From YalePressWiki

Contents 1 Chapter 1: Introduction to Network Technology and Market Structure 1.1 The Simple Argument for Concentrated Market Structure 1.2 The Conduct of Network Operators in Concentrated Market Structures 2 Notes

[edit] Chapter 1: Introduction to Network Technology and Market Structure

In electricity, natural gas, and telecommunications markets, the configuration for the delivery of services, whether in terms of shares of sales, revenues, or product, is determined by network technologies. In each of these industries, a product is received at a network’s nodes and transported on links to hub switches, where it is directed to other links to the retailer or consumer. The business consists of the service functions of collection, transmission, and distribution, which are neither necessarily integrated backward into making the product nor forward into retail delivery of that product to the consumer. Although the language varies depending on the industry, collection occurs in each case, whether in taking electricity from the generating plant or withdrawing natural gas from the ground or from a liquefied natural gas facility, or receiving a phone call on a local exchange system. Similarly, in each case delivery to the final consumer begins at local exchange facilities and ends at a lamp or kitchen stove or telephone. Between these endpoints, networks gather individually generated signals or molecules of energy at a node, and switch and/or transmit them by links to a final node for delivery to an individual or organization for final disposition.

One can describe this system of nodes and links as servicing electricity, gas, and telephone call delivery. The configuration of each network is of course different—the links of gas pipelines are high-pressure tubes of thirty-six inches or more in diameter; those of power transmission systems are insulated wires of less than six inches; and those of telephone long distance systems are optical fibers with diameters that are a fraction of an inch. Even so, there is also important uniformity. The technologies have all improved over time, with recent advances connecting larger numbers of providers and users, reducing link lengths, or even eliminating physical links altogether, while still providing connection services in physical hubs. The newer links or pipes of one sort or another deliver throughput at an exponentially higher rate per unit space. In general, the greater the volumetric throughput, the lower the unit costs of providing service, given that capital costs of tubes are linear with respect to surface area, but throughput is exponential with respect to interior volume. There are also throughput cost reductions in switching, but these are not generic, although they have been significant in scale as a result of the application of digital technologies.

Of course, gains from scale are limited to some extent in each of the three industries. There is not just one single gas transmission line between the largest area of gas reserves (in East Texas and Louisiana) and the largest concentration of consumers (in the Northeast seaboard) because there is no extant technology for a single pipe with a diameter large enough to contain all the required throughput. A single hub-spoke system for delivering all the electric power demanded at major population centers in northern California is technically feasible but not economically the most efficient, given that redundancy can be useful for responses to forced outages; and the legacy multiple systems operate now with short-run variable costs that are less than the long-run marginal cost of a single system. Thus there has been a case in larger markets for two or three independent networks transmitting to the city gate. In telephone communications, the limits on a single node/hub long distance system in the entire country have been due to regulation, not to throughput volume; new technology has replaced links with “wireless” and nodes with “switchless” alternatives. The result has been that unless variable costs of incumbents have been close to zero then smaller independent systems have become more efficient. Neither an incumbent nor an entrant in long distance/local exchange service markets in the past decade has established a system capable of carrying all the traffic; here, particularly with long distance service providers, economies of scale have not been determinative so that three or four overlapping service providers have continued to provide service to the city gate.

Service on all these networks takes place in real time. The ability to store electricity is extremely limited. Natural gas is delivered almost immediately at the wholesale node as retail distribution service providers call for it in their pipes, with a quarter of delivery volumes from underground storage near delivery nodes and the rest from “line pack” in long distance pipes. Individual callers do not wait for a connection in a telephone network, except on rare occasions when there is no dial tone because the system is fully occupied. Data transmission can be postponed for brief periods, but messages and data that support personal and business decisions cannot be stored. In all three industries, for the most important services—those most in demand—the consumer does not turn to extensive standby inventory when service is denied. These market structural conditions establish competitiveness of service provisions that are different from those in other service industries.

[edit] The Simple Argument for Concentrated Market Structure

The conditions make the case for a limited number of individual company systems, specifically relative to market size. The argument is that there are limited links, and newer technologies that characteristically allow additional links to connect to a single hub without commensurate increases in hub capital investment in switching. Also, costs do not increase proportionate to higher quality of service—when continuity of service is the relevant measure of quality, larger size provides increases in redundancy, lessening the frequency of (unexpected) outage. Compare a dual-service network of the simplest sort, with hub A linked to delivery nodes B and C to a larger network with an added link from B to C (see Figure 1). As an alternative, imagine that an independent network owns the switch at B and link to C. If a break in link AB shuts down service from A to B, the larger single or joint network can restore service by switching the signal through AC to CB. But the joint system including an independent BC network would do the same by encountering transaction costs.

