui (t) is the investment of firm i at time t and δ > 0 is the decay tain (i, j = 1, . . . , N, i = j)
rate of individual capacity. si (t) and ki (t) denote the firm’s polluting 1
 τ 
emissions and R&D effort respectively, and τ is the tax rate. λ̂ii = p−
ρ +δ ρ +η
The emissions of a firm follow the dynamics
 λ̂i j = 0
ṡi (t ) = xi (t ) − zki − h k j − ηsi (t ) (3)
−τ
j=i μ̂ii =
ρ +η
where z is a positive parameter, η > 0 is the natural decay rate of
emissions, and parameter h ∈ [0, z) measures the spillover effect re-
μ̂i j = 0. (10)
ceived from rivals’ R&D activity. Inserting into (8) yields the following equilibrium controls (i =
The total instantaneous volume of emissions at the industry level 1, . . . , N)

is S(t ) = Ni=1 si (t ). Therefore the social welfare function at any time 1
τ 
can be defined as û∗i = p−
ρ +δ ρ +η

N 
N
zτ
SW (t ) = πi (t ) + CS(t ) − S(t ) + τ si (t ). (4) k̂∗i = −γ. (11)
i=1 i=1
ρ +η

2
In the original formulation of the model, xi (t) is firm i’s sales volume, and ui (t)
its advertising investment. However, one can think of these variables as representing,
respectively, installed capacity (with each firm selling at full capacity at any time) and
the instantaneous investment to increase it.
3
In this respect, a remark is in order: note that, in general, the objective functional
π i has to be multiplied by the general multiplier λ0 to allow for the abnormal case (see
e.g. Leitmann, 1981). However, in the current model that abnormal case can be ruled
out as can be readily shown.
 G. Feichtinger et al. / European Journal of Operational Research 249 (2016) 1131–1138 1133

Using these expressions for the state equations we obtain (i = the picture along two dimensions: one is obviously τ , as is usually
1, . . . , N) the case in environmental economics, the other is p, which is a spe-
1

τ  cific feature of the present model. Here, the mark-up is fixed, and this
x̂i = p− fact can be interpreted as a consequence of a price cap imposed by a
δ(ρ + δ) ρ +η
1
 1
 τ   zτ   public authority. The research question we are about to assess in the
ŝi = p− − − γ (z + h(N − 1)) . following section is the following: is the portfolio of policy instru-
η δ(ρ + δ) ρ +η ρ +η ments {p, τ } going to modify the apparently monotone behaviour of
(12) aggregate R&D efforts K∗ outlined in Corollary 4? And, if so, in what
direction?
Since all firms are assumed to be a priori symmetric, we define the
steady state values as û := ûi , k̂ := k̂i , x̂ := x̂i and ŝ := ŝi . Due to the
4. Environmental policy and aggregate investment
economic meaning of the model, we have to assume that the con-
trols and the states are non-negative for all t ∈ [0, ∞). The following
The bearings of p and τ on aggregate R&D incentives can be ap-
Lemma provides assumptions such that the non-negativity is fulfilled
preciated by addressing the issue in the following terms. It is already
in equilibrium.
known (see Benchekroun & Long, 1998, 2002, inter alia) that there
Lemma 1. The steady state variables of the state and control variables exists a level of Pigouvian taxation driving the industry to the first
of every player i (i = 1, . . . , N) are non-negative if the following assump- best which would be obtained under social planning. Call this tax
tions on the parameters are fulfilled rate τ SP (p, N). This tax rate must maximise the steady state level of
the social welfare function, defined as
γ (ρ + η)
p(ρ + η) ≥ τ ≥ (A1)
z SW ∗ (τ ) = Nπ ∗ (τ ) + CS∗ (τ ) − N(1 − τ )s∗ (τ ) (15)

