PLAUSIBILITY OF BERTRAND COMPETITION

MANIPUSHPAK MITRA
RUPAYAN PAL
ARINDAM PAUL
P. M. SHARADA

Abstract. We consider a three stage game where a partially privatized firm (Firm 1) competes
with a private firm (Firm 2). In Stage 1, the government determines a level of privatization for
Firm 1 to maximize welfare. In Stage 2, the two firms simultaneously and credibly determine
the mode of competition. In Stage 3, the two firms compete in the market with differentiated
products. With imperfect substitutes, the no privatization Bertrand equilibrium where the
government adds full weight to welfare and firms compete in prices is the sub-game perfect
Nash equilibrium (SPNE) outcome. With complements, strictly partial privatization Bertrand
equilibrium where the government adds non-zero weight to both welfare and own profit of Firm
1 and firms compete in prices is the SPNE outcome. Therefore, when a partially privatized
firm competes with a profit maximizing firm in a regulated market with differentiated products,
competing in prices is the unique equilibrium outcome.

JEL Classification: D43, H44, L13

Keywords: Partially private firm, Bertrand competition, Cournot competition

1. Introduction

Singh and Vives [23] considered a two-stage game for a differentiated product duopoly market
where both firms are profit maximizers. In the first stage, the firms credibly announce to play
in either quantity or price strategies. If the goods are substitutes (complements), then it is
shown that Cournot (Bertrand) competition is the SPNE outcome of this two stage game. For
the substitutes case, Cheng [5] gives a nice diagrammatic exposure to analysis and showed that
under very general assumptions this result is true. However, it is not always the case that
we find profit maximizing firms operating in a differentiated product market and competing in
quantities. In reality, we often find the co-existence of welfare maximizing public firm and profit
maximizing private firms. Examples include the telecom sector, banking industry, airlines, postal
services, health sector, and education sector (see for example Backx, Carney and Gedajkovic [3],
Badertscher, Shroff and White [4], Doganis [9] and La Porta, Lopez-De-Silanes and Shleifer [16]).
Our results show that the moment we have a regulated partially privatized firm competing with
a profit maximizing firm, competing in prices (and not quantities) is the equilibrium outcome.
We first add an earlier (first) stage to the two-stage game of Singh and Vives [23] and Cheng
[5]. In the first stage, a benevolent government decides how much weight a partially privatized
firm must attach to its own profit and social welfare assuming that the competing firm is a profit
maximizer. We show that in such a set-up, when the goods are substitutes we uniquely end
up in the co-existence of welfare maximizing public firm and profit maximizing private firms,
that is, no privatization Bertrand equilibrium is the SPNE outcome of this game where the
1
2 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

government sets zero (full) weight to profit (social welfare) of the partially privatized firm and
both firms compete in prices (that is, Bertrand competition). When the goods are complements
we uniquely end up in an SPNE outcome which we call strictly partial privatization Bertrand
equilibrium where, in Stage 1, the government adds non-zero weights to both Firm 1’s own profit
and social welfare and, in Stage 2, firms choose to play price strategies. Our results show that
the moment we have a regulated partially privatized firm competing with a profit maximizing
firm, competing in prices (and not quantities) is the equilibrium outcome and this conclusion
holds under very general demand specifications.
The first stage regulatory instrument of the government is the weight θ (lying in the closed
interval [0, 1]) attached to the profit of the partially privatized firm and the residual weight
(1 − θ) attached to the welfare of the society. According to Vives [26], when both firms are
profit maximizers, then with Cournot competition there is less of a profit loss with price under-
cutting than with Bertrand competition. However, when we have one partially privatized firm,
then there exists a critical value of weight (θ > 0) such that for each weight below this critical
weight, there exists a critical price of Firm 2 firm below which Vives’s [26] argument holds and,
more importantly, above this critical price the reverse argument holds, that is, with Bertrand
competition there is less of a profit loss with price under-cutting than with Cournot competition.
It is precisely this feature that drives our main result when the goods are substitutes. Our result
with complements follows intuitions similar to Singh and Vives [23].
Our results hold under very general demand specifications. In particular, for substitute goods,
our result hold under the set of assumptions made by Cheng [5] and with an additional assump-
tion on welfare which is general enough and was used in Ghosh and Mitra [13]. To prove our
results we have at times made use of Cheng’s [5] geometric approach and we have also used the
line integral techniques used in Ghosh and Mitra [12], [13].
The paper is organized as follows. We conclude this section with a brief discussion of the
related literature. In Section 2, we introduce the basic framework and our assumptions with
imperfect substitute goods. In Section 3, we explain the three stage game. In Section 4, we
present out main theorem with imperfect substitutes. In Section 5, we present our result with
complement goods. In Section 6, we address the robustness of our game with quadratic utility.
Finally, Section 7 is the appendix section where we provide the proofs of all the results.

1.1. Related literature. The classic work by Singh and Vives [23] endogenizes price and quan-
tity strategies with profit maximizing firms in a differentiated product market. This was later
generalized by Cheng [5] by providing an elegant geometric approach. There are papers that
deal with Bertrand Cournot comparison with profit maximizing firm in a differentiated products
market with general demand specifications (see for example Cheng [6], Häckner [14], Okuguchi
[22] and Vives [26]).
In this paper we apply two stage of endogenization. The first stage endogenization is the
objective function of the partially privatized firm and, like Singh and Vives [23] and Cheng [5],
the second stage endogenization is price and quantity strategies. The first stage endogenization
of adding positive weights on welfare in a firms objective function seems natural in the context
of partially privatized firms (see, for example, the papers in the mixed-oligopoly literature by
Anderson, De Palma, and Thisse [1], Ghosh and Mitra [12], [13], Matsumura [18] and Mat-
sumura and Ogawa [19]). This literature focuses on mixed markets where both private and
partially privatized (or public) firms coexist. In the early stages of industrialization of develop-
ing economies, there is often an upper bound on the extent of private ownership. Even when
 PLAUSIBILITY OF BERTRAND COMPETITION 3

a foreign firm tries to enter a domestic market, it has to take the permission of the domestic
government which can then inflict an objective on the foreign entrant which is different from
profit maximization. If we assume that the government cares about social welfare and private
firm cares about profit, then it seems plausible to assume that the partially privatized firm max-
imize a weighted combination of profit and welfare. Therefore, objectives different from profit
is quite important and prevalent in the industrial organization literature. A paper with a very
general objective function that allows for altruism and informational asymmetry is by Heifetz,
Shannon and Spiegel [15]. However, Heifetz, Shannon and Spiegel [15] do not allow for either
privatization based enodogeneity (like Stage 1 of our three stage game) or price-quantity based
endogeneity (like Stage 2 of our three stage game).1
With quadratic utility function there is a growing literature that studies the coexistence of
partially privatized firm and a private firm in a differentiated product market. With quadratic
utility, only Stage 1 endogeneity like ours was addressed by Fujiwara [11] and by Ohnishi [21].
In Fujiwara [11], it is argued that under Cournot competition it is optimal to choose a positive
weight θ > 0 for the partially privatized firm. In Ohnishi [21], it was argued that under Bertrand
competition it is optimal to choose zero weight θ = 0 for the partially privatized firm. Our
analysis shows that, in general, if we also endogenize mode of competition along with θ, then
Cournot competition (Fujiwara’s [11] analysis) is never achieved as an equilibrium outcome.
With quadratic utility, only Stage 2 endogeneity like ours was addressed by Matsumura and
Ogawa [19] with an added assumption that one firm is fully public (that is, θ is exogenously
fixed at 0). Matsumura and Ogawa [19] argued that Bertrand competition is the SPNE of the two
stage game regardless of whether goods are substitutes or complements. We show that Bertrand
competition is the SPNE of the three stage game, which allows for endogenous determination of
the level of privatization θ. Moreover, our results hold for a very general demand specification.

2. Preliminaries and the assumptions with imperfect substitutes

We consider an economy with a competitive sector producing the numeraire good (money)
y and with a imperfectly competitive sector where two firms operate. Each firm produces a
differentiated good. For any firm i ∈ {1, 2}, let pi and qi denote Firm i’s price and quantity
respectively. For convenience we define the following notations. Let <+ represent the non-
negative orthant of the real line <. For any x = (x1 , x2 ) ∈ <2+ and any y = (y1 , y2 ) ∈ <2+ ,
x 6= y means either x1 6= y1 or x2 6= y2 , x ≥ y means x1 ≥ y1 and x2 ≥ y2 , and, x >> y means
x1 > y1 and x2 > y2 . We assume a representative consumer who maximizes V (q, y) := U (q) + y
subject to p1 q1 + p2 q2 + y ≤ M where q = (q1 , q2 ) ≥ (0, 0), p = (p1 , p2 ) >> (0, 0) and M denotes
income of the representative consumer. For any function G : <2+ → <, define for any i ∈ {1, 2},
∂G(x) ∂ 2 G(x) ∂ 2 G(x)
Gi (x) := ∂xi , Gii (x) := ∂x2i
and for any i, j ∈ {1, 2} such that i 6= j, Gij (x) :=∂xj ∂xi and
Gij (x) = Gji (x). Similarly, for any firm i ∈ {1, 2} and any firm specific function Hi : <2+ → <,
∂Hi (x) ∂ 2 Hi (x)
define for any j, k ∈ {1, 2}, Hi,j (x) := ∂xj , Hi,jk (x) := ∂xj ∂xk .

Assumption 1. For all i, j ∈ {1, 2} such that i 6= j and for all q >> (0, 0), (i) Ui (q) > 0, (ii)
Uii (q) < 0,(iii) Uij (q) < 0 and (iv) |Uii (q)| > |Uij (q)|.

1Even when we have fully privatized firms, we know from the managerial-delegation literature that managers max-
imize a weighted combination of profit and quantity/revenue/welfare and it is compatible with profit maximization
(see Fershtman and Judd [10], Miller and Pazgal [20], Sklivas [24] and Vickers [25]).
4 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

Given V (q, y) is quasi-linear, there is no income effect and hence the representative consumer’s
optimization is to select q to maximize U (q) − p1 q1 − p2 q2 . Utility maximization yields the
inverse demand function pi = Ui (q) := FiQQ (q) for all q ≥ (0, 0) and for each i ∈ {1, 2}.
QQ QQ
Using Assumption 1 it follows that Fi,i (q) = Uii (q) < 0 and Fi,j (q) = Uij (q) < 0 for i 6= j.
From Assumption 1(iv) we know that the demand system is invertible. Therefore, given any
price vector p = (p1 , p2 ) >> (0, 0), we get the direct demand function qi = FiP P (p) for each
i ∈ {1, 2}. Let |D| := U11 (q)U22 (q) − (U12 (q))2 > 0. Given Assumption 1, it also follows that
P P (p) = U (q)/|D| < 0 and F P P (p) = −U (q)/|D| > 0 for i, j ∈ {1, 2} with i 6= j. For
Fi,i jj i,j ij
any i ∈ {1, 2}, any quantity qi ≥ 0, the level set Qi (qi ) = {p | p >> (0, 0), FiQQ (p) = qi }
generates iso-quantity curve for Firm i in the price space. Due to Assumption 1, the slope of
dp P P (p)/F P P (p) > 0. By Assumption 1, own effect
the iso-quantity curve at qi = q̄i is dpji |q̄i = −Fi,i i,j
dominates cross effect implying that Q1 is steeper than Q2 in the price space (see Cheng [5]).
We assume symmetric total cost of both the firms and it is given by C(y) = cy where c > 0
and y ≥ 0. When both firms choose quantity as a strategic variable, profit of Firm i is given
as πiQQ (q) = (FiQQ (q) − c)qi for i, j = 1, 2 with i 6= j. The profit function of Firm i when both
chooses price as a strategic variable is given by πiP P (p) = (pi − c)FiP P (p) for all i, j = 1, 2 with
i 6= j.
QQ
Assumption 2. For any i, j = 1, 2 with i 6= j and any q >> (0, 0), (i) πi,ij (q) < 0 and (ii)
QQ QQ
πi,ii (q) + |πi,ij (q)| < 0.
P P (p) > 0 and (ii)
Assumption 3. For i, j = 1, 2 with i 6= j and any p >> (0, 0), (i) πi,ij
P P (p) + |π P P (p)| < 0.
πi,ii i,ij

Assumption 1, Assumption 2 and Assumption 3 are very standard and these are satisfied
by any standard demand function when products are gross-substitutes (see Cheng [5]). Let
CS = U − p1 q1 − p2 q2 denote the consumer surplus and π = π1 + π2 = (p1 − c)q1 + (p2 − c)q2
denote the aggregate profit. The (social) welfare is given by W = CS + π = U − c(q1 + q2 ). The
welfare function when both firms choose quantity as a strategic variable is given by W QQ (q) =
U (q) − c(q1 + q2 ) with WiQQ (q) = FiQQ (q) − c, WiiQQ (q) = Fi,i
QQ
(q) < 0, and, WijQQ (q) =
QQ
Fi,j (q) < 0. The welfare function when both firms choose price as a strategic variable is
given by W P P (p) = W QQ (F1P P (p), F2P P (p)) = U (F1P P (p), F2P P (p)) − c(F1P P (p) + F2P P (p)) with
WiP P (p) = (pi − c)Fi,i
P P (p) + (p − c)F P P (p).
j j.i

Assumption 4. For any i, j ∈ {1, 2} with i 6= j, (i) WiiP P (p) < 0; (ii) WiiP P (p) + WijP P (p) < 0.

