1

1	UNITED STATES FEDERAL TRADE COMMISSION
2	and
3	UNITED STATES DEPARTMENT OF JUSTICE
4
5
6
7	SHERMAN ACT SECTION 2 JOINT HEARING
8	UNDERSTANDING SINGLE-FIRM BEHAVIOR:
9	MONOPOLY POWER SESSION
10	THURSDAY MARCH 8, 2007
11
12
13
14
15	HELD AT:
16	UNITED STATES FEDERAL TRADE COMMISSION
17	601 NEW JERSEY AVENUE, N.W.
18	WASHINGTON, D.C.
19	9:30 A.M. TO 4:30 P.M.
20
21
22
23
24	Reported and transcribed by:
25	Susanne Bergling, RMR-CLR

2

1	MODERATORS:
2	THOMAS J. KLOTZ
3	Attorney, Policy Studies
4	Office of the General Counsel, Federal Trade Commission
5	and
6	GREGORY J. WERDEN
7	Senior Economic Counsel
8	Antitrust Division, Department of Justice
9
10	PANELISTS:
11
12	Andrew Chin
13	Robert H. Lande
14	Richard Schmalensee
15	Alan H. Silberman
16	Michael A. Williams
17
18
19
20
21
22
23
24
25

3

1	C O N T E N T S
2
3	Introduction
4	Presentations:
5	Andrew Chin
6	Robert H. Lande
7	Richard Schmalensee
8	Alan H. Silberman
9	Michael A. Williams
10	Moderated Discussion
11	Conclusion
12
13
14
15
16
17
18
19
20
21
22
23
24
25

4

1	P R O C E E D I N G S
2	- - - - -
3	MR. KLOTZ: Good morning. I am Tom Klotz, an
4	attorney in the Office of General Counsel at the Federal
5	Trade Commission, and I am one of the moderators for
6	this morning. My co-moderator is Greg Werden, Senior
7	Economic Counsel at the Antitrust Division of the
8	Department of Justice.
9	Before we get into the substance of the program,
10	I want to go through a couple of preliminaries. First,
11	I want to thank our colleagues at the Department of
12	Justice for jointly presenting this program, and on
13	behalf of the Federal Trade Commission, I would like to
14	thank each of the panelists for agreeing to participate
15	with us today.
16	As I cover a couple of housekeeping matters, I
17	would ask first of all that you turn off any cell
18	phones, BlackBerries or other devices that would make
19	noise and that would interrupt our panel. Second, the
20	restrooms are outside the double doors. Just go across
21	the lobby, and there are signs that will help direct you
22	to the appropriate place.
23	Third, particularly for visitors, in the
24	unlikely event that the building alarms go off, we ask
25	that you please proceed calmly and quickly as

5

1	instructed. If we leave the building, we will go out
2	the exit on New Jersey Avenue, past the guard's desk,
3	and just follow the group of people across the street to
4	await further instructions.
5	Finally, given the format of the program, we ask
6	that you not make comments or ask questions during the
7	session, and we will proceed from there.
8	Yesterday, we began the program on monopoly
9	power and market definition, and today we are going to
10	continue that discussion, and at this point, I will turn
11	things over to Greg Werden.
12	DR. WERDEN: Thank you.
13	This is the last of our three sessions on
14	monopoly power. This session is focused in particular
15	on technology markets, with all the possible meanings of
16	that term, and single-brand markets. I want to join my
17	FTC colleague in thanking the panelists for appearing
18	here today and to thank the staffs of the two agencies
19	for doing quite a bit of work in organizing these
20	sessions.
21	These are sessions in a continuing process of
22	hearings that the Antitrust Division and the Federal
23	Trade Commission began last June on the law and policy
24	concerning single-firm conduct addressed under Section 2
25	of the Sherman Act. The materials from these hearings