As another example, an electricity transmission system connecting many more sources of available generation would access more alternative sources should any one facility fail. Then, in many but not all instances, in a system with more product sources, only interruptions of numerous plants large relative to the size of the market would lead to a break in real-time service. That condition most likely has come to affect decisions that now reduce scale in electricity, gas, and telecommunications transmission services.

The transmission grid of the western states illustrates the scale and geographic spread of current separate networks in electricity. The head end in generation sends the electricity, often at remote facilities, to be distributed to multiple hubs, where it is purchased at wholesale and switched at retail to households and industries by a retailer (see Figure 2). Within California, for example, the transmission capacity, consisting of two or three high-voltage systems of power lines, provides link capacity traded off for more product from entry nodes of more adjacent generating plants. As the market expands, with more subscribers at any node, higher capacity lines are added, necessarily increasing the potential for more transmission in the incumbent networks when linkage goes out of service.

These hypothetical conditions in electric power transmission result in service being concentrated in at most two or three networks at the receiving hub for large population centers. If there is only one network serving the receiving node in a retail market, there are two and possibly three independent transmission grids, each having separate sets of nodes and links integrated by switches into a regional network that transports power from locations far from the nearest retail receiver node. Pacific Gas and Electric Corporation, for example, provides retail service based on power received from Oregon, Washington, and Canada through separate grids, not all of which are owned by one supplier (see Figure 2).

In addition, this retailer takes hydropower from Oregon and Washington from one network, fossil- and nuclear-generated power from Arizona from a second network, and natural-gas–fired, nuclear, and hydroelectric power from a third network linked to northern and southern California. All these major generating sources, along with multiple smaller sources, are connected by transmission links owned by the three large retail distribution companies but operated by the California Independent System Operator (CAISO) connecting plants throughout the state. These overlapping link systems reduce concentration in power generation to the extent that plants of multiple owners offer alternative power supplies.

In gas transmission networks, there are three major subnational markets. Within each, three to five independent pipelines provide service over links from source gas fields in the Southwest, the Rocky Mountains, and Canada to the highly populated areas of the north and west. In the Northeast there are the equivalent of five equal-sized pipeline gas supply sources in the major population centers, while less populated states in northern New England have from one to less than two equal-sized service suppliers. In the Midwest, the largest population states have four or five equal-sized pipelines providing service while less populated states, such as Wisconsin, have between two and three equivalent- sized pipelines. California has the equivalent of three to four equal-sized suppliers, but there are only one or two in the other states of the West and Northwest.^[1]

In long distance telecommunications, there are fewer than a half-dozen independent networks in what are two national markets, for “mass” residential service and “business” service. The telecommunications markets differ from those in electricity and gas in that the largest service provider, AT&T, was losing share, initially in the 1980s and early 1990s, to the second and third largest providers and losing even more share to smaller, more specialized entrants in the late 1990s and early 2000s. The first incumbent, AT&T, had a messageminute share of 79 percent in 1987, for the two markets together, following federal court implementation of an antitrust divestiture decree beginning in 1984. That share declined over fifteen years, to 35 percent in 2002. The share of the second largest service supplier, MCI WorldCom, increased from 9 percent in 1987 to 25 percent in 1998, then fell to 22 percent in 2002. Sprint, the third largest, at 6 percent in 1987, stabilized between 9 and 10 percent in 2002. These three incumbents in the 1990s collectively lost share to regional or specialized entrants on the fringes of the national mass and business markets—to regional wholesalers seeking to provide transport by leasing links from AT&T, MCI, and Sprint and to extend wholesale service in private lines of large industrial users. Thus, a rough approximation of market-wide concentration would have the largest service supplier provide twice the share of the second, and the same for the second and third, but together constituting more than 80 percent of service volume, with a fringe of complementary smaller firms.

Despite individual differences, then, the characteristic market structure in these industries has been that of “fewness,” an attribute requiring each supplier to take account of the effect of the terms and conditions of others on its demands for service. The conduct of the limited number of providers has to be necessarily interactive: the service demands of any one carrier determined by offerings of others. Given that the entry and exit of large-scale network-level capacity has been limited and the major service providers have not changed, the interaction patterns in demand have become repetitive. The first power grid or gas pipeline or long distance telecommunications carrier has remained the largest. It has interconnected with the second and third largest service providers and has contested new entrants that were to offer limited service at the margin. These incumbents have been only semi-independent of each other, by and large, with repetitive demands and interconnection, so that they each adjust the same pricing strategies to the prices and conditions of service of the two or three other network suppliers.

