I. Networks, Competition, and Microsoft[1]
Much of my work with Stephen Margolis has examined the logic and empirical support for concepts of lock-in and path dependence. These topics have now entered the realm of public policy in what may be the most important antitrust action, or at least the most famous, since the Standard Oil case almost one hundred years ago.
Network effects occur when the value consumers receive from a product increases as the number of users of that product increases. The archetypal example would be a fax machine, or telephone, where the product has virtually no value to a consumer if only that consumer has one, but whose value begins to increase as additional users adopt the product.
As long as other factors, such as increasing costs, do not overpower network effects, larger firms or networks will have advantages over smaller firms or networks. Following through the logic of the model, this leads, naturally enough, to a single winning firm or network.[2] Network effects share an important characteristic with economies of scale. In particular, both tend to advantage larger firms in an industry, although the former does it by increasing demand and the latter by decreasing costs.
A portion of this literature claims that network effects tend to keep a leading firm’s position intact, even if there is a superior product available, a result otherwise known as lock-in.[3] We need to make clear that our definition of ‘lock-in’ does not just imply that costs of switching make it uneconomic to switch. If it is uneconomic to switch, we do not want the switch to occur. Instead, our definition of lock-in implies that a superior product is not adopted even when all costs (except coordination costs) are less than the benefits. In case it appears that we are setting up a straw man, the reader should note that it is exactly this form of lock-in that generated so much excitement about the typewriter keyboard fable in the profession.[4] Our previous investigations of supposed real world cases of lock-in, however, forced us to conclude that were as yet no known real-world instances.[5]
The first use of these concepts against Microsoft, in the mid 1990s, was the claim the network effects in software would lead to lock-in, or the wrong choice of software product.[6] The more recent use of network effects has modified this claim very slightly. For example, Franklin Fisher, in his testimony against Microsoft, claims that network effects enhance Microsoft's monopoly power. Judge Jackson, in his finding of fact, claimed that the “chicken-and-egg” nature of software markets let to an “application barrier to entry” meaning that incumbent operating system producers were protected from competitors by the fact that network effects and coordination costs would be working against the challenger. The implication would seem to be that the challenger might not be superior, but that the barrier to entry prevents the equal or inferior product from ever getting a chance to seriously compete with the incumbent.
This concept of entrenched incumbents is, apparently, a beguiling idea because it has seduced a large portion of the economists who have encountered it. This is so in spite of the fact that the economic theory is itself ambiguous, as demonstrated by the concept of excess momentum (Farrell and Saloner, 1985). Perhaps an even more telling criticism is the fact that it is unclear that there is any empirical support for this claim. Nonetheless, this concept has achieved considerable acceptance, perhaps in part riding on the coattails of the popularity of the concept of network effects, which itself has been the focus of only a very limited and suspect empirical literature, particularly as regards software markets. [7]
Since it has been claimed that software manifests network effects, and since increasing returns in production (decreasing average cost as output increases) are also likely in the software industry, the conditions would seem to be in place for software to be an incubator for lock-in and/or inertia. In the following section we will examine some implications of these theories. The analysis of software markets that follows examines evidence regarding the general predictions of these theories and some specific allegations of Microsoft’s ‘monopolist’ behavior in these markets.
II. Implications of Network Effects
Network effects are thought to be common in many technology industries. A virtual mountain of papers have been written about these effects in spite of the lack of evidence of their overall strength, such is its theoretical appeal. In the Microsoft case it has just been assumed that network effects were pervasive, which does not seem out of line with many academic articles. Legal standards, however, probably should require more stringent standards than current academic fads since so much is at stake, but the standards, in antitrust at least, have not been historically high. Just because there is so much uncorroborated agreement about network effects, however, does not mean that it is so. It is important that we try to verify these beliefs. Here we suggest some implications of network effects. If these implications are not found in software (or other network) markets, then we believe that the importance of network effects in software or other markets should be reexamined.
1. Price: The Neglected Implication
We start with this not because it is easily testable, but because it has been entirely ignored in the literature and because it is about as direct an impact as one can find.
Network effects increase the demand for the product. Ceteris Paribus, as network effects get stronger, prices should rise. This should be true for individual firms as they increase their market shares and also for the market as a whole as the size of the network increases, at least as network effects are usually modeled. This is in complete contrast with the impact of economies of scale, which by lowering the costs of the firm, should lower the price of the product for any given level of monopoly power exerted in the market.
It is somewhat surprising, then, that this clear implication of network effects has not received much attention in the literature.[8] The higher prices that should be caused by the increase in demand brought about, in turn, by larger networks, the relationship at the very core of this theory, has been neglected.
Because so many factors are involved in setting market prices, it may be difficult to generate a definitive test relating prices to network size, but the general claim of higher prices as networks get larger should be observable in some fashion. The reader should keep this in mind as actual prices in various markets are examined.
2. Winner-Take-All (or most)
The implication of network effects that has received most attention is the winner-take-all (or winner-take-most) result coming from increasing returns. A potential problem in testing this implication is that it applies equally well to old-fashioned economies of scale. For many high-tech products, particularly software, economies of scale seem very likely, thus making it difficult to separately distinguish the impact of network effects.
Another difficulty is that software has a characteristic that we refer to as ‘instant scalability’ meaning that the output can be increased very rapidly without the usual additional costs associated with sudden increases in output. This instant scalability is due to the fact that reproduction equipment is not specialized. Instant scalability will also be found for compact disc music recordings and long-playing records prior to that.
Note that this is not the same as zero or low reproduction costs. Printed books using specialized plates can be reproduced at very low marginal cost, but the scale of output is limited to that one printing location using those plates since it is time consuming to create other plates.
Software has this attribute because disc-reproducing machines are not specialized for any particular software title and thus it is considerably easier to increase output of a software title than to increase output of automobiles or even books. In a world where output can be changed so quickly, market shares can adjust to meet every whim of consumers. The market need not have unfilled orders or higher prices; no cabbage patch dolls or Furby’s here.[9] This provides yet another potential explanation for winner-take-all outcomes in software. The difference, however, is that market shares can change dramatically here since a product that better strikes a fancy will quickly come to dominate the market. The software markets that we look at appear to change in a manner consistent with this view of the world.
