The Economics of Networks
by
Prof. Nicholas Economides
Stern School of Business
New York University
and
Center for Economic Policy Research
Stanford University
http://edgar.stern.nyu.edu/networks/
BASICS
Networks are composed of links that connect nodes.
It is inherent in the structure of a network that many
components of a network are required for the provision of a typical service.
Network components are complementary to each other. Typically
networks provide substitute services where each substitute is composed of
two or more complements.
Figure 1: An Information Superhighway
Figure 2: A simple star network.
For example, in Figure 2, a phone call from A to B is composed of AS (access to the switch of customer A), BS (access to the switch of customer B), and switching services at S. Despite the fact that goods AS and BS look very similar and have the same industrial classification, they are complements and not substitutes.
CLASSIFICATION
Networks where services AB and BA are distinct are named
two-way networks.
Two-way networks include railroad, road, and many telecommunications
networks.
When one of AB or BA is unfeasible, or does not make economic
sense, or when there is no sense of direction in the network so that AB
and BA are identical, then the network is called a one-way
network.
For example, broadcasting and paging are one-way networks.
The crucial economic relationship in both one-way and two-way
networks is the complementarity between the pieces of the network. This
relationship is also often observed between different classes of goods in
non-network industries. Economides and White (1993) point out that a pair
of vertically-related industries (Figure 3) is formally equivalent to a
one-way network.
The classification in network type (one-way or two-way)
is not a function of the topological structure of the network.
Rather, it depends on the interpretation of the structure to represent a
specific service.
Figure 3: A simple local and long distance
network.
For example, the network of Figure 3 can be interpreted
as a two-way telephone network where SA represents a local switch in city
A, Ai represents a customer in city A, and similarly for SB and
Bj. In this network, there are two types of local phone calls
AiSAAk and BjSBBR, as well as long distance phone call AiSASBBj.
We can also interpret the network of Figure 3 as an Automatic
Teller Machine network. Then a transaction (say a withdrawal) from bank
Bj from ATM Ai is AiSASBBj.
Connections AiSAAk and BjSBBR may be feasible but there is no demand for them.
Figure 4: A pair of vertically-related markets.
HISTORY
Traditionally, networks were analyzed under the assumption
that each network was owned by a single firm. Economic research focused
on efficient use of the network structure as well as on the appropriate
allocation of costs.
In the 70s, partly prompted by the antitrust suit against
AT&T, there was a considerable amount of research on economies of scope
and on the efficiency gains from joint operation of complementary components
of networks.
Once one of the most important networks (the AT&T telecommunications
network in the US) was broken to pieces, economic research focused in the
80s and 90s on issues of interconnection and compatibility.
Similar research on issues of compatibility was prompted
by the reduced role of IBM in the 80s and 90s in the setting of technical
standards in computer hardware and software.
Importance of interconnection and interoperability also
came from very significant reductions in costs in telecommunications.
Costs of transmission have fallen dramatically with the
introduction of fiberoptic lines.
Switching costs have followed the fast cost decreases of
microchips and integrated circuits.
These cost reductions have transformed the telecommunications
industry from a natural monopoly to an oligopoly, with the implied fragmentation.
The Telecommunications Act of 1996 recognizes the telecommunications
network as a "network of interconnected networks".
NETWORK EXTERNALITIES
Fundamental property of networks: they exhibit positive
consumption externalities called network externalities.
A good exhibits network externalities if "the value
of a unit of the good increases with the expected number of units
to be sold."
SOURCES OF NETWORK EXTERNALITIES
Direct: The addition
of the n+1th node in Figure 1 creates 2n new goods. Typical in two-way networks.
Indirect: (In Figures
2 or 3) higher demand for good A creates more varieties of A and of complementary
good B thus increasing the number of varieties in the market and the utility
that each consumer receives at constant prices. If more varieties mean more
competition, consumers may have added benefits. Typical for one-way networks.
Financial exchange networks also exhibit indirect network externalities. The exchange brings together the two complementary goods, "willingness to sell at price p" (the "offer") and "willingness to buy at price p" (the "counteroffer") and creates a composite good, the "exchange transaction". Increasing size (or thickness) of an exchange market increases the expected utility of all participants by reducing the variance of price. Economides and Siow (1988).
THE "MACRO APPROACH"
COMPATIBILITY
Let the willingness to pay for n units of the good when
ne units are expected to be sold be p(n; ne). This
is a decreasing function of its first argument because the demand slopes
downward. p(n; ne) increases in ne; this captures
the network effect.
The fulfilled expectations demand p(n, n) may not be monotonic.
Figure 4 shows the construction of a typical fulfilled expectations demand.
Figure 5: Construction of the fulfilled expectations demand.
The fulfilled expectations demand is increasing for small
n if either one of three conditions hold: (i) the utility of every
consumer in a network of zero size is zero, or (ii) there are immediate
and large external benefits to network expansion for very small networks,
or (iii) there is a significant density of high-willingness-to-pay consumers
who are just indifferent on joining a network of approximately zero size.
