Multiproduct Pricing Made Simple
Mark Armstrong John VickersOxford University
September 2016
Armstrong & Vickers () Multiproduct Pricing September 2016 1 / 21
Overview
Multiproduct pricing important for:
unregulated monopolyoligopolymost effi cient prices which cover fixed costs or generate tax revenue(Ramsey prices)optimal regulation when costs are private information
Key feature is that firm(s)/regulator must decide about pricestructure as well as overall price level
This paper:
derives simple formulas using notion of consumer surplus as function ofquantitiesdemonstrates equivalence between symmetric Cournot equilibria andRamsey pricesdescribes generalized form of homothetic preferences so that pricingdecisions can be decomposed into “relative”and “average”decisionsfirms then have good incentives to choose relative quantities
Armstrong & Vickers () Multiproduct Pricing September 2016 2 / 21
Some (old) literature
Baumol & Bradford (1970): principles of Ramsey pricing“plausible that damage to welfare minimized if quantities areproportional to the effi cient quantities”
Gorman (1961): conditions on preferences to get linear Engel curvesBergstrom & Varian (1985), Slade (1994), Moderer & Shapley(1996): what does an oligopoly maximize?
we show it maximizes a Ramsey objective (and vice versa)
Bliss (1988) and Armstrong & Vickers (2001): multiproductcompetition with one-stop shopping
firms first decide how much surplus to offer customers, then solveRamsey problem of maximizing profit subject to this constraint
Marketing literature: patterns of cost passthrough in retailingown-cost passthrough is positive, cross-cost passthrough ambiguous
Baron & Myerson (1982): optimal regulation of single-product firmwith unobserved costs
we can sometimes extend this to the multiproduct caseArmstrong & Vickers () Multiproduct Pricing September 2016 3 / 21
General framework
There are n products
quantity of product i is xivector of quantities is x = (x1, ..., xn)
Consumers have quasi-linear preferences
there is representative consumer with concave gross utility u(x), whomaximizes u(x)− p · x when price vector is pinverse demand function is pi (x) ≡ ∂u(x)/∂xi or in vector notationp(x) ≡ ∇u(x)total revenue with quantities x is r(x) = x · ∇u(x)so consumer surplus with quantities x is
s(x) ≡ u(x)− x · ∇u(x)
Armstrong & Vickers () Multiproduct Pricing September 2016 4 / 21
Ramsey monopoly problem
Products supplied by monopolist with convex cost function c(x)
Ramsey objective with weight 0 ≤ α ≤ 1 is
[r(x)− c(x)] + αs(x) = [u(x)− c(x)]− (1− α)s(x)
α = 0 corresponds to profit maximizationα = 1 corresponds to total surplus maximization
First-order condition for maximizing Ramsey objective is
p(x) = ∇c(x) + (1− α)∇s(x)
price above [below] cost for product i if s is increasing [decreasing] in xiwhen c(x) is homogeneous degree 1 and α ≈ 1 Ramsey problem issolved by x ≈ αxw , where xw is effi cient quantity vector (withp = ∇c)so equiproportionate quantity reductions a good rule of thumb forsmall deviations
Armstrong & Vickers () Multiproduct Pricing September 2016 5 / 21
Ramsey quantities as weight on consumers varies
x1
x2
Armstrong & Vickers () Multiproduct Pricing September 2016 6 / 21
Cournot competition
Consider symmetric Cournot market where each multiproduct firmhas cost function c(x)
then symmetric equilibrium (if it exists) has first-order condition fortotal quantities x
p(x) = ∇c( 1m x) +1m∇s(x)
this coincides with optimal quantities in the Ramsey problem ofmaximizing
u(x)−mc( 1m x)− (1− α)s(x)
when α = m−1m
TheoremIf m firms have the same convex cost function, there exists a symmetricCournot equilibrium in which quantities maximize the Ramsey objectivewith α = m−1
m . There are no asymmetric equilibria.
