Janusz A. Ordover (NYU)
and Robert D. Willig (Princeton University)

The Progress and Freedom Foundation Conference Competition, Convergence and the Microsoft Monopoly
Washington, DC; February 5, 1998

The raging antitrust debates surrounding Microsoft all include, and must face up to, one underlying question: whether antitrust should stay away from high-technology markets.2 There are serious arguments in the affirmative that are based on unique characteristics endemic to high-technology markets. These include:

• High-technology markets move too rapidly and unpredictably for government intervention to succeed in achieving its intended goals.

Each of these arguments has its own truth. Most striking is the importance of technological progress to humankind and our economy throughout the past century, even before the prominence of today's "high-tech." Despite the challenges of understanding today's information revolution while it is still in progress, high-technology markets are pivotal to continued growth and prosperity -- indeed one study (reported by Amano and Blohm in the Wall Street Journal, 10/17/96) ascribed the entire growth of the U.S. economy in 1996 to the Internet market.

However, the importance of high-technology markets does not in itself indicate that antitrust should be ignored. Indeed, there is an equally compelling argument that antitrust is particularly vital to our economy when it is employed to protect competition in such critical sectors. Nevertheless, there are important lessons from the arguments against the application of antitrust to high-technology that must shape the way it is applied:

• Antitrust may be crucial to protect long-run competition in innovation and technological development, and this should be its focus in high-technology markets.3

In sum, the stakes in applying antitrust to technology-driven markets are extremely high. In light of the importance to the economy of the competitive performance of high-technology markets and its vulnerability to antitrust intervention, it is crucial for antitrust to guard high-tech competition while respecting its unique traits. It is the thesis of this paper that some of the special characteristics of high-technology markets make them vulnerable to unconstrained monopolization that would profitably suppress the competition and innovation that would otherwise benefit the economy.

The scenario at the center of our analysis (and at the center of many of the expressions of antitrust concerns over high-technology markets) focuses on a firm that controls a "bottleneck" component of systems that may include other components that could potentially be offered competitively by stand-alone suppliers. The classic antitrust example of such a bottleneck is a railroad bridge or terminal controlled by one railroad, but needed by the other railroads to haul freight in competition with the bottleneck-holder. Here, the "system" includes the bottleneck facility together with the potentially competitive line haul services that make up the complementary components of the system. The antitrust concern is that the bottleneck-holder will use its control over the "essential facility" to foreclose rivals from competing for the supply of the complementary services. Today, of course, the leading example of an alleged bottleneck is the operating system software Windows 95 controlled by the Microsoft Corporation. It is claimed that Windows 95 is a bottleneck component of systems that include applications software such as word processors or Internet browsers, because end-users can only be supplied the computing services they demand by employment of applications software in conjunction with use of Windows 95.

In view of the economic characteristics of operating systems software (OSS), it would not be surprising if Windows 95 were a genuine bottleneck. By its very function, OSS embodies a set of standards or protocols for the ways that the hardware components of a computer work together, that applications software interacts with the computer hardware and with the software functions of the OSS, and that applications software interfaces with the end-user. Such standards and protocols must be embodied in the designs of compatible applications software and computer hardware, as well as in the working habits of end-users. There are substantial-to-enormous fixed first-copy costs to write applications software and to design and write OSS, as well as significant fixed costs to users learning the particular OSS computing environment. Consequently, the entire OSS system -- the compatibility features of applications software, hardware designs, and end-users' training and habits -- might constitute a natural monopoly, because the excess costs of two or more alternative systems might overwhelm any of the benefits of diversity. Moreover, the commitments and switching costs of the end-users, hardware manufacturers, and software writers associated with the installed bases of hardware, software, and use patterns may create substantial barriers to the entry of entrepreneurs seeking to sell alternatives to an established OSS.

To assert lack of surprise about operating system software functioning as an economic bottleneck is not to claim the inevitability of such a conclusion. The benefits from a new innovative approach to an OSS might overcome the barriers to entry and the cost disadvantages of breaking into a natural monopoly, and so genuine actual or potential competition might episodically or continuously face the supplier of even a highly popular OSS. Alternatively, the standards and protocols of a highly popular OSS might become freely available to those writing alternatives to it, so that the bottleneck-holder loses control of the elements underlying the characteristic natural monopoly and entry barrier traits, and competition for the OSS function replaces the bottleneck monopoly. Another possibility is that several OSSs coexist, perhaps serving primarily different users with different tastes and needs, but nevertheless each threatening to capture the users largely served by one of the other OSSs.

Even if a firm controls a true bottleneck, it does not necessarily follow that it will employ its control to stifle competition over other system components or to foreclose rivals from participating in the supply of systems to end-users. Instead, the benchmark case is the "perfect squeeze," in which the bottleneck-holder charges either end-users or suppliers of other components so much for the use of the bottleneck component that the full net value of the system accrues to the bottleneck-holder as profit, and other players are held ("squeezed") to no return in excess of their opportunity costs. In this case, the bottleneck-holder has profit incentives to arrange system participation so that systems are configured with optimal efficiency, including the participation of rivals in the supply of system components if they bring lower incremental costs or superior designs. Here, as a matter of economics, there is no problem for antitrust to solve, since market incentives conduce to efficiency, and only inefficient rivals of the bottleneck-holder are left out of equilibrium market participation. Any apparently high levels of profits that accrue to the bottleneck holder cannot be distinguished from efficient rewards for owning the intellectual property embodied in the bottleneck.

When the benchmark "perfect squeeze" case does not hold, there may be rational incentives for the bottleneck-holder to control the bottleneck so as to disadvantage or disable or foreclose rivals for a variety of different reasons.4 Rivals that participate in supplying components of the system may become stronger competitors to the bottleneck-holder in other, non-coincident markets as a result of economies of scale or scope between the primary market and these other markets. Then, the bottleneck-holder may nevertheless profit from excluding the rival from efficiently participating in the systems market, because the consequent additional market power in the non-coincident market may generate sufficient additional profits to overcome the losses from tolerating an inefficient configuration of the systems sold to end-users. Similarly, rivals that participate in the supply of system components may simultaneously provide end-users with an alternative system that bypasses the bottleneck, and that, accordingly, constrains the dominant firm's ability to earn profits. Then, inefficient foreclosure of the rival may so weaken the rival's ability to offer an alternative system - one that includes the bottleneck-bypass component and competes against the dominant firm -- that these gains in market power for the dominant firm are worth the losses from the inefficient foreclosure. In another scenario, the bottleneck-holder is prevented from implementing its profit maximizing (discriminating) pricing strategy as a result of the alternatives offered to end-users by an efficient supplier of other system components. This effect provides yet another motive for the bottleneck-holder to employ its control to disadvantage the rival.