We believe that this concept of instant scalability may have been at work for some time and is worthy of further examination to determine how important its impacts may be. We hope some readers will join us in putting this concept to some empirical testing.
Winner-take-all results, therefore, are consistent with network effects, but also with economies of scale and instant scalability.
The concept of lock-in would seem to apply to software if it applies anywhere. Network effects should be stronger in most software markets than would have been expected in the other suspected havens of lock-in -- typewriters or videorecorders, say.
If large firms have advantages over small firms, then we have the possibility of what we have defined elsewhere (1995) as third degree lock-in or inertia. Obviously, this is not to say that an incumbent remains forever entrenched, but instead that when a better product comes along it might not get adopted even if there were economic advantages to its adoption. This is to be distinguished from weaker cases of lock-in where real costs of switching make it uneconomical to switch to another product.
Confirmation of the lock-in hypothesis requires finding better products that are not adopted.[10] Testing for inertia would require that we know the optimal switching speed and compare that to the actual speed with which a switch is made. Testing for inertia obviously requires more detailed knowledge than testing for lock-in, and considerably more detail that we are likely to find. Instead, we ask the reader to compare the speed of changes in software markets to speed of changes elsewhere, admittedly an imperfect measure.
As already mentioned, Judge Jackson in his Finding of Fact converted the lock-in concept to a barrier to entry. As we all know, barrier-to-entry is a slippery concept, meaning anything from a high cost of entry to a cost incurred by the challenger but not the incumbent. The latter definition has always seems correct to us, and in that case it is unclear how network effects, economies of scale, or any of the other factors at work favor the incumbent relative to the challenger. The incumbent had to coordinate consumers to adopt his product, whether it was first in the market or replaced a previous incumbent. In either case, getting consumers to come on board is a cost imposed on all market entrants, late or early and does not necessarily favor early firms.
In contrast to much of the path dependence literature which focuses on excess inertia, we note the possibility that the speed at which market leaders replace one another might be greater in software markets than in many ‘ordinary’ markets due to the instant scalability feature of software. In fact, the software markets that we examine below exhibit extremely rapid changes in market leadership. Although we are not aware of any precise benchmarks with which to make comparisons to other industries, it appears on casual inspection that changes in market share occur at a far more rapid rate in software markets than in other markets.
Tipping is a term that has been used rather vaguely in the literature and in Microsoft antitrust testimony. A market is considered to have ‘tipped’ when one product or standard has become dominant. A market is tipping when it is sliding toward the eventual domination by a single firm. It is generally claimed that tipping occurs when a product has generated sufficient momentum with regard to network effects that its domination is inevitable. Although there is considerable vagueness in the literature, the general idea seems to be that after a period of struggle, one firm breaks out from the pack and tipping ensues.
We assume that tipping has some meaning other than just market dominance. Since network effects only strengthen as market share increases, there should be a pattern to market share growth. First, when a firm is smaller than it competitors, growth should be slow since network effects should be working against the firm. Later, when consumers anticipate that a firm will become larger than its competitors, market share increases should come more easily, since network effects are self-reinforcing. This tipping point, or critical mass, should mark an acceleration in market share growth. In such instances, increases in market share for leading products would occur even without any advantages in product quality. As we will see, there is no evidence for markets tipping in this manner.
III. Software Markets: Background
We will focus on a few key software application markets. We are particularly interested in markets where Microsoft has achieved or failed to achieve a dominant status. We examine spreadsheets and personal finance as two markets where Microsoft has had very different success, and Internet browsers since this market was so crucial to the case.
When faced with software choices, whether changing to a completely different product, or just upgrading a current choice,[11] the logic of the choice facing the consumer is straightforward. First, the consumer must evaluate the increased value that would be brought about by using the new software relative to remaining with the current software. This extra value might come from new features, increased speed, better compatibility with other users, and so forth. On the other side of the ledger go the costs of purchasing the software, the costs of the transition (presumably users will be less productive as they learn to use the new software), and the costs of any file incompatibility with previous software (the cost of imperfect access to old data).
Software has several characteristics that will make it appealing for consumers to remain with the same vendor. First and foremost is a concern with backwards-compatibility–a compatibility with the learned skills associated with previous versions of a program, and compatibility with file formats that were created with previous versions of the program. If the adoption of a new spreadsheet or wordprocessor removed any ability to read the consumer’s old files, it is very unlikely the consumer would make the change. That is why virtually all software products try to provide some degree of backward compatibility with older versions. Note that this is a real cost and it is not a form of inefficient lock-in for the consumer to avoid changing if the benefits to changing are less than this cost.
It is also common for software products to provide some ability to read files produced by other products so as to entice the users of other products to switch and to increase the value of the product to users who exchange files. Some software products try to lower the consumer’s costs of switching by mimicking aspects of the user interface of other products.
There are, of course, numerous reasons why a user might have an interest in changing their software applications. Even if they are happy with their current product, it might be that the user needs to move to a new operating system that doesn’t run the old product (say from PC to Mac). Or, as is more commonly the case, the user might be looking for some additional features missing from his current product.
Another factor that influences consumers’ choices and is likely to be related to large changes in market shares are paradigm changes, or market displacements. These are external changes in the markets that provide motivations for large numbers of current users to switch to a new product. These changes might be a design alteration in the nature of the product itself, or a change in the operating environment of the product. Examples would be the removal of copy-protection, allowing the product work in networked environments, the movement from 8 bit to 16 or 16 to 32 bit architectures, the growth of the Internet, the invention of macros, graphical systems, voice recognition, and so forth. If consumers find these changes compelling, they will be willing to switch products if necessary in order to take advantage of the new environment. Two such displacements were the movement from DOS programs to Windows programs and the creation of Office Suites.[12]
How does one go about measuring the quality of a software product? We have decided to leave the examination of the software quality the to experts writing for (mainly) computer magazines. There are, however, still some issues that require judgement calls. First, many magazines do not give numerical scores to competing software products. In many instances, particularly during the 1980s, magazines compared features of various products without making judgements. When judgements were made, magazines often picked a ‘winner’ or ‘editors choice’ without indicating the quality differential between winners and losers, and without differentiating among the losers.