6

1	are being made available on the agencies' web sites.
2	Submissions of panelists, their slides, and ultimately
3	transcripts, although they run a little behind, are
4	being made available. The sessions are being also
5	videotaped. I am not sure whether they will be
6	available for sale or not, but you might want to put
7	your orders in.
8	Our panelists today, in the order that they will
9	be speaking, are first Richard Schmalensee, who is the
10	John C. Head, III Dean and Professor of Economics and
11	Management at Sloan School at MIT. I am sure everybody
12	is very familiar with Dick's contributions to industrial
13	organization and antitrust policy, and he will speak
14	with particular experience from some work that he has
15	done in technology markets in recent decades.
16	Second, we have Mike Williams, director of ERS
17	Group, formerly, a long time ago, a colleague of mine at
18	the Antitrust Division at the Department of Justice.
19	If he arrives, we will then have third Andrew
20	Chin, Associate Professor of Law at the University of
21	North Carolina, who worked a little bit with Judge
22	Jackson on the Microsoft case, a little behind the
23	scenes, we learned about that recently.
24	Then Bob Lande, Venable Professor of Law at the
25	University of Baltimore School of law, frequent

7

1	commentator on antitrust policy issues and long ago with
2	the Federal Trade Commission.
3	And finally, Alan Silberman, a partner at
4	Sonnenschein Nath & Rosenthal, LLP, a long-time
5	practitioner of antitrust law who will be bringing the
6	practitioner perspective to these issues.
7	With that, I will add that we unreasonably
8	refuse to allow audience participation in any way, shape
9	or form, but we will allow people to submit written
10	comments for the record if they want.
11	I now turn it over to Dick Schmalensee.
12	DR. SCHMALENSEE: Okay, thanks, Greg, and thank
13	you for having me. This is a set of semi-disconnected
14	comments on markets that are experiencing or could be
15	experiencing rapid technological change.
16	Now, there are a number of basic features of
17	in these markets. Greg pointed out that occasionally
18	witnesses in these hearings go over well-known ground,
19	and I am going to do a little bit of that today, but I
20	think we do that to make sure everybody remembers that
21	this is well-known ground.
22	In markets with rapid technological change, you
23	expect to see market power because that is the reward to
24	innovation. So, you would be surprised in a market
25	where there is a lot of innovation going on if you did

8

1	not see some market power, because that is the return
2	for the investment. To find monopoly power, the issue
3	is typically durability of that market power. Is this
4	the blink of an eye in a Schumpeterian world, or is this
5	something that is likely to endure long enough to be an
6	issue?
7	Typically we address the issue of durability by
8	looking at entry barriers, but entry barriers usually
9	involve me-too entry, of a similar product. The hard
10	part -- and it is a hard part, though I am not making a
11	pitch that it is ubiquitous or inevitable is that in
12	markets with rapid technological change, entry may take
13	a rather different form than the incumbent's product
14	even if matching the incumbent's product is difficult.
15	So, in markets like that, when rapid technological
16	change is possible, the key to market performance is
17	competition to innovate, is competition on technology or
18	dynamic competition.
19	Unfortunately, I do not have any solutions to
20	this. This is a cautionary tale. If you ignore the
21	special features of these markets, you will tend to find
22	monopoly power where, in fact, it is relatively
23	transient. If you exaggerate those features, you will
24	tend to think it is transient when it is not. And there
25	are no bright lines that I can think of for reasons I

9

1	will discuss.
2	So, I am going to focus on three issues. The
3	first is, the difficulty of thinking about whether rapid
4	technological change is of the disruptive sort. Let me
5	be clear that technological change comes in various
6	flavors. If you think about microprocessors, there has
7	been enormous technical change, but nothing truly
8	disruptive for some time; very rapid increases in
9	performance, but incremental change; no one innovation
10	has radically disrupted things. Other markets have been
11	marked by rapid, disruptive change. Both pose problems,
12	and the tricky part is predicting whether disruptive
13	change is likely.
14	Then I want to talk about network effects
15	briefly. This is, I think, relatively well-understood
16	stuff. Finally, then I want to say a little bit about
17	something have been interested in for the last several
18	years: Two-sided businesses, which I do not think of as
19	two-sided markets. I will spend a little time on that.
20	So, if there is Schumpeterian competition,
21	competition for the market, the kind of competition that
22	in the Microsoft case we noted had occurred with some
23	regularity in the early years of PC software when
24	dominant products losted their positions, then short-run
25	market power is less of a concern. You still worry,