Comparative statics for m straightforward when c(x) is CRS.Armstrong & Vickers () Multiproduct Pricing September 2016 7 / 21
Sketch proof of existence of Cournot equilibrium
When α = m−1m , the Ramsey objective when firm j chooses quantity
vector x j is1m r(Σjx
j ) + m−1m u(Σjx j )− Σjc(x j )
which has symmetric solution x j ≡ x , sayIn particular, choosing y = x maximizes the function
ρ(y) ≡ 1m r([m− 1]x + y) +
m−1m u([m− 1]x + y)− c(y)
A Cournot firm’s best response when its rivals each supply x is tochoose quantity vector y to maximize
π(y) ≡ y · p([m− 1]x + y)− c(y) ≤ ρ(y)− m−1m u(mx)
(inequality follows from concavity of u)Hence
π(x)− π(y) ≥ ρ(x)− ρ(y) ≥ 0and it is an equilibrium for each firm to supply x
Armstrong & Vickers () Multiproduct Pricing September 2016 8 / 21
Homothetic consumer surplus
TheoremConsumer surplus s(x) is homothetic in x if and only if
u(x) = h(x) + g(q(x))
where h(x) and q(x) are both homogeneous degree 1
“If”: We have
p(x) = ∇h(x) + g ′(q(x))∇q(x)
sor(x) = h(x) + g ′(q(x))q(x)
and hence consumer surplus is
s(x) = g(q(x))− g ′(q(x))q(x)
which depends only on q(x)
Armstrong & Vickers () Multiproduct Pricing September 2016 9 / 21
Homothetic consumer surplus
We can write quantities in “polar coordinates” form
x = q(x) · xq(x)
x/q(x) is homogeneous degree 0, depends only on the ray from originq(x) measures how far along that ray x liesrefer to q(x) as “composite quantity”and x/q(x) as “relativequantities”we know consumer surplus s(x) depends only on the q(x) coordinate
Three degrees of freedom in the family: q(x), h(x) and g(q)
this is a much wider class than those where consumer surplus ishomothetic in pricessuch preferences must have homothetic u(x), so h ≡ 0
Armstrong & Vickers () Multiproduct Pricing September 2016 10 / 21
Examples
Linear demand: An example with linear h(x) is
u(x) = a · x − 12xTMx ; p(x) = a−Mx
where a > 0 and M is a positive definite matrix, so that
h(x) = a · x , q(x) =√xTMx , g(q) = − 12q
2
Logit demand: An example with linear q(x) has demand function
xi (p) =eai−pi
1+∑j eaj−pj
which corresponds to the utility function
u(x) = a · x +∑ixi log
q(x)xi︸ ︷︷ ︸
h(x )
+ g(q(x))
where q(x) ≡ ∑j xj and g(q) = −q log q − (1− q) log(1− q)Armstrong & Vickers () Multiproduct Pricing September 2016 11 / 21
Examples
Strictly complementary products:
In some natural settings consumption requires prior purchase of a baseproduct (“access”)Suppose all who buy access (e.g. to theme park) get gross utility U(y)from complementary services (rides) y .If x1 consumers acquire access, and each then buys complementaryservices y = x2/x1, where x2 is the total supply of complementaryservices, gross utility has the form
u(x1, x2) = x1U(y) + g(x1)
= x1U(x2/x1) + g(x1)
This is an example with q(x) = x1 so consumer surplus is simply afunction of the number of consumers that buy accessSo services are priced at marginal cost
Armstrong & Vickers () Multiproduct Pricing September 2016 12 / 21
Consumer optimization
Consumer surplus with price vector p and quantities x is
u(x)− p · x = g(q(x))− q(x)p · x − h(x)q(x)
expressed in terms of the two coordinates q(x) and x/q(x)for all q(x) surplus is maximized by minimizing [p · x − h(x)]/q(x)this determines the optimal relative quantities given price vector p, sayx∗(p)let
φ(p) ≡ minx≥0
:p · x − h(x)
q(x)
which is concave in p and x∗(p) ≡ ∇φ(p)
Armstrong & Vickers () Multiproduct Pricing September 2016 13 / 21
Consumer optimization
The consumer then optimizes her composite quantity, say Q
Q(φ) maximizes g(Q)−Qφ(p)
and so consumer demand as function of p is
x(p) = Q(φ(p))x∗(p)
Thus φ(p) is the “composite price”
all prices with the same φ(p) induce the same composite quantity q(x)inverse demand for composite quantity is φ = g ′(Q)
Armstrong & Vickers () Multiproduct Pricing September 2016 14 / 21
Market analysis
Suppose a monopoly or oligopoly supplies the n products
with constant-returns-to-scale cost function c(x)
We know Cournot equilibrium coincides with Ramsey optimum, sostudy the latter objective which is
αg(q(x)) + (1− α)q(x)g ′(q(x))− q(x)c(x)− h(x)q(x)
again, a function of the two coordinates q(x) and x/q(x)regardless of composite quantity, choose relative quantities x∗ tominimize [c(x)− h(x)]/q(x)so relative quantities the same for all α in Ramsey problemthis implies relative price-cost margins also the same for all α