To overcome some of these limitations, we used more than one method of comparison. These ranking generated with either of these methods are presented as timelines, which can then be compared with market shares.
When magazines numerically rated individual software products, we used these ratings. When numerical scores were given we report these scores, normalized so that the highest rated product always receives a ‘10’. In cases where reviewers rated software by qualitative category such as ‘excellent’, ‘good’, and so forth, we converted these categories into numerical equivalents. In those cases where no overall rating was given, but only ratings for particular characteristics (ease of use, speed, and so forth) we take the average value of the characteristics to get an overall quality ranking for the software application relative to its competitors. In these cases we assigned numerical scores to these ratings [excellent=10, good=7.5, fair=5, poor=2.5] and summed the ratings across categories to arrive at a score for the package as a whole.[13]
For those reviews that provide clear winners, but no overall evaluation, we count the number of reviewers who state that a particular product is the best. We refer to the number of reviews make such judgements as the number of ‘wins.’ This latter measure, even though it provides no information about the remaining products, provides a relatively large number of data points.
To avoid clutter we often remove from the diagrams products that were not important players. Occasionally, therefore, the winner in a particular review may be one of those packages that we have removed, but the reader can easily tell its presence from the fact that the highest score listed will be less than 10. In fact, this occurred quite infrequently.
Although both measures were constructed for each market, only the measure of wins is reported in this paper. Interested readers will need to refer to our book for complete coverage.
There are also choices regarding the measurement of market share. It can be measured either in units, or in revenues. Most of the time it makes little difference, but we report shares based on revenues unless otherwise specified. Since we usually report prices as well as market shares, the reader can determine unit based market shares for themselves.
Some computers come with a set of software products installed. Although the consumer may not use these preinstalled products, market share statistics treat these preinstalled products as though they represent actual purchase decisions and use.[14] In that sense, market share statistics might sometimes be misleading indicators of actual use. Fortunately, it appears that OEM sales are not likely to influence our results too strongly.
OEM sales, while increasingly important, are still a minority component of application markets. Office Suites are one of the more commonly included products yet the importance of OEM sales is not high. Writing in 1996, IDC analyst Mary Wardley stated:[15]
OEM sales have suddenly grown to represent a significant percent of overall sales for office suites and thus spreadsheets… However, OEM sales have had little to no impact on Microsoft so far, and IDC believes their impact will remain negligible throughout the forecast period.
The data on OEM sales of Office Suites in the two years for which they are available, 1996 and 1997, reveal OEM sales to have a limited impact, as illustrated in Figure 1. Although OEM sales are responsible for about 30% of unit sales in these two years, they are responsible for less than 10% of revenues. And for market leader Microsoft, they are responsible for only about 10% of sales and 7% of revenues.
It is worth noting, in light of the role given to Microsoft’s putative attempts to control OEMs in the government’s case against Microsoft, the latter’s limited use of OEM sales.
Finally, statistics on market shares and prices were generated using data collected by data collection companies Dataquest and IDC. We concluded after a brief examination that list prices don’t seem to tell us very much. For one thing, they have not varied very much. Second, it must be remembered that list prices neglect the impact of office suites. Third, list prices do not allow us to control for the important role of upgrades, which are less expensive than first time purchases. Also, the list prices fail to account for the units sold to OEMs, which have a far lower price. For all these reasons, we will not focus on list prices but instead on the actual transaction prices received by software producers.
Spreadsheets are one of the mainstays of computer applications. Spreadsheets allow the manipulation of numbers and formulas, and their conversion to charts. The leading spreadsheet is Excel, from Microsoft. It is one of the categories of applications where questions arise about the causes of Microsoft's success.
A. The Evolution of the PC Spreadsheet Market
Credit for the invention of the spreadsheet goes to Dan Bricklin and Bob Frankston who created VisiCalc for the Apple II. VisiCalc had a list price of $250. Lotus 1-2-3 was introduced in January 1983 at a price of $495. It was immediately acknowledged to be a better product than VisiCalc. In December of 1982 Gregg Williams wrote in Byte (page 182) that 1-2-3 had "many more functions and commands than VisiCalc" and that 1-2-3 was "Revolutionary instead of evolutionary". PC World called it "state of the art".
In October of 1983 PC World reported that 1-2-3 was outselling VisiCalc.[16] VisiCalc was removed from the market in 1985 after being purchased by Lotus. Users of VisiCalc were offered upgrades to 1-2-3.[17] In terms of market share, Lotus was to remain dominant for almost a decade.
Unfortunately we do not have detailed data on this early market. Still, from what we have pieced together, it is clear that a superior product, Lotus 1-2-3, was able to quickly wrest market share away from VisiCalc.
By 1985 most spreadsheets were meant to work with the IBM PC. Lotus 1-2-3 was the champ, with other contenders such as Computer Associates (Easy Planner; $195), Ashton-Tate (Framework; $695), Software Publishing (PFS Plan; $140), and IBM (PlannerCalc $80 and Multiplan 1.0; $250). Microsoft had Excel for the Mac (at $495) and Multiplan 2.0 ($195) for the PC.
Soon, many new spreadsheets arrived on the seen. Some were clones of 1-2-3 at lower prices (VP Planner, The Twin, and VIP Professional). Other spreadsheets, such as Javelin and SuperCalc tried to differentiate themselves by improving upon 1-2-3 in some manner, but although these alternatives received some praise, they did not receive universal acclaim and did not make much of a splash in the market.
Excel first appeared in 1985, but for the Macintosh only. Jerry Pournelle, a well-known columnist for Byte (and science fiction author), wrote (incorrectly but nonetheless prophetically): "Excel will make the Mac into a serious business machine."[18]
In late 1987 Microsoft ported Excel to the PC (running under an early version of Windows) and Borland introduced Quattro for DOS. Microsoft’s election not to produce a DOS version of the program was something of a gamble. The success of Microsoft Windows was far from assured until version 3.0, which became available in 1990. The fact that PC users did not flock to earlier versions of Windows most likely reduced the sales of Excel since users would have had to load Windows to run the spreadsheet and then return to DOS for other applications. Also, many of the features of Excel would have worked best with a mouse, and it was rare for PC’s to come with a mouse.