Assumption 4 was used in Ghosh and Mitra [13]. Assumption 4 (i) is necessary to satisfy the
second order condition of any welfare maximizing firm.

2.1. Closed forms: We consider two very standard utility specifications.

Quadratic Utility Function: Suppose that the utility function of the representative consumer
is given by
1
(2.1) U (q) = a(q1 + q2 ) − (q12 + q22 + 2γq1 q2 ),
2
where a (> c) is a preference parameter, γ (−1 < γ < 1) is the product differentiation parameter
(see Dixit [7] and Singh and Vives [23]). A positive (negative) value of γ indicates substitute
 PLAUSIBILITY OF BERTRAND COMPETITION 5

(complement) goods. We first restrict attention to substitute goods case. One can show that
the quadratic utility function given in (2.1) satisfies all our assumptions (that is, Assumption 1
to Assumption 4) when the goods are substitutes.
CES Utility Function: Suppose that the utility function of the representative consumer is
given by
(2.2) U (q) = [q1s + q2s ]γ , sγ, γ, s ∈ (−∞, 1),
1
where σ = 1−s measure the elasticity of substitution (see Dixit and Stiglitz [8] and Vives
[27]). Goods are substitute if γ, s ∈ [0, 1] and complement if γ, s ∈ [−∞, 0]. We first restrict
attention to substitute goods case. One can show that the CES utility function satisfies the first
three assumptions (that is, Assumption 1 to Assumption 3). If the marginal cost c ≥ 1, then
Assumption 4 always holds. If c < 1 and 1−2s+γs2 > 0, then also conditions in Assumption 4 are
satisfied by the CES utility functions given in (2.2). A large family of CES utility functions satisfy
Assumption 4(i) which guarantees the satisfaction of the second order conditions. Moreover, if
Assumption 4(i) is satisfied by any CES utility function, then Assumption 4(ii) is also satisfied.

3. The three stage game

We assume that Firm 1 is partially privatized (maximizing a weighted sum of welfare and
its own profit) and Firm 2 is a private firm (maximizing its own profit). Therefore, the payoff
function of Firm 1 is V1 := θπ1 + (1 − θ)W (where θ is the privatization ratio) and that of Firm 2
is π2 . Specifically, if Firm 1 is fully nationalized (fully privatized), then θ = 0 (θ = 1) and Firm
1 maximizes social welfare (its own profit). For any given weight θ ∈ (0, 1), Firm 1 maximizes
the weighted sum of its own profit and social welfare. We consider a three stage game Γ and
the stages of the game are as follows.

• Stage1: The government decides the level of privatization (θ ∈ [0, 1]) in order to maxi-
mize social welfare.
• Stage 2: Each firm decides (simultaneously and independently) whether to adopt a
price strategy or a quantity strategy (see Table 1).
• Stage 3: Firm 1 and Firm 2 compete in the market.

We solve the game using backward induction. Given the first stage choice of θ, let the
optimal price and quantity of Firm i be pXY i (θ) and qiXY (θ) assuming Firm 1 adopts strategy
X and Firm 2 adopts strategy Y where X, Y ∈ {P, Q}. We denote the consequent profit of
the Firm i at the optimal choice and contingent on XY by π̄iXY (θ) = πiQQ (q1XY (θ), q2XY (θ)) =
πiP P (pXY XY
1 (θ), p2 (θ)). Similarly, the consequent welfare at this optimal choice contingent on
XY is W̄ XY (θ) = W QQ (q1XY (θ), q2XY (θ)) = W P P (pXY XY
1 (θ), p2 (θ)). So the optimal pay-off of
Firm 1 and Firm 2 contingent on XY are V̄1 (θ) = θπ̄i (θ) + (1 − θ)W̄ XY (θ) and π̄2XY (θ)
XY XY

respectively. With this specification, in the second stage firms play the following stage game.

Table 1. Stage 2 of Γ
XX
XXX Firm 2
XXX Price Quantity
Firm 1 XXX
Price V̄1P P (θ), π̄ P P (θ) V̄1P Q (θ), π̄2P Q (θ)
Quantity V̄1P Q (θ), π̄2P Q (θ) V̄1QQ (θ), π̄2QQ (θ)
6 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

For any X, Y ∈ {P, Q}, any x1 ∈ {p1 , q1 }, any y2 ∈ {p2 , q2 }, and, any θXY ∈ [0, 1], a profile of
strategies (θXY , (X, xXY
1 (θ
XY )), (Y, y XY (θ XY ))) is a sub-game perfect Nash equilibrium (SPNE)
2
of Γ if it induces a Nash equilibrium in every sub-game of Γ. That is, in Stage 3, given θXY ,
(xXY
1 (θ
XY ), y XY (θ XY )) is a Nash equilibrium choice vector given XY (that is, xXY (θ XY ) and
2 1
y2 (θ ) are respectively the optimum choice of X by Firm 1 given y2XY (θXY ) and the optimum
XY XY

choice of Y by Firm 2 given xXY 1 (θ

XY )). Moreover, in Stage 2, X is a best response of Firm 1

against Y of Firm 2 and Y is a best response of Firm 2 against X of Firm 1 given θXY . Finally,
θXY induces XY in Stage 2 and maximizes W̄ XY (θ) in Stage 1. Moreover, there does not exist
θ that induces ZW and yields a higher welfare than W̄ XY (θXY ). We define four possible types
of equilibria of Γ.

(i) Let (θP P , (P, pP1 P (θP P )), (P, pP2 P (θP P ))) be a Bertrand equilibrium with θP P . If θP P =
0, then we call it the no privatization Bertrand equilibrium. If θP P ∈ (0, 1), then we call
it the strictly partial privatization Bertrand equilibrium.
(ii) Let (θQQ , (Q, q1QQ (θQQ )), (Q, q2QQ (θQQ ))) be a Cournot equilibrium with θQQ .
(iii) Let (θP Q , (P, pP1 Q (θP Q )), (Q, q2P Q (θP Q ))) be a Type 1 equilibrium with θP Q .
(iv) Let (θQP , (Q, q1QP (θQP )), (P, pQP 2 (θ
QP ))) be a Type 2 equilibrium with θ QP .

4. The main result with imperfect substitutes

Theorem 1. Suppose Assumption 1, Assumption 2, Assumption 3 and Assumption 4 hold.

The strategy combination (θP P = 0, (P, pP1 P (θP P )), (P, pP2 P (θP P ))), that is, no privatization
Bertrand equilibrium, is the unique SPNE outcome of Γ.

Before going to the proof of Theorem 1 we illustrate the relevant reaction functions that
will be helpful for our analysis. If both firms compete in prices, then for any θ ∈ [0, 1], let
SV P1 P (θ) = {p | p >> (0, 0), V1,1
P P (p, θ) = 0} be the reaction function of Firm 1 in the price

space. Given, Assumption 3 and Assumption 4, SV P1 P (θ) is invertible. Hence, we can represent
0
it as p1 = SV1P P (p2 , θ). In Figure 1, we represent p1 = SV P1 P (p2 , 0) by the S1P P S1P P curve
0
and we represent p1 = SV P1 P (p2 , 1) by the R1P P R1P P curve and, for any θ ∈ (0, 1), the curve
p1 = SV P1 P (p2 , θ) must lie between the curves p1 = SV P1 P (p2 , 0) and p1 = SV P1 P (p2 , 1). The
P P (p) = 0}.
reaction function of Firm 2 is locus of all points in the set RP2 P = {p | p >> (0.0), π2,2
P P
By Assumption 3, we know that R2 is a positively sloped curve with slope less than unity
(see Cheng [5]) hence it is invertible. Therefore, we can represent it as p2 = R2P P (p1 ). In Figure
0
1, we represent p2 = R2P P (p1 ) by the R2P P R2P P curve. If both firms compete in quantities,
then for any θ ∈ [0, 1] the reaction function of Firm 1 is the locus of all points in the set
SV QQ QQ
1 (θ) = {q | q >> (0, 0), V1,1 (q, θ) = 0}. By Assumption 1 and Assumption 3, it is possible
QQ QQ QQ QQ
to show that V1,11 (q, θ) < 0, V1,12 (q, θ) < 0 and |V1,11 (q, θ)| > |V1,12 (q, θ)|. Hence, in the (q1 , q2 )
plane, the SV QQ
1 curve is negatively sloped and that its slope is more than unity in absolute sense.
Therefore, we can represent it as q1 = SV1QQ (q2 , θ). The reaction function of Firm 2 is locus of
all points in the set RQQ
2
QQ
= {q | q >> (0, 0), π2,2 (q) = 0}, again by Assumption 2 in the (q1 , q2 )
plane RQQ 2 is strictly decreasing with slope less than unity in absolute sense (see Cheng [5]), hence
it is invertible. Therefore we can represent it as q2 = R2QQ (q1 ). The graphs of RQQ 2 and SV QQ
1 (θ)
in price space are respectively P(RQQ 2 ) = {p | π QQ
2,2 (q) = 0 and q i = Fi
P P (p) ∀ i = 1, 2} and
QQ
P(SV QQ (θ)) = {p | V1,1 (q, θ) = 0 and qi = FiP P (p) ∀ i = 1, 2} and their respective equations in
 PLAUSIBILITY OF BERTRAND COMPETITION 7

implicate form are F1P P (p) − SV1QQ (F2P P (p), θ) = 0 and F2P P (p) − R2QQ (F1P P (p)) = 0. In Figure
1, the set of points in P(SV QQ
1 (0)) is represented by the line p1 = c. Like Cheng [5], one can
show that the set of points P(RQQ P P P P 0 . One such representation is
2 ) must lie above the R2 R2
the r2 r20 curve in Figure 1.

p2 p1 = c 0 0
R1P P S1P P p1 = p2

D 0
A0 S2P P
r20
0
R2P P

r2

B0 C
B

R2P P
p2 = c

S2P P
p1
o SP P R1P P
1

Figure 1. The case of imperfect substitutes.

The proof of Theorem 1 follows from the next six lemmas.

Lemma 1. For any weight θ ∈ (0, 1), π1,1 P P (pP P (θ), pP P (θ)) > 0 and for any Firm i with
1 2
i ∈ {1, 2}, WiP P (pP1 P (θ), pP2 P (θ)) < 0.
Lemma 2. For any θ ∈ (0, 1),

∂q1QQ (θ) ∂q2QQ (θ)

(i) ∂θ < 0 and ∂θ > 0.
∂pQQ
1 (θ) ∂pP P (θ) ∂pP Q
(θ)
(ii) ∂θ > 0, and, for any i = 1, 2, i
∂θ > 0 and i
∂θ > 0.
8 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

Lemma 1 states that with price competition and given any θ ∈ (0, 1), at any equilibrium price
vector (pP1 P (θ), pP2 P (θ)) it is always optimal for Firm 1 to increase (decrease) price given Firm
2’s price remains at pP2 P (θ) when Firm 1 is a profit (welfare) maximizer. Lemma 2 provides the
standard comparative static results.
Lemma 3.