1
 1
 1τ
 zτ −1  where
p− −γ −z +1≥N (A2) (a − p)Nx∗ (τ )
h δ(ρ + δ) ρ +η ρ +η CS∗ (τ ) = (16)
2
Condition (A1) guarantees non-negativity of the controls (see is consumer surplus, calculated postulating the existence of a linear
(11)). Non-negativity of x̂i is implied by (A1) and that of ŝi by (A2). and decreasing market demand function  p = a − Nx∗ in which a >
The analysis of the Jacobian matrix of the system shows that 0 is consumers’ reservation price; indeed,  p is the price that would
Proposition 2. The unique equilibrium (x̂, ŝ, û∗ , k̂∗ ) is a saddle point. prevail if the mark-up were unregulated. Moreover, (15) accounts for
the additional fact that the revenue produced by Pigouvian taxation,
From the adjoint equations it is easy to show that μi j (t ) = λi j (t ) = Nτ s∗ (τ ), is redistributed to consumers as a windfall.
0. Due to the structure of the system it is possible to derive an analyt- Then, τ SP (p, N) can be easily calculated by solving the necessary
ical expression of the stable path, i.e. condition ∂ SW ∗ (τ )/∂τ = 0, satisfied by the unique tax rate:4
xi (t ) = x̂ + (xi0 − x̂)e−δt τ SP ( p, N)
1 (η + ρ)[p(δ − ρ)η − (δ + ρ)(aη − 2) + 2δ(δ + ρ)2 (z + h(N − 1))z]
si (t ) = ŝ − (xi0 − x̂) (e−δt − e−ηt ) + (si0 − ŝ)e−ηt = . (17)
η+δ 2δη[1 + (δ + ρ) z2 ]
2

μii (t ) = μ̂ii
Now observe that
λii (t ) = λ̂ii (13)
∂ K ∗ (τ SP ( p, N))

Now we are able to define a sufficient assumption that the con-  ,

= 0 in N = max 1, N (18)
trols and the states are positive for all t ∈ [0, ∞).
∂N

 = 2δγ η + 4cδ(δ + ρ) (γ η + h − z) + p(ρ − δ)η + (δ + ρ)(aη − 2)

2 2
Lemma 3. Let (A1) and (A2) hold. Then the following assumption is suf- N ,
4hδ(δ + ρ) z
2
ficient to ensure that all controls and states are positive over the whole
planning horizon.
and
xi0 ≤ x̂, si0 ≤ ŝ, δ ≤ η. ∂ 2 K ∗ (τ SP ( p, N)) 2h(δ + ρ) z2
2
= > 0, (19)
Proof. u∗i (t ) > 0 and k∗i (t ) > 0 are trivial by the signs of the adjoint ∂ N2 η[1 + (δ + ρ)2 z2 ]
variables in (13). xi (t) > 0 is implied by xi0 ≤ x̂ (see (13)). For si (t) we
showing that, if N  ≥ 2, then in correspondence of N  the aggregate
obtain from (13)
R&D level K∗ (τ SP (p, N)) is indeed being minimised. Hence, in this sce-
1  is admissible, a U-shaped
si (t ) = ŝ(1 − e−ηt ) + si0 e−ηt + (x̂ − xi0 ) (e−δt − e−ηt ) (14) nario no inverted-U may arise (at most, if N
η+δ curve obtains), since:
Since η > 0 the first and the second term are trivially (strictly)
Lemma 5. If p is given, the equilibrium aggregate R&D effort is convex
positive. δ ≤ η implies e−δt − e−ηt ≥ 0 for all t ≥ 0 (equality is the
in N.
case only for t = 0). Thus also the third term of the above expression
for si (t) is non-negative.  On the basis of the above Lemma, it seems that Pigouvian taxa-
tion is in itself insufficient to deliver an inverted-U R&D curve: the
The foregoing analysis has a seemingly not-so-intriguing ancillary  ≥ 2 (if so, then for a limited
opposite shape does in fact appear if N
implication:

Corollary 4. Since K ∗ = Nk∗ is everywhere increasing in N for all τ > 4

The second order condition is satisfied, as
(η + ρ)γ /z, the behaviour of aggregate R&D is Arrovian for any given  
N 1 + (δ + ρ) z2
2
Pigouvian policy allowing for a positive investment. ∂ 2 SW ∗ (τ )
= − <0
∂τ 2 (δ + ρ)2 (η + ρ)2
However, there is more to it, which can be shown to emerge as
soon as one admits the reasonable possibility for regulation to enter always.
1134 G. Feichtinger et al. / European Journal of Operational Research 249 (2016) 1131–1138

number of firms the Schumpeterian hypothesis is confirmed, while Now we can differentiate K∗ (·) w.r.t. N, obtaining:
for sufficiently large number of firms the Arrovian position prevails).
∂ K ∗ (·) (ϒ1 + ϒ2 )ϒ3 + ϒ4 − ϒ5
What if p is set by the government for some purpose? Suppose = (27)
first that a public agency is in charge of regulating the mark-up of this
∂N ηϒ32
industry having in mind objectives such as the entry process, con- where
sumer surplus or the volume of industry emissions. Be that as it may,
the resulting regulatory measure can be defined as p = p(N), so that ϒ1 ≡ z(2 − aη) − γ η
the mark-up is a function of industry structure. Substituting p(N) into
τ SP (p, N), the optimal tax rate is then defined in terms of industry
ϒ2 ≡ 2zδς [z + h(N − 1) − γ N]
structure (as well as the parameters of the model), and can be rela-
belled as τ SP (N). Then, the aggregate R&D effort at the steady state
equilibrium writes as follows: ϒ3 ≡ [1 + zδ(z(δ + 3ρ) − h(N − 1)(δ − ρ))] (28)
N[zτ SP (N) − γ (ρ + η)]
K ∗ (τ SP (N)) =
ρ +η
ϒ4 ≡ hzδ(δ − ρ)[S̄δη2 (δ − ρ)
with
+ 2Nz(1 + z(h(N − 1) + z)δ(δ + ρ))]
∂ K ∗ (τ SP (N)) zτ SP (N) − γ (ρ + η) + Nz · ∂τ SP (N)/∂ N
= (20)
∂N ρ +η
ϒ5 ≡ Nhzδη(δ − ρ)[γ + z(a + 2δγ ς)]
and
Then, differentiating (27) w.r.t. N, we have the following:
∂ 2 K ∗ (τ SP (N)) z[2 · ∂τ SP (N)/∂ N + N · ∂ 2 τ SP (N)/∂ N2 ]
= (21)
∂ N2 ρ +η ∂ 2 K ∗ (·)
If there exists a value of N at which (20) is nil, this is implicitly iden-
∂ N2 
tified by ϒ32 ϒ2 − 2ϒ3 (ϒ4 − ϒ5 ) − ϒ3 (ϒ1 + ϒ2 )ϒ3 − ϒ4 + ϒ5
= (29)
ηϒ33
∂τ SP (N) γ (ρ + η) − zτ SP (N)
= <0 (22)
∂N Nz in which ϒ j ≡ ∂ ϒ j /∂ N, j = 2, 3, 4, 5.
as zτ SP (N) > γ (ρ + η) in order for the equilibrium R&D effort to be The equation ∂ K ∗ (·)/∂ N = 0 has two roots:
positive. Looking back at (17 ), it appears that (i) as long as p is not a