Thus began the market struggle between Microsoft, Borland and Lotus. Reviews found windows-based Excel to be superior to 1-2-3, with DOS-based Quattro in second place. 1-2-3's market share was unable to stand up to the assault of superior products and Excel's market share eventually surpassed 1-2-3 and it has been firmly in first place ever since.
In the early and mid 1980s the closest competitor to 1-2-3 in terms of reviews appears to be SuperCalc. PC Magazine, in its "BEST OF 1986" review had this to say: "if market dominance were based on rational criteria, Computer Associates' SuperCalc 4 would certainly replace 1-2-3 as the leading spreadsheet program. After all, it can do anything that 1-2-3 can do and adds some notable features of its own."[19] But although various spreadsheets had attributes that were sometimes considered superior to 1-2-3’s, there was no general consensus that any alternative was superior. For example, in October of 1987 Michael Antonoff in Personal Computing stated: "SuperCalc, VP-Planner, and Twin, lack the elegance of 1-2-3 in links to applications."[20]
In late 1987 the quality of the competitors began to change. In a portentous statement, PC Magazine, stated: "Microsoft Corp. has just unleashed a spreadsheet that makes 1-2-3 look like a rough draft."[21] This reference was to Excel 1.0 for the PC, a port of its Macintosh Program.
According to reviewers, Lotus appeared to fall behind its competitors in terms of functionality and usability. When comparing 1-2-3 to Excel, a reviewer states "Excel offers a lot in the form of tantalizing features missing from the current version of 1-2-3."[22] Quattro was called "a powerful spreadsheet with more features than 1-2-3 Release 2.01, yet fully compatible and a better price."[23]
These were not isolated opinions. Reviewers in general had a very high opinion of Excel in the late 1980s.[24] Clearly, Excel was thought to be the best spreadsheet.
The reason for Lotus’ loss of prestige is illustrated better in Figure 2, which represents the number of times each spreadsheet won a comparison review in a year, or was bestowed in a magazine with the claim that it was the best product. Excel clearly was the winner over the ten-year period. Between 1989 and 1994 Quattro also managed a fair share of wins, although since 1994 Excel has monopolized the victories. The remarkable feature of this chart, however, is that over the entire ten year period, Lotus 1-2-3 just barely avoids a shutout, managing but a single win out of the numerous instances where a magazine reviewer declared a winner.[25]
The message from these data is quite clear: Excel was clearly the leading spreadsheet in terms of capabilities. The main reservation about Excel was its need for powerful hardware. This last requirement was due to Excel’s entirely graphical interface. As is the case with virtually all graphical software applications, Excel was slower to perform tasks than were DOS spreadsheets, even though it could show results that non-graphical (DOS) based applications could not. Once the hardware had caught up to the software (and Windows itself improved) there were no serious challengers to Excel’s superiority
This brings us to another consideration: what about price? If we are interested in explaining changes in market share we should expect both price and product quality to be important. If there are serious differences in the prices of competing products, a lower quality but lower prices product might have a larger market share than a higher quality higher priced product.
The pattern of average prices received by the manufacturer is illustrated in Figure 3. Borland’s price discount strategy is clearly revealed. Note also that Lotus kept its prices similar to Excel’s even in the face of the latter’s increasing market share (as we shall see shortly) and superior reviews. Lotus only began to clearly undercut Microsoft’s price in 1996, well after it had fallen below Excel in terms of market share.
But the big story is the stunning fall in prices. By 1997, the typical price (received by the vendor) for a spreadsheet had fallen to approximately $50, or a fall of over 80% from the typical price in 1988.
D. Changes in Market Share: Analysis
All of the above evidence indicates that if markets chose better products, that Lotus should have lost market share and market dominance. After all, for a considerable period of time it was not the top ranked product. The winner should have been Excel. Is this what happened? Absolutely.
Figure 4.provides market share data (based on revenues) that reveals 1-2-3 losing its dominant position to Excel.
Excel had other factors increasing its likelihood of success. Not only was it the best Windows spreadsheet, but it was the first Windows spreadsheet. It was also part of the first Office Suite, and it was part of the best Office Suite. Office suites and Windows were both strategic gambles by Microsoft that paid off handsomely.
Given the superiority of Excel, this result should be viewed as support for the view that markets choose the better products.
The market share results do not provide any empirical confirmation for the idea of markets tipping. When a product has a small market share, network effects should work to hinder increases in market share, but beyond some breakout point network effects should work in favor of market share increases. This would tend to imply some form of bend or kink, where positively sloped market share curves are flatter to the left of the breakout point, and steeper to the right. Instead what we find is a fairly steady increase in Excel’s market share and a fairly steady decrease in 1-2-3’s. This may mean that network effects are just not a very important factor in this market, or that their impact on market share is more complex than indicated in the concept of tipping. Either way, this does not bode well for the accuracy of the government's expert economists who discussed the concept of tipping with little or no skepticism.
There are some other issues worthy of examination. In a world of instant scalability, why did it take five years and not one, or two, for Lotus to be dethroned? We can hypothesize some answers. Users know that there is a tendency for software products to leapfrog each other in capabilities. It normally will not make sense for a consumer to switch products every time one product exceeds the capabilities of another, because of the costs of switching. Instead, it is rational for the consumer to expect that the next upgrade of a product that has fallen behind, but which has a history of technological innovation, to contain the features missing from the current version and perhaps even a few extra capabilities. Therefore, consumers will only switch when they can not wait for the missing feature to appear in the next version of their current product, or when it becomes apparent that their current vendor has fallen behind the curve and will not be able to provide a product meeting their needs in the foreseeable future. It probably took two failed generations of products for Lotus to convince the market that it was not going to catch up to Excel anytime soon. Still, Excel average 8 market share points a year, which by almost any other standards is quite amazing.
