Spatial competition

7 November 2008

1 Principle of minimal differentiation

• In what must be one of the most influential paper in Industrial Organization1 ,
Harold Hotelling of Columbia University proposed a model to study the loca-
tional choices of firms, which has subsequently been applied to study the extent
to which firms choose to differentiate their products from each others’.

• In the Hotelling model, there is a “linear city” of length 1 that lies on the
abscissa of a line, and N consumers are uniformly distributed along this interval.
The city begins at point 0, and ends at point 1.

• There is a single homogeneous product. Firms selling the product may locate
at any point on the city. Each consumer will buy one unit of the good, and will
buy from the firm locating closest to her to minimize the necessary travelling.
Firms choose where to locate in the city to compete for customers.
1
Hotelling, H., 1929, “Stability in Competition,” Economic Journal, 39. 41-57.

Figure 1: The linear city

1
 Figure 2: Firm A at 1/4; Firm B at 3/4

• We begin with the simplest case of two firms, called firms A and B. The task is
to analyze the equilibrium locations of the two firms. First suppose A locates
at point 1/4 and B at 3/4 as in the diagram below. Is this equilibrium?

• Obviously, all consumers located between 0 and 1/2 travel shorter distances to
A than to B, and all located between 1/2 and 1 travel shorter distances to B
than to A. Hence each firm grabs one half of the market.

• What if A moves to point 1/2? All consumers to the left of 1/2 still find it best
to purchase from A. Now those located between 1/2 and 5/8 find it better to
switch to A. The firm’s market increases from 1/2 to 5/8 by moving from 1/4
to 1/2 given B’s location at 3/4.

• In fact, given B’s location at 3/4, A’s market share goes up by moving ever
closer and closer to B’s location. As long as A is to the left of B, moving
towards B by a distance of d raises A’s market on the left side by d while
only shrinks its market on the right side by 12 d. This is the case because all
consumers to the left of A will be closer to A than to B, whereas only 1/2 of
the consumers between A and B are closer to B. Hence given B’s location at
3/4, A’s best response would be to locate at a shade to the left of B.In general,
at any locations of B to the right of 1/2, A’s best response is to locate at a
shade to the left of it.

• On the other hand, if B is at some point to the left of 1/2, the best response of
A is a shade to the right of B’s location.

• Firm B’s best response function may be defined similarly.

• Neither situation depicted above can be equilibrium. Given that A is at just to

the left of 3/4, locating at 3/4 yields B a market share of 1/4. Jumping to the
left of A however raises B’s market share to approximately 3/4. On the other
hand, given that A is just to the right of 1/4, B can raise its market share by
jumping to the right of A.

2
 Figure 3: Firm A’s best response to Firm B at 3/4

Figure 4: Firm A’s best response to Firm B at 1/4

• The only equilibrium is at where the two firms locate back to back at 1/2.

• In this case, A has all the market to its left, and B has all the market to its
right. When A locates just a shade to the left of 1/2, B can not hope to increase
its market share beyond 1/2 by moving to any other points on the road. When
B locates just a shade to the right of 1/2, A can not hope to raise its market
share any further by changing its location from just a shade to the left of 1/2.
Hence the pair of locations must constitute an equilibrium.

• The model may be applied to understand firms’ choices of product differentia-

tion. The locational distribution of consumers along the road may be thought
of as the distribution of consumers’ taste along some product characteristic
spectrum.

• Take soft drinks for example. Some consumers have tastes for very sweet soda.
Some other people (those on diet in particular) however prefer drinks that taste
almost close to water. Some prefer somewhere between the two extremes. We
may interpret point 0 as representing soda that is close to 100% pure sugar solu-
tion and point 1 as representing pure water with minimal sugar. The Hotelling
model predicts that the two soft drink giants, Coca-Cola and Pepsi, would both

3
 Figure 5: NE

offer a product catering to the medium customers, with minimal product differ-
entiation. This is not a bad prediction at all. The casual consumers can hardly
distinguish Coca-Cola from Pepsi-Cola. TVB and ATV offer programming that
are essentially identical. HKU and CUHK offer almost identical program of
studies in many disciplines. The election platforms as to “livelihood” issues of
the DAB party and the Democratic party are no different from each other. It
would take some very strange commuter to prefer City Bus over New World
First Bus.

• The Hotelling equilibrium is a manifestation of what may be called a principle

of minimal differentiation, with firms choosing not to differentiate their brands
at all. The minimal differentiation equilibrium arises from the tendency that
firms benefit from moving to the center of the market, where the demand is
greatest.

• The inquisitive students may have noticed that the prediction is completely
contrary to the prediction of our previous analysis that firms should differentiate
their brands from each other as much as possible. The reason for the difference
is that we have in the analysis above abstracted from price competition by
assuming that firms only compete in terms of product characteristics, but not
by prices at all.