1 + z2 δ(δ + 3ρ) ς
function of industry structure, ∂τ SP (p, N)/∂ N > 0; and (ii) the deriva- N± = 1 + ± √ (30)
tive of the optimal tax w.r.t. N may become negative only if p is indeed
hzδ(δ − ρ) hz2 δ(δ 2 − ρ 2 ) 2
a decreasing function of N. where
Then, using (22), (21) becomes: 
2ς 
N+ − N− = >0 (31)
∂ 2 K ∗ (τ SP (N)) 2[γ (ρ + η) − zτ SP (N)] + zN2 · ∂ 2 τ SP (N)/∂ N2 hz2 δ(δ 2 − ρ 2 )
=
∂ N2 N(ρ + η)
for all δ > ρ , provided  ≥ 0 in such a way that N± ∈ R, with
(23)
 ≡ Shzδ 2 η2 (δ − ρ)2 + [1 + zδ(h(δ − ρ) + z(δ + 3ρ))]
whose sign determines whether the solution to ∂ K ∗ (τ SP (N))/∂ N = 0
is a maximum or a minimum. × [4zδς 2 − γ η(δ − ρ) + z(aηρ − δ(aη − 4))
A sensible way of modelling the role of price regulation rests on − 2zγ δης (δ 2 − ρ 2 )] (32)
considering that, in general, τ SP (p, N) - although maximising steady
For future reference, define  ≡ 1 + zδ(h(δ − ρ) + z(δ + 3ρ)) and
state social welfare - does not ensure the minimisation of the exter-
nality or the attainment of any given cap S̄ targeted by the public  ≡ 4zδς 2 − γ η(δ − ρ) + z(aηρ − δ(aη − 4))
agency in charge of the environmental policy.
− 2zγ δης (δ 2 − ρ 2 ), (33)
If indeed the government wants to reduce emissions to a given
level S̄ , it must set the regulated price at the level solving Ns∗ = S̄, which allw us to formulate the following
which is
  Lemma 6. If δ > ρ ,  ≥ 0 for all
2S̄δ 2 η2 1 + ς 2 + N[2 + 2δ(h(N − 1) + z)(2ς − δ ) − aη ]
pSP (S̄) = (24)  
ηN[1 − δ z((δ + 3ρ)z + h(N − 1)(ρ − δ)z)] ·
S ≥ max 0, − .
hzδ 2 η2 (δ − ρ)
2
where ς ≡ z(δ + ρ), and
If instead δ ∈ (0, ρ ),  ≥ 0 for all
 ≡ γ η + [γ η − h(N − 1) − z]ς 2 ;  
(25)
 ≡ 1 + δς [h(N − 1) + z]. ·
S ∈ 0, .
hzδ 2 η2 (δ − ρ)
2
Price (24) is in fact a function of N and we may further investigate the
bearings of pSP (S̄) on the equilibrium R&D effort of the industry. Lemma 6 says that (i) if the efficiency of natural carbon sinks is
The adoption of such a regulated price delivers higher than the discount rate, the solutions N± to ∂ K ∗ (·)/∂ N = 0 are
K ∗ (τ SP ( pSP (S̄), N), pSP (S̄)) real if S is large enough; (ii) if instead the opposite applies, S must be
low enough in order for N± ∈ R.
= [δη2 (δ − ρ)S̄ − η(az + γ + 2zδγ ς )N The expressions N± can be substituted into (29) to verify that
+ 2z(1 + δς (z + h(N − 1)))N]/η[1 + zδ(z(δ + 3ρ)  √
∂ 2 K ∗ (·)  4 2hzδς 2
− h(N − 1)(δ − ρ))]. (26) = −  <0 (34)
∂ N2 N=N η ς
+
 G. Feichtinger et al. / European Journal of Operational Research 249 (2016) 1131–1138 1135

and 4000
 √
∂ 2 K ∗ (·)  4 2hzδς 2
=  >0 (35)
∂ N2 N=N η ς
3500
−

The foregoing analysis produces the following: 3000

Proposition 7. If δ ∈ (0, ρ ) and S ∈ [0, /(hzδ 2 η2 (δ − ρ)2 )],
then
 ≥ 0 and N+ < 0 < N− . In this parameter range, K∗ (·) is convex in N, 2500
taking its unique minimum at N = N− .
If instead δ > ρ and S ≥ max{0, − /(hzδ 2 η2 (δ − ρ)2 )}, then  2000

K
≥ 0 and N− < 0 < N+ . In this parameter range, K∗ (·) is concave in N and
takes its unique maximum at N = N+ . In the remainder of the parameter 1500
space, N± ∈/ R.