Each of these theories of motives for anticompetitive foreclosure might apply to Microsoft. The Netscape browser might, for example, be viewed as a strong competitor to Microsoft's browser in an incipient, non-coincident market, for example, for financial services offered over the Internet. Then, even though MS may forgo profits by selling Windows 95 that excludes the Netscape browser, the strategy might recoup these losses and more from the resulting additional market power enjoyed in the non-coincident market. The pricing of MS Office suite might assist in effecting profitable price discrimination -- charging more to high-willingness-to-pay business end-users than to low-willingness-to-pay household end-users. Then, MS would have an incentive to foreclose a competing applications suite from the Windows 95 environment in order to protect its ability to price discriminate via supra- competitive margins on the MS Office suite. Finally, the Netscape browser might be viewed by Microsoft as a precursor to a competitive attack on its OSS monopoly based on network intelligence migrating to the desktop. Then, the non-coincident market is just the market for OSS at a later date, and Microsoft would be motivated to forgo maximal profits today by foreclosing Netscape in order to protect future OSS returns against a resurgence of Netscape.

Given the possibility that Microsoft would be motivated to foreclose or weaken a components rival like Netscape or Corel, under any of the theories just described, the next question is how an OSS bottleneck-holder might leverage its control to accomplish the weakening of its intended victim. Perhaps an OSS bottleneck-holder could design the OSS to render the rival applications software ineffective or degraded in its performance or just unattractive in its use by the end-customer, either absolutely or in comparison to the applications offered by the OSS vendor. This kind of strategy might be implemented under the cover of new positive features of the OSS that were incompatible with rival applications or that rendered the previous functionality of the OSS incompatible with the rival applications. Perhaps the OSS could be sold with an application software module integrated with it, either physically or commercially, that substituted for the rival's product, thereby undermining the demand for rival's offering. In each instance, the decreased demand for and appeal of the rival's product would help raise the bottleneck-holder's profit by diminishing the rival's ability to (i) compete through the effects of lost economies of scale and scope in non-coincident markets, or (ii) arbitrage away profitable price discrimination, or (iii) contribute to bottleneck bypass options. With all these possibilities and more, the antitrust challenge is to sort the dangerously anticompetitive conduct and practices from those that are innocuous or procompetitive but harm rivals sufficiently to motivate them to invest in antitrust complaints, lobbying and litigation. It may be entirely efficient for the bottleneck-holder to arrange systems in ways that preclude some or all rivals, and in a perfect squeeze situation, the bottleneck-holder would find it profitable to do so. Yet, the bottleneck-holder might find it profitable to exclude a rival even though the rival is efficient, in order to weaken competition in a market related through a common software platform, user interface, or other proprietary common component. How is antitrust enforcement to distinguish the pro- from the anti-competitive?

II. An Economic Test for Anticompetitive Pricing of Access and Technology Choices

We now delineate a three-pronged test that can be used to structure the assessment of the competitive effects of pricing, bundling/packaging, and technological decisions by a bottleneck-holder -- i.e., a firm with monopoly power over a "component" that rivals need in order to offer a viable system.5 The analysis of our test builds from stylized facts of the operating system software and browser "markets" in order to capture some of the key features of the competitive issues in the latest round of litigation against Microsoft.

The three-pronged test examines whether the actual strategic decisions adopted by a firm make business sense (i.e., are profitable) irrespective of their effect on the economic viability of rivals, or whether these strategic decisions are only profitable because they destroy rivals' ability to compete and, thereby, enable the firm to earn additional monopoly profits in some relevant market. In stylized models, the test has strong economic welfare properties in the sense that pricing, bundling/packaging, and technological decisions that the test does not find to be anticompetitive do improve aggregate economic welfare. Some commentators have noted, however, that the test may be too "lenient" in some particular circumstances where it permits conduct that harms competitors and also might lower aggregate long-run welfare (see an example below). Nevertheless, the test can be viewed as conservative in that it is designed to avoid the suppression of conduct that is consistent with the operation of genuine competition. We now describe the three prongs of the test:

• Prong 1: Analysis of the likelihood and the sources of monopoly profits from exclusion.
  The first prong of the test directs the analyst to determine whether the allegedly anticompetitive exclusionary conduct creates a dangerous probability of monopolization of some relevant non-coincident market or markets. In this step, the analysis first focuses on the competitive conditions in the primary market. If the firm engaged in the allegedly exclusionary practice does not have bottleneck market power in the primary market, then any further inquiry is unnecessary and would likely be detrimental to overall economic incentives to compete and innovate. Thus, if there were two or three reasonably well-matched vendors of operating system software, and if competition for the market were potent and unimpeded, then licensing practices and software design choices of any one of them would likely not merit antitrust scrutiny. Also at this stage, the analyst must rigorously identify the various non-coincident markets in which the dominant firm can potentially gain market power and earn monopoly profits from its conduct. The non-coincident market(s) could include (i) the primary market, but at some future date;6 (ii) the same product market, but at some other geographic location; or (iii) another product market. Finally, the challenged conduct must be shown to be the cause of a dangerous probability of monopolization by virtue of its impact on the ability of rivals to compete. (For example, if the challenged conduct amounts to the placement of a removable, or even a non-removable, browser icon on the OSS opening screen, a nexus between this behavior and monopolization must be established for an inquiry to proceed.) Because this prong forces the complainant into a coherent statement of the economic predicates for allegations of anticompetitive conduct, it substantially restricts the domain of action by the government and private plaintiffs in challenging business practices in high-technology industries. As we noted in our introduction, in high technology markets, subjecting the circumstances of antitrust action to restraining discipline is, on balance, desirable.