[edit] The Conduct of Network Operators in Concentrated Market Structures

These network structures with a limited number of large suppliers fit the economic theory classification of oligopoly firms and markets. Without trepidation we proceed to compare the descriptive propositions in oligopoly theory as to price, costs, demand elasticities, and concentration of supply with the market behavior of companies operating the networks. Given the limited number of market participants, sellers have the potential to act with some knowledge of each other’s conduct to set prices. Numerous oligopoly theories describe markets in which the conduct or strategies of one supplier are affected by those of other large service suppliers. These specific theories describe resulting price levels, each based on the extent of collusion among the network operators. Here, we consider three of these theories centering on different price levels relative to costs.

In addition, we seek to describe the extent to which the regulatory process specific to partial deregulation has also affected the price and production decisions of the networks. This requires that we take two steps: (1) select the oligopoly models that best describe behavior, and (2) determine the extent to which the regulatory process distorts that behavior. The first step is to observe and analyze how the volume and pricing data compare with those expected from monopoly, and from oligopoly, for example, Cournot and Bertrand oligopoly model behavior. The next step is to assess whether the behavior of suppliers in the market deviates from “pure” model behavior because of, or in association with, the interventions of the relevant regulatory agencies. The principle of Occam’s Razor applies, suggesting that the most fitting construct is the one that is simple and transparent, while describing observed patterns of behavior.

The behavioral indicator of central interest is the representative price-cost margin. If price is the instrument for generating revenues, and price-cost margin is the source of earnings from revenues, then interactive behavior of market participants that takes margin levels to the level of long-run marginal cost earns a rate of return on investment equal to that for alternative (risk-adjusted) investments. In this context, with three or fewer sources of network services, interactive behavior can generate price-cost margins that earn returns to sustain replacement and expansion of capacity. Fiber-optic or gas throughput capacity in relevant markets could be at levels that provide service expansions consistent with growth in service demands. The issue is whether price caps of federal and state commissions could push those margins to lower levels, cutting off growth of capacity.

To proceed, we construct an estimate of unit cost and an estimate of the price-cost margin to compare with one or the other levels descriptive of the various types of oligopoly. The price-cost margin so observed is constructed as a ratio of price, known as the Lerner index, equal to the difference between price (<math>p</math>) and marginal cost (<math>mc</math>), divided by price, or [<math>(p – mc) / p</math>]. The value of the Lerner index has a range of 0 to 1. For a market to be considered free of any manifestations of interactive behavior, the Lerner index has to equal 0.^[2] When there are few firms, but a lack of coordination in pricing, this result is called a Bertrand oligopoly, in which prices recover only short-run marginal costs. Such behavior prevails in markets when price is the determinant of share, when products are close to identical, and when collective predetermination by implicit or explicit agreement on the price level is not achieved. That is, each firm’s pricing is independent of that of other suppliers, ruled by discounting any positive margin until it is eliminated.

But when interactive strategy does prevail, the Lerner index does not always approach 0. It is possible for the few large service providers to achieve prices higher than marginal costs, through collective management of some type of agreement in what is called a Cournot oligopoly. Consider a scenario in which the large incumbents are able to assume that others will stay the same, whatever price they set. The determining condition for the resulting price level is then the concentration among providers in the market, as measured by the Herfindahl- Hirschman index (HHI) (the sum of the squares of market shares). The more concentrated the market, that is, the higher the value of the HHI, the more price will exceed marginal cost.^[3] The argument is that individual restraint on increasing market share implies implicit but collective control to restrict supply, and that the extent of restriction is greater if there are fewer suppliers. The response pattern of each service provider, to choose not to respond to an expansion or restrictive initiative of another provider by changing its offering of service, works to restrict the aggregate level of service to Cournot oligopoly levels.