10

1	properly, about an incumbent's ability to use short-run
2	power to stifle that dynamic competition, but if
3	competition is healthy, the fact that a software product
4	sells for well over its marginal cost is not
5	problematic.
6	The problem is that that kind of competition
7	often comes in bursts. If you look at the automobile
8	industry early on, it is really quite extraordinary,
9	right? You had steam, you had electric, you had the
10	invention of the starter, you had the innovation of the
11	closed body, you had all kinds of things going on, and
12	then quiet. There is a great quote in Alfred P. Sloan's
13	book, My Years with General Motors, to the effect that
14	by the mid-1920s, the automobile and the industry were
15	set, and that is about right. Sloan was writing in the
16	late fifties.
17	You could argue that was an industry with rapid
18	technological change for a time and then it was not.
19	There was innovation after the 1920s: Engines got
20	better as did many other things, but nothing disruptive
21	happened. So, if you were trying to make policy in the
22	auto industry in 1910, you would have this question of
23	how long will this healthy dynamic competition continue,
24	and there would have been no easy answer.
25	It is also hard -- and this is troubling in

11

1	these markets -- by the nature of disruptive innovation
2	to predict its direction and source. Most of us, I
3	hope, can remember when the Walkman owned the carrying
4	around music business. It was wiped out not by somebody
5	who did anything with tape but by a very different
6	approach based on disk drives. The difficulty with
7	looking at who is spending what on innovation, which I
8	think is a useful thing to do, is that it may miss the
9	radical, the novel.
10	Now, again, this is a call for skepticism.
11	There are two possible errors. One is ignoring the
12	disruptive that is being developed over here in the next
13	room out of sight of the industry players, and the other
14	is reading my alma mater's alumni publication Technology
15	Review, too closely and becoming convinced that every
16	technology they talk about is going to come to market
17	tomorrow and disrupt its industry. Both are wrong, and
18	finding the truth is hard. Ignoring the potential for
19	disruptive innovation, however, gives you the bias of
20	assuming the status quo is forever.
21	In a number of markets marked by rapid
22	technological change, network effects can lead some
23	firms to high shares. If you have a snapshot in which
24	network effects have led to a dominant position, that
25	snapshot is consistent with a world of vigorous

12

1	Schumpeterian competition, in which the next hot product
2	may displace the leader. Think word processors in the
3	early days. WordStar dominates; WordPerfect comes along
4	and is better, and wham, WordPerfect owns the market.
5	Why? Network effects. So, a snapshot in which
6	WordPerfect owns the market is consistent with vigorous
7	Schumpeterian competition. It is also consistent with
8	its absence. So, just looking at the leader's share,
9	just looking at its apparent dominance, just looking at
10	the network effect, does not tell you whether there is
11	dynamic competition in the market. You have to look
12	beyond the snapshot.
13	One important thing that I would point out is
14	that network effects build large shares, build
15	apparently dominant positions, through expectations.
16	You can have a large share because everyone expects you
17	to have a large share. PCs wiped out Wang word
18	processors very quickly. WordPerfect took over from
19	WordStar very quickly, and Word took over from
20	WordPerfect very quickly. These things happened
21	rapidly, but -- and again, I will come back to my
22	cautionary note -- it is hard to predict the pace of
23	that kind of change.
24	There was discussion in the Microsoft trial of
25	software as a network-based service. This idea was in