Monopolist has good incentives to choose its relative quantities
sole ineffi ciency stems from it supplying too little composite quantity
Armstrong & Vickers () Multiproduct Pricing September 2016 15 / 21
Market analysis
Let
κ = minx≥0
:c(x)− h(x)
q(x)
then optimal composite quantity Q maximizes
αg(Q) + (1− α)Qg ′(Q)− κQ
which satisfies the Lerner formula
g ′(Q)− κ
g ′(Q)= (1− α)η(Q), where η(Q) ≡ −Qg
′′(Q)g ′(Q)
TheoremAs α increases (or number of Cournot competitors increases), compositequantity increases, composite price decreases, each individual quantityincreases equiproportionately, and each price-cost margin contractsequiproportionately
Armstrong & Vickers () Multiproduct Pricing September 2016 16 / 21
Cost passthrough
Suppose c(x) ≡ c · x , so that κ = φ(c) and x∗ = x∗(c)all vectors c with the same “composite cost”φ(c) induce seller tosupply same composite quantity and same composite price
If ci increases, φ(c) increases, and so composite quantity decreasesalong with consumer surplus
so our class not rich enough to permit the “Edgeworth paradox”, wherea higher cost for a product induces firm to reduce all prices
When h(x) = a · x the Ramsey prices are
pi =ci − (1− α)η(Q)ai1− (1− α)η(Q)
So with constant η there is zero “cross-cost”passthrough in prices(though quantities are affected unless demands are independent)For instance, profit-maximizing (α = 0) prices and quantities withlinear demand (η = −1) are
p = 12 (a+ c)
Armstrong & Vickers () Multiproduct Pricing September 2016 17 / 21
Optimal monopoly regulation
Suppose monopolist has private information about its vector ofconstant marginal costs, c
regulator puts weight 0 ≤ β ≤ 1 on profit relative to consumer surpluscan make transfer to firm to encourage higher quantity
Look for situations where firm is given discretion over choice ofrelative quantities
cf. Armstrong (1996) and Armstrong & Vickers (2001)
Consider hypothetical case:
regulator knows the effi cient relative quantities x∗ corresponding to thefirm’s costs, but not the (scalar) average level of costs, κ = φ(c)set of c with the same x∗ = x∗(c) is a straight line
Armstrong & Vickers () Multiproduct Pricing September 2016 18 / 21
Optimal monopoly regulation
This scalar screening problem can be solved as Baron & Myerson
suppose regulator’s prior for κ on this iso-x∗ line has CDF F (κ | x∗)and density f (κ | x∗)then optimal quantities for type-κ firm are
Q(
κ + (1− β)F (κ | x∗)f (κ | x∗)
)︸ ︷︷ ︸
composite price
× x∗
each firm supplies the effi cient relative quantities x∗
If this regulatory scheme does not depend on x∗ it is valid even whenregulator cannot observe x∗
i.e., if distribution for cost vector c is such that φ(c) and x∗(c) arestochastically independent the regulation problem can be solvedincentive scheme depends only on the firm’s composite quantityfirm has freedom to choose its relative quantities
Armstrong & Vickers () Multiproduct Pricing September 2016 19 / 21
Optimal monopoly regulation
To illustrate, suppose u(x) =√x1 +
√x2, so q(x) = (
√x1 +
√x2)2,
h(x) ≡ 0, g(Q) =√Q and
κ = φ(c) =1
1c1+ 1
c2
method works if 1c1 +1c2and c2
c1are stochastically independent
e.g., if each 1ciindependently comes from exponential distribution (so
ci comes from inverse-χ2 distribution) optimal prices are
pi = ci × [1+ (1− β)κ(1+ κ)]
even very high-cost firms produce (so no “exclusion”), thoughregulated prices can be above unregulated monopoly pricesprice for one product increases with other product’s cost, even thoughcosts are i.i.d.
Armstrong & Vickers () Multiproduct Pricing September 2016 20 / 21
Conclusions
Insightful to consider multiproduct pricing problems by way ofconsumer surplus as a function of quantities
Ramsey-Cournot equivalence result
(Surprisingly?) broad class of demand systems with consumer surplusas a homothetic function of quantities is quite tractable
Relative quantities are effi cient
Multiproduct cost passthrough results
Conditions identified for optimal monopoly regulation to focus on thelevel but not the pattern of prices
Armstrong & Vickers () Multiproduct Pricing September 2016 21 / 21