(i) There exists a unique θ1 ∈ (0, 1) such that π̄2P P (θ) R π̄2P Q (θ) if and only if θ Q θ1 .
(ii) There exist θ3 ∈ (0, 1) such that V̄1P P (θ3 ) = V̄1QP (θ3 ) and, for any θ ∈ (0, θ3 ), V̄1P P (θ) >
V̄1QP (θ).
(iii) There exist a unique θ4 ∈ (0, 1) such that π̄2QP (θ) R π̄2QQ (θ) if and only if θ Q θ4 .

Assume that Firm 1 chooses price strategy. Lemma 3 (i) states that there exist a unique
θ1 ∈ (0, 1) for which Firm 2 is indifferent between choosing price strategy and quantity strategy.
Moreover, if θ < θ1 , then price strategy is optimal for Firm 2, and, if θ > θ1 , then quantity
strategy is optimal for Firm 2. Next, assume that Firm 2 chooses price strategy. Lemma 3
(ii) states that there exist θ3 ∈ (0, 1) for which Firm 1 is indifferent between choosing price or
quantity strategy. Moreover, if θ < θ3 , then Firm 1 chooses price strategy. Lemma 3 (iii) states
that when Firm 1 chooses quantity strategy, there exist an unique θ4 ∈ (0, 1) for which Firm
2 is indifferent between choosing price strategy and quantity strategy. For any θ < θ4 , price
strategy is optimal and, for any θ > θ4 , quantity strategy is optimal.
The cut-off point θ1 (θ4 ) is associated with the case where Firm 1 chooses price (quantity)
strategy. These cut-off points in Lemma 3 (i) and (iii) reflects the reversal in the cost of adopting
price strategy for Firm 2 compared to quantity strategy. For the privatization weights below
these cut-off points the reverse intuition of Vives [26] holds. To prove Lemma 3 (i) and (iii) we
use the line integral technique used in Ghosh and Mitra [12], [13].2
Lemma 4. Under price competition in Stage 3, the resulting welfare W̄ P P (θ) is maximized at
θ = θP P = 0. Moreover, at θP P = 0, the government can induce price strategy for both firms.

Lemma 4 indicates that no privatization Bertrand equilibrium is a possible SPNE outcome

of Γ provided we can rule out the other modes of competition.
Lemma 5. There is no θP Q ∈ [0, 1] such that (θP Q , (P, pP1 Q (θP Q )), (Q, q2P Q (θP Q ))) is an SPNE
outcome of Γ.
Lemma 6. There is no θQP ∈ [0, 1] such that (θQP , (Q, q1QP (θQP )), (P, pQP
2 (θ
QP ))) is an SPNE

outcome of Γ.

Lemma 5 and Lemma 6 rule out the possibilities of Type 1 and Type 2 equilibria. Let
us generate the Cournot equilibrium path in the (p1 , p2 ) space by varying θ from 0 to 1 and
plotting the corresponding price vector. See the path B 0 A0 in Figure 1 where B 0 corresponds to
(pQQ QQ 0 QQ QQ
1 (0), p2 (0)) and A corresponds to (p1 (1), p2 (1)). The next lemma captures the exact
behavior of the Cournot equilibrium path as we vary θ.
Lemma 7. Let (pQQ QQ QQ 0 QQ 0
1 (θ), p2 (θ)) and (p1 (θ ), p2 (θ )) be any two points along the Cournot
equilibrium path. If (pQQ QQ QQ QQ QQ 0 QQ 0
1 (θ), p2 (θ)) is closer to (p1 (1), p2 (1)) than (p1 (θ ), p2 (θ )) in
0
terms of arch length of the path, then θ > θ .
2For a more formal approach to the technique of line integral see Apostol [2].
 PLAUSIBILITY OF BERTRAND COMPETITION 9

Lemma 7 can be explained in terms of the B 0 A0 segment of the r2 r20 in Figure 1. For each
point in the segment B 0 A0 , we can associate a (pQQ QQ
1 (θ), p2 (θ)) vector. Lemma 7 states that
as we move along the B A segment of the r2 r2 curve (starting from B 0 and ending at A0 ), the
0 0 0

underlying θ increases. Finally, to complete the proof of Theorem 1, we need to eliminate the
possibility of quantity competition. Given Lemma 7 identifies the properties of the Cournot
equilibrium path in terms of θ, we can use this path along with the cut-off point θ4 (identified
in Lemma 3 (iii)) to establish the impossibility of quantity competition. Hence, we have Lemma
8.
Lemma 8. There is no θQQ ∈ [0, 1] such that (θQQ , (Q, q1QQ (θQQ )), (Q, q2QQ (θQQ ))) is an SPNE
outcome of Γ.

It is possible that there exists a level θ for which the welfare under quantity competition
is higher than the welfare associated with no privatization Bertrand equilibrium. But, due to
Lemma 8, it is not possible for the government to induce quantity competition with such a θ.

5. The result with complement goods

To obtain the equilibrium outcome when the goods are complement we use the following
assumptions.
Assumption 5. For all i, j ∈ {1, 2} such that i 6= j and for all q >> (0, 0), (i) Ui (q) > 0, (ii)
Uii (q) < 0,(iii) Uij (q) > 0 and (iv) |Uii (q)| > |Uij (q)|.
QQ
Assumption 6. For any i, j = 1, 2 with i 6= j and any q >> (0, 0), (i) πi,ij (q) > 0 and (ii)
QQ QQ
πi,ii (q) + |πi,ij (q)| < 0.
P P (p) < 0 and (ii)
Assumption 7. For i, j = 1, 2 with i 6= j and any p >> (0, 0), (i) πi,ij
P P (p) + |π P P (p)| < 0.
πi,ii i,ij

Assumption 8. For any i, j ∈ {1, 2} with i 6= j, (i) WiiP P (p) < 0; (ii) WiiP P (p) − WijP P (p) < 0.

Assumption 5, Assumption 6 and Assumption 7 are very standard and these are satisfied by
any standard demand function when the goods are complements (see Vives [27]). Assumption
8 (i) is necessary to satisfy the second order condition of any welfare maximizing firm.
Before stating our result, the implications of these assumptions in terms of reactions functions
in the price plane is explained using Figure 2. In particular, we are interested in the function
p2 = R2P P (p1 ), the set P(RQQ
2 ) for Firm 2 and, for θ ∈ {0, 1}, we are interested in the function
p1 = SV1 (p2 , θ) and the set P(SV QQ
P P 0
1 (θ)) for Firm 1. In Figure 2, the curve R2 R2 represents
the reaction function of Firm 2 when Firm 1 chooses price strategy, that is, p2 = R2P P (p1 ). By
Assumption 7, it is decreasing in p1 with an absolute slope less than unity. Given Assumption
P P (p) < 0 implying that in the region above the R R0 curve π P P (p) < 0 and in the region
7(ii), π2,22 2 2 2,2
below the R2 R20 curve we have π2,2P P (p) < 0. Therefore, given π P P (p) = (p − c)F P P (p) < 0, in
2,1 2 2,1
the region above the R2 R20 curve, the iso-profit curve of Firm 2 is decreasing and in the region
below the R2 R20 curve, the iso-profit curve of Firm 2 is increasing. In each point in the set
P(RQQ PP PP
2 ), Firm 2 maximizes profit π2 (p) subject to q1 = F1 (p). Hence, each point in the set
QQ
P(R2 ) is a point of tangency between the iso-profit curve of Firm 2 and the iso-quantity curve
of Firm 1. By Assumption 5, the iso-quantity curve of Firm 1 is negatively sloped implying that
10 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

the tangency of the iso-quantity curve of Firm 1 with the iso-profit curve of Firm 2 must lie
above the R2 R20 curve. Therefore the set of points in P(RQQ 0
2 ) lie above the R2 R2 curve. Finally,
0
as we move along the R2 R2 curve towards the p2 axis, the profit of the Firm 2 increases since
dπ2P P (p1 ,R2P P (p1 )) ∂π P P (p ,RP P (p ))
dp1 = 2 1∂p12 1 < 0. In Figure 2, the R1 R10 curve is the reaction function of
Firm 1, that is, p1 = SV1P P (p2 , 1) for θ = 1. By Assumption 7, it is decreasing and the slope is
greater than unity. One can also show that each point in the set P(SV QQ 1 (1)) lies to the right
of the R1 R10 curve. By definition, p1 = c represents the set of points in the set P(SV QQ
1 (0)). In
0 P P
Figure 2, the S1 S1 curve represents the function p1 = SV1 (p2 , 0) and it satisfies the following
condition.
PP PP
(5.1) (p1 − c)F1,1 (p) + (p2 − c)F2,1 (p) = 0.
By Assumption 5, F1,1 P P (p) < 0, F P P (p) < 0 and |F P P (p)| > |F P P (p)| and hence using (5.1) it
2,1 1,1 2,1
0
follows that the S1 S1 curve must lie between the p1 = c line and the p1 + p2 = 2c line (see line
P P 0 in Figure 2). Similarly, the S2 S20 curve represents the locus of points satisfying W2P P (p) = 0
and this curve lies between the p2 = c and the P P 0 lines. In Figure 2, point B is the intersection
point between the R1 R10 curve and the R2 R20 curve representing the Bertrand equilibrium point
for θ = 1. Since both firm are facing symmetric demand and identical cost conditions, point B
lies on the p1 = p2 line. Point A is the point of intersection between the S1 S10 curve and the
R2 R20 curve representing the Bertrand equilibrium for θ = 0. Given any θ ∈ [0, 1], the function
p1 = SV1P P (p2 , θ) lies between the S1 S10 curve and the R1 R10 curve. Therefore, for any θ, the
equilibrium price vector (pP1 P (θ), pP2 P (θ)) must belong to the segment AB of the R2 R20 curve.
Proposition 1. Suppose Assumption 5, Assumption 6, Assumption 7 and Assumption 8 holds.
There exists θP P ∈ (0, 1) such that the strictly partial privatization Bertrand equilibrium strat-
egy combination (θP P , (P, pP1 P (θP P )), (P, pP2 P (θP P ))) is the unique SPNE outcome of Γ.

6. On the structure of Γ and on cost asymmetry

Following Kreps and Scheinkman’s [17] argument on the importance of game form, we check
how important our three stage game Γ is in driving Theorem 1 and Proposition 1. We do this
robustness check with quadratic utility function given by (2.1). Firstly, if we interchange Stage
1 and Stage 2 of Γ, then, in case of imperfect substitutes, we have no privatization Bertrand
equilibrium as the unique SPNE outcome, and, in case of complements, we have strictly partial
privatization Bertrand equilibrium with θP P = −γ(1 + γ)/(4 + 3γ) ∈ (0, 1) as the unique SPNE
outcome.3 Secondly, keeping everything else unchanged, suppose in Stage 1 of Γ we replace
the objective function of the government by V1 := θπ + (1 − θ)W where θ ∈ [0, 1]. For both
imperfect substitutes and complements, no privatization Bertrand equilibrium is the unique
SPNE outcome. Finally, ceteris paribus, suppose in Stage 1 we replace the objective function of
the government by V1 := π + (1 − θ)CS where θ ∈ [0, 1]. In case of substitute we have strictly
partial privatization Bertrand equilibrium with θP P = γ/(4 + γ − 6γ 2 − 3γ 3 ) as the unique
SPNE outcome provided the goods are ‘sufficiently’ differentiated. Specifically, this result holds
for γ ∈ (0, γ̂) where γ̂ ≈ 0.8. In case of complements, we have no privatization Bertrand
equilibrium as the unique SPNE outcome and it holds for all γ ∈ (−1, 0).
Suppose Ci (q) = ci qi is the total cost function of Firm i for i = 1, 2 and assume that c1 6= c2 .
If the difference in the marginal costs of the two firms is ‘large enough’, then results can change
3This result can be generalized to include other forms of utility specifications as well.
 PLAUSIBILITY OF BERTRAND COMPETITION 11

p2 S10 SV 0 (θ∗ )
p1 = c 1 R1 p1 = p2

R2

A
C
D
B
P

R20
S2
p2 = c

S20

p1
o S1 SV1 (θ∗ ) P0 R10

Figure 2. The case of complements.

for imperfect substitutes (see Zanchettin [28]). Keeping the game Γ unchanged, if we assume
cost asymmetry, then, with quadratic utility function given by (2.1), we have the following
results. In case of imperfect substitutes, if γ(3 − γ 2 )/2 < α1 /α2 < 1/γ, then we have no
privatization Bertrand equilibrium as the unique SPNE outcome of Γ where αi = a − ci > 0
for all i = 1, 2. In case of complements, for any γ ∈ (−1, 0), we have the strictly partial
privatization Bertrand equilibrium as the unique SPNE outcome of Γ with θP P = −γ(1 −
2 2

γ )(α2 − γα1 )/ (4 − 3γ )(α1 − γα2 ) − γ(α2 − γα1 ) ∈ (0, 1).
Therefore, with differentiated products and ‘bounded’ cost difference, price competition is
an inescapable equilibrium outcome when a regulated partially privatized firm competes with a
private firm in a three stage game like ours. What will happen if we have three or more firms is
still an important open question.
12 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

7. Appendix

Proof of Lemma 1: We use two steps to prove the result.