13

1	the air then, it was being discussed by the engineers,
2	but it has taken a long time to happen. Could you know
3	it was going to take a long time to happen? Maybe;
4	maybe not. But that seemed to me to be a relevant
5	question. Google now has an online service offering
6	that may actually be serious. There has not been
7	anything terribly serious until now.
8	Finally, let me talk about multi-sided
9	businesses, my third topic. There are a whole set of
10	businesses that fit this two-sided market paradigm. If
11	you think of businesses that bring different customer
12	groups together, there are indirect network effects, and
13	the Coase theorem fails. This means that a wheat market
14	that brings buyers and sellers together really does not
15	quite do this if it is just buyers and sellers, because
16	you know that the price structure does not matter,
17	right? You can tax the buyer; you can tax the seller;
18	the end result is the same.
19	An important point here is that the term
20	"two-sided markets" is, a misnomer, because it is not
21	necessarily a characteristic of a market; it is a
22	characteristic of a business model. This is a strategy.
23	You could have some firms competing with two-sided
24	models with firms that do not. Two-sided medels apply,
25	as Rochet and Tirole pointed out, to a wide variety of

14

1	businesses: Obviously marriage brokers; media bring
2	eyeballs and advertisers; shopping malls bring customers
3	and stores.
4	In case of securities exchanges, one thinks of
5	the group as buyers and sellers, but, in fact, if you
6	look closely, it is providers and consumers of
7	liquidity. A number of exchanges have what are called
8	"maker-taker" models where, in fact, if you post a
9	standing order and somebody comes in and takes you up on
10	it, you are paid. So, it is a more complicated thing
11	than buyers and sellers. And payment cards, of course,
12	connect merchants and consumers.
13	This class of business strategies has become
14	more important recently because software platforms are
15	in a number of settings a natural way to build a
16	business like this. The Windows platform is an obvious
17	one. It links applications developers, not all of whom
18	work for Microsoft, and end users. The firm that has
19	the platform, Microsoft or Apple, needs to court its
20	developers, and its end users.
21	I want to make a few points about these business
22	models, based in part of a book David Evans and I have
23	coming out from the Harvard Business School Press this
24	spring. First, one of the surprising features is how
25	often in practice pricing is quite asymmetric; that is

15

1	to say, all the money is made from one of the groups.
2	Theory does not predict this.
3	In credit cards, if you pay on time and do not
4	have an annual fee, you do not pay anything to use a
5	credit card. The merchant pays. But, of course, for
6	any two-sided business, all the groups it deals with
7	need to be treated as customers, even if they are not
8	directly the source of profits.
9	One can have competition involving firms with
10	the same business model; that would be overlapping
11	platforms. One can have a platform competing with a
12	single-sided business, i.e., a business that targets
13	only one customer group, or one can have a competition
14	involving intersecting platforms that target only some
15	groups in common. This would happen if I target groups
16	A and B, and you target groups B and C. These potential
17	patterns of competition, complicate assessment of market
18	power.
19	The business in these cases is not just sales to
20	the profitable side. So, if you think about the
21	business that the credit card companies are in as sales
22	to merchants, you fundamentally misunderstand what is
23	going on. The money is directly made on the merchant
24	side, but, in fact, the consumer who carries the card is
25	just as important as the merchant that takes the card.

16

1	That is an obvious mistake one would not make in this
2	setting, but it is less obvious elsewhere.
3	Think about video game console makers. They
4	also have to court game developers, because if there are
5	not games for the consoles, the consoles do not sell.
6	So, they are in the business of dealing with both
7	groups, not just selling consoles. And, in fact,
8	consoles, as we know, are not the source of profit in
9	that business.
10	A two-sided business also has to worry about
11	competition from different business models. Satellite
12	radio is a single-sided business by and large. I mean,
13	it is not heavily advertising-dependent, yet it deals
14	with the same listeners that broadcast FM deals with.
15	Broadcast radio deals with those listeners with
16	two-sided models, advertisers and consumers; satellite
17	radio, consumers only.
18	Google and magazines compete for advertisers,
19	but they do it in different ways. Magazines use content
20	to assemble eyeballs; Google uses search to assemble
21	eyeballs or, better, to assemble focused eyeballs.
22	Craig's List has kind of wiped out newspaper want-ads;
23	it is again, a very different model.
24	The price-cost margin is pretty useless in
25	assessing the market power of two-sided businesses