• Prong 2: Profit sacrifice.
  The second prong of the test requires a comparison of the profit flows from the actual challenged conduct versus a particular carefully specified alternative but less exclusionary course of conduct. It is critical that the profits from the exclusionary strategy be calculated on the assumption that the excluded rival remains viable as a competitor. If under the assumption of continued viability of the rival, the exclusionary strategy earns the firm less profit than the alternative, less exclusionary strategy, then the firm is sacrificing profits that it could earn but for the adverse effects of the exclusionary strategy on present and future competition in the relevant markets.
  
  For example, if consumers value having choices of alternative Internet browsers, then the owner of a bottleneck, such as the operating system software, should possibly welcome the presence of such alternatives in the market because doing so would increase the value of and consumers' willingness to pay for the OSS. Then, exclusion of rivals' alternative browsers might entail a predatory sacrifice of profits, inasmuch as the exclusion would not be profitable if the rivals were nonetheless to remain viable competitors. Yet, insofar as the exclusion would in fact eliminate the rivals as competitors, it might very well be a profitable, if anticompetitive, strategy.

• Prong 3: Recoupment of the forgone profits.
  The final step in the test is the determination of whether the exclusionary strategy is more profitable than the less exclusionary strategy because it has led (or will likely lead) to the already-established diminution of competition in the relevant markets7. The rationales for this prong of the test are straightforward. First, this prong tests the theory of sacrifice that underlies the second prong in the different and more realistic context in which the rival is competitively disabled by the challenged conduct. The change in the assumed competitive viability of the rival must swing the profit comparison from favoring the less exclusionary to favoring the more exclusionary conduct. This prong also protects the dominant firm from being forced into making strategy choices that accommodate the rival. In particular, a firm should be free to price or invest in a fashion that happens to cause rival's exit or foreclose rival from the market, if doing so would be more profitable than a more accommodating strategy, whether the rival remains or is excluded.
  
  For example, if a closed system generates significant efficiencies and consumers do not much value choice, then a closed system may be the right solution, irrespective of its effect on the viability of rivals. On the other hand, if a chosen closed system is costly to develop and lessens current demand for the bottleneck, but enables the firm to extract additional profits in the non-coincident market following the demise of the rival, then this strategic decision is only profitable because the sacrificed profits are more than recouped elsewhere (or at some future time) in the non-coincident market. If, in contrast, analysis of evidence showed that the alleged sacrifice of profit could not be recouped following the demise of the complaining rival, perhaps because of rivalry from other sources of competition, then the third prong of the test would not be satisfied, indicating that the entire theory of the case was not sufficiently coherent to warrant intervention.
  
  It has been suggested that the last prong of the test can create the error of permitting anticompetitive behavior to the detriment of aggregate welfare. Consider the following market scenario:8 Two firms, A and B, compete in the "Internet cookie" (IC) market. ICs sell for $10 a piece. Firm A invests $X million and reduces its operating cost by $2 per IC. The investment is of such a nature that it also increases Firm's B operating costs; Firm B cannot compete and irreversibly exits the IC market.9 Following B's exit, Firm A raises the price to $50. Prong 2 of the test directs the analyst to examine whether the $X million investment would be profitable even if the rival were to remain viable. In the present context, rival's "viability" can be taken to mean B's continued ability to constrain the price of cookies at the pre-existing level. Hence, if Firm A could profitably undertake the investment under the counterfactual assumption that B is viable, the investment would satisfy our test, despite the fact that consumers are worse off as a result. This is not a shortcoming of the test. Note that Firm A's behavior would not be viewed as anticompetitive if it were to lower its operating costs by $2 without at the same time increasing its rival's costs. Having lowered its costs, Firm A could potentially induce B's exit and subsequently raise the IC price to $50, as in the original example. Our test reveals that the induced exit of the rival is incidental to the profitability of the R&D investment. Consequently, it is appropriate to regard A's conduct as no different from a cost-reducing R&D expenditure that profitably "kills" the rival but does not affect the rival's unit costs. This is what our test does. Should the competitive assessment of Firm A's behavior change if we were to learn that Firm A chose not to pursue a "less exclusionary" strategy that would not have affected B's costs but would have lowered A's operating costs by only $1? Under the analytic framework of our test, the answer to this question is no. The reason for this conclusion is that the lowering of A's operating costs by $2 makes the exclusionary strategy more profitable for A regardless of whether or not firm B is disabled as a competitor. Thus, the more exclusionary strategy does not entail a predatory sacrifice of profit, even when it is compared to the less exclusionary strategy that would lower A's costs by $1. The absence of predatory sacrifice indicates that the chosen strategy is apt to be consistent with the efficient configuration of industry supply, even though it fails to protect rivals who might offer valuable competition despite their inefficiency. Our test might, under circumstances like these, fail to prosecute firms for strategies that do happen to harm rivals, because the strategies are consistent with competition. In this way, it may be concluded that our test, properly applied, is conservative in its respect for the competitive process.
  

III. Applying the Test to Commercial Bundling of Software

In order to further clarify the test and to use it to illuminate public policy prescriptions for antitrust enforcement, we first apply it in the context of some stylized models of commercial bundling. These models represent the incentives and opportunities facing an OSS monopolist to exclude a competing browser when the operating system software is designed in such a manner that it can (i) readily accommodate the competing browser, and (ii) the browser produced by the OSS monopolist, and its "icon," can be "disabled" without affecting the performance of the operating system software. Purely for mnemonic purposes, we refer to this scenario as the Win95 software environment.

Later, we consider these issues in the scenario in which the operating system software is inextricably linked with the browser software through technological bundling. In that case, it is practically impossible to remove the bottleneck-holder's browser because the operating system software and the browser software are indistinguishable. Hence, it is impossible to disable the "primary" browser and, moreover, the competing browser may not be fully compatible (or at all compatible) with the operating system software. That scenario entails genuine integration between the technologically bundled operating system software and the browser. Purely for mnemonic purposes, we refer to that scenario as the Win98 software environment.