This is not a sufficient condition, however. As implied by Bertrand oligopoly, high concentration by itself does not imply a high price-cost margin. For example, a market may have such effective product substitutes that the product may be found to have a high elasticity of demand. (Demand elasticity measures the extent to which demand responds to a change in price by the coefficient <math>e</math>, equal to the ratio of the percentage change in quantity demanded to the percentage change in price.) In this case, greater buyer responsiveness to price increases will reduce the price-cost margin. The more extensive the practice of switching to other products in response to price increases and of going back in response to price decreases, the lower will be the expected value of the Lerner index in the context of Cournot strategic behavior.^[4]

With the same level of concentration and price elasticity, the extent of direct interactions in a firm’s decision making related to price or throughput levels determines whether it is functioning as a Bertrand or Cournot oligopoly. This conduct, or strategic behavior, of the service provider is the conjectural variation (denoted by <math>v</math>), that is, the extent to which service levels of other providers change in response to a change.^[5] When service providers increase or decrease capacity and available throughput together, <math>v</math> is positive, and price-cost margins exceed Cournot levels. When interactive responses are not present, <math>v</math> is 0 and the price-cost margin is characteristic of Cournot oligopoly. When responses are in the opposite direction, and <math>v</math> is negative, then margins are characteristic of a Bertrand oligopoly. If a provider takes a network out of service, others in the market are faced with two options: maintain their current levels of output (<math>v_i = 0</math> for firm <math>i</math>) or cancel out the reduction by increasing output (<math>v < 0</math>). That range is from Cournot (<math>v = 0</math>) to Bertrand (<math>v < 0</math>) behavior, causing price-cost margins to range from an increase due to higher concentrations with Cournot to no change with Bertrand.

The relationship between market behavior and theory in this formulation is particularly amenable to testing with data available from public sources. From the two assumptions that the firm maximizes profits (marginal revenues equal marginal costs) and that demands are interactive (quantity demanded of firm 1 changes when that of firm 2 changes), then with market price <math>p</math> and quantity <math>q</math>, but firm price <math>p_i</math> and output <math>q_i</math>, market and firm marginal cost (or constant unit variable cost) <math>c_i</math>,

(1) <math>p - c_i = q_i \cdot p (1 + v_i) / eq \,\!</math>

is the first-order condition for the profit-maximizing firm production level. For the representative enterprise, when rewritten as the Lerner index,^[6]

(2) <math>(p - c_i) / p = (q_i / q) (1 + v_i) / e \,\!</math>

Here, when the left side of Equation 1 is multiplied and divided by <math>q_i</math>, then <math>(pq_i - c_i / q_i) / pq_i</math> is an approximation of the firm’s price-cost margin, and the right side is <math>q_i / q</math>, the firm’s share of market output (or sales, if multiplied and divided by market price) multiplied by the firm’s conjectural variation (the change in market output divided by the change in this firm’s output, [<math>1 + v_i</math>]) all divided by the market elasticity of demand, <math>e</math>.

For the market as a whole, summing across firms,

(3) <math>(pq - \sum c_i q_i) / pq = HHI (1 + v) / e \,\!</math>

where the left side of Equation 3 is the market-wide price-cost margin and the right side is the HHI times the average conjectural variation divided by market elasticity.

The equation provides magnitudes of three determinants of the Lerner index: concentration measured by HHI, conjectural variation, <math>v</math>, and demand elasticities, <math>e</math>. These can be approximated for both firms and markets in the network industries for comparison with the behavior of few providers in Cournot or Bertrand oligopoly. More important, however, the left side of Equation 3 divided by the concentration measure and multiplied by the elasticity measure can be used to approximate (<math>1 + v</math>) to trace changes over long periods in the interactive behavior of the networks. The index of market concentration may decline and the elasticity of demand may increase, leading to Lerner estimated margins that were lower in later years. If the Lerner index declines by more than would be consistent with these adjustments, then reductions in conjectural variation, from reduced interfirm cooperation, would be indicative of less Cournot-like behavior and more Bertrand-type behavior. Such a movement may have occurred when regulation was breaking down or partial deregulation was being established.^[7]

It is important to note that Bertrand and Cournot oligopolies do not derive from overt collusion among few service providers. There is no overt collusion unless explicit interactions have been worked out between firms before prices are set and contracts made for delivery. But actions of one can cause repetitive reactions that have been construed as “tacit” collusion, as follows: “Tacit collusion occurs when firms are able to coordinate their behavior simply by observing and anticipating their rivals’ pricing behavior. Because all firms recognize their mutual interdependence and the advantages of coordination, a firm might well anticipate that any increase in its price will be matched by its rivals. Firms will adopt a course of action—raise their price—in the knowledge that it is mutually beneficial if all firms adopt the same course of action.”^[8] To distinguish this from overt collusion, there is no “association of firms that explicitly agrees to coordinate its activities.”^[9] Such an organization has been described in my book The Trunk-Line Railroad Cartels and Interstate Commerce Commission Before 1900. Four railroads with networking from the upper Midwest to the East Coast developed a formal company to solve two interrelated problems: to reach agreement on transport prices and/or shares in a forthcoming traffic season, and to agree on the means to enforce the prices and shares in order to prevent discounting that would shift shares away from the agreement. Enforcement worked when those firms loyal to preset price levels were rewarded by compensation from organization funds while other railroads deviated from the preset levels, making deviant behavior unprofitable and therefore less likely.^[10]