17

1	because of asymmetric pricing, how do you compute the
2	price-cost margin? Think about a video game console
3	maker. Video game consoles are sold at a loss or at
4	break-even, depending on the maker and the year, but
5	that is not where the money comes from. The money comes
6	typically from sales of games you make yourself and
7	license fees from independent people like Electronic
8	Arts that make games to run on your console.
9	So, what is the price-cost margin? It is not
10	the loss on the consoles, and as to the royalties, there
11	is no cost or a very tiny cost associated with the
12	royalties you get from Electronic Arts. So, it is very
13	hard to figure out how to do a price-cost margin with
14	these businesses, and if you leap into some calculation,
15	it will likely be misleading.
16	As to market definition, the Guidelines approach
17	can be hard to adapt. The problem is multiple groups
18	and different models. In video games, the money is made
19	from the games. In contrast, in games that run on PCs,
20	the PC software platform vendor, does not make anything
21	from the game developers. So, games are not a source of
22	profits in the PC gaming, but they are the source of
23	profits for consoles. How do you think about a price
24	reduction or a price increase for purpose of market
25	definition -- which price?

18

1	Another problem is posed by feedback effects.
2	If you sell to A and B, you go through a hypothetical
3	price increase to A that reduces demand from A, but, of
4	course, if there are indirect network effects, that will
5	make the platform less attractive to B. There will be a
6	reduction of demand on the B side, which in turn will
7	make the platform less attractive to A, and so on.
8	Now, it is not hard to write down the
9	mathematics. It is just hard to think about how you
10	would do the calculation correctly in practice. The
11	existence of this sort of feedback effect does not mean
12	there is a death spiral with quantities driven to zero
13	-- things converge typically. The point is just that
14	you have to be very careful, and the typical Guidelines
15	approach is not well-suited to market definition in
16	these contexts, nor do we have data that lets us measure
17	those kinds of externalities.
18	Finally, and this is a cute feature of these
19	businesses, you must have both groups. The simplest
20	case is singles bars. For a heterosexual singles bar,
21	you really have to get both men and women in the door,
22	and if you have to spend a lot of money to persuade one
23	group or the other to come, it does not matter if you
24	have dominance, so to speak, on the other side.
25	Competition for the patronage of men or the patronage of

19

1	women, depending on the market, can eliminate profits.
2	So, you have to look at both sides, because again, the
3	key to these businesses is the need to balance, and the
4	need to balance means competition on either side can
5	dissipate profits.
6	Now, this is not obviously a presentation that
7	gives you answers, but I have tried at least to pose
8	some important questions. I wish I could be more
9	upbeat, but sometimes life is hard.
10	Thank you very much.
11	(Applause.)
12	DR. WERDEN: Mike Williams.
13	MR. WILLIAMS: Okay, thanks a lot, Greg.
14	So, I am going to talk about technology markets
15	in a different sense than Dick just talked about them.
16	I am going to talk about technology markets as they are
17	defined in the FTC and DOJ IP Guidelines, and those
18	technology markets are really literally markets for
19	ideas. So, they are markets for intellectual property.
20	They are not markets for widgets or even software. They
21	are markets for intellectual property.
22	I will start with just a few of the more
23	prominent cases. I think the main take-away from this
24	overview slide is some of the bigger cases is just that,
25	number one, there have been a number of them. Number

20

1	two, economists always get in trouble for making
2	predictions, but I think it is a fairly safe prediction
3	that there is going to be more, and probably
4	disproportionately more, as obviously intellectual
5	property is so critical to future markets.
6	Another quick take-away from this is that I have
7	put in quotes after each case what the technology was
8	that was being disputed, and I think another thing to
9	draw from this is that there are certainly a lot of
10	examples where the technology in question was
11	intellectual property for what we would traditionally
12	call high technology industries, but there is also
13	intellectual property for very mundane things.
14	For example, the DOJ versus American National
15	Can case, the laminated tube-making was -- at least in
16	part the intellectual property was the patents that
17	protected a certain way of making toothpaste tubes. So,
18	you can have intellectual property for high technology
19	things and intellectual property for very ordinary
20	things.
21	I will not spend a lot of time on this slide.
22	This is literally just the language right out of the IP
23	Guidelines. So, what is a technology market? It
24	consists of intellectual property that is licensed and
25	its close substitutes; that is, the technologies or