1. Bundling and the Perfect Price Squeeze.

1. Conventional economics teaches that a firm with market power in one market has no generally applicable incentive to expand its market power from that market (which we call the primary market) to other markets (which we call non-coincident or third markets). The core of the argument is that such an expansion (leveraging) of market power is costly to the firm because it requires that the dominant firm dissipate its primary market profits in order to "purchase" market share and market power in the non-coincident market. The premise behind the conventional reasoning is that such a sacrifice of primary profits is rarely, if ever, profitable when the forgone profits are counted against the additional profits from leveraging.10 To see how this argument works in the present context, and how it can be modified for realistic complications, we start with the simplest possible model of the OSS and Internet browser markets.

Assume that U* is the utility to each consumer from an "old" PC that has neither the current state-of-the-art OSS nor the Internet browser. U* is the reservation utility level (in dollar terms) that the consumer needs to receive from the new system if the consumer is to purchase it (and possibly give up the current system). U1 is the gross level of utility in dollar terms for each consumer from the new system with both the latest OSS (call it Win95) and the compatible browser (call it IE) loaded on the PC; let p0 be the stand-alone price for the Win95 OSS, and let p1 be the standalone price for IE; finally, pS is the price of a bundle that includes both pieces of software. If there were no rival in the browser market, the owner of the operating system software bottleneck (call it MS) could either bundle the products or not without any effect on consumer welfare or on the state of competition in any relevant market. Whatever it does, the total price for the bundled software and for the components must satisfy the following constraint in order for consumers to be willing to buy and for software profits to be maximized:

(1) p0 +p1 = pS = U1 - c - U*,

where c is the cost (and the price) of the PC hardware. In this simple case, bundling does not generate any more profits than would unbundled pricing. If consumers' levels of willingness to pay for the various pieces of software differed, bundling could be an effective means of generating maximal revenues from consumers, in part by effectively matching consumers' tastes with the combinations of software they choose to acquire. As such, however, it should not create competitive concerns.11

In order to deal with the competitive effects of bundling, we introduce another browser vendor, for mnemonic purposes named Netscape. Assume that consumers derive gross utility (in dollars) of U2 from the system that combines the competitive browser (call it NB) and Win95. We define U12 to be gross utility from a PC that includes Win95 plus IE and NB. Realistically, U12 > U1, U2 so that consumers prefer a fully-loaded PC to the one that only has one browser on it. However, IE and NB are substitutes so that:

(2) U12 - U0 <(u1 U0) + (U2 U0),

where U0 is the base level of utility from a PC with Win95 with no browser installed. Inequality (2) asserts that the incremental value of both browsers together is less than the sum of the incremental values of the two browsers, because the two replicate at least some of each other's capabilities. It follows that the incremental value of a particular browser added to the other browser is smaller than the incremental value from adding that particular browser to a base PC with no browser installed:12

(3) U12 - Ui = incremental value to consumers from having browser j = IE or NB, installed on a PC with browser i

In this setting, the competitive concern is that MS could commercially bundle Win95 and IE and exclude NB from the primary market defined as licensing of browser software to PC original equipment manufacturers ('OEMs"). Absent other considerations, such as effects on competition in non-coincident market(s), MS has no profit motive to exclude NB from being installed on PCs. Since consumers are willing to pay something for having the second browser, MS can extract more for its bundled (or unbundled) offering than it would be able to if it were somehow to prohibit OEMs from installing the competing browser.13 If NB has already been developed and is readily available, the maximum system price is:

(4) pS = U12 - c - U* - r2 > U1 - c - U*,

for some small positive r2. Here, r2 is the maximum amount that NB can collect from consumers who pay pS for the MS software package and c for the PC hardware. Since MS controls the package price, it can squeeze p2, the price of NB, down to the point where

(5) p2 = r2.

MS can set the residual as low as it wants, in particular, at the level that makes NB indifferent between selling and not. Here, then, bundling enables MS to implement a perfect price squeeze. Now consider the market outcome if bundling is not allowed. First, it is not entirely clear what a prohibition against bundling means in this setting. Perhaps each PC comes with IE already installed, but each consumer may have to use the "Add/Remove" function in Win95 to activate the browser. Alternatively, consumers may have to purchase IE and/or NB as separate pieces of software and install them or have them pre-activated on their PCs.14

Second, if the antitrust authorities did not permit MS to do commercial bundling, what kind of pricing freedom would MS be accorded? It is important to recognize that if MS can set any price it wants for Win95, than it can readily implement the same outcome (i.e., receive the same profits) as it would if commercial bundling were allowed. Indeed, without bundling, the price of the OSS bottleneck is:

(6) p0 = U12 - c - U* - (r1 + r2),

where r1 + r2 is the maximum revenue that the two browsers can earn in the aggregate. MS can set the ceiling on that number as low as it wants, subject only to the constraint that the marginal costs of reproduction of the two browsers are covered. Since these marginal costs are (essentially) zero, the browser prices are driven essentially to p1 = p2 = 0. This is the same outcome as that under the perfect squeeze.

In other words, as long as MS can appropriate all the surplus available in the market by appropriately setting the price for its bottleneck OSS, Win95, it not only does not have an incentive to exclude NB but, rather, it has all the incentive to include NB in all the boxes! This is so because MS can appropriate all the value of the OSS together with the value of the applications that it enables, so that MS has financial incentives in the primary market to maximize efficiently that total value.15

Recent newspaper reports indicate that Netscape is now "giving away" its Netscape browser. This is the price that would emerge if the bottleneck were priced at the profit-maximizing level and browser vendors were to engage in Bertrand-style competition against each other for the right to have their products installed on a PC. There are two assumptions to remember here: First, the competing browser is already available (it was developed before the squeeze). Regarding this assumption, it is important to note that it cannot be an equilibrium for an owner of the bottleneck to extract all the quasi-rent from the vendors of each new generation, unless these vendors can recover their R&D investments from other sources. Consequently, if the bottleneck owner wants to maintain the incentives for the vendors of rival browsers to innovate, it must leave them with enough return to recover not only the production and distribution costs, but also the "first copy" R&D costs. The second assumption is that monopoly over browsers does not generate any additional monopoly profits in other markets. We now relax the second assumption, and generate very different implications.