The above Proposition illustrates the existence of parameter con- 1000

stellations wherein the aggregate advertising effort in steady state is
non-monotone in the number of firms, taking the form of either a
U-shaped curve or an inverted U-shaped curve w.r.t. the number of
firms in the industry. However, it is also interesting to single out the
regions in which the curve in question is indeed monotone. These are
identified in the following:
Corollary 8. In the parameter regions where  < 0, N± ∈ / R, and there-
fore ∂ K∗ (·)/∂ N has the same sign as ρ − δ. This entails that K∗ (·) is mono-
tone in N.
Proof. This result can be easily proved noting that the coefficient of
N2 the numerator of ∂ K∗ (·)/∂ N in (27) is indeed 2h2 z2 δ 2 ς(ρ − δ),
while the denominator of ∂ K∗ (·)/∂ N, i.e., expression ηϒ32 , is positive.
Hence, when  < 0,
• K∗ (·) decreases monotonically in N for all δ > ρ ;
• K∗ (·) decreases monotonically in N for all δ ∈ (0, ρ ).
This concludes the proof. 
This amounts to saying that, when  < 0, aggregate green R&D has
a Schumpeterian (resp., Arrovian) flavour when discounting is low
(resp., high) enough.
A special case where K∗ (·) is monotone in N is the following. From
(26), we see that if h = 0, then
K ∗ (τ SP ( pSP (S̄), N), pSP (S̄))|h=0
= [S̄zδη2 (δ − ρ) + 2Nz(1 + δ(δ + ρ)z2 )
− Nη(az + γ (1 + 2z2 δ(δ + ρ)))]/[η(1 + δ z2 (δ + 3ρ))], (36)
which is necessarily monotone in N. In particular:
Corollary 9. In the special case in which technological spillovers are ab-
sent, take
S̄ > max
 
N[η(az + γ (1 + 2z2 δ(δ + ρ))) − 2z(1 + z2 δ(δ + ρ))]
× 0,
zδ(δ − ρ)η2
to ensure K ∗ (·)|h=0 > 0. Then,
∂ K ∗ (·)|h=0 2z(1 + z2 δ(δ + ρ)) − η(az + γ (1 + 2z2 δ(δ + ρ)))
=
∂N η(1 + z2 δ(δ + 3ρ))
≷0
for all
2z(1 + z2 δ(δ + ρ))
η≶η
≡ . graphs in Fig. 2, where the qualitative effect on the inverted-U shape
az + γ (1 + 2z2 δ(δ + ρ))
Corollary 9 entails that, if the individual firm’s abatement capabil-
ity is unaffected by the rivals’, the behaviour of aggregate green R&D
as N changes is Arrovian (resp., Schumpeterian) if the environment’s
recycling rate is sufficiently low (resp., high). i.e., it is as if the indus-
try were complementing the natural absorption activities if the lat-
ter are not particurlarly effective (which corresponds to the Arrovian
500
0
1000 2000 3000 4000 5000 6000 7000
N
Fig. 1. The inverted-U curve.
case), and conversely (which instead corresponds to the Schumpete-
rian case).
Now it is appropriate to provide a numerical example illustrating
the arising of an inverted U-shaped curve. Fixing parameter values
a = 150; h = 1/10; z = 2/5; S̄ = 12 × 103 ;
δ = 2/3; γ = 3/2; η = 2; ρ = 1/10, (37)
aggregate R&D steady state investment K ∗ (τ SP ( pSP (S̄), N), pSP (S̄))
can be drawn as in Fig. 1, where the concavity of industry effort
emerges clearly and K ∗ (τ SP ( pSP (S̄), N),SP (S̄)) is maximised at N =
N+  822 (while N− < 0). In correspondence of these numerical val-
ues, ∂ p(S)/∂ N  −0.0097 and from 22 and 23 we have ∂τ SP (N)/∂ N 
−0.0059 and ∂ 2 K ∗ ( · )/∂ N2  −0.0938.
Something more can be said about the effects of the size of the
population of firms. Concerning the supply side, we have
k∗ (τ SP ( pSP (S̄), N),SP (S̄)) = 0 at N  7348
(38)
x∗ (τ SP ( pSP (S̄), N),SP (S̄)) = 0 at N  7497
which implies that there exists a non-negligible range of N,
namely, (7349, 7389) in which the individual R&D effort drops
to zero but firms still produce and sell to consumers. As instead
to the welfare performance of this industry, one can check that
SW ∗ (τ SP ( pSP (S̄), N),SP (S̄)) is maximised at N ࣃ 3522. This result (at
least in the numerical example based on the above values) illustrates
a situation in which consumer surplus matters more than the envi-
ronmental externality, so that the industry structure that maximises
welfare is a lot more fragmented than that maximising the aggregate
volume of green R&D.
The inverted-U is relatively stable with respect to the model pa-
rameters. However, the level and the position of the curve differs.
Table 1 summarizes the dependence. The first column includes the
effect on the peak, the second column the effect on the level of
the curve and the third one links the whole effect to each of the five
is illustrated.
4.1. Discussion
Concerning the concavity of K ∗ (τ SP ( pSP (S̄), N),SP (S̄)) with re-
spect to N, the foregoing analysis seems to imply that the arising of
inverted-U curves is the consequence of the pressure of regulatory
1136 G. Feichtinger et al. / European Journal of Operational Research 249 (2016) 1131–1138