A second question that might arise in the mind of the reader is why Quattro never surpassed 1-2-3. It appears that we can categorize Quattro as a superior product to 1-2-3, even if it was inferior to Excel. There are two reasons, however, why Quattro might not have surpassed an inferior 1-2-3.
First, if a 1-2-3 consumer was planning to switch, it would make more sense to switch to the number 1 product than to the number 2 product. Quattro’s only advantage over Excel was its lower price.[26] Instant scalability gave Microsoft the chance to meet the demand of all defecting Lotus users without incurring increases in average cost. In markets without this feature, the number two firm likely would pick up some of the slack due to constraints in production for the leading firm.
The second point is that between 1988 and 1995, when Quattro garnered superior reviews to Lotus the gap with Lotus did diminish. It fell from a ratio of about 7:1 in sales to a ratio of 2:1. After that, however, Quattro no longer enjoyed clear superiority over 1-2-3.
Would the Lotus hegemony over spreadsheets have ended without the advent of Windows and Office Suites? Certainly some GUI was going to replace DOS, and spreadsheets were going to have to have to be graphical in nature. Given that Lotus was unable to produce a high quality graphical spreadsheet, it would have lost the market to someone. Microsoft would have been a strong candidate to dethrone Lotus in any graphical environment because of Excel’s total dominance in the Macintosh market.[27] If the market had gone to the Macintosh, where Excel was clearly the dominant product Lotus would have lost. If OS/2 had predominated, there is every reason to believe that the OS/2 version of Excel would have replaced 1-2-3. And if some third party graphical operating system had prevailed on Intel based machines, there is every reason to suspect that Microsoft would have produced a better product than Lotus because of its familiarity with graphical products.
Personal finance software allows individuals to balance their checkbooks, track their investments, and plan for the future. The software has become very easy to use and very inexpensive. The major players are Quicken by Intuit, Microsoft Money, and Managing Your Money by Meca. This market is particularly interesting because it is a market in which Microsoft is a major player but not the leader in the market.
Network effects are probably smaller for these products than for many other software markets. Consumers generally are not interested in being able to switch files with other individuals.[28] With the advent of the Internet, however, the ability of software to be compatible with the offerings provided by one’s banking institution has probably increased the level of network effects somewhat.
Personal finance products were introduced in the early to mid 1980s. “Andrew Tobias’ Managing Your Money” was the original market leader, to be replaced by Quicken.
Figure 5 shows the wins for the various products in the market. In the late 1980s Managing Your Money was initially considered the best and most powerful product in the category. When Quicken was introduced it received reviews that were less positive because it was not as powerful as Managing Your Money. It basically limited individuals to balancing their checkbooks and checking their budgets, but it did this quite well.[29] It also was less than one third the price of Managing Your Money.[30]
Over time Intuit improved Quicken, adding more sophisticated features, and as Figure 5 reveals, by the early 1990s it was considered at least the equal of Managing Your Money and by the mid 1990s Quicken was clearly considered the best product. Managing Your Money still appealed to those more interested in power, e.g. tracking sophisticated investments as opposed to just ordinary stocks and bonds. Microsoft Money, which had recently appeared, appealed to those interested in simplicity.[31]
Figure 6 provide information on the market shares of the three leading products. This appears to be another case of one dominant firm followed by another dominant firm, what we call serial monopoly. A market leader with 60% of the market is displaced within three years by another firm which then achieves a market share of greater than 60%. The rapidity with which Quicken overtook Managing Your Money is astonishing, although the source of the data requires some caution on our part.[32] Still, market shares changed so rapidly in this market that the concepts of inertia or lock-in seem entirely out of place. Also absent is any hint of tipping in this market.
There are some differences in this market from the results found in other markets, however. It appears that Quicken became the leading product in the market when it had merely matched and before it had surpassed the quality of Managing Your Money. Figure 7, which represents the prices for the leading products, provides a possible explanation for Quicken's ascendancy before it surpassed the quality of Managing Your Money. The very high relative price of Managing Your Money is clearly apparent prior to 1992. Unlike many other software markets, this market caters to individuals as opposed to businesses. The retail price of Managing Your Money, being three times the price of Quicken, might well have deterred price conscious individuals from choosing it when the two products were similar in quality. This could well have led to the rapid demise of Managing Your Money.
By 1992, when Managing Your Money finally lowers its price to match the other programs in the personal finance software category, it has lost its quality advantages. Also, Managing Your Money was hurt, as were so many other products that we have examined in other markets, by its inability to provide a timely Windows version of the product; it was not until 1994 that a Windows version of Managing Your Money arrived.
We can conclude that consumers preferred Quicken's low price, simplicity, and checkbook handling to the sophisticated financial management and much higher price of Managing Your Money.
In 1991 Microsoft introduced its Money program for Windows. Unlike many of the market leaders in other applications markets, Intuit had a Windows product as early as Microsoft. In fact, its initial share of the Windows market was actually greater than its share of the DOS market. Quicken's market share in the Window’s market ranged between 70-80% in the early 1990s, and was some ten to twenty points higher than its overall market share.
Microsoft has gained market share throughout the decade, but not from Quicken. Instead, its Windows product has replaced products in the DOS market. Microsoft’s share of the Windows market has remained virtually unchanged, having actually fallen slightly since 1991. Microsoft also has had a lower price than its competitors since 1992, perhaps taking its cue from some reviewers who argued that since it was less powerful than its competitors it should have a lower price.
Quicken's retention of its market leadership is not surprising given its high quality as indicated in successful reviews. According to Microsoft’s critics, however, Microsoft should have been able to leverage its ownership of the operating system to achieve a dominant position, independent of the quality of its software. The fact that Microsoft was unable to do so casts further doubt upon these claims.
Part of the reason for Quicken's success in the Windows market may have been Intuit’s experience in the Macintosh market where Quicken was also the leading program, giving Intuit a clear understanding in how to write a successful GUI application.[33] This appears to be a common theme—success in the Windows market is presaged by success in the Macintosh market. Of course, it is entirely logical that it should be so.