B. Bottleneck-holder Incentives with Non-Coincident Market.

MS's incentives for pricing access for NB could change dramatically, however, if there is a non-coincident market in which the amount of profit that the bottleneck-holder can realize depends on whether or not NB is available.16 The presence of the non-coincident market does not change pricing incentives, nor does it affect the decision whether or not to "bundle" Win95 with IE, provided that NB has already been developed and is being made available to the OEMs as well as in the non-coincident market. Hence, the owner of the bottleneck cannot gain any additional profits from engaging in pricing or licensing strategies that exclude NB from selling to OEMs. MS loses in the primary market from such a strategy because U12 > U1, and it makes no additional profits in the non-coincident market because the rival remains active there. This analysis has demonstrated that when the rival cannot be dislodged and can enhance the value of the bottleneck to consumers, there is no incentive to engage in exclusionary conduct, and the owner of the bottleneck has the right incentives to allow access. Thus, if we observe that the owner of the bottleneck engages in exclusionary behavior, we can conclude preliminarily that some of the assumptions underlying the perfect squeeze model are not satisfied in the relevant market. In particular, we must examine whether the exclusionary behavior is motivated by the goal of diminishing competition for the provision of the next generation of software (such as browsers). This we consider next.

C. Bundling and Regulatory Price Floors

A prohibition on bundling may be interpreted as a regulatory constraint on the price that an owner of a bottleneck component may charge for the "competitive" component, such as an Internet browser. As we have seen above, if there is no floor on the price of the competitive element, the bundling outcome can be replicated even if the vendor technically does not offer a bundle. There are several possible price floors for p1. Setting p1 at variable cost is consistent with the Areeda-Turner rules for predatory pricing. Since here, the variable cost is essentially zero, we are back in the case examined in the preceding section. Alternatively, p1 could be fixed at average cost. This is very inefficient because such pricing could unduly repress the demand for browsers.17 One plausible price floor, which also is a price ceiling, requires that the browser be offered at

(7) p1 ? [U12 - U2] - r1.

The term in brackets is the incremental value from the IE in a PC that already has NB installed. The deduction from this level, r1, is a number that ranges from zero to the amount equal to the expression in the brackets, and it falls within the purview of the same governmental intervention that prohibits bundling for the sake of this analysis. With p1 determined by eq. (7), Bertrand-type competition between the browser vendors, given the stand-alone price of Win95, p0, leads to:

(8) p1 = [U12 - U2] - r1,
(9) p2 = [U12 - U1] - r2.

That is, each browser commands a price that is at, or just below, the incremental value to consumers from having that browser available on the PC, when the other browser is already installed.18 Moreover, given these prices, the profit-maximizing price for Win95 is:

(10) p0 = U12 - U* - c - [p1 + p2],

which is the remaining consumer's surplus from buying a PC that has both browsers installed on it, as opposed to buying the old system. From equations (8) - (10), we can see that as r1 is increased, the price of the competing browser falls, and the price of Win95 increases. That is, the lower the regulated price floor for IE, the higher the profit-maximizing price for Win95.

We also note that the lower the value of the residual, r1, the closer the prices of the browsers to the social values created by them. Indeed, when the residuals ri are both equal to zero, each browser sells for the incremental social value that it creates in the primary market. Moreover, if browser i = 1,2 cannot recover the R&D costs when pi = U12 - Uj, j = 1,2, then the development of the browser is not in the social interest, at least from the narrow perspective of the primary market, i.e., the market for browsers licensed to PC OEMs.

D. Bundling and Incentives to Invest in Browser Development

When MS bundles Win95 and the IE, it can hold NB to a price as close to zero as possible. This profit-maximizing strategy is not anticompetitive because it does not threaten the viability of NB, which, by assumption, has already been developed and is being made available to the OEMs. Indeed here, under our three-pronged test, the perfect price squeeze is not anticompetitive.

In the event that the (next version of the) NB has not yet been developed, bundling could foreclose its availability, even if consumers would benefit from the browser coming to market. Recall that if MS bundles, it can set pS, the system price, at the level that captures all the consumer benefit from having NB installed alongside the IE, so that p2 falls to zero. Will NB invest in the next generation of the browser given the anticipated price squeeze? The answer is likely to be negative, unless NB can earn sufficient revenues from non-coincident market(s).

Note that MS's revenues in the primary market actually fall if NB does not have the (new) browser available. This is because U12 - U1 is positive. Thus, in order to induce NB to innovate, the bundled system price must leave enough surplus for NB to anticipate covering its R&D costs. In particular, for the innovation to be socially worthwhile, NB should require no more than

(11) max rS = U12 - U1,

in order to undertake the innovation. If the Netscape product is indeed anticipated to be socially worthwhile, in the sense described above, there is an opportunity for a Pareto-improving arrangement: MS bundles its product and commits to leaving enough of the anticipated "surplus" created by NB on the table for Netscape to be able to finance the development of (the new version of) its browser.

Incentives to negotiate such a deal change when there is another market in which the two browsers compete. In that market, MS's profits depend on whether NB is available or not. If IE and NB are both available, each one gets its share of duopoly profits in that market. If NB is not available, MS secures monopoly profits from the non-coincident market. Then, a bundled price may be gainfully set by MS deliberately to make it unprofitable for NB to invest; for example, if rS is set significantly below the magnitude in eq. (11). As prescribed by our three-pronged test, a lower - and thus less exclusionary -- price for the software package would enable Netscape profitably to develop its browser and would be more profitable than the actual ps but for the fact that without NB, MS can reap additional profits from the non-coincident market. Thus, if NB were to remain viable in the non-coincident market, MS would not have selected the exclusionary level of the bundled price that it did.

This analysis shows commercial bundling itself is not the source of competitive concerns. Rather, bundling may be the "vehicle" for the exclusionary conduct -- that is, the refusal of the bottleneck-holder to negotiate a deal at the compensatory price for access to Win95 (and thus to PC OEMs). Bundling may merely cover up the fact that IE is implicitly priced too low. In such a bundle, IE is implicitly priced below the level that is commensurate with its contribution to consumer welfare, given by U12 - U2, and, possibly, at a level that if anticipated would not justify the R&D costs.