21

1	goods that are close enough substitutes significantly to
2	constrain the exercise of market power. So, the main
3	thing to take away there is certainly sort of the
4	primary intellectual property that we are thinking of is
5	generally patents, but you may have a circumstance where
6	other technology -- and by "other technology," it could
7	just be know-how, it does not necessarily have to be
8	patented -- and then goods. You can certainly imagine a
9	circumstance where there is an allegation that somebody
10	has market power over a certain kind of intellectual
11	property embodied in patents, but there may be a
12	physical product that is a good substitute for that
13	technology.
14	So, three general points that I just want to
15	touch on in this short talk. What are some of the
16	challenges that you face when you try to define the
17	markets? What are some of the challenges you face when
18	you try to assign market shares? And what are some of
19	the challenges you face when you try to determine
20	whether or not a firm has market or monopoly power in a
21	technology market?
22	So, the first thing to recognize is that these
23	are all derived demands. Nobody wants to license
24	intellectual property just for the heck of it. You want
25	to license it to do something with it, to make a product

22

1	that can then be sold. So, you can obviously, going
2	back all the way to the 19th Century, Alfred Marshall's
3	Four Laws of Derived Demand can help you organize your
4	thoughts about when a putative market for intellectual
5	property may or may not qualify in terms of actually
6	meeting the Horizontal Merger Guidelines test for an
7	actual antitrust market.
8	Again, it really boils down to, is the demand
9	for this intellectual property inelastic? Is it
10	inelastic enough that a hypothetical monopolist would
11	find it profitable to raise price? And I should mention
12	that the Intellectual Property Guidelines are quite
13	clear that even though the idea of a market for patents
14	or a market for intellectual property is a new
15	construct, the basic market definition methodology in
16	the Horizontal Merger Guidelines is still quite
17	applicable.
18	So, what are some of the practical problems you
19	face when you try to define a technology market in this
20	sense? One is that firms generally do not license their
21	patents one at a time. They will generally license
22	their entire portfolio. A portfolio generally has a lot
23	of complementary technologies within it. As I am sure
24	you are aware, a lot of big companies have hundreds if
25	not thousands of patents. The patents generally are

23

1	not -- I mean, you would be surprised if they were
2	substitutes, right? I mean, the whole point that they
3	are patenting different things, and they tend to be
4	complements, but they tend not to be sold one at a time.
5	Another way to think about it is, I have often
6	found a good way to organize your thoughts when you are
7	asking kind of what data are available, what do I have,
8	is to ask, what is the perfect data set? What would I
9	really like to have, and then what can I actually get?
10	So, if you said, "Well, what is the perfect data set for
11	thinking about technology markets," what you would
12	really like to see is each patent licensed separately so
13	you could look at the patents across portfolios,
14	across -- in other words, suppliers of intellectual
15	property -- and each patent licensed at an explicit
16	price.
17	So, you could use the royalty revenues, but in
18	most circumstances, we do not have either one of those
19	things. They generally get licensed in a bundle, in a
20	portfolio, that has substitutes and complements all
21	mixed together, and they generally do not have their
22	license revenues broken out certainly by patent or even
23	in many circumstances -- I will get to this in a
24	minute -- in many circumstances, no money changes hands,
25	because many companies do these in royalty-free