2 Efficient locations
• Before analyzing how allowing for firms to compete by prices as well will change
the conclusion, we shall first check whether the minimal differentiation equilib-
rium is efficient? To answer the question, we return to the Hotelling model and
ask what may be the efficient locations of the two firms.

• But efficient in what sense? For anyone individual consumer, she would cer-
tainly like the firms to locate as close to her as possible. This way, she travels

4
 the least. When interpreted as a model of product differentiation, each con-
sumer would like to be able to purchase a brand that is as close as possible to
her taste. It would be best for consumers located at point 0 to have the firms
located just at this point. On the other hand, those at point 1 is better off if
the firms choose to locate there instead. If we treat all consumers equally, the
appropriate objective would be to choose the locations of firms to minimize the
average distance travelled.

• But where

total distance travelled by all consumers = number of consumers

×average distance travelled,

and with the number of consumers fixed at some N, minimizing the average
distance travelled is equivalent to minimizing the total distance travelled. In
turn, the total distance traveled can be stated as
⎛
A+ B−A
ZA Z 2
⎜
T.D. = N × ⎜
⎝ (A − s) f (s) ds + (s − A) f (s) ds
0 A
⎞
ZB Z1
⎟
+ (B − s) f (s) ds + (s − B) f (s) ds⎟
⎠
A+ B−A B
2

where f (s) denotes the density of consumers at point s. Assuming a uniform

distribution in [0, 1], f (s) = 1. Taking foc with respect to A and B :
B−A
ZA ∙ ¸µ ¶ A+Z 2 ∙ ¸µ ¶
B−A 1 B −A 1
ds+ A + −A 1− − ds− B − A − 1− = 0,
2 2 2 2
0 A

∙ ¸ ∙ ¸ ZB Z1
B−A 1 B−A 1
A+ −A − B −A− + ds − ds = 0.
2 2 2 2
A+ B−A B
2

Simplifying yields {A = 1/4, B = 3/4}.

• All consumers to the left of 1/2 purchase from A and all to the right of 1/2
from B. Suppose we move A a tiny little bit to the left by say some ε. All
consumers to the left of A benefit, as they need only travel shorter distances.
Since there are 1/4N so many consumers to the left of A, the savings in total
distance travelled is 1/4N × ε. However, all consumers between 1/2 and 1/4 are
worse off, as each will have to travel ε farther. The increases in total distance
travelled is 1/4N × ε. Thus, the gains and the losses just cancel out. We cannot

5
 cut down the total distance travelled by relocating A from 1/4, given that B is
at 3/4. Similarly, we can rule out any possible savings in total distance travelled
by relocating B from 3/4 given that A is at 1/4. Then the pair of locations
{A = 1/4, B = 3/4} must minimize the total distance travelled and therefore
the average distance travelled.

• Efficiency call for firms to locate away from each other to a certain extent in
contrast to equilibrium, where the firms would stick to each other as close as
possible. There is too little product differentiation in equilibrium, compared to
the ideal of efficient product differentiation.

• In equilibrium with the firms located right at the median consumer, the in-
terests of consumers with less common tastes are not well taken care of. The
competition for market shares make firms concentrate all their efforts on serving
the median consumers. Efficient product differentiation balances the interests
of all consumers and call for greater product diversity.

• Why may the competition for market shares result in an equilibrium that devi-
ates from efficiency? As a general principle, equilibrium deviates from efficiency
when there exists some externality. That is, when a firm’s actions inflict dam-
ages on others. The externality in the Hotelling model is obvious; when a firm
chooses to move closer to its competitor’s location, it “steals” market share
from the firm. The firm’s gain is at the expense of the its competitors. There
can thus be excessive incentives from an efficiency point of view for firms to
locate close to each other.

3 Competition with three brands

• A major defect of the baseline Hotelling model is that no equilibrium exists
when the analysis is generalized to competition among 3 brands.

• To see how that is so, assume that there are now 3 firms, called A, B and C.
Suppose to begin with A and B locate back to back at 1/2. What may be C’s
best response? It may grab approximately one half of the market by locating
either a shade to the left of A or a shade to the right of B. In the next figure,
we assume that C is at the left of A.

• If that is the case, A is in deep trouble, being sandwiched between C and B; it

will have almost no customers. The best response will be to relocate to either
the left of C or to the right of B.

• It is easy to see that the three firms would never settle down to any equilibrium
locations. There is always one firm sandwiched between the other firms, and
this firm can raise its market share by relocating.

6
 Figure 6: Hotelling model with three firms

• Nevertheless, equilibrium again exists when there are 4 firms.