It is not an appropriate solution to the problem of concerns raised by commercial bundling to mandate a regulated price floor as in section III(1)(B), above. The economic costs of misjudging the efficient level for the regulated price would be potentially enormous, both in terms of static misallocation of resources as well as in the potential harm to incentives to innovate. And it is difficult to imagine a source of empirical guidance on which antitrust authorities could rely in such a policy quest. Yet, it is important to realize that properly directed intervention regarding exclusionary bundling must, ultimately, address the issues of pricing of access along the lines developed in our three-pronged test.

It is also important to note that if there is enough money for Netscape in the third market, then bundling in the primary market is "merely" a way for MS to make more money, and constraining this drive would have few benefits and a substantial downside from misjudgments. Obviously, there is no reliable way for antitrust authorities to discern how much money MS requires for its applicable R&D investment program, or its source.

We submit that a preferable public policy approach is for the authorities to stand ready to assess the validity of complaints by applying our three-pronged test. The test is founded on caution and respect for undistorted market forces, since it does not judge conduct unless there is a finding of bottleneck monopoly power exercised in a manner that disables competition that would otherwise be viable. Only then would the test inquire whether the dominant firm refused to permit access at a price that would be fully compensatory outside of the non-coincident domain where monopoly power is in danger of being created by exclusion.

IV. Technological Bundling and Access to the Win98 Software Environment

We now must consider the case in which the operating system monopolist develops a new product that, in effect, combines both the operating system software and the browser software into one technologically inseparable product. We call this integrated product Win98. One can envision two versions of this possibly metaphorical Win98: (i) the "closed version," such that no other browser can effectively be integrated with the software; and (ii) the "open version," which allows other browsers to be integrated with the operating system software. What distinguishes the Win98 product from the Win95 product is that the former cannot be effectively unbundled. There is no way to remove IE files from the operating system software without rendering it useless. This does not mean, however, that Win98 cannot accommodate competing browsers. Whether it can or not depends on the details of the design. It is also a possible design choice that determines how well the competing browsers will function with the new system.

From the antitrust perspective, the main issue is whether the choice of product design - the choice between, for example, a "closed" version of Win98 and an "open" version of Win98 - is consistent with procompetitive objectives or is instead driven by the expectation of additional profits from induced monopolization of non-coincident market(s). This is not a simple question to answer. Some antitrust scholars have suggested that design choices should not be subject to antitrust scrutiny.19 We agree that product design choices should not routinely be subject to antitrust scrutiny. Firms should be allowed to innovate and develop new products, even if such new products and design choices induce rivals to exit from the marketplace. As Mr. Gates correctly noted, the government is not very good at designing software and should not dictate design choices. Nonetheless, from the social welfare perspective, in certain circumstances, some design choices may have far-reaching implications for the state of competition in non-coincident markets.20 Hence, in some limited market settings, design choice may be an appropriate target for antitrust scrutiny. In particular, from the public policy standpoint, there may be a trade-off between an exclusionary design that generates significant gross consumer benefits, but that predictably would lead to long-lasting monopoly power in some other market, and a possibly less exclusionary design that was rejected by the bottleneck-holder because it potentially enables competition to thrive. 21

In this section, we discuss this issue, and conclude that the three-pronged test articulated above leads to welfare optimal outcomes for pricing access and for design choices. While this test cannot be shown to be analytically perfect in all circumstances, we wish to emphasize that it is possible to offer a structured approach to these difficult issues, rather than relying on such vague and possibly irrelevant criteria as whether Win98 (or Win95) and the IE are one product or two separate products. Indeed, it is clear that even if Win95 and IE 3.0 are determined to be two distinct products, the available information indicates that Win98 will be a fully integrated product. Consequently, for technological reasons, unbundling along the traditional lines cannot necessarily be accomplished.

1. Compensatory Pricing of Access When Only Closed Win98 is Available.

To explain our approach, we first assume that only a closed version of Win98 is available. We also assume for simplicity that consumers do not derive any additional utility from having both browsers installed on the PC, and that they may be willing to purchase a system consisting of an OSS from a third party vendor (UNIX) and an NB. However, we also assume that consumers may view such a combination as somewhat inferior to a combination of Win95 and the NB. One way to model this assumption is that Netscape must incur an additional cost of d in order to make the UNIX+NB combination acceptable to buyers as being on par with the Win98 combination.22 With this assumption, it follows that MS can charge at most

(12) p0 ? m + d,

where m is the unit production cost of the old monopoly component, if it is required to sell the product.

Here, it can be shown that if the new technologically integrated and incompatible system, Win98, is superior to the old, then it can profitably displace all the sales of the old system, even if Win95 is, in principle, available to Netscape at a compensatory price. The compensatory price for Win95, p0, is a price such that MS is indifferent between selling an additional unit of Win95 (under license to an OEM or a final consumer) and selling an additional unit of Win98 at the current market price, pS. Conversely, if any compensatory price arrangements make it infeasible for the closed Win98 to displace sales of the old OSS in conjunction with NS, then it follows that the old system is the superior one for social welfare. Thus, there can be a real social benefit from a cautious readiness of antitrust to consider investigating exclusionary campaigns of system closure without willingness to deal on compensatory terms.

Let us review briefly the public policy concerns resulting from the introduction of new software that is totally incompatible with the existing competing product of the rival and is so designed that the rival cannot offer a product, even after expending significant resources, that would work in a manner satisfactory to consumers when combined with the integrated software product. The paramount public policy concern may seem to be that the innovation drives out a rival whose products were desired by the public. In addition, there is the concern that the innovation could discourage the rival from future investments in R&D. But why should the exit of a rival generate public policy concerns? In fact, the driving engine behind innovation is often the Schumpeterian quest for the gains from monopoly-like uniqueness of creation, however transitory such a monopoly may ultimately prove to be. Here, the public policy concern arises from the fact that the rival needs the product(s) of the dominant firm as a part of its offering. In such a setting, the dominant firm can potentially manipulate the terms of access (or deny access altogether), not only to disadvantage the rival but also to harm consumers.