Central to this theory of conduct in oligopoly is that overt collusion leads to higher price-cost margins. Cournot non-collusion results in conjectural variation values near 0 (Cournot), whereas overt collusion seeks the best collective (or monopoly) price, that at the Lerner index <math>(p - mc) / p = -1 / e</math> at the monopoly level.

With these specifications of price-cost margin levels associated with types of oligopoly conduct, we can compare margins estimated for network service providers in their operations over the past fifteen years for periods of complete regulation with those for periods of partial deregulation. Estimated price-cost margins then can imply changes in patterns of conjectural interplay due to changes in regulations. Changes in Lerner margins can be explained by those in HHI, <math>v</math>, and <math>e</math> and, of not least importance, changes in regulation working through these determinants.

There are of course numerous complexities in firm performance due to specific practices in the delivery of these three kinds of service. If the pattern is Cournot-like, then the largest incumbent has set its capacity, and capacity utilization parameters, so that prices in the ensuing bid period do not disturb market shares. Alternatively, for the Bertrand description to be a more likely fit for the oligopoly, prices have to determine shares directly, and a “one-shot” round of price setting is the set pattern.

It makes a great difference how oligopolists have performed in the context of specific regulations. Over the past fifteen years, the context for each of these industries included some aspects of phased deregulation, a magnitude change from public utility control of prices to partial decontrol, excluding only certain “high” prices. The timing of this process has varied in these three industries, and some services are almost completely free of regulatory price controls, while others are still subject to regulatory constraints.

[edit] Notes

1. ↑ As discussed later in the chapter, a key measure of market effectiveness is a measure of the concentration of ownership, calculated as the Herfindahl- Hirschman index (HHI). The use of the term equivalence indicates that the HHI index, the sum of the squares of market shares, is equal to 1 / n for n equal-sized firms. When, for example, HHI equals 0.20, it is equivalent to five firms of equal size.
2. ↑ The price-cost margin has been used as the empirical measure of market power for more than sixty years. See Abba Lerner, “The Concept of Monopoly and the Measurement of Monopoly Power”; Keith Cowling and Michael Waterson, “Price-Cost Margins and Market Structure”; Roger A. Clark and Stephen W. Davies, “Market Structure and Price-Cost Margins”; William Landes and Richard Posner, “Market Power in Antitrust Cases”; Dennis Carlton and Jeffrey Perloff, Modern Industrial Organization, 2nd ed., pp. 352–354, 360–366. Carlton and Perloff cite numerous empirical studies that have used the price-cost margin to measure monopoly power.
3. ↑ If there is one source of supply, a monopoly, HHI is 1; if there are two equal-sized firms, then HHI (the sum of the squares of their shares) is 0.50 and the duopoly price-cost margin (equal to the inverse of the demand elasticity) is reduced to 0.50 of the previous level. This relationship between the price-cost margin and HHI is the monopoly margin times HHI, with HHI between 1 and 0 when the interactive strategies of the firms is what is termed Cournot (see below).
4. ↑ See, for example, Luis Cabral, Introduction to Industrial Organization, Chapter 9.
5. ↑ The conjectural variation can be represented as the partial derivative of all other firms’ output with respect to a change in that firm’s output. See Stephen Martin, Advanced Industrial Economics, Chapter 2; James Brander and Anming Zhang, “Market Conduct in the Airline Industry: An Empirical Investigation,” p. 56.
6. ↑ For this particular form of the Lerner margin equation, I am indebted to Dr. Richard Lee Schmalensee in correspondence dated February 24, 2006.
7. ↑ Values of the conjectural variation parameter <math>v</math> that are sufficiently large to prove the existence of overt collusion are discussed in the text below.
8. ↑ Jeffrey Church and Roger Ware, Industrial Organization: A Strategic Approach, p. 314.
9. ↑ Dennis Carlton and Jeffrey Perloff, Modern Industrial Organization, 3rd ed., p. 121.
10. ↑ Paul W. MacAvoy, The Economic Effects of Regulation: The Trunk-Line Railroad Cartels and the Interstate Commerce Commission before 1900, Chapter 2.