For the purposes of our study, the important finding is that we have another case where an incumbent was quickly replaced, but only after the incumbent lost quality advantages. We also have an instance where Microsoft’s ownership of the operating system has not allowed it to dislodge the dominant product, in spite of its considerably lower pricing. This reinforces our findings that achieving product quality parity or superiority is a requirement for increased market share.
Browsers are products that allow PC owners to access the World Wide Web (WWW). The software category is relatively new since the WWW is a relatively recent phenomenon. As late as October of 1994, this market was largely unformed.
Mosaic (created by the University of Illinois’ National Center for Supercomputing Applications) was the first successful browser. As PC Magazine reported in early 1995:
NCSA's Mosaic was the first Windows-based Web browser--the killer app that started the stampede to the Web. Today, largely as a result of its stellar success, the browser field is in furious ferment, with new products released weekly and existing ones updated sometimes hourly. Browsers are even making it into operating-systems: IBM includes a browser with OS/2 Warp, Version 3, and Microsoft plans one for Windows 95.[34]
The first magazine reviews of browsers that we found were in mid 1994. Netscape, which was formed by some former programmers of Mosaic, introduced Navigator in late 1994. As far as we can tell, magazine reviews considered Navigator to be better than Mosaic from the day of its introduction. Both products were free. Internet Explorer, Microsoft’s entry in this market, makes its first (beta) appearance in our magazine reviews in September of 1995.
Since the browser ‘market’ began as freeware, and largely continues as freeware, it is a somewhat unusual market in that there are no direct revenues. This is not, however, any different than the over-the-air television market, where stations do not generate any revenues from their viewers. Instead, television stations pay for their programming by selling advertising, and browser companies appear to have adopted a similar model. Revenues can come either from the sale of servers, which send out the information retrieved by browsers, or by selling advertising from the sites that browsers initially go to.[35] Although consumers can easily change the website that is the starting location for the browser, continued patronage of the original location (which is believed likely to occur) would provide considerable advertising revenues to the company that created the browser.[36]
The market for browsers is of great interest because of its role in the DoJ antitrust proceedings against Microsoft. The claims made by Justice are that Microsoft’s inclusion of its browser with the operating system has foreclosed Netscape’s ability to generate market share, independent of the quality of the browsers. The government is claiming that Microsoft’s bundling of the browser into Windows 98 has caused the erosion of market share that has been Netscape’s recent fate.
Given our findings in other software markets that market share is strongly related to product quality, we would expect a similar relationship to hold here. In fact, since all browsers have essentially the same price (zero), there are fewer other factors at work that might disturb this relationship. [37]
In light of results from other markets, we can try to answer the following question: has Microsoft’s browser achieved a market share incommensurate with the share that might be expected given the relative quality of its browser?
Figure 8 indicates the wins in the browser ‘market’ since 1995. Because this market is so new, the time span is compressed, with intervals of six months as opposed to the one or two year intervals in prior charts. Still, the trend is very clear. Netscape Navigator was clearly considered the best product until the second half of 1996, after which Internet Explorer became the superior product.[38]
The time frame is compressed once again in Figure 9 which provides information on
the market shares of the two products.[39]
Market share here is measured differently than in other markets. Instead of
measuring sales, which is the flow of new products into an already installed
base, market share here reflects installed base. However, the information on
use also indicates that most browsers in use are of very recent vintage, so
that stock and flow magnitudes are likely to be very similar.
The actual increase in Internet Explorer 's market share has been a matter of some contention. Microsoft relied on a survey that indicated a relatively small increase in Internet Explorer's market share, and the government relied on browser usage data which tend to reveal a considerably larger increase in Internet Explorer's market share.
There are serious difficulties impeding an unambiguous measurement of browser usage, including the fact that many ISP's 'cache' popular web sites and that some ISP's, such as AOL put their own brand on someone else's browser.
Nevertheless, Internet Explorer has increased its market share. The Zona survey, from which Figure 9 is based, indicates one of the larger increases in Internet Explorer's market share.
This increase in Internet Explorer's market share has occurred concomitantly with Internet Explorer’s quality improvements relative to Netscape Navigator. Internet Explorer barely grew at all, even though it was freely distributed with Windows and on the Internet, until after July of 1996. This would be the same period of time where Internet Explorer achieved parity with Netscape for the first time. This is also the first time that Internet Explorer appears to have taken market share away from Netscape which had continued growing its market share even in the face of Microsoft providing a large number of free copies of Internet Explorer.
The level of superiority of Microsoft Internet Explorer after 1997, as indicated in the reviews summarized in Figure 8 has in our examinations of other markets always led to increased market share and even market dominance. Is there any reason to expect different results in this market?
There are several reasons that we might expect more rapid market share changes in this market than in the typical software market. First, network effects seem rather small in this market since there is little interactivity between users that might cause users to care who else uses their make of browser. Second, compatibility with prior versions seems small, since learning the idiosyncrasies of a particular browser takes far less time than learning a spreadsheet, word processor, or desktop-publishing program. Further, there is less reason to worry about incompatibility with the files created by these programs since generally the only files that remain through multiple generations of browsers are the list of ‘favorite’ sites, and these are fairly easily translated. Finally, the price of browsers is zero, and users presumably already have access to the free downloads available on the web, since browsers are of no value without Internet access. Therefore, the concept of instant scalability, meeting shifts in market demand on a moment’s notice, is even more pronounced here than in other software markets.
Thus, we should expect that market shifts would occur more rapidly in the browser market than almost any other software market that we have encountered.
There are two other markets where market shares changed by fifty or sixty percentage points within two years, one being personal finance software (the other being low-end desktop publishing). In the browser market we find a change in market share of approximately forty points within a two year interval from July of 1996 to July of 1998. Is the difference in browser quality such that we would expect this type of change?
We are not in a position to give a definitive answer. As noted, we should expect lower costs of switching browsers than most other software, which would enhance the speed of market share shifts. Whether the quality differential is sufficiently great to support this level of market share change we cannot know with certainty.
But given the low costs of switching browsers and the extent of the quality differentials, and given what we have learned about market share changes in software markets, Microsoft’s increase in market share in the browser market seems well within reasonable levels. Our analysis indicates that the change in market share that has occurred in the browser market could have been explained entirely by quality differentials, with no appeal to the factors that the government has focused on in its case against Microsoft, i.e. ownership of the operating system or exclusionary contracts.