We have shown, however, that a supplier of technologically integrated software can induce exit of a rival, even if it notionally continues to provide the old bottleneck software at a compensatory price, provided that the new, exclusionary software is superior to the prior offerings. Stated another way, if the innovator can drive the rival out of the market with the new product while, at the same time, offering access to the "old" bottleneck at compensatory prices, then the displacement of the old system by the new increases social welfare, treating R&D costs for the new system as unrecoverable sunk costs. From this discussion it follows that if the dominant firm refuses to deal with rival(s) on compensatory terms, the motivating factor could be the desire to monopolize a non-coincident market, including the "primary" market at some future date.23

B. Access in the Open Win98 Environment.

We now proceed to consider the possibility that a more open version of the technologically integrated OSS bottleneck could be made available instead of the closed version described above. An open version may be less desirable for consumers and more costly to manufacture and develop.24 On the other hand, it is equally plausible that the development costs are not significantly different, or possibly even lower than for the closed, fully integrated system, and that the loss to consumers from having an open system may not be substantial; in fact, it is also possible that consumers may prefer an open system. Here, it is key to recognize that the alternative to the closed system is not a situation in which competitors have free access to the proprietary operating system. Instead, the alternative entails the vendor of the operating system having the discretion to charge compensatory prices for access, including the possibility of charges and contracts for the provision and development of desirable APIs.

Here too, we submit that the best available structure for antitrust analysis is our three-pronged test. It would permit consideration of the question of whether the bottleneck-holder had reasonably dealt with rivals' needs for access, in a manner that permitted the bottleneck-holder to insist on compensation for incremental costs and forgone profits in the primary markets and other markets in which any resulting monopolization is omitted from the baseline. This approach does not dwell on the constructs of tying and bundling per se, but rather on the core competitive issue. It has the potential to influence conduct prospectively with win/win outcomes for all parties when the gains from deliberate monopolization are not included: the innovators are protected by the compensatory pricing standard, access to the bottleneck is encouraged in ways that do not undermine the competitive rewards from innovation and entrepreneurial investment, and consumers receive the benefits of innovations induced by the pro-competitive access policy and a more open set of interfaces available for access to them.

V. What Are the Government's Concerns Expressed in the Current Microsoft Cases? 25

Finally, following all we have said about metaphorical software suppliers and theoretical antitrust issues, it is time to turn briefly to consideration of the government's apparent concerns over Microsoft's conduct -- in particular, what roles are played by access and bundling in the government's theories of the cases, and what elements of monopoly power are driving the fears for competition?

In its December 1997 action, the government sued Microsoft for violating the terms of the August, 21, 1995 Final Judgment.26 In particular, the government claimed that Microsoft had violated parts of section IV(E)(i) of the Final Judgment, which prohibits Microsoft from requiring OEMs to license other Microsoft products as a precondition for the right to license and install Microsoft's operating system products. The government focused on the fact that Microsoft allegedly refused to license Windows 95 (Win95) without its Internet Explorer (IE). Allegedly, the licensing agreements also prohibited OEMs from "disassembling" the package so that IE had to remain as a component of the OSS package.

The government alleged that Microsoft's licensing practices amounted to a classic tying arrangement, i.e., conditioning the license to Win95 on the licensee's also accepting the license to the IE. The government's pleadings and the district court's December 11, 1995 decision do not make clear which aspects of Microsoft's licensing practices were particularly troublesome or why they in fact constituted a classic tying arrangement. Unlike a classic tying arrangement, the fact that the IE was preinstalled with Win95 did not preclude an OEM from installing any other browser. The Government did not allege, as it did in its earlier complaint against Microsoft (Microsoft I), that Microsoft's licensing terms for Win95 + IE made it uneconomic for an OEM also to install another browser on a PC or that Microsoft explicitly prohibited OEMs from installing other browsers "on top of" Win95. The Government did not allege that Microsoft's licensing terms for Win95 + IE made it impossible for Netscape (or any other browser manufacturer) to license its browser to OEMs. And, finally, the Government did not allege that Microsoft designed Win95 so as to degrade the performance of competing browser products. Hence, at least on its face, the Government did not make it clear how (or why) Microsoft's licensing practices created a dangerous probability that Microsoft would extend or lever its alleged monopoly in the operating system software market to some other market, or markets.

In fact, an important part of the government's case, and certainly the part that appealed to the court, appears to be the claim (or "prognostication", according to the court) that by stifling competition in the browser market, Microsoft will be able to maintain its current dominance in the operating system software market. According to the government, a system comprised of an Internet browser and application software(s) could be a potent threat to more familiar software "systems" that comprise Windows and various applications software. Consequently, the government portrayed Microsoft's licensing practices as not merely designed to create a monopoly in the browser market, in which competition now reigns, but rather as a means of defending OSS dominance against future alternatives.

Microsoft responded to the Government's allegations by arguing that Win95 and the IE are one product - an integrated software product - that cannot be separated without compromising the functionality of the Win95 operating system itself. Moreover, argued Microsoft, section IV(E)(i) of the Final Judgment expressly states that the prohibition against tying "in and of itself shall not be construed to prohibit Microsoft from developing integrated products," such as Win95 that also includes the IE.

After the court removed some IE files from Win95 by utilizing the "Add/Remove" function of Win95 without rendering Win95 unusable, the question could be legitimately posed whether the Government suit was really about the fact that the IE icon stares at the user from the first screen, or whether there was something more to the Government concerns! Setting aside the legal squabbles regarding the interpretation of the modified Final Judgment, the core of the government's concern must be the future of competition not only in the Internet browser market but also in the market for OSS software itself.

VI. Concluding Remarks

It is our view that if this interpretation is somewhat accurate, then the key issues for today and for the coming iterations of the government-business conflict do indeed center on access, rather than on bundling or tying, or on what appears on the display when the machines first ship. If these elements of conduct matter, then it is through their influences on access that such effects would occur. We maintain that the first prong of our proposed test would provide appropriate discipline to the enforcement community to resist the temptation to second-guess and intervene in business decisions that are unlikely to disable competitive, efficient firms. We correspondingly maintain that adherence to our test would provide an appropriate spur for attention, instead, to mutually beneficial and serious business-like negotiations concerning genuine needs for access to OSS bottlenecks and what would constitute appropriate compensatory terms.