Step 1: In Stage 3, if the firms compete in prices, then, for any θ ∈ [0, 1], the equilibrium choice
vector (pP1 P (θ), pP2 P (θ)) is unique.
Proof of Step 1: In the third stage, given p2 , Firm 1 chooses p1 to maximizing V1P P (p, θ) =
θ.π1P P (p)+(1−θ)W P P (p) and, given p1 , Firm 2 chooses p2 to maximize π2P P (p) = (p2 −c)F2P P (p).
The first order conditions are the following:
PP
(7.1) V1,1 (p, θ) = θF1P P (p) + (p1 − c)F1,1
PP PP
(p) + (1 − θ)(p2 − c)F2,1 (p) = 0,
and
PP
(7.2) π2,2 (p) = F2P P (p) + (p2 − c)F2,2
PP
(p) = 0.
Using Assumption 3 (ii) and Assumption 4 (i) it follows that V1,11 P P < 0 and π P P < 0. Therefore,
2,22
second order conditions for maximization are satisfied. Since π2,12 P P > 0, Firm 2’s reaction
P P | > |π P P | implies that the slope of the reaction
function is increasing in (p1 , p2 ). Moreover, |π2,22 2,12
function of the Firm 2 is less than unity. The sign of V1,12 P P is still ambiguous. If for some
P P P P P P
(p1 (θ), p2 (θ)), V1,12 > 0, then by Assumption 4 (ii), the slope of the reaction function of the
Firm 1 must be greater than unity implying that the intersection of this reaction function with
∂V P P
Firm 2’s reaction function is unique since, along the ∂p1 1 = 0 curve, given any p2 we have
∂V P P
only one p1 , the locus of the function ∂p1 1 = 0 will never intersect Firm 2’s reaction function
twice. If for some (pP1 P (θ), pP2 P (θ)), V1,12
P P = 0, then at that point Firm 1’s reaction function has

a slope of ∞. Given that the slope of the reaction function of Firm 2 is increasing (and is less
than unity), we have a unique best response for Firm 1 given any p2 implying uniqueness of the
equilibrium point. Finally, if for some (pP1 P (θ), pP2 P (θ)), V1,12
P P < 0, then it is obvious that we

will have a unique intersection.

P P (pP P (0), pP P (0)) > 0.
Step 2: π1,1 1 2

Proof of Step 2: At θ = 0, the equilibrium price vector (pP1 P (0), pP2 P (0)) satisfy following first
order conditions
(7.3) (pP1 P (0) − c)F1,1
PP PP
(p1 (0), pP2 P (0)) + (pP2 P (0) − c)F2,1
PP PP
(p1 (0), pP2 P (0)) = 0,
and
(7.4) (pP2 P (0) − c)F2,2
PP PP
(p1 (0), pP2 P (0)) + F2P P (pP1 P (0), pP2 P (0)) = 0.

By definition q2P P (0) := F2P P (pP1 P (0), pP2 P (0)) > 0 and, by Assumption 1 it follows that
F2,2P (pP1 P (0), pP2 P (0)) <
P 0. Therefore, from (7.4) we have pP2 P (0) > c. Given pP2 P (0) >
c, and by Assumption 1, F2,1 P P (pP P (0), pP P (0)) > 0 and F P P (pP P (0), pP P (0)) < 0. There-
1 2 1,1 1 2
P P
fore, from (7.3) we get p1 (0) > c. By Assumption 1 we also have F2,1 P P (pP P (0), pP P (0)) =
1 2
P P (pP P (0), pP P (0)) < |F P P (pP P (0), pP P (0))|. Hence, given pP P (0) > c and pP P (0) > c, from
F1,2 1 2 1,1 1 2 2 1
condition (7.3) we get pP2 P (0) > pP1 P (0) > c. Using pP2 P (0) > pP1 P (0) > c and using the fact
that the demands are symmetric with own effect dominant cross effect we have,
(7.5)
F1P P (pP1 P (0), pP2 P (0)) > F1P P (pP1 P (0), pP1 P (0)) = F2P P (pP1 P (0), pP1 P (0)) > F2P P (pP1 P (0), pP2 P (0)).
 PLAUSIBILITY OF BERTRAND COMPETITION 13

Finally,
PP PP
π1,1 (p1 (0), pP2 P (0)) = (pP1 P (0) − c)F1,1
PP PP
(p1 (0), pP2 P (0)) + F1P P (pP1 P (0), pP2 P (0))
= −(pP2 P (0) − c)F2,1
PP PP
(p1 (0), pP2 P (0)) + F1P P (pP1 P (0), pP2 P (0))
> (pP2 P (0) − c)F2,2
PP PP
(p1 (0), pP2 P (0)) + F1P P (pP1 P (0), pP2 P (0))
> (pP2 P (0) − c)F2,2
PP PP
(p1 (0), pP2 P (0)) + F2P P (pP1 P (0), pP2 P (0))
= 0.
Here the first equality is by definition, the second equality is due to (7.3), the first inequality
P P (pP P (0), pP P (0)) > F P P (pP P (0), pP P (0)) and last inequality is due to (7.5).
follows from −F2,1 1 2 2,2 1 2
This proves Step 2.
To complete the proof we also use Figure 3. Given any θ, its (unique) corresponding equi-
librium price vector (pP1 P (θ), pP2 P (θ)) is the intersection of the reaction function of Firm 1
p1 = SV1P P (p2 , θ), and the reaction function of the Firm 2 p2 = R2P P (p1 ). By condition
(7.2), R2P P (c) > c and 0 < dR2P P (p1 )/dp1 < 1 implying that p2 = R2P P (p1 ) must intersect
the p1 = p2 line from above (see Figure 3). Thus, to the left of the p1 = p2 line along Firm
2’s reaction function p2 = R2P P (p1 ) we have p2 > p1 . Moreover, by symmetry of the firms, at
θ = 1 we have pP1 P (1) = pP2 P (1). Hence, the intersection point of the curve p2 = R2P P (p1 ) and
the line p1 = p2 is also the intersection point of the curves p2 = R2P P (p1 ) and p1 = SV1P P (p2 , 1).
By Step 2, the intersection point of p2 = R2P P (p1 ) and p1 = SV1P P (p2 , 0) must lie to the
left of p1 = SV1P P (p2 , 1) and, for any θ ∈ (0, 1), p1 = SV1P P (p2 , θ) is bounded between
p1 = SV1P P (p2 , 0) and p1 = SV1P P (p2 , 1) (given Assumption 3 and Assumption 4(i)). As a result,
every equilibrium price vector (pP1 P (θ), pP2 P (θ)) must belongs to the segment of p2 = R2P P (p1 )
that lie between intersection of p1 = SV1P P (p2 , 0) and p1 = SV1P P (p2 , 1), that is, the over braced
segment P P 0 in Figure 3.

0 0
p2 p1 = c S1P P R1P P p1 = p2

0
R2P P
P0
PP
P

R2P P

p2 = c

o p1
S1P P R1P P

Figure 3

The P P 0 segment in Figure 3 must lie completely to the left of p1 = SV1P P (p2 , 1) implies
π1,1 1 2 > 0. The P P 0 segment in Figure 3 lies completely to the right of p1 =
P P (pP P (θ), pP P (θ))
14 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

SV1P P (p2 , 0) implies W1P P (pP1 P (θ), pP2 P (θ)) < 0. Finally, for p2 > p1 > c, the P P 0 segment in
Figure 3 must lie completely above the p1 = p2 line implies W2 (pP1 P (θ), pP2 P (θ)) < 0. 
∂q QQ ∂q QQ
Proof of Lemma 2: To prove ∂θ 1
< 0 and ∂θ 2
> 0, we first differentiate the two functions
QQ QQ QQ QQ QQ QQ
V1,1 (q1 (θ), q2 (θ), θ) = 0 and π2,2 (q1 (θ), q2 (θ)) = 0 with respect to θ and then solve for
∂q1QQ ∂q2QQ
∂θ and ∂θ . This results in
QQ
QQ QQ ∂V1,1
∂q1QQ π2,22 (q (θ)) ∂θ
=− ,
∂θ |AQQ |
and
QQ
QQ QQ ∂V1,1
∂q2QQ π2,12 (q (θ)) ∂θ
= ,
∂θ |AQQ |
QQ
∂V1,1
where for any θ ∈ [0, 1], q QQ (θ) := (q1QQ (θ), q2QQ (θ)), ∂θ
QQ
= π1,1 − W1QQ = q1QQ (θ)F1,1
QQ
<0
QQ
QQ QQ QQ QQ ∂q1 ∂q2QQ
and |AQQ | = V1,11 π2,22 − V1,12 π2,12 > 0. Hence, we have ∂θ < 0 and ∂θ > 0.
∂pQQ QQ ∂q1QQ QQ ∂q2QQ
∂q2QQ ∂q1QQ QQ QQ
Observe that 1
∂θ = F1,1 ∂θ + F1,2 ∂θ / ∂θ
∂θ = −π2,12
and that (q)/π2,22 (q) =
QQ QQ QQ QQ
dR2 (q1 )/dq1 . From Assumption 1 and Assumption 2 we have F1,1 + F1,2 (dR2 (q1 )/dq1 ) < 0.
∂q1QQ ∂pQQ
  QQ
QQ QQ ∂q1
Hence, using the earlier result ∂θ < 0, we get ∂θ
1
= F1,1 + F1,2 (dR2QQ (q1 )/dq1 ) ∂θ >
0.
∂pP P (θ)
For any θ ∈ [0, 1], define pP P (θ) := (pP1 P (θ), pP2 P (θ)). To show i∂θ > 0 for i = 1, 2, we
P P (pP P (θ), pP P (θ), θ) = 0 and π P P (pP P (θ), pP P (θ)) = 0 with
first differentiate the functions V1,1 1 2 2,2 1 2
respect to θ. This gives
PP PP ∂pP1 P (θ) PP PP ∂pP P (θ) PP PP
(7.6) V1,11 (p (θ)) + V1,12 (p (θ)) 2 = −(π1,1 (p (θ)) − W1P P (pP P (θ))),
∂θ ∂θ
and
PP ∂pP1 P (θ) ∂pP P (θ)
(7.7) π2,12 (pP P (θ)) PP
+ π2,22 (pP P (θ)) 2 = 0.
∂θ ∂θ
∂pP P
1 (θ) ∂pP P
2 (θ)
Solving for ∂θ and ∂θ from (7.6) and (7.7) we obtain
P P (p)(W P P (pP P (θ)) − π P P (pP P (θ)))
∂pP1 P (θ) π2,22 1 1,1
= ,
∂θ |AP P |
and
P P (p)(π P P (pP P (θ)) − W P P (pP P (θ)))
∂pP2 P (θ) π2,12 1,1 1
= .
∂θ |AP P |
The term |AP P | = V1,11 P P (pP P (θ))π P P (pP P (θ)) − V P P (pP P (θ))π P P (pP P (θ)) is positive due to
2,22 1,12 2,12
Assumption 3 and Assumption 4. Moreover, using Lemma 1 it also follows that for every θ ∈
PP
P P (pP P (θ), pP P (θ)) − W P P (pP P (θ), pP P (θ)) > 0. Hence, for each θ ∈ (0, 1), ∂p1
(0, 1), π1,1 1 2 1 1 2 ∂θ > 0
∂pP P
and 2
∂θ > 0.
∂pP Q
Next we prove that ∂θ i
> 0 for i = 1, 2. Suppose, given any q2 , Firm 1 chooses p1 to
PQ PQ
maximize V1 (p1 , q2 ) = θπ1 (p1 , q2 ) + (1 − θ)W P Q (p1 , q2 ) and, given any p1 , Firm 2 chooses
 PLAUSIBILITY OF BERTRAND COMPETITION 15

q2 to maximize π2P Q (p1 , q2 ) = (F2P Q (p1 , q2 ) − c)q2 4. The first order condition of Firm 1 is
PQ
(7.8) V1,1 (p1 , q2 ) = θF1P Q (p1 , q2 ) + (p1 − c)F1,1
PQ
(p1 , q2 ) = 0.
The first order condition of Firm 2 is
PQ
(7.9) π2,2 (p1 , q2 ) = (F2P Q (p1 , q2 ) − c) + q2 F2,2
PQ
(p1 , q2 ) = 0.