24

1	exchanges. So, those are challenges that you face when
2	you try to think about how to define these markets.
3	Assuming that you have managed to define a
4	technology market in this sense, now we face the
5	challenge of assigning market shares. So, you are in a
6	world where, I guess the first thing to say is, what is
7	the principle? What is it we are trying to accomplish
8	when we assign market shares? Going back to the
9	Horizontal Merger Guidelines, the answer, of course, is
10	we are looking for a statistic that gives us the best
11	indicator of a firm's future competitive significance.
12	That is what a market share is supposed to tell us.
13	So, I mentioned earlier that you do not have
14	royalty payments generally, so what are the normal ways
15	in which we would think about assigning market shares?
16	You might do it on the basis of output, you might do it
17	on the basis of revenues, sales and so on, but most of
18	the time we do not have royalty payments, because, for
19	example, like cross-licensing, we do not have the
20	ability to disentangle all of the IP within a portfolio
21	because they were packaged as a portfolio and sold as a
22	portfolio.
23	Of course, unfortunately, the whole notion of a
24	capacity or a shipment does not make any sense in this
25	context. There is no capacity constraint to an idea.

25

1	So, those are challenges.
2	So, what have people done to try and assign
3	market shares in technology markets? I think there is
4	basically two approaches that have been offered. One is
5	sort of what your Bayesian priority would be if you had
6	a really diffuse knowledge, which would just be I really
7	am not sure what to do, I am just going to say it is
8	1/N. Now, I say that is an advantage because it is
9	simple to compute, because that is conditional on
10	agreeing what N is, and, of course, reasonable people in
11	any particular case might have fundamental disagreements
12	about what N is, because again, think about N can be
13	patents, it can be just know-how, and it can be physical
14	products that arguably compete in the same technology
15	market.
16	When would 1/N be a good statistic? When would
17	it tell you the likely future competitive significance
18	of a given firm in a technology market, the answer would
19	be -- and this quoted out of the IP Guidelines -- is
20	does 1/N give you a good estimate for the ability of
21	firms to produce close substitutes at comparable costs?
22	So, another way to say it is, suppose for the
23	sake of argument we had four different patent
24	portfolios, four different providers of intellectual
25	property. If each of those patent portfolios provided

26

1	the downstream manufacturers that were actually going to
2	bend the metal and make a product with the intellectual
3	property, do each of those four patent portfolios give
4	the downstream manufacturers the ability to produce
5	close substitutes at comparable costs?
6	If you thought that was right, then 1/N probably
7	would be a good statistic, because you are saying that
8	each of those four patent portfolios is reasonably equal
9	in terms of what their probable future competitive
10	significance is, because they all seem to be about
11	equally valuable in the sense that if they were
12	purchased by one of these downstream manufacturers, the
13	downstream manufacturer, arguably in this hypothetical,
14	would be somewhat indifferent between which of the four
15	patent portfolios it used, because each of them, by
16	hypothesis, is reasonably good at enabling the
17	downstream manufacturer to produce close substitutes at
18	comparable costs.
19	There are some disadvantages to the 1/N method,
20	namely, the flip side, which is, what if the four patent
21	portfolios are not equally valuable to the downstream
22	manufacturers? Of course, that is -- at least that is
23	what my prior is, is that these patent portfolios are
24	very heterogenous animals. You know, one firm has got
25	200 patents; one has got one. Of course, in principle,

27

1	the one patent could be more valuable than the 200
2	patents, you just do not know, but you would be
3	surprised if each of the four patent portfolios in my
4	simple little example were equally valuable to the
5	downstream firms.
6	I mean, I think going into it, at least my prior
7	is it is more likely that they are highly differentiated
8	in terms of their fundamental value to downstream firms
9	in terms of making the products that can then be sold.
10	So, the patent portfolios are highly differentiated.
11	Another aspect that comes up in this is that if
12	you think about the IP suppliers, there is actually two
13	things that they do. They provide ideas, they provide
14	patented technology, but they also work with the firms
15	that bend the metal, and so if you think, for example,
16	about firms that license technology to make memory
17	chips, for example, they license the idea, but they also
18	work closely with the companies that try to actually
19	make the computer chips, because if you think about it,
20	they are the ones who in some sense know more about how
21	the product is supposed to work.
22	Now, they may not have the same engineering
23	expertise that the downstream manufacturer has, but a
24	complementary service that they are offering is, how do
25	you actually implement my idea? Of course, the IP