Endnotes

1. Janusz Ordover is presently consulting for Microsoft Corporation.
2. See, "What To Do About Microsoft," Business Week (April 20, 1998), 112-26, for a brief summary of some the issues. Professor N. Economides maintains a website http:// raven.stern.nyu.edu/networks/ms/top.htme that contains much of the up-to-date information on the subject of the Department of Justice investigation of Microsoft.
3. We have articulated this view in Janusz A. Ordover and Robert D. Willig, "Antitrust for High-Technology Industries: Assessing Research Joint Ventures and Mergers," Journal of Law and Economics, vol. 28 (2), May 1985, 311-34. See also, Janusz A. Ordover and William J. Baumol, "Antitrust Policy and High-Technology Industries," Oxford Review of Economic Policy, vol.4 (4), Winter 1988, 13-34.
4. See, Janusz A. Ordover and Robert D. Willig, "An Economic Definition of Predation: Pricing and Product Innovation," Yale Law Journal, vol. 91 (1), November 1981, 8-53, for a full articulation of these reasons.
5. For prior developments and applications of the test, see Ordover and Willig, op. cit, n. 4, supra; Janusz A. Ordover, Alan O. Sykes, and Robert D. Willig, "Predatory Systems Rivalry: A Reply," Columbia Law Review, vol. 83 (5), June 1983, 1150-66; Janusz A. Ordover and Robert D. Willig, "Economists' View: The Department of Justice Draft Guidelines for the Licensing and Acquisition of Intellectual Property," Antitrust Magazine, vol. 9 (2), Spring 1995, 29-36. See also, Statement of the Department of Transportation's Enforcement Policy Regarding Unfair Exclusionary Conduct in the Air Transportation Industry, Docket No. OST-98-3713.
6. For example, in a standard predatory pricing case, the predator prices aggressively "now" in order to monopolize the relevant product market at some "later" time. According to the government, Microsoft's licensing practices would have an adverse effect not only on competition in the browser market but also on future rounds of competition in the OSS market.
7. That is, before this prong of the test is reached, the fact finder must conclude that the dominant firm's business conduct has or will likely harm competition.
8. This and the next example were given to us by the Principal Deputy Assistant Attorney General, Antitrust Division, Douglas Melamed. It is worth noting here that if B's exit were not irreversible, then the inquiry would be terminated by the first prong of the test.
9. In the Melamed example, Firm A re-routes the supply of water, which is a necessary input into the production of cookies, from the rival.
10. See, e.g., P. Rey and J. Tirole, "A Primer on Foreclosure," February 22, 1996, for a discussion of the traditional argument and its limitations.
11. When consumers differ in their willingness to pay for the two types of software, bundling may or may not increase aggregate profits. See, W. J. Adams and J. L. Yellen, "Commodity Bundling and the Burden of Monopoly," Quarterly Journal of Economics, vol. 90, 1976, 475-98 (1976).
12. It may be useful to treat Ui as including the value to the owner of browser i of having its browser on the desktop screen. In particular, because the OEM market is competitive, the payments that the owner of the browser makes to OEMs to display its browser will be passed on to consumers. Hence, Ui includes the intrinsic utility to the consumer from a browser plus the transfer payment from the browser manufacturer.
13. Importantly, Bernheim and Whinston obtain precisely the same result in a rigorous model of exclusive dealing. They show that if the profits of the monopolist retailer (the bottleneck owner) and of the two manufacturers are jointly maximized when both goods are sold and when there are no contracting externalities (their equivalent of what we call the perfect price squeeze) then exclusive dealing will not arise. See, B. Douglas Bernheim and Michael D. Whinston, "Exclusive Dealing," Journal of Political Economy, vol. 106 (1), February 1998, 64-103. We bring out additional analogies with their paper later on.
14. No doubt, some consumers may find such a prospect appealing; however, most would likely prefer to have the browsers pre-installed.
15. This is precisely the result obtained by Bernheim and Whinston, op. cit.,in a more general setting.
16. For example, one may think of markets for financial services over the Internet; or content "markets."
17. On this point, see Ordover and Willig, Antitrust Magazine (1995), in which we discuss volume-sensitive pricing of operating system software.
18. Bernheim and Whinston, op. cit., show that such prices would emerge in a setting with perfect information, no "contractual externalities," and no vertical integration.
19. P. Areeda and H. Hovenkamp, Antitrust Law (1966); J. Gregory Sidak, "Debunking Predatory Innovation," Columbia Law Review , vol. 83 (No.5), June 1983, 1121-49.
20. Non-coincident markets could include the "primary"market in the future. For example, the Government has alleged in the latest lawsuit against Microsoft that the Netscape browser coupled with applications could in the future challenge Microsoft's current position as a leading supplier of operating system software. Apparently, browsers and browser-based applications may become a plausible substitute for a Windows standard.
21. For example, the less exclusionary system might make it possible for competitors to get access to APIs on terms that do not render them uncompetitive.
22. d can also be interpreted as an hedonic difference in the two OSS products.
23. There is an additional point: even if the dominant firm can drive the rival out by charging compensatory prices for the bottleneck, it still may be inefficiently and anticompetitively sacrificing profits by doing so, inasmuch as it invested more in the needed R&D than warranted by the incremental profits in the primary markets. The expenditures on R&D may have been motivated instead by the additional monopoly profits anticipated from the impacts on exclusion in non-coincident markets.
24. For example, it may take much more programmer effort and testing to write software code that would allow NB to function almost as effectively on top of Win98 than it would to write code that does not have this complexity.
25. We chose not to update the discussion here. No doubt, by the time this article appears, the issues will have shifted again. We only note that some of the concerns regarding Microsoft's contracts with the Internet service and content providers have been obviated by Microsoft's own actions. In any case, these concerns are not particular to high-technology industries.
26. United States v. Microsoft Corp., Civ. Action 94-1564 (D.D.C., filed Dec. 11, 1997). Our recitation of the facts relies on the court's opinion in this case.

May 31, 1998 Draft [JAO]