Observe that Firm 1’s reaction function is p1 = F1QQ (SV1QQ (q2 , θ), q2 ) and Firm 2’s reaction
function is q2 = F2P P (p1 , R2P P (p1 )). For any θ ∈ [0, 1],
(7.10) pP1 Q (θ) = F1QQ (SV1QQ (q2P Q (θ), θ), q2P Q (θ)),
and
(7.11) q2P Q (θ) = F2P P (pP1 Q (θ), R2P P (pP1 Q (θ))).
∂pP Q
∂q2P Q
Differentiating (7.10) and (7.11) with respect to θ and then solving for 1
∂θ and ∂θ we get,
QQ ∂SV QQ
∂pP1 Q F1,1 (q) ∂θ1
= ,
∂θ |AP Q |
and
∂SV1QQ dR2P P
 
QQ P P (p) + F P P (p)
∂q2P Q F1,1 (q) ∂θ F2,1 2,2 dp1
= ,
∂θ |AP Q |
∂SV1QQ QQ QQ QQ
where ∂θ = −q1 F1,1 /V1,11 (q) < 0 and |AP Q | > 0.5 Hence, given F1,1 (q) < 0, we get
∂pP Q
∂q2P Q ∂pP Q
P P (p) + F P P (p)(dRP P /dp ) implies that
1
∂θ > 0 . Finally, ∂θ / ∂θ
1
= F2,1 2,2 2 1

∂pP2 Q PQ ∂pP1 Q PQ ∂q2P Q

= F2,1 (p1 , q2 ) + F2,2 (p1 , q2 )
∂θ ∂θ ∂θ
 P Q
dR2P P
 
PQ PQ PP PP ∂p1
(7.12) = F2,1 (p1 , q2 ) + F2,2 (p1 , q2 ) F2,1 (p) + F2,2 (p)
dp1 ∂θ
dR2P P ∂pP1 Q
= > 0.
dp1 ∂θ

Proof of Lemma 3:
Proof of (i): In the price space, given any θ ∈ (0, 1), if Firm 2 chooses price strategy, then
Firm 1’s reaction function is p1 = SV1P P (p2 , θ), and, if Firm 2 chooses quantity strategy, then
Firm 1’s reaction function is the set of points P(SV QQ 1 (θ)) and can be written in implicit form
P P QQ P P
as F1 (p) − SV1 (F2 (p), θ) = 0. Given Firm 1 chooses price strategy, Firm 2’s reaction
function is p2 = R2P P (p1 ). Fix a θ ∈ [0, 1]. Consider p1 (t) = tpP1 P (θ) + (1 − t)pP1 Q (θ) defined for

4where F P Q (p , q ) is demand function of firm i, for i = 1, 2

i 1 2  
QQ
dRP P

5Specifically, |AP Q | = 1 − F QQ (q) dSV1 + F QQ (q) F P P (p) + F P P (p) 2
1,1 dq2 1,2 2,1 2,2 dp1
QQ QQ QQ
! ! !
dRP P dSV dSV R2PP dRP P dSV1
U11 U22 +U11 | dp2 | dq1 |−U12 | dq1 |−U12 dp U11 U12 1− dp2 1−| dq2
|
1 2 2 1 1
= |D|
> |D|
> 0.
16 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

P P (p) at p = pP Q (θ) with endpoints

each t ∈ [0, 1]. Applying the line integral on the function π2,2 2 2
pP1 P (θ) and pP1 Q (θ) we get
Z 1
PQ PP PQ PQ PQ
PP PP PP
π2,2 (p1 (θ), p2 (θ)) − π2,2 (p1 (θ), p2 (θ)) = (p1 (θ) − p1 (θ)) PP
π2,12 (p1 (t), pP2 Q (θ))dt.
0

The point (pP1 Q (θ), pP2 Q (θ)) is on p2 = R2P P (p1 ). Hence, P P (pP Q (θ), pP Q (θ))
π2,2 1 2 = 0. As a result
we have
Z 1
(7.13) π2,2 (p1 (θ), pP2 Q (θ))
PP PP
= (pP1 P (θ) − pP1 Q (θ)) PP
π2,12 (p1 (t), pP2 Q (θ))dt.
0
R1 PQ
By Assumption 3 (i) PP
0 π2,12 (p1 (t), p2 (θ))dt > 0 implying pP1 P (θ) R pP1 Q (θ) if and only if
P P (pP P (θ), pP Q (θ)) R 0. Observe first that
π2,2 1 2

(7.14) lim π2,2 (p1 (θ), pP2 Q (θ)) = π2,2

PP PP
(p1 (0), pP2 Q (0)) > 0.
PP PP
θ→0

Condition (7.14) holds since from the first order condition of profit maximization and welfare
maximization we have c = pP1 Q (0) < pP1 P (0) < pP2 P (0) and since R2P P is increasing, that is,
pP2 Q (0) < pP2 P (0) therefore (pP1 P (0), pP2 Q (0)) lie below the R2P P hence implies (7.14). Also
observe that
(7.15) lim π2,2 (p1 (θ), pP2 Q (θ)) = π2,2
PP PP
(p1 (1), pP2 Q (1)) < 0.
PP PP
θ→1

Condition (7.15) holds since pP1 P (1) = pP2 P (1) < pP2 Q (1) implies that the point (pP1 P (1), pP2 Q (1))
lies above the R2P P . Conditions (7.14) and (7.15) implies that there exist θR ,θS with θR ≤ θS
such that for any θ ∈ (0, θR ) and any θ ∈ (θS , 1) we have pP1 P (θ) > pP1 Q (θ) and pP1 P (θ) < pP1 Q (θ)
P P (pP P (θ ), pP P (θ )) − π P P (pP P (θ ), pP P (θ )) = 0 and applying line
respectively. Therefore, π2,2 1 S 2 S 2,2 1 R 2 R
integral to this equality with end points (pP1 P (θS ), pP2 P (θS )) and (pP1 P (θR ), pP2 P (θR )) we get

(7.16)
Z 1 Z 1
(pP1 P (θS ) − pP1 P (θR )) PP
π2,12 (p1 (t), p2 (t))dt + (pP2 P (θS ) − pP2 P (θR )) PP
π2,22 (p1 (t), p2 (t))dt = 0.
0 0
By Assumption 3 and Lemma 2 (ii) it follows that for condition (7.16) to hold we must have
pP1 P (θS ) > pP2 P (θS ) > pP2 P (θR ) > pP1 P (θR ) if θR < θS . But for each θ ∈ [0, 1] we have pP2 P (θ) ≥
pP1 P (θ). Therefore, pP1 P (θS ) > pP2 P (θS ) is a contradiction, hence we have θS = θR = θ1 . Thus ∃
a unique θ1 ∈ (0, 1) such that pP1 P (θ) R pP1 Q (θ) if and only if θ Q θ1 .
Observe that, for any given p1 , firm 2’s optimal profit remains the same regardless of whether it
chooses price strategy or quantity strategy, which is given by π2P Q (p1 ) = π2P P (p1 ) = (R2P P (p1 ) −
c)F2P P (p1 , R2P P (p1 )) = π̄2 (p1 ). Now,
dR2P P (p1 ) P P PP
 
dπ̄2 (p1 )
= F2 (p1 , R2P P (p1 ))+(R2P P (p1 )−c) F2,1 P P (p , RP P (p )) + F P P (p , RP P (p )) dR2 (p1 )
dp1 dp1 1 2 1 2,2 1 2 1 dp1
P P P P P P dR2P P (p1 ) P P P P P P P P P P


= (R2 (p1 )−c)F2,1 (p1 , R2 (p1 ))+ dp1 F2 (p1 , R2 (p1 )) + F2,2 (p1 , R2 (p1 ))(R2 (p1 ) − c) >
0, since R2P P (p1 ) − c > 0 (by condition 7.2), F2,1 P P (p , RP P (p )) > 0 (by Assumption 1) and
1 2 1
p2 = R2P P (p1 ) satisfies firm 2’s first order condition π2,2 P P (p , p ) = 0, and the proof is complete.
1 2

Proof of (ii): Consider the difference V̄1P P (θ) − V̄1QP (θ) at θ = 0. It is easy to check that
V̄1P P (0)− V̄1QP (0) = W P P (pP1 P (0), pP2 P (0))−W P P (pQP QP
1 (0), p2 (0)) > 0. In particular, whatever
 PLAUSIBILITY OF BERTRAND COMPETITION 17

be the shape of the locus of W1P P (p) = 0, starting from the point (c, c) as we move along
that locus by increasing p2 , the welfare has to fall (see Figure 4) and, since the transformed
QQ
reaction function (π1,1 (q) = 0) of Firm 1 in price space must lie above p1 = R2P P (p2 ), we have
pP2 P (0) > pQP PP PP QP QP PP
2 (0) and (p1 (0), p2 (0)) and (p1 (0), p2 (0)) lie on the locus of W1 (p) = 0.

p2 p1 = c S10 p1 = p2

{p | W1P P (p) = 0}

p2 = c

p1
o S1

Figure 4

Now consider V̄1P P (θ)−V̄1QP (θ) at θ = 1. We have, V̄1P P (1)−V̄1QP (1) = π1P P (pP1 P (1), pP2 P (1))−
π1QP (pQP QP
1 (1), p2 (1)) < 0 since for a profit maximizing firm quantity strategy strictly dominates
price strategy. Since V̄1P P (θ) − V̄1QP (θ) is a continuous function of θ the result follows.
Proof of (iii): For this proof we restrict our attention to the quantity space (q1 , q2 ). Given any θ ∈
(0, 1), if Firm 2 chooses quantity strategy, then Firm 1’s reaction function is q1 = SV1QQ (q2 , θ).
If Firm 2 chooses price strategy, then Firm 1’s reaction function is p1 = SV1P P (p2 , θ). If
we transform p1 = SV1P P (p2 , θ) to the quantity space, then we can be write it implicitly as
F1QQ (q) − SV1QQ (F2QQ (q), θ) = 0. Given Firm 1 chooses quantity strategy, Firm 2’s reaction
function is q2 = R2QQ (q1 ).