6000 4000

3500
5000

3000

4000
2500

3000 2000
K

K
1500
2000

1000

1000
500

0 0
1000 2000 3000 4000 5000 6000 7000 1000 2000 3000 4000 5000 6000 7000 8000
N N

4000 4000

3500 3500

3000 3000

2500 2500

2000 2000
K

1500 1500

1000 1000

500 500

0 0
1000 2000 3000 4000 5000 6000 7000 8000 1000 2000 3000 4000 5000 6000 7000 8000
N N

4000

3500

3000

2500

2000
K

1500

1000

500

0
1000 2000 3000 4000 5000 6000 7000
N

Fig. 2. Sensitivity of the inverted-U curve (curve shifts along the arrow for the corresponding increasing parameter).

policy (possibly, as is the case in our model, of the adoption of multi-
ple tools at the same time, to pursue different although - in some way
- related objectives). This could be a plausible explanation for the lack
of analogous outcomes in the vast literature discussing the bearings
of industry structure on aggregate R&D, that has been produced so far
in IO.
The acquired wisdom on the matter, delivering monotone predic-
tions in one way or the other, can be quickly summarised as follows.
The Schumpeterian hypothesis claims that market power is the driver
of innovation, and therefore monopoly should be expected to stand
out as the market form producing the highest R&D incentives. This
argument rests on the so-called efficiency effect, whereby a monop-
olist can at least replicate the behaviour of any oligopolistic or per-
fectly competitive industry. Adhering to this view, one should expect
to observe aggregate R&D to decrease monotonically in the number
of firms. The opposite perspective is based on Arrow’s replacement ef-
fect, whereby a monopolist has a lower incentive to innovate than a
competitive industry (or any oligopoly in between) because, even if
 G. Feichtinger et al. / European Journal of Operational Research 249 (2016) 1131–1138 1137

Table 1
Dependence of the inverted-U curve on the model parameters.
Setting (41) equal to zero, we obtain
Effect on the peak Effect on curve level
∂τ γ (η + ρ) − z(τ + N · ∂τ /∂ p · ∂ p/∂ n)
= (43)
ρ Move to the right Ambiguous: increase for low N, Top left panel ∂N Nz
decrease for high N
η Move to the right Ambiguous: decrease for low N, Top right panel
This can be substituted into (42), which can also be further simplified
increase for high N using additional pieces of information that we can draw from expres-
γ No/marginal Increase Middle left panel sion (17), whereby
a No/marginal Increase Middle left panel
z Move to the left Decrease Middle right panel ∂ 2τ ∂ 2τ ∂ 2τ
= = = 0. (44)
S Move to the right Decrease Low panel ∂ N2 ∂ N∂ p ∂ p2
δ Move to the right Decrease Low panel
h Move to the left Decrease middle Right panel Hence, (42) simplifies as follows:

∂ 2 K ∗ 2[γ (η + ρ) − zτ ] + zN2 · ∂τ /∂ p · ∂ 2 p/∂ n2

= (45)
the innovation is patentable, the monopolist’s benefit reduces to re- ∂ N2 N(ρ + η)
place itself by acquiring the patent, while a smaller firm operating Observing (45), we may note that
initially under much less favourable conditions might gain monopoly
power by getting to the patent office before any of its rivals does.5 γ (η + ρ) − zτ = −(η + ρ)k∗ < 0 (46)
The large subsequent literature has alternatively confirmed one view and ∂τ /∂ p ࣥ 0 for all δ ࣥ ρ - which again can be easily deduced from
or the other, with the exception of Aghion et al. (2005), where the (17). Accordingly, we may take a final step and rewrite (45) in a more
only available model showing a non-monotone result accompanies intuitive form:
an empirical evidence with analogous properties. It is worth stress-
ing that most, if not all of this literature relies on theoretical models
∂ 2 K ∗ zN2 · ∂τ /∂ p · ∂ 2 p/∂ n2 − 2(η + ρ)k∗
= (47)
where policy instruments are either absent or taken as exogenously ∂ N2 N(ρ + η)
given. which is negative for all
How can we justify or interpret the arising of a concave aggregate  
R&D effort in the presented model? The source of this effect must ∗ zN2 · ∂τ /∂ p · ∂ 2 p/∂ n2
k > max ,0 . (48)
be found in some aspect that the previous literature has overlooked, 2(η + ρ)
such that the outcome is a non-monotone mixture of Arrow’s replace-
ment effect appearing first, to be replaced by Schumpeter’s efficiency If we confine our attention to the parameter region defined by δ >
effect. The present model has several special features. First of all, a ρ , which is what we have done to generate the inverted-U curve ap-
patent system is left out of the picture. Additionally (i) individual ef- pearing in Fig. 1 - then ∂τ /∂ p > 0; therefore, in this range ∂ 2 p/∂ N2 <
forts spill over to rivals; and (ii) innovation is green, which amounts 0 suffices to ensure that K∗ is indeed concave w.r.t. N for all k∗ > 0.
to saying that R&D is spurred by emission taxation. That is, we are Having said that, two natural questions arises, namely, (i) should
treating a particular type of investment which would be altogether we conclude that it is altogether impossible to reproduce the same
nil without an equally specific policy. Yet, Pigouvian taxation per se is result if regulation is assumed away in differential games investi-
not an explanation of the arising of an inverted-U curve, as we know gating some form of R&D for either process or product innovation?
from Corollary 4 and Lemma 5. In particular, the latter would imply The few existing examples (see Cellini & Lambertini, 2002, 2009, for
a U-shaped curve, not the opposite. Hence, the responsibility of our instance)6 indeed yield monotone outcomes, but are by no means
result must be imputed to the remaining policy instrument, the reg- general; (ii) shall we deem the usual assumption of a linear mar-
  ket demand responsible for monotone outcomes? In fact, empirical
ulated price pSP S̄ . From (11), we have that the aggregate effort is
research (Hausman, 1981; Varian, 1982, 1990, inter alia) has shown
 zτ 
that most markets are characterised by non linear demand functions,
K∗ = N −γ (39)
ρ +η which are best approximated by isoelastic curves. These extensions
are left for future research.
In (39) we can plug τ = τ SP ( p, N) from ( 17); however, τ SP (p, N) being
linear and increasing in N, this yields a convex relationship between
5. Concluding remarks
K∗ and N. Therefore, the source of the inverted-U curve is not Pigou-
vian taxation. Whatcreates
 it is the additional policy measure regu-
We have characterised green R&D incentives for firms operating
lating price, i.e., pSP S̄ from (24), using which we can rewrite (39) as
in an industry where production pollutes the environment the gov-
follows:
  ernment regulates the mark up and adopts a Pigouvian tax policy
zτ ( p(N), N) to decrease emissions and stimulate the introduction of clean tech-
K∗ = N −γ (40)
ρ +η nologies. The model delivers a thus far rare result, in the form of an
inverted-U aggregate R&D expenditure at equilibrium. The implica-
Now observe that
tion of our analysis seems to be that the empirical evidence concern-
∂ K ∗ z[τ + N(∂τ /∂ n + ∂τ /∂ p · ∂ p/∂ n)]
= −γ (41) ing the emergence of inverted-U curves is a consequence of some