The first condition is straightforward and applies directly to all two-way
networks. The other two conditions are a bit more subtle, but commonly observed
in networks and vertically-related industries.
When the fulfilled expectations demand increases for small
n, we say that the network exhibits a positive critical mass under
perfect competition.
Perfect competition is inefficient. Under perfect competition,
the marginal social benefit of network expansion is larger than the benefit
that accrues to a particular firm under perfect competition. Thus, perfect
competition will provide a smaller network than is socially optimal.
Decentralization of the welfare-maximizing solution?
A monopolist who is unable to price-discriminate will support
a smaller network and charge higher prices than perfectly competitive firms.
This is despite the fact that the monopolist has influence
over expectations and he recognizes that, while no perfectly competitive
firm has such influences.
m compatible Cournot oligopolists support a network of
a size between monopoly (m = 1) and perfect competition (m = 4). Monopolistic competition in Figure 6.
Figure 6
OLIGOPOLY UNDER INCOMPATIBILITY
Total output is lower if there is some incompatibility.
Issues of coalition formation, where a coalition is identified
with a "standard":
Non-cooperative competition.
A firm benefits from a move to compatibility if
(i) the marginal externality is strong
(ii) it joins a large coalition
(iii) there is no big increase in competition
A coalition benefits from a firm joining its "standard" if
(i) the marginal externality is strong
(ii) the joining firm is large
(iii) there is no big increase in competition
Clearly, in both cases there can be conflict between the
last two incentives, and this will help define the size of coalitions at
equilibrium.
Sometimes a non-cooperative equilibrium does not exist.
Expectations play a crucial role.
COORDINATION TO TECHNICAL STANDARDS WITH ASYMMETRIC TECHNOLOGIES
THE "MICRO" APPROACH
The micro approach starts with an analysis of the specific
micro-structure of a network.
After identifying the physical aspects of a network, such
as nodes and links, we identify the goods and services that are demanded
on the network.
We distinguish between the case where only end-to-end services
are demanded and the case when there is also demand for some services that
do not reach from end to end.
MIX AND MATCH: COMPATIBILITY VS. INCOMPATIBILITY
Figure 7: Mix-and-match compatibility.
The mix-and-match literature does not assume
a priori network externalities; however, it is clear that demand
in mix-and-match models exhibits network externalities.
In Figure 7, the incentive for compatibility of a vertically integrated firm (producing A1 and B1) depends on the relative sizes of for each combination of complementary components.
Reciprocal compatibility, (i.e., simultaneous compatibility
between A1 and B2, as well as between A2
and B1) increases demand (be allowing for the sale of A1B2
and A2B1) but also increases competition for the individual
components.
Therefore, when the hybrid demand is large compared to the own-product demand (including the case where the two demands are equal at equal prices), a firm has an incentive to want compatibility. When the demand for hybrids is small, a firm does not want compatibility.
Thus, it is possible, with two vertically integrated firms,
that one firm wants compatibility (because it has small own-product demand
compared to the hybrids demand) while the other one prefers incompatibility
(because its own-product demand is large compared to the hybrids demand).
In theses cases, the presumption is that opponents will
not be able to counteract and correct all incompatibilities introduced by
an opponent, and therefore, in situations of conflict we expect that incompatibility
wins.
One-way compatibility.
For more than two firms, see the earlier discussion on
coalition formation around "standards".
Figure 8: Compatibility decisions are less flexible than vertical integration decisions
CHANGES IN THE NUMBER OF VARIETIES AS A RESULT OF COMPATIBILITY DECISIONS
Figure 9
Suppose that the A industry is duopoly and the B industry
is monopolistic competition.
Under compatibility, each firm of type B produces one variety, compatible with A1 and A2 (Figure 9).
Under incompatibility, each firm of type B produces two varieties, one compatible with A1 and one with A2 (Figure 10).
Figure 10
Under incompatibility, each B-type firm incurs higher fixed
costs; it follows that ceteris paribus the number of B-type brands
will be smaller under incompatibility.
If industry demand is not sensitive to increases in the number of varieties of composite goods n (and does not increase much as n increases), then equilibrium
profits of an A-type firm decrease in the number of firms;
therefore profits of an A-type firm are higher at the smaller number of
firms implied incompatibility, and an A-type firm prefers incompatibility.
Conversely, when consumers have a strong preference for
variety and demand for composite goods increases significantly in n, equilibrium
profits of an A-type firm increase in the number of firms; therefore its
profits are higher at the larger number of firms implied by compatibility,
and an A-type firm prefers compatibility.
QUALITY COORDINATION IN MIX-AND-MATCH
Suppose that the quality levels of the components are qA
and qB, while the quality level of the composite good is qAB
= min (qA, qB).
Consider vertical integration of A and B. Economides and
Lehr (1994) show that an integrated monopolist provides a higher quality
than the two independent monopolists. In bilateral monopoly, marginal increases
in quality have a bigger impact on price. Being able to sell the same quality
at a higher price than under integrated monopoly, the bilateral monopolists
choose lower quality levels, which are less costly. Despite that, because
of double marginalization, prices are higher than in integrated monopoly,
a lower portion of the market is served, and firms realize lower profits.