Fix a θ ∈ [0, 1]. Consider q(t) = tq1QP (θ) + (1 − t)q1QQ (θ) defined for each t ∈ [0, 1]. Applying
QQ
the line integral on the function π2,2 (q) at q2 = q2QQ (θ) with end point q1QP (θ) and q1QQ (θ) we
have
Z 1
QQ QP
π2,2 (q1 (θ), q2QQ (θ)) − π2,2
QQ QQ
(q1 (θ), q2QQ (θ)) = (q1QP (θ) − q1QQ (θ)) QQ
π2,12 (q1 (t), q2QQ (θ))dt.
0

The point (q1QQ (θ), q2QQ (θ)) is on q2 = R2QQ (q1 )

implying QQ QQ
π2,2 (q1 (θ), q2QQ (θ)) = 0. Hence
Z 1
QQ QP QQ QP QQ QQ
(7.17) π2,2 (q1 (θ), q2 (θ)) = (q1 (θ) − q1 (θ)) π2,12 (q1 (t), q2QQ (θ))dt.
0
R1 QQ
By Assumption 2 (i), 0 π2,12 (q1 (t), q2QQ (θ))dt < 0. Therefore, q1QP (θ) Q q1QQ (θ) if and only
QQ QP
if π2,2 (q1 (θ), q2QQ (θ)) R 0. Let r1P P (q1 ) be the transformed price reaction of Firm 1. Given
q2QP (1) < q2QQ (1), we have r1P P (q2QP (1)) > R1QQ (q2QP (1)) > R1QQ (q2QQ (1)) implying q1QP (1) >
18 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

q1QQ (1). Hence (q1QP (1), q2QQ (1)) must lie above the R2QQ curve. Thus, we have
QQ QP
(7.18) lim π2,2 (q1 (θ), q2QQ (θ)) = π2,2
QQ QP
(q1 (1), q2QQ (1)) < 0.
θ→1
Also observe that
QQ QP
(7.19) lim π2,2 (q1 (θ), q2QQ (θ)) = π2,2
QQ QP
(q1 (0), q2QQ (0)) > 0
θ→0
Condition (7.19) holds since the price welfare reaction function of the Firm 1 in the quantity
space must intersect the R2QQ curve to the left of F1QQ (q) = c. Therefore, (q1QP (0), q2QQ (0)) lies
below the R2QQ curve. Condition (7.18) and (7.19) implies that there exist θR 0 , θ 0 with θ 0 ≤ θ 0
S R S
0 ) and θ ∈ (θ 0 , 1) we have q QQ QP QQ QP
such that for all θ ∈ (0, θR S 1 (θ) > q1 (θ) and q1 (θ) < q1 (θ)
QQ QQ 0
respectively. Therefore, π2,2 (q1 (θS ), q2QQ (θS0 ))−π2,2
QQ QQ 0
(q1 (θR ), q2QQ (θR
0 )) = 0 and applying line

integral to the equality with end points (q1QQ (θS0 ), q2QQ (θS0 )) and (q1QQ (θR
0 ), q QQ (θ 0 )) yields
2 R
(7.20)
Z 1 Z 1
QQ 0 QQ 0 QQ QQ 0 QQ 0 QQ
(q1 (θS ) − q1 (θR )) π2,12 (q1 (t), q2 (t))dt + (q2 (θS ) − q2 (θR )) π2,22 (q1 (t), q2 (t))dt = 0.
0 0
By Assumption 2 and Lemma 2 (i), it follows that for condition (7.20) to hold with θR 0 < θ0 ,
S
QQ 0 QQ 0 QQ 0 QQ 0
we must have q1 (θR ) > q2 (θS ) > q2 (θR ) > q1 (θS ). But we know that for all θ ∈ [0, 1],
q2QQ (θ) < q1QQ (θ) and we have a contradiction. As a result we must have θS0 = θR 0 = θ . Thus ∃
4
a unique θ4 ∈ (0, 1) such that q1QP (θ) Q q1QQ (θ) if and only if θ Q θ4 .
 
Observe that, for any given q1 , firm 2’s optimal profit is π̄2 (q1 ) = F2QQ (q1 , R2QQ (q1 )) − c R2QQ (q1 ),
regardless of whether it chooses quantity strategy or price strategy, i.e., π2QQ (q1 ) = π2QP (q1 ) =
π̄2 (q1 ). Now,
dR2QQ (q1 )
  QQ
dπ̄2 (q1 ) QQ QQ QQ QQ QQ QQ QQ dR2 (q1 )
dq1 = [F2,1 (q 1 , R2 (q 1 ))+F2,2 (q 1 , R2 (q 1 )) dq1 ]R2 (q 1 )+ F2 (q 1 , R2 (q 1 )) − c dq1
QQ dR2QQ (q1 )
= F2,1 (q1 , R2QQ (q1 ))R2QQ (q1 ) + dq1
QQ
[F2,2 (q1 , R2QQ (q1 ))R2QQ (q1 ) + F2QQ (q1 , R2QQ (q1 )) − c] =
QQ
F2,1 (q1 , R2QQ (q1 ))R2QQ (q1 ) < 0, since F2,1 QQ
(q1 , R2QQ (q1 )) < 0 (by Assumption 1) and q2 =
R2QQ (q1 ) satisfies firm 2’s first order condition π2,2 QQ
(q) = 0, and the proof is complete. 
Proof of Lemma 4: In Stage 1, when both firms choose price strategy, the government chooses
θ to maximize welfare. Given W̄ P P (θ) = W P P (pP1 P (θ), pP2 P (θ)), differentiating with respect to
θ we get,
∂ W̄ P P ∂pP P (θ) ∂pP P (θ)
(7.21) = W1P P (pP P (θ)) 1 + W2P P (pP P (θ)) 2 .
∂θ ∂θ ∂θ
∂pP P (θ)
By Lemma 2(ii), i∂θ > 0 and by Lemma 1 WiP P < 0. Therefore, from equation (7.21)
∂ W̄ P P
∂θ < 0. So the optimal choice of θ in Stage 1 under price competition is θ = 0.
Given the optimal choice θ = 0 < θ1 , from Lemma 2(i) it follows that it is optimal for Firm 2
to choose price strategy when Firm 1 chooses price strategy in Stage 2. Again, given the Stage
1 optimal choice θ = 0 < θ3 , from Lemma 3(ii) it follows that it is optimal for Firm 1 to choose
price strategy when Firm 2 chooses price strategy in Stage 2. Therefore, in Stage 2, if Firm i
chooses price strategy, then Firm j(j 6= i) also chooses price strategy. Since these choices depend
on the Stage 1 choice of θ, the second stage choice of price strategy by both firms is sequential
rational. Therefore, the result follows. 
 PLAUSIBILITY OF BERTRAND COMPETITION 19

Proof of Lemma 5: We prove Lemma 5 using the following figure.

p2 p1 = c 0 0
R1P P S1P P p1 = p2

0
R2P P

C
A B E
R2P P

p2 = c

p1
o S1P P R1P P

Figure 5. Impossibility of Type I equilibrium

0
In Figure 5, the curve R1P P R1P P is the locus of π1,1 P P (p) = 0 (that is, it is the locus of
6 0
p1 = SV1 (p2 , 1)). By Assumption 3 the curve R1 R1 P is increasing in the price plane with
P P P P P
0
slope greater than unity and hence must lie to the right of the p1 = c line. Since S1P P S1P P is also
0
the locus of W1P P = 0, it must be lie between the p1 = c and p1 = p2 lines. Similarly, R2P P R2P P
is the locus of the function p2 = R2P P (p1 ) and, by Assumption 3, it is always increasing in the
price plane with slope less than unity and hence must lie above the p2 = c line. Therefore, the
0 0
intersection point of R1P P R1P P and R2P P R2P P is the Bertrand equilibrium point C for θ = 1 and
by Assumption 3 this point is unique. Since firms have identical cost and demand conditions,
point C must lie on the p1 = p2 line. By Assumption 3 and Assumption 4 the intersection
0 0
of R2P P R2P P and S1P P S1P P is the Bertrand equilibrium point (B) for θ = 0 and, by Step-2 of
the Lemma 1, the point B must lie to the left of point C on R2P P . Since we do not impose
any restriction on the locus of P(S1P P (1)), it can take any shape and can intersect the curve
0 0
R2P P R2P P more than ones but its locus must lie to the left of the R1P P R1P P curve (see Cheng
0
[5]). Hence, any intersection point between R2P P R2P P and the locus of P(S1P P (1)) must lie to
0
the right of point C on the R2P P R2P P curve.
The line p1 = c is the locus of P(S1P P (0)). If Firm 1 select price strategy, then Firm 2’s optimal
0
reaction is to react along the R2P P R2P P curve (see Singh and Vives [23]) in the price space. Again,
given some θ ∈ [0, 1] if Firm 2 select quantity strategy, then Firm 1 optimally reacts (in terms
6In this Figure 5, we draw all curve as straight line for simplicity of the figure.
20 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

of prices) according to the locus of P(S1P P (θ)) in the price space. Since P(S1P P (θ)) must lie
between the line p1 = c and the locus of P(S1P P (1)), for any given θ, when Firm 1 chooses price
0
strategy and Firm 2 chooses quantity strategy the equilibrium point lie on the R2P P R2P P curve
and it must lie on or to the right of point A.
By Lemma 3(i), when Firm 1 chooses price strategy, there exist a θ1 ∈ (0, 1) at which
Firm 2 is indifferent between choosing price strategy and quantity strategy, and, for θ < θ1 (>
θ1 ), it chooses price (quantity) strategy. Hence, at θ1 the Bertrand equilibrium price vector
(pP1 P (θ1 ), pP2 P (θ1 )) and Type-1 equilibrium price vector (pP1 Q (θ1 ), pP2 Q (θ1 ) induces same profit
0
for Firm 2. The point (pP1 P (θ1 ), pP2 P (θ1 )) is the intersection point of R2P P R2P P and the locus
0
of p1 = SV1P P (p2 , θ1 ) and the point (pP1 Q (θ1 ), pP2 Q (θ1 )) is intersection point of R2P P R2P P and
0
locus of P(S1P P (θ1 )) in the price space. Since along the R2P P R2P P curve, any two distinct
points generate distinct profits, we must have pPi P (θ1 ) = pPi Q (θ1 ). Hence, at (pP1 P (θ1 ), pP2 P (θ1 )),
0
the locus of p1 = SV1P P (p2 , θ1 ) and the locus of P(S1P P (θ1 )) intersect on the R2P P R2P P curve
in the price space and the intersection point is unique by Lemma 3 (i). Since the locus of
0 0
p1 = SV1P P (p2 , θ1 ) must lie between S1P P S1P P and R1P P R1P P and since θ1 ∈ (0, 1), the point
(pP1 P (θ1 ), pP2 P (θ1 )) must lie at the interior on the segment BC of the R2P P curve. Without loss
∂pP Q
of generality, let E be that point. By Lemma 2 (ii) ∂θ i
> 0, any point on the segment AE
excepting point E corresponds to θ < θ1 . Hence, we cannot find any selection θ in Stage 1 for
the government that can induce any (p1 , p2 ) combination that lie in this segment of AE (except
point E). Finally, the government won’t induce any point on or to the right of E since each
0
such point (on the R2P P R2P P ) generates less welfare than at point B. Since the point B can be
induced by choosing θ = 0 (by Lemma 4), the result follows. 
Proof of Lemma 6: We use Figure 6 below to prove this result. In this Figure 6 we introduce
two new curves. The first is the iso-welfare curve corresponding to welfare level of point B
0
(that is, the welfare level W̄ P P (0)). The second is the S2P P S2P P curve which is the locus of
W2P P (p) = 0. Point B is Bertrand equilibrium for θ = 0 and by Lemma 4 this point can be
induced by choosing θ = 0. If the resulting welfare in any equilibrium point corresponds to
Firm 1 choosing quantity strategy and Firm 2 choosing price strategy yields a welfare less than
welfare level corresponding to point B, then the government will never simultaneously induce
Firm 1 to choose quantity strategy and Firm 2 choose price strategy. By Assumption 4 (i),
0 0
in the regions above and below both S1P P S1P P and S2P P S2P P curves, the iso-welfare curve is
0 0
upward sloping and in the region lying between the S1P P S1P P curve and the S2P P S2P P curve, the
iso-welfare curve is downward sloping and the Bertrand equilibrium point at θ = 0 (that is, point
0 0
B) lies on the S1P P S1P P curve and is located above the S2P P S2P P curve. Therefore, to the left
of point B the iso-welfare curve is increasing and to the right of point B it is decreasing. Since
a consequence of welfare maximization in terms of quantity choice yields (p1 = c, p2 = c) under
the optimal choice, it is the unique global maximum of W P P (p). Therefore, the upper contour
set ΩPWP = {p | W P P (p) ≥ W̄ P P (0)} of B is the region shaded in gray in Figure 6 that always
includes point (c, c) as an interior point. When Firm 1 chooses quantity strategy and Firm 2
chooses price strategy, then the reaction function of Firm 1 is the locus of p1 = SV1P P (p2 , θ)
0 0
lying between the R1P P R1P P and the S1P P S1P P curves. The reaction function of Firm 2 is the
0
locus of the set P(RP2 P ) that lies completely above the R2P P R2P P . Therefore, when Firm 1
chooses quantity strategy and Firm 2 chooses price strategy, the equilibrium point are must
0 0
belong to the region lying between the R1P P R1P P curve and the S1P P S1P P curve and must also
0
lie above the R2P P R2P P as shown in the Figure 6 by the dotted region (where the boundary is
 PLAUSIBILITY OF BERTRAND COMPETITION 21

p2 p1 = c 0 0
R1P P S1P P p1 = p2

D
0
S2P P
0
R2P P

C
B
W̄ P P (0)