This is another market which doesn’t seem to exhibit either inertia or tipping. Since network effects should be small in this market, it might not be a very strong test of lock-in or the tipping hypothesis, however.
Currently, Internet Explorer appears to have built up a considerable lead in market share, at least as far as usage, although some data services do not agree.[40] As long as Internet Explorer retains its advantage in quality, our analysis of other markets indicates that we can be confident that Internet Explorer will continue to take share away from Netscape – whether or not it is included in the operating system.
VII. Microsoft, Monopoly, and Consumer Harm
Monopolists cause harm to consumers by reducing output and raising price. Although there has been much talk about Microsoft’s impact on innovation, there is no clear relationship between monopoly and innovation, nor any data to test such assertions. The question of whether Microsoft acts like a monopolist and whether it causes harm to consumers in the traditional, well-tested context should be crucial to the current DoJ case. What is the evidence?
In the sections that follow we examine the price history in various software markets. Putting several disparate pieces of this pricing puzzle together makes, we believe, a persuasive case for concluding that Microsoft has not harmed consumers.
A. Word Processor and Spreadsheet Prices
Figure 10 reports average prices for the two largest markets, spreadsheets and word processors (in the PC, or IBM compatible, market). The most relevant feature of this chart is the very large overall fall in prices, as already noted.
This fall in prices is not constant throughout the period, however. From 1986 until 1992, prices were either constant or rising slightly in the word-processing market. From 1986 until 1990 prices were essentially constant in the spreadsheet market. Beginning in 1991 or 1992 prices fell in a very dramatic fashion. What is the proper interpretation of these price declines?
It will suffice for the moment to describe the early period as the Lotus era in spreadsheets and the WordPerfect era in word-processing. Although any firm, including Microsoft, could have charged whatever price it wanted when it had a small share of the market, it is likely the case that the most effective strategy was to follow the price set by the leader in the market. This conclusion is reinforced by our finding that, except in markets catering to individuals, price differentials were rather ineffective at generating additional market share.
This chart provides information on price changes, but monopoly and competition differ in the level of prices, not in the way that they change price. However, if the market were moving from a less competitive to a more competitive market equilibrium, prices would be expected to fall during the transition. If Lotus and WordPerfect tried to use their large market shares to generate short run monopoly profits, and Microsoft didn’t, we would expect prices to fall as the markets stopped taking its cues from WordPerfect and Lotus and instead started to take their cues from Microsoft. Once the new regime was in place, prices should have stabilized at their new lower level, ceteris paribus.
If Microsoft believed that Lotus and WordPerfect lost their dominant position because they failed to act competitively, it might have chosen a competitive price even after it achieved a dominant market shares. That appears to be what has happened since there is no evidence of prices rising even after market shares above 99% are achieved, as in the Macintosh spreadsheet market, and this pattern of behavior on Microsoft’s part occurred in some other markets such as midrange desktop publishing which we do not cover in this paper.
By itself, this evidence might be suggestive, but would fall short of a sufficient basis on which to draw conclusions. Combined with the evidence in the next two sections, however, there is a compelling case to be told in favor of Microsoft.
B. Microsoft’s Overall Impact on Prices
One question that naturally arises in regard to pricing is how the markets examined in this study compare to markets overall. After all, it is possible that the pricing pattern in figure 4 was reproduced in many other markets and had nothing to do with Microsoft’s increased market position. We were able to perform a somewhat crude test of this claim. Dataquest provides consistent market definitions for fourteen software markets for the contiguous period 1988-1995.[41] We calculated the average price in each category for each year as if this ‘average price’ represented changes in prices for the underlying products.[42]
Next, we categorized markets into three main groups: one group of markets where Microsoft has a product, one group where Microsoft has no product, and a third group where the products compete with some function of Microsoft’s operating system.[43] All prices are normalized to their 1988 levels to simplify comparisons.[44] The results are shown in Figure 11.
The results are rather striking. Although it appears that software prices in general have fallen in price over this period of time, some software prices have fallen far more than others. In particular, those categories where Microsoft participates, directly or indirectly, have had far more dramatic declines than in other markets, falling by approximately 60% compared to the relatively paltry 15% fall in prices for software in markets completely devoid of Microsoft’s influence.
C. PC – Macintosh Price Comparison
The data allow one more interesting price comparison. That is for the same product in two markets with very different market shares. We are here talking about Microsoft Word and Microsoft Excel in both the PC and the Macintosh market.
Figure 12 provides market shares for these two products in the PC and Macintosh markets. Examination reveals that Microsoft achieved very high market shares in the Macintosh market even while it was still struggling in the PC market. On average, Microsoft’s market share was about forty to sixty percentage points higher in the Macintosh market than in the PC market in the 1988-1990 period.[45] It wasn’t until 1996 that Microsoft was able to equal in the PC market its success in the Macintosh market. These facts can be used to discredit a claim sometime heard that Microsoft only achieved success in applications because it owned the operating system, since Apple, not Microsoft, owned the Macintosh operating system and Microsoft actually competed with Apple products in these markets.
Microsoft’s market share in the late 1980s in the Macintosh market is generally higher than the market shares achieved by Lotus or WordPerfect in the PC market. Microsoft’s market share by the mid 1990s in the Macintosh market was considerably higher than any leading firm in the DOS era. To a structuralist Microsoft would appear to enjoy a monopoly in the Macintosh market. One might expect, therefore, that its prices in the Macintosh market to be at least as great as its prices in the PC market. Additionally, since it was trying to increase its market share in the PC market in the late 1980s one might think that its prices there would be considerably lower than where it already has a ‘monopoly’ market share.[46]
Those holding a structural view of monopoly, therefore, might expect Microsoft to charge a higher price in the Macintosh market and a lower price in the PC market.[47]
Although there is some imprecision in the data,[48] the price comparisons suggests the exact opposite of this. PC-Excel, far from being cheaper, in fact averages 13% higher prices than its Macintosh cousin during the period 1988-1992. PC Word’s price, on average, was more than eighty percent above the price for Macintosh Word prior to 1993. After that, the prices are virtually the same. To double check this result we went to price advertisements in computer magazines from PC Connection and Mac Connection, a retailer selling to both markets, and compared prices. Excel on the PC was consistently about 33% higher than for the Macintosh version.[49] A similar result holds for Microsoft Word.