Thus, lack of vertical integration (unbundling) leads
to a reduction in quality. This is not because of lack of coordination
between the bilateral monopolists in the choice of quality, since they both
choose the same quality level.
NETWORK EXTERNALITIES AND INDUSTRY STRUCTURE: INVITATIONS TO ENTER
In the presence of strong network externalities an innovator-monopolist
needs to commit to a high output, but may be unable to do so.
If he licenses the technology to a number of firms and
invites them to enter and compete with him, market output will be higher;
and since the level of market output depends mainly upon other firms, the
commitment to high output will be credible.
The invitation to enter has two effects; a competitive
effect and a network effect.
The competitive effect is an expected increase in competition
because of the increase of the number of firms.
The network effect tends to increase the willingness to
pay and the market price because of the high expected sales.
Economides (1994), (1995) shows that, if the network externality
is strong enough, the network effect is larger than the competitive effect,
and therefore an innovator-monopolist invites competitors and even subsidizes
them on the margin to induce them to increase production.
INTERCONNECTION OR FORECLOSURE BY A LOCAL MONOPOLIST?
Figure 11:
AB is a bottleneck facility.
If firm I owns links 1 and 2, how much should firm II (that
owns link 3) pay for services on link 1?
Lack of validity of the "Efficient Component Pricing
Rule" (ECPR).
Figure 12: Intermodal competition.
In Economides and Woroch (1992), the integrated firm has
the opportunity to foreclose the opponent, but it prefers not to. In fact,
the integrated firm is better off by implementing a vertical price squeeze
on the opponent, and charging a significantly higher price to the opponent
for the use of the monopolized link than it charges itself.
They also find that vertical disintegration is not desirable
for the monopolist. This result is in contrast to Bonnano and Vickers (1988)
because of the absence of two-part contracts in Economides and Woroch (1992).
Even in simple networks, there may be relations among firms
that are neither purely vertical nor purely horizontal. Thus, the conventional
wisdom about vertical and horizontal integration fails.
In the model of Figure 12, Economides and Woroch (1992) consider the case where link ST is owned by a
firm that owns a vertically-related link (either AS or
BT), or is owned by an independent firm.
Clearly, the strategic structure of the game remains unaffected when link ST changes hands between two firms that also own a link that is vertically related to ST.
Therefore, if ST has a fixed cost, it is a liability to
such a firm; each firm would like the opponent to own it. However, if the
link is owned by a third party, it is has a positive value because of its
monopoly position in the chain. Thus, each original owner has an incentive
to sell ST to a third party.
Economides and Salop (1992) discuss pricing in various
ownership structures in the model of Figure 6. They call the ownership structure
of this figure, where each firm produces a component of each type, parallel
vertical integration. They also consider the independent ownership
structure, where each of the four components is owned by a different firm.
In both of these structures, no firm is purely vertically or purely horizontally
related to another firm. Starting from independent ownership, or starting
from parallel vertical integration, a merger to joint ownership,
where all components are produced by the same firm, can either increase
or decrease prices.
Thus, simple prescriptions against mergers may easily fail.
SEQUENTIAL GAMES
Perfect competition
Multiple equilibrium paths
Adoption paths much steeper than without externalities:
application to FAX in US and Japan
History matters
Strategic advantage of first mover: QWERTY example
Oligopoly
Coordination problem with two technologies
Player 2
New Technology Old Technology
Ì))))))))))))))
New Technology Î (a, b) (c, d)
Player 1
Old Technology (e, f) (g, h)
Ï
))))))))))))))Í
Network externalities for both technologies mean that a > c, e; b > d, f; g > c, e; h > d, f. If both firms are worse off when they are not coordinated, both the "New Technology" (i.e., (N, N)) and the "Old Technology" (i.e., (O, O)) will arise as equilibria. Clearly, one of the equilibria can be inefficient. If the (O, O) equilibrium is inefficient and is adopted, Farrell and Saloner (1985) call the situation excess inertia. Similarly, if the (N, N) equilibrium is inefficient and it is adopted, the situation is called excess momentum.
In a two-period model consumers may be differentiated according to their willingness to pay for the change of the technology 2. They may
(i) they never switch
(ii) switch in period 2 if other users have switched in period 1 -- jumping on the bandwagon;
(iii) switch in period 1
Consumers of low 2 choose strategy 1, of intermediate 2 choose strategy 2, of high 2 choose strategy 3.
Consumers would like to coordinate themselves and switch
in the first period (thereby getting the bandwagon rolling) but are unable
to do so, thus creating excess inertia.
This inertia can be reduced through communication among
the consumers, though contracts, through coordination in committees, or
through new product sponsorship and special introductory pricing.
OPEN ISSUES WITH INSUFFICIENT RESEARCH
Determination of Standards in Markets with More than Two
Participants
Joint Determination of Compatibility and Degree of Vertical
Integration
Markets for adapters
Markets for add-ons
Multiperiod Dynamic Games with Network Externalities
Foreclosure and Predation Games
The Information Superhighway