R2P P
p2 = c

S2P P
p1
o SP P R1P P
1

Figure 6. Impossibility of Type II equilibrium

not included for the BC segment). Hence, the set in which the Type II equilibrium can occur is
P P (p) > 0, W P P (p).π P P (p) ≤ 0}. Since, due to Assumption 4, the S P P S P P 0 curve
E QP = {p | π2,2 1 1,1 1 1
can never bend back, the set ΩPWP and the set E QP must be disjoint (since only intersection of
the boundary of E QP and ΩPWP is point B and which is not in E QP ). Hence, given any Type II
equilibrium price vector, we can always find a θ, in particular θ = 0, that will ensure Bertrand
competition in the Stage 2 and generate higher welfare. Hence Type II equilibrium is ruled
out. 
Proof of Lemma 7: To prove Lemma 7 let us consider Figure 7 where in Figure 7a we consider
the quantity space and in Figure 7b we consider the price space. In Figure 7a, the curves R1 R10 ,
RC and R2 R20 corresponds respectively to the function q1 = SV1QQ (q2 , 1), q1 = SV1QQ (q2 , 0) and
q2 = R2QQ (q1 ). Each curve is negatively sloped and both R1 R10 and R1 C curves have an absolute
slope of more than unity and the R2 R20 curve has an absolute slope of less than unity. Given
symmetry, if θ = 1, then both firms are identical and hence we have q1QQ (1) = q2QQ (1). Hence,
22 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

the intersection point of R1 R10 and R2 R20 must lie on the q1 = q2 line (see point A in Figure 7a).
For any point on the R1 C curve we have p1 = c and for any point on the R1 R10 curve we have
p1 > c excepting at point R1 where we have q1 = 0 and hence we also have p1 = c. Since by
Assumption 1 own effect on indirect demand is negative, the R1 C curve must lie to the right of
the R1 R10 curve . Consider point B (in Figure 7a)which is the point of intersection between the
R1 C and the R2 R20 curves. Point B must lie to the right of point A and both A and B are on
∂q1QQ
R2 R20 . Point B is the Cournot equilibrium vector (q1QQ (0), q2QQ (0)). By Lemma 2, ∂θ < 0 and
∂q2QQ ∂q2QQ dRQQ ∂q1QQ
∂θ > 0. One can also show that ∂θ = dq21 ∂θ (see the proof of Lemma 2 (i)). Since for
QQ QQ
any θ ∈ [0, 1] the equilibrium point (q1 (θ), q2 (θ)) must lie on the R2 R20 curve. Therefore, for
all θ ∈ [0, 1], (q1 = q1QQ (θ), q2 = q2QQ (θ)) is the parametric representation of the AB segment of
R2 R20 with A (B) representing the quantity vector corresponding to θ = 1 (θ = 0). As θ varies
from 0 to 1 we move from point B to point A along R2 R20 as shown by the arrows in Figure 7a.

q2 p2 p1 = c
q1 = q2 p1 = p2

R1
A0 r20
r2

R2 B0

A p2 = c
B
q1 p1
o R10 C R20 o
(a) Quantity Space (b) The Price Space

Figure 7. Translating quantity reaction function graphically.

Consider Figure 7b and let the curve r2 r20 represent the set P(RQQ 0
2 ). Point A and B in
0

Figure 7b correspond to the points A and B respectively of Figure 7a. Since for any Cournot
equilibrium the resulting price vector must satisfy p2 ≥ p1 ≥ c, the segment B 0 A0 must lie
0
between the p1 = c line and the p1 = p2 line and above the R2P P R2P P curve (see Cheng [5]).
By Assumption 1 and Assumption 2, the Cournot equilibrium quantity vector (q1QQ (θ), q2QQ (θ))
is unique for each θ implying that (pQQ QQ
1 (θ), p2 (θ)) is also unique. Therefore, for the segment
B 0 A0 , given any p1 we must get a single p2 and this segment can be represented as a function
p2 = r2QQ (p1 ) defined for p1 ∈ [c, pQQ QQ
1 (1)]. For each p1 , r2 (p1 ) is always well-defined and
given continuity of AB segment, the B A segment is also continuous. Starting from B 0 if we
0 0

move towards A0 along the segment B 0 A0 , the underlying θ increases since the B 0 A0 segment
∂pQQ
has a functional representation it cannot be backward bending. Hence, given 1
∂θ > 0, pQQ
1 (θ)
increases along the segment B 0 A0 when we start from B 0 . 
Proof of Lemma 8: To prove this Lemma 8 we consider Figure 8. Given any θ ∈ [0, 1],
if q QQ (θ) is Cournot equilibrium quantity vector, then q2QQ (θ) = R2QQ (q1QQ (θ)) and q1QQ (θ) =
SV1QQ (q2QQ (θ), θ) and the resulting price of Firm i is pQQ
i (θ) = Fi
QQ QQ
(q (θ)). For the Cournot
 PLAUSIBILITY OF BERTRAND COMPETITION 23

equilibrium price vector (pQQ QQ QQ QQ QQ

1 (θ), p2 (θ)) ∈ P(R2 ) ∩ P(SV 1 (θ)). The graph P(R2 ) must
lie above R2P P in the price space and P(SV QQ1 (θ)) is bounded between p1 = c and graph
QQ
P1 (SV 1 (1)). Again, since the firms face identical demand and cost conditions, the Cournot
equilibrium price vector must lie in E QQ = {p | π1,1
P P (p) > 0, p > c and p ≥ p }. Therefore,
1 2 1
the region A in Figure 8 represents the set E QQ ∩ ΩPWP . This region A represents the set of
points where Cournot equilibrium can occur and resulting welfare is higher compared to point
B. If P(RQQ PP
2 ) ∩ ΩW = ∅, then, in Stage 1, the government’s optimal choice of θ can never
induce Cournot competition since, by choosing θ = 0, the government can improve the level of
welfare.
If P(RQQ PP
2 ) ∩ ΩW 6= ∅, then can the government induce quantity competition by choosing θ
in such a way that the resulting price vector (pQQ QQ
1 (θ), p2 (θ)) ∈ E
QQ ∩ ΩP P ? Consider the sets
W
QQ QQ
E≥ = {p | p ∈ E QQ and p1 ≥ pP1 P (0)} and E< = {p | p ∈ E QQ and p1 < pP1 P (0)}. Observe
QQ QQ QQ QQ
that E≥ ∩ E< = ∅ and E≥ ∪ E< = E QQ . We consider two exhaustive cases.

QQ
Case 1: E QP ∩ E< = ∅.
QQ
Case 2: E QP ∩ E< 6= ∅

For Case 1, E QP lies to the right of the vertical line p1 = pP1 P (0) (see Figure 8). Since
QQ
E QP ⊂ E QQ , we must have E QP ⊂ E≥ . Given E QP ∩ P(RQQ 2 ) 6= ∅, E
QP ∩ ΩP P = ∅ (by
W
Lemma 6) and the continuity of the graph of P(RQQ 2 ) in the price plane, there exists exactly
QQ
one compact set SP (⊂ P(R2 )) such that (a) the interior of SP is contained in the complement
set of E QP ∩ ΩPWP , (b) we can find (p1 , p2 ) in the intersection of the boundaries of the sets SP
and ΩPWP , and, (c) we can find another (p1 , p2 ) in the intersection of the boundaries of the sets
SP and E QP . Using Lemma 7 we can now say that each θ for which (pQQ QQ
1 (θ), p2 (θ)) in the
interior of SP is higher compared to every θ such that (pQQ QQ PP QQ
1 (θ), p2 (θ)) ∈ ΩW ∩ P(R2 ) and
QQ QQ QQ
is lower compared to every θ such that (p1 (θ), p2 (θ)) ∈ E QP ∩ P(R2 ). By Lemma 3 (iii),
(pP1 P (θ4 ), pP2 P (θ4 )) ∈ E QP . Hence, for every θ such that (pQQ QQ PP
1 (θ), p2 (θ)) ∈ ΩW ∩ P(R2 ),
QQ

θ < θ4 . Thus, it is impossible for the government to induce Cournot competition by choosing θ
such that resulting price vector belongs to ΩPWP ∩ P(RQQ 2 ).

For Case 2, the entire E QP does not lie to the right of the vertical line p1 = pP1 P (0) (see Figure
QP QP
9). Consider the set E< = {(p1 , p2 ) | (p1 , p2 ) ∈ E QP , p1 < pP1 P (0)}. If E< ∩ P(RQQ
2 ) = ∅
the analysis is similar to Case 1 and Cournot competition cannot be sustained. Finally, if
QP
E< ∩ P(RQQ QQ PP
2 ) 6= ∅, then given P(R2 ) ∩ ΩW 6= ∅, ΩW ∩ E
PP QP = ∅ and continuity of the graph

of P(R2 ) in the price plane, we can find at least one S ⊂ P(RQQ

QQ P
2 ) for which we have three
mutually exclusive sets S , S and S such that S ∪ S ∪ S c = S P , S P a ⊂ E QP , S P b ⊂
P a P b P c P a P b P

(ΩPWP ∩ E QP ) and S P c ⊂ ΩPWP . Assume that there are M such S P s’. Denote a representative S P
as SmP where m ∈ {1, 2, ...., M }. Therefore, for each S P we have a set S P 0 ⊂ E QP . Can we find
m m
(pQQ
1 (θ 4 ), pQQ
2 (θ 4 )) ∈ Sm
P a ? The following argument shows that the answer is no. By Lemma

7, along the graph of the set Sm P a in the price plane, p is increasing (along the segment B 0 A0
1
in Figure 7b) and it must contain at least two points in the boundary of E QP each of which
corresponds to Type I equilibrium price vector for θ = 0. Along the graph Sm P a , the behavior of

pQP QQ QQ Pa
1 (θ) is shown in Figure 10. Suppose (p1 (θ4 ), p2 (θ4 )) ∈ Sm . By Lemma 3 (ii) θ4 is unique
and by Lemma 2, p1 (θ) is increasing in θ. Therefore (p1 (θ4 ), pQQ
QQ QQ
2 (θ4 )) is unique. Hence, if
24 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

p2 p1 = c 0 0
R1P P S1P P p1 = p2

D
0
S2P P
0
R2P P

C
B
W̄ P P (0)
A

R2P P
p2 = c

S2P P
p1
o SP P R1P P pP1 P (0)
1

Figure 8. Characterization of Cournot equilibrium

(pQQ QQ Pa
1 (θ4 ), p2 (θ4 )) ∈ Sm , then there dose not exist any k ∈ {1, 2, ...., M } with k 6= m such that
QQ QQ
(p1 (θ4 ), p2 (θ4 )) ∈ SkP a .
¯ denote the length of the OT segment in Figure 10. Given (pQQ (θ4 ), pQQ (θ4 )) ∈ Sm
Let OT P a,
1 2
θ4 > OT ¯ , then for pQQ (θ) = pQP (θ) at θ = θ4 either pQQ (θ) has slope
¯ is not possible. If θ4 = OT
1 1 1
QQ QQ QQ QQ
of ∞ at θ4 (which impossible since V1,11 π2,22 − V1,12 π2,12 6= 0) or pQQ
1 (θ) should intersect pQP
1 (θ)
twice which is again a contradiction due to uniqueness of θ4 (see Lemma 3 (ii)). If θ4 < OT ¯ ,
QQ QP QQ
then (given p1 (θ) = p1 (θ) holds for at most one θ) the only possibility is that p1 (θ) is
tangent to the lower segment of pQP 1 (θ) at θ = θ4 which is again a contradiction since, in that
case, we can find at least one θ > θ4 such that pQP QQ
1 (θ) > p1 (θ). 
Proof of Proposition 1: We use four steps to prove the result.
Step (i): The value of θ that maximizes W̄ P P (θ) must belongs to (0, 1).
 PLAUSIBILITY OF BERTRAND COMPETITION 25

p2 = c 0
S1P P
R1P P p1 = p2

0
R2P P

R2P P

p1 = c
(c, c) R1P PpP1 P (0)