In other words, Microsoft was not charging high prices in the market where it clearly has a structural monopoly. If anything, the prices in the market where it was dominant were lower than in the markets where it was competing. After Microsoft had come to dominate the PC market, it might have been expected to raise prices, but it lowered them dramatically. We cannot attribute this result to some idiosyncratic difference between PC and Macintosh markets since Microsoft equalized the prices in the two markets after gaining dominance in both. What might be going on, then?
One answer, that appears consistent with all our findings, is that Microsoft worries about competitors even when it has a very large market share. Such concern about potential entrants might explain why Microsoft has not lost any markets it has gained. The decline of DOS leaders might have been partly due to an erroneous lack of such concern.
The concept of network effects has been a cudgel in the government's case against Microsoft. The government's use of network effects requires that certain assumptions be true if the case is to make economic sense. This paper has opened a window into the workings of software markets. To our knowledge, ours is the first systematic examination of these markets. The results have run counter to much of what passes for conventional wisdom.
The lack of evidence for market inertia and tipping should raise red flags to analysts in these markets who have treated these untested claims as established facts. The rapidity of market share changes should cause believers in market inertia to take pause. Also, the concept of tipping, which seems so reasonable in principle, seems to have no support if tipping is to mean that self-reinforcing mechanisms accelerate rates of change.
The rapid changes in market share and the close relationship between these changes and product quality that was demonstrated in this paper for three markets was repeatedly found in our more complete analysis. The large and rapid changes in market shares are counter to the claim that dominant firms become locked-in. We have seen instances of a once-dominant firm that faded into obscurity including Managing Your Money, Lotus 1-2-3 (and elsewhere, WordPerfect, PageMaker, Ventura, First Publisher and Prodigy). In some instances, the product that replaced these onetime leaders was a product from Microsoft, in other instances it was from some other company. In each instance, however, the product that replaced the incumbent was of higher quality and often cheaper.
The evidence makes clear that the most important factor for replacing an incumbent firm is to produce a better product. Other factors are undoubtedly also important, but just getting a product to the starting line of a new race was unlikely to be successful if the product was not of high quality. Sometimes price played a role, although it appeared to have a surprisingly small role (perhaps because price changes were often mimicked by other firms). Also of interest was the finding that being number two was of little value. Virtually all the shift in market share seemed to go to the number one product as measured by quality.
The other important finding is the amazing track record of Microsoft in terms of producing better products at low prices. Although it has been suggested that Microsoft did not earn its large market shares in applications, we have found the evidence to be at complete variance with this suggestion. In those instances where Microsoft moved from a low to a high market share, its products were clearly better than the market leader’s. When its products were not superior, it did not make inroads against the market leader.
It has been suggested that Microsoft has used it monopoly position to refrain from lowering prices. Any claim that Microsoft harms consumers through monopolistic pricing is wholly at odds with the evidence we have uncovered. Microsoft appears to have a regime of low prices in markets it has come to dominate. Naturally, Microsoft also had to hurt its competitors by these actions.
It is true that the government has focused on the operating system and not on applications, although the states involved in the case had attempted to bring the application market into the case and seemed only to lack sufficient time to do so. Evidence on Microsoft's pricing of the operating system, although largely given behind closed doors, does not seem to be contrary to our findings in applications markets and does not seem capable of answering the basic question of monopoly power. As we have stated before, monopoly does not imply rising prices, only high prices.
We conclude that the use of network effects in antitrust
cases appears to be unwarranted in the software industry. We also believe that
Judge Jackson was mistaken in his findings of fact, which clearly signal that
he is going to find Microsoft guilty of monopoly behavior. Since there is no
evidence in any other industry indicating that network effects have the
pernicious effects that have been so often attributed them, we believe that it
is unwise for network effects be granted any role in antitrust whatsoever.
Bibliography
Brynjolfsson Erik and Chris F. Kemerer, Network Externalities in Microcomputer Software: An Econometric Analysis of the Spreadsheet Market, MANAGEMENT SCIENCE, 42: (12) 1627-1647 DEC 1996.
David, Paul A. ''Clio and the Economics of QWERTY.'' American Economic Review 75 (May 1985): 332-37.
Farrell J., and G. Saloner. 1985. Standardization, compatibility, and innovation. Rand Journal 16:70–83.
Katz, M. L., and C. Shapiro. 1985. Network externalities, competition, and compatibility. American Economic Review 75:424–40.
Kobayashi, B., and L. Ribstein. 1996. Evolutions of spontaneous uniformity: Evidence from the evolution of the limited liability company. Economic Inquiry 34:464–83.
Liebowitz, S. J, and S. E. Margolis. 1990. The fable of the keys. Journal of Law and Economics 33:1–26.
¾¾¾. 1995. Path dependence, lock-in and history, Journal of Law, Economics, and Organization 11:205–26.
Liebowitz, S. J. and Margolis, S. E. Winners, Losers, and Microsoft: Competition and Antitrust in High Technology, Independent Institute, 1999, Oakland CA.
Lopatka, J. E., and W. Page. 1995. Microsoft, monopolization, and network externalities: Some uses and abuses of economic theory in antitrust decision making. The Antitrust Bulletin 40:317–69.
Reback, Gary, Susan Creighton, David Killam And Neil Nathanson, With The Assistance From Garth Saloner And W. Brian Arthur, "technological, Economic And Legal Perspectives Regarding Microsoft's Business Strategy In Light Of The Proposed Acquisition Of Intuit, Inc." Upside Magazine, February 1995.
Roe, M. J. 1996. Chaos and evolution in law and economics. Harvard Law Review 109:641–68.
Williamson, O. E. 1993a. Contested exchange versus the governance of contractual relations. Journal of Economic Perspectives 7:103–8.