• Still that no equilibrium exists for a certain number of firms is troubling, ren-
dering the model unusable in some important situations.

• However, this shortcoming is not insurmountable. We have so far restricted to

analyzing what is called “pure strategy” equilibrium in game theory. Once we
allow for “mixed strategies”, equilibrium does exist for 3 firms. We shall not go
into the analysis of mixed strategy equilibrium here for lack of time.

• A more fundamental shortcoming is that we have so far assumed that firms may
compete only in terms of locations. When applied to product differentiation, the
assumption restricts firms to compete only in terms of product characteristics.
In practice, firms may certainly compete in terms of price as well. To this we
now turn.

4 Competition in price and in product character-

istics
• Consumers must be willing to substitute proximity for a lower price for firms to
be able to attract more customers by lowering prices. For otherwise the market
size of a firm is invariant to the price charged given its location.

• To model the substitution between price and proximity, we now assume that the
consumer will incur a traveling cost of τ (d) when buying from a firm located a

7
 Figure 7: Point x and traveling expense

distance d away from the consumer’s location. For simplicity, we assume that

τ (d) = d2 t.

• To fix ideas, we begin the analysis by assuming that the two firms are located at
the extremes of the city, at point 0 and point 1 respectively. Then a consumer
located at point x will incur a traveling cost of tx2 buying from firm A and
t (1 − x)2 from firm B.

• The total costs of buying from the two firms are equal to respectively:

qA (x) = pA + tx2 ,
qB (x) = pB + t (1 − x)2 .

A consumer who is indifferent between buying from the two firms is located at
a x that satisfies
pA + tx2 = pB + t (1 − x)2 .
Solving for x,
pB − pA + t
x= .
2t
• Consumers located on the left of x incur a lower total cost buying from firm A,
while those to the right do so buying from firm B. Thus firm A’s demand will
be given by
pB − pA + t
DA (pA , pB ) = N .
2t
Assume each firm incurs a constant marginal cost of c. Firm A’s profit function
is
pB − pA + t
π A (pA , pB ) = (pA − c) N ,
2t

8
 Holding a fixed belief on pB , firm A chooses a pA to maximize profit:
∂π A
= 0 ⇒ pB + c + t − 2pA = 0.
∂pA
Solving for pA yields firm A’s best response function:
c + t + pB
pA = .
2
• We may similarly derive firm B’s best response function:
c + t + pA
pB = .
2
• Solving the two best response functions for pA and pB yields the NE in prices:
pA = pB = c + t.

• In equilibrium with the two firms charging the same price, each will grab one-
half of the market; profits are:
1 t
π A = π B = (c + t − c) = .
2 2
• As a model of product differentiation, the brands are more “differentiated” for
the consumer when the traveling cost is higher. For then the consumer is less
willing to substitute proximity for a lower price. In this case, firms can make
use of the consumers’ inflexibility to charge a higher price. At t = 0, prices will
be driven down to the marginal cost of production. The absence of product
differentiation leads to a very intense price competition that would eventually
drive prices down to the lowest possible level.
• Because our primary interest is in the firms’ choices of product differentiation,
we would like to know how the equilibrium prices would vary with the firms’
locations. We have looked at one polar case—the one in which firms are located
as far as possible from each other (maximal differentiation). The other polar
case is that in which they produce the same brand, i.e. they are located at the
same point, and so their brands are perfect substitutes. Then each consumer
must incur the same traveling cost buying from either firm. Each consumer’s
choice of buying from which firm is thus governed simply by a comparison
between the prices charged: pA and pB . In this case, the model degenerates into
a homogeneous product oligopoly in which the price competition will be most
intense. As we have seen previously, the equilibrium price will eventually be
driven down to the lowest level that is equal to the marginal cost of production.
• In general, the locations could be within the two extremes. The competition for
market shares tend to draw firms together, whereas the desire to soften price
competition tends to drive firms apart from each other. Where the equilibrium
shall take place depends on the relative strength of the two forces.

9
5 Summary
• To summarize then, we have identified two effects governing the pattern of
differentiation:

1. firms benefit from differentiating from each other to soften price competi-
tion.
2. firms benefit from moving to where demand is.

• There could well be other factors affecting incentives to differentiate. For ex-
ample, quite often, firms in the same line of business agglomerate in the same
areas.

• One probable reason is that search by consumers encourage firms to gather

together. In a world with at least some amount of product differentiation other
than locations, consumers find it convenient to search next door if they do
not find their preferred item in a given shop. When firms selling differentiate
brands locate close to each other, the search costs for consumers are lower and
the potential benefits of going shopping are greater. This may then encourage
consumers to shop more often, raising the aggregate demand for all firms, even
though agglomerating together may intensify price competition.

10