Figure 9

pQP QQ
1 (θ), p1 (θ)

pQP
1 (θPk = 0)

pQP
1 (θAk = 0)

θ
O T

Figure 10

Proof of Step (i): The first order condition of Stage 1 under the assumption that firms select
price strategy in Stage 2 is given by

∂ W̄ P P ∂pP P (θ) ∂pP P (θ)

(7.22) = W1P P (pP P (θ)) 1 + W2P P (pP P (θ)) 2 .
∂θ ∂θ ∂θ
26 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

∂pP P
1 (θ) ∂pP P
2 (θ)
Like Lemma 2, when goods are complement one can show that ∂θ > 0, ∂θ < 0, and,
∂pP P
2 (θ) dR2P P (p1 ) ∂pP P
1 (θ)
∂θ = dp1 ∂θ .
Therefore, from condition (7.22) we get,

∂ W̄ P P dR2P P P P P P
  PP
PP PP ∂p1 (θ)
(7.23) = W1 (p (θ)) + W2 (p (θ)) .
∂θ dp1 ∂θ
At θ = 0, the price vector (pP1 P (0), pP2 P (0)) corresponds to point A in the Figure 2. At
point A we have W1P P (pP1 P (0), pP2 P (0)) = 0 (since, point A must lie on the S1 S10 curve),
W2P P (pP1 P (0), pP2 P (0)) < 0 (since, point A must lie above P P 0 ) and, by Assumption 7, we
PP
also have (dR2P P (p1 )/dp1 ) < 0. Hence, at θ = 0, ∂ W̄∂θ > 0. At θ = 1, the price vector
(pP1 P (1), pP2 P (1)) corresponds to point B in the Figure 2 where we have W1P P (pP1 P (1), pP2 P (1)) <
0 (since point B must lie to the right of S1 S10 ), W2P P (pP1 P (1), pP2 P (1)) < 0 (since point B must
lie above S2 S20 ), W1P P (pP1 P (1), pP2 P (1)) = W2P P (pP1 P (1), pP2 P (1)) (since point B must lie on
p1 = p2 line and the welfare function is symmetric) and (applying Assumption 7) we also have
PP PP PP
−1 < (dR2P P (p1 )/dp1 ) < 0. Thus, at θ = 1, ∂ W̄∂θ < 0. Given ∂ W̄∂θ > 0 at θ = 0, ∂ W̄∂θ < 0
2W P P
at θ = 1 and given the second order condition d dθ 2 < 0 it follows that the optimal stage 1
choice of θ is some θ∗ that lies in the open interval (0, 1). Hence, at θ = θ∗ the equilibrium price
vector (pP1 P (θ∗ ), pP2 P (θ∗ )) must belong to the interior of the segment AB (say some point like
D in Figure 2) where the iso-welfare curve is tangent to the R2 R20 curve.
Step (ii): For any θ ∈ [0, 1] the set of points SV P1 P (θ) must lie to the left of set of points in the
set P(SV QQ
1 (θ)).

Proof of Step (ii): If, in Stage 2, Firm 2 chooses price strategy, then Firm 1’s reaction function
is given by
PP PP
(7.24) V1,1 (p, θ) = (p1 − c)F1,1 (p) + θF1P P (p) + (1 − θ)F2,1
PP
(p) = 0.
If, in Stage 2, Firm 2 chooses quantity strategy, then Firm 1’s reaction function is
QQ
(7.25) V1,1 QQ
(q, θ) = F1QQ (q) − c + θq1 F1,1 (q) = 0.
QQ
If (p̂1 , p̂2 ) is a solution to (7.25), then p̂1 −c+θF1P P (p̂1 , p̂2 )F1,1 (F1P P (p̂1 , p̂2 ), F2P P (p̂1 , p̂2 )) = 0 im-
QQ
plying that p̂1 − c = −θF1P P (p̂1 , p̂2 )F1,1 (F P P (p̂1 , p̂2 ), F2P P (p̂1 , p̂2 )). At (p̂1 , p̂2 ), V1,1
P P (p̂ , p̂ , θ) =
1 2
 1 
P P P P P P P P QQ P P P P
(p̂1 − c)F1,1 + θF1 + (1 − θ)F2,1 = θ 1 − F1,1 F1,1 F1 + (1 − θ)F2,1 < 0. Therefore, given
any θ ∈ [0, 1], we have SV1P P (p̂2 , θ) < p̂1 , that is, all points satisfying p1 = SV1P P (p2 , θ) must
lie to the left of all points in P(SV QQ
1 (θ)).

Step (iii): In Stage 2, choosing price is a dominant strategy for Firm 2.

Proof of Step (iii): When Firm 1 chooses price strategy, the curve R2 R20 is the reaction function
of Firm 2. Therefore, the singleton set SV P1 P ∩ RP2 P must lie to the left of all points in the set
P(SV QQ PP
1 (θ)) ∩ R2 . Therefore, at any given θ, if Firm 1 chooses price strategy, then it is always
optimal for Firm 2 to choose price strategy.
When Firm 1 chooses quantity strategy, the set of points P(RQQ 2 ) represent the reaction
function of Firm 2 in terms of prices. Again like Lemma 7 one can show that if we generate
the Cournot equilibrium path in the price space by changing θ from 0 to 1 and plotting the
corresponding price vector then along that Cournot equilibrium path as we move from price
vector (pQQ QQ QQ QQ
1 (0), p2 (0)) to price vector (p1 (1), p2 (1)) underlying θ increases and like Lemma
 PLAUSIBILITY OF BERTRAND COMPETITION 27

∂pQQ
2(ii) one can also show that ∂θ >
1
0. Hence along that Cournot equilibrium path p1 also in-
QQ
∂π2 P P < 0, implying dπ QQ (F P P (p), RQQ (F P P (p)))/dp
creases. By Assumption 5, ∂q1 > 0 and F1,1 2 1 2 1 1
∂π2QQ P P < 0 . Therefore, along that Cournot equilibrium path profit of the Firm 2 decreases
= ∂q1 F1,1
as we move from (pQQ QQ QQ QQ
1 (0), p2 (0)) to (p1 (1), p2 (1)). Since, by Step (ii) for any θ the set of
point SV P1 P (θ) lie to the left of the set of points in P(SV QQ
1 (θ)), the profit associated with the
QQ QQ
point in the singleton set P(SV 1 (θ)) ∩ P(R2 ) is less than profits from all point in the set
SV P1 P (θ) ∩ P(RQQ2 ). Therefore, at any given θ, if Firm 1 chooses quantity strategy, then it is
also optimal for Firm 2 to choose price strategy.
Step (iv): In Stage 2, if Firm 2 chooses price strategy, then Firm 1 also chooses price strategy.
Proof of Step (iv): On the region lying above the set T := {p | p2 ≥ c, p1 +p2 ≥ 2c}, V1,2P P (p, θ) =
P P P P P P
θπ1,2 (p) + (1 − θ)W2 (p) < 0. Therefore, V1 (p, θ) is decreasing in p2 for points in the set
SV P1 P (θ) ∩ T . Again, the reaction function of Firm 2 given Firm 1 chooses price (that is, the
R2 R20 curve in Figure 2) lies above the line p2 = c and each point in this reaction function lies
below all points in the set P(RQQ 2 ). Moreover, all points on the AB segment (see Figure 2) is
contained in T . For all points on the AB segment we have dV1P P (SV1P P (p2 , θ), p2 )/dp2 < 0.
Therefore, at the intersection point of the R2 R20 curve and the p1 = SV1P P (p2 , θ) curve, we have
higher value function V1P P (p, θ) compared to any points in the set SV P1 P ∩ P(RQQ 2 ). Hence,
given Firm 2 chooses price strategy, it is always optimal for Firm 1 to choose price strategy.
Hence, Step (iv) follows.
By Step (iii) and Step (iv), given any θ ∈ [0, 1] price competition is the only Nash equilibrium
of the sub-game starting from Stage 2. By Step (i), at θ∗ (∈ (0, 1)) the government maximizes
welfare under price competition. Therefore, price competition with strictly partial privatization
with θ = θ∗ , specifically, the strategy combination (θP P = θ∗ , (P, pP1 P (θP P )), (P, pP2 P (θP P ))) is
the unique SPNE of Γ. 

References
[1] Anderson, S. P., de Palma, A., Thisse, J.-F., 1997. Privatization and efficiency in a differentiated industry.
European Economic Review 41, 1635-1654.
[2] Apostol, T. M., 1969. Calculus, Volume II (Second edition). Wiley, New York.
[3] Backx, M., Carney, M., Gedajkovic, E., 2002. Public, private and mixed ownership and the performance of
international Airlines. Journal of Air Transport Management 8(4), 213-220.
[4] Badertscher , B., Shroff, N., Hal, D., White, H. D., 2013. Externalities of public firm presence: Evidence
from private firms’ investment decisions. Journal of Financial Economics, 109(3), 682-706.
[5] Cheng, L., 1985. Comparing Bertrand and Cournot equilibria: a geometric approach. The RAND Journal
of Economics 16, 146-152.
[6] Cheng, L., 1985. Inverting system of demand functions. Journal of Economic Theory 37, 202-210.
[7] Dixit, A., 1979. A model of duopoly suggesting a theory of entry barriers. The Bell Journal of Economics
10, 20-32.
[8] Dixit, A., Stiglitz, J. 1977. Monopolistic Competition and Optimum Product Diversity. The American
Economic Review 67, 297-308.
[9] Doganis, R., 2001. The Airline Business in the 21st Century, Routledge, London.
[10] Fershtman, C., Judd, K., 1984. Equilibrium incentives in oligopoly. American Economic Review 77, 927-940.
[11] Fujiwara, K. (2007). Partial privatization in a differentiated mixed oligopoly. Journal of Economics, 92(1):51-
65.
[12] Ghosh, A., Mitra, M., 2010. Comparing Bertrand and Cournot in mixed markets, Economics Letters 109,
72-74.
28 MANIPUSHPAK MITRA RUPAYAN PAL ARINDAM PAUL P. M. SHARADA

[13] Ghosh, A., Mitra, M., 2014. Reversal of BertrandCournot Rankings in the Presence of Welfare Concerns,
Journal of Institutional and Theoretical Economics 170 (3), 496519.
[14] Häckner, J., 2000. A note on price and quantity competition in differentiated oligopolies. Journal of Economic
Theory 93, 233-239.
[15] Heifetz, A., Shannon, C., Spiegel, Y., 2007. What to maximize if you must. Journal of Economic Theory
133, 31-57.
[16] La Porta, R., Lopez-De-Silanes, F., Shleifer, A., 2002. Government Ownership of Banks. Journal of Finance
57(1), 265-301.
[17] Kreps, D., Scheinkman, J.-A, 1983. Quantity precommitment and Bertrand competition yield Cournot
outcomes. Bell Journal of Economics 14, 326-337.
[18] Matsumura, T., 1998. Partial privatization in mixed duopoly. Journal of Public Economics 70, 473-483.
[19] Matsumura, T. and Ogawa, A. 2012. Price versus quantity in a mixed duopoly. Economics Letters,
116(2):174-177.
[20] Miller, N. and Pazgal, A. 2001. The equivalence of price and quantity competition with delegation. RAND
Journal of Economics 32, 284-301.
[21] Ohnishi, K. (2010). Partial privatization in price-setting mixed duopoly. Economics Bulletin, 30(1):309-314.
[22] Okuguchi, K., 1987. Equilibrium prices in the Bertrand and Cournot oligopolies. Journal of Economic
Theory 42, 128-139.
[23] Singh, N. and Vives, X., 1984. Price and quantity competition in a differentiated duopoly. The RAND
Journal of Economics 15, 546-554.
[24] Sklivas, S. D., 1987. The strategic choice of managerial incentives. The RAND Journal of Economics 18,
452-458.
[25] Vickers, J., 1985. Delegation and the theory of the firm. Economic Journal, Supplement 95, 138-147.
[26] Vives, X., 1985. On the efficiency of Bertrand and Cournot equilibria with product differentiation. Journal
of Economic Theory 36, 166-175.
[27] Vives, X., 1999. Oligopoly pricing: old ideas and new tools. MIT Press, Cambridge and London.
[28] Zanchettin, P., 2006. Differentiated duopoly with asymmetric costs. Journal of Economics and Management
Strategy 15, 999-1015.