∂N ρ +η form of regulation that modifies the aggregate behaviour of the in-
and dustry as compared to the predictions of theoretical models where
     2  regulation is either totally exogenous or just assumed away. Whether
∂τ ∂τ ∂ p ∂2τ ∂2τ ∂ p ∂2τ ∂p ∂ 2 p ∂τ ours is a special (and fortunate) case or instead an indication of some
z 2 + +N +2 + +
∂ 2 K∗ ∂n ∂ p ∂n ∂ n2 ∂ n∂ p ∂ n ∂ p2 ∂n ∂ n2 ∂ p
=
general rule previously overlooked, is a question left for future re-
∂ N2 ρ +η search.
(42)
6
One could also address in the same spirit other dynamic models whose focus is
5
A full account of this discussion is in Tirole (1988, ch. 10) and Reinganum et al. on the investment in advertising to expand the demand level or goodwill stock, as in
(1989). Cellini and Lambertini (2003a, 2003b) and the references therein.
1138 G. Feichtinger et al. / European Journal of Operational Research 249 (2016) 1131–1138
References
Aghion, P., Akcigit, U., & Howitt, P. (2013). What Do We learn from Schumpeterian Growth
Theory?. Harvard University.Mimeo, Department of Economics.
Aghion, P., Bloom, N., Blundell, R., Griffith, R., & Howitt, P. (2005). Competition and
innovation: an inverted-u relationship. Quarterly Journal of Economics, 120, 701–
728.
Aghion, P., & Howitt, P. (1998). Endogenous Growth Theory. Cambridge, MA: MIT Press.
Arrow, K. (1962). Economic Welfare and the Allocation of Resources for Invention. In
R. Nelson (Ed.), The rate and direction of industrial activity. Princeton, NJ: Princeton
University Press.
Benchekroun, H., & Long, N. V. (1998). Efficiency inducing taxation for polluting
oligopolists. Journal of Public Economics, 70, 325–342.
Benchekroun, H., & Long, N. V. (2002). On the multiplicity of efficiency-inducing tax
rules. Economics Letters, 76, 331–336.
Cellini, R., & Lambertini, L. (2002). A differential game approach to investment in prod-
uct differentiation. Journal of Economic Dynamics and Control, 27, 51–62.
Cellini, R., & Lambertini, L. (2003a). Advertising in a differential oligopoly game. Journal
of Optimization Theory and Applications, 116, 61–81.
Cellini, R., & Lambertini, L. (2003b). Advertising with spillover effects in a differential
oligopoly game with differentiated goods. Central European Journal of Operations
Research, 11, 409–423.
Cellini, R., & Lambertini, L. (2009). Dynamic R&D with spillovers: competition vs coop-
eration. Journal of Economic Dynamics and Control, 33, 568–582.
Dragone, D., Lambertini, L., & Palestini, A. (2010). The Leitmann-Schmitendorf adver-
tising game with n players and time discounting. Applied Mathematics and Compu-
tation, 217, 1010–1016.
Feichtinger, G. (1983). The nash solution of an advertising differential game: General-
ization of a model by Leitmann and Schmitendorf. IEEE Transactions on Automatic
Control, 28, 1044–1048.
Gilbert, R. (2006). Looking for mr Schumpeter: where are we in the competition-
innovation debate?. In J. Lerner, & S. Stern (Eds.), Innovation Policy and Economy.
NBER: MIT Press.
Hausman, J. A. (1981). Exact consumer’s surplus and deadweight loss. American
Econonomic Review, 71, 662–676.
Lambertini, L. (2013). Oligopoly, the environment and natural resources. London: Rout-
ledge.
Leitmann, G. (1981). The Calculus of Variations and Optimal Control (Mathematical Con-
cepts and Methods in Science and Engineering). Springer.
Leitmann, G., & Schmitendorf, W. E. (1978). Profit maximization through advertising:
a nonzero sum differential game approach. IEEE Transactions on Automatic Control,
23, 646–650.
Reinganum, J. (1989). The timing of innovation: research, development and diffusion.
In R. Schmalensee, & R. Willig (Eds.), Handbook of industrial organization, Amster-
dam, North-Holland: Vol. 1.
Schmutzler, A. (2010). Is competition good for innovation? a simple approach to an
unresolved question. Foundations and Trends in Microeconomic Analysis, 5, 355–428.
Schumpeter (1934). The theory of economic development. Oxford: Oxford University
Press.
Schumpeter, J. A. (1942). Capitalism, socialism and democracy. New York: Harper.
Tirole, J. (1988). The theory of industrial organization. Cambridge, MA: MIT Press.
Varian, H. R. (1982). The nonparametric approach to demand analysis. Econometrica,
50, 945–973.
Varian, H. R. (1990). Goodness of fit in optimizing models. Journal of Econometrics, 46,
125–140.
Vives, X. (2008). Innovation and competitive pressure. Journal of Industrial Economics,
56, 419–469.