28

1	suppliers could differ quite generally in their ability
2	to work with the downstream manufacturers; their ability
3	to actually get their ideas implemented. So, even
4	though you might have four equally valuable patent
5	portfolios, one of the firms might be much better at
6	working with the downstream firms to turn their ideas
7	into real products.
8	The last bullet, I will not really go over, it
9	frankly, it just takes too long to explain, and
10	colleague of mine and I have -- Ashish Nayyar -- an
11	article that is just devoted to that particular subject,
12	but I do not have time to get into that just now. So,
13	1/N is one approach.
14	A second approach is to say I am going to look
15	at in some sense how manufacturers have voted with their
16	dollars. In other words, if I cannot directly observe
17	and assign market shares based because I do not have
18	royalties, the patents are not licensed individually, I
19	am going to look at how manufacturers have voted with
20	their dollars to pick amongst, for example, these four
21	patent portfolios.
22	If I look at what the manufacturers have picked,
23	who has been successful in the marketplace? Has one
24	manufacturer been much more successful than the other
25	manufacturers because it used firm one's patent

29

1	portfolio instead of firm two's? So, if you think about
2	it, that is kind of the mirror image of what we are
3	trying to observe, that is kind of the mirror image of
4	how that technology has played out in the marketplace.
5	Has one technology proven, based on the choices of
6	manufacturers and ultimately the choices of consumers,
7	to be more valuable than another set of technology?
8	So, an advantage to that is that it arguably
9	captures the differentiated nature of the portfolios,
10	because one will probably be better than another, but as
11	with all these things, there is some disadvantages to
12	it. Suppose you have -- and this is common -- suppose
13	you have a manufacturer deciding that he needs to
14	license technology from two of the intellectual property
15	providers. Well, now, how are you going to assign
16	shares now? You have got two of the four, in my
17	example, patent portfolio providers. Both of their
18	technologies are being purchased by one manufacturing
19	firm to produce one product. Well, now you have a
20	problem. How are you going to sign, using this kind of
21	mirror image approach, how are you going to assign those
22	sales to one of the two patent portfolio providers or to
23	the firms competing in the technology market?
24	Finally we get to really the last question,
25	which is how are we going to measure monopoly power in a

30

1	technology market? As with most instances in antitrust
2	economics, there is kind of two ways to think about
3	monopoly power or how we would investigate monopoly
4	power. One is structural, and one is performance.
5	So, from a structural perspective, remember, by
6	this point we have defined a market as best we could, we
7	have assigned shares as best we could, given all these
8	problems that I have talked about, and you are going to
9	get some measure of market concentration. Now, it might
10	be an interesting statistic, you might view it with a
11	lot of skepticism, but you will have some measure of
12	market concentration, and then you would look at, again,
13	kind of a traditional factor, barriers to entry.
14	Now, the barriers to entry tend to take kind of
15	a different nature in a technology market. There is
16	different kinds of things that firms have to do, invent
17	around the IP, defend against patent infringement
18	claims. If you are an entrant into a technology market,
19	one of the things you might well have to do is indemnify
20	people buying your technology against patent
21	infringement claims from, say, an incumbent provider of
22	technology. So, that gives you kind of a structural way
23	to think about how one might study the existence of
24	monopoly power in technology markets.
25	Then finally, a different way to think about it

31

1	is, can I study the performance of these markets and
2	gain any insight as to whether or not these firms or one
3	firm seems to have monopoly power? I think in some
4	circumstances it might be possible to look at changes in
5	royalty rates. I wrote in the parenthetical, "assume
6	marginal costs are not possible to measure but
7	constant." So, it is very difficult to know what the
8	marginal cost of a patent is.
9	I mean, in one sense, on a forward-looking
10	basis, really the marginal cost of a patent is the cost
11	of enforcing it, because the costs of coming up with it
12	are all sunk, so we may not know what the marginal costs
13	are, but if we are willing to make perhaps a rogue
14	assumption that those costs are constant, then changes
15	or increases in royalty rates might be informative.
16	Then finally, there are certain circumstances
17	where IP gets licensed with what are called tie-ins or
18	tie-outs or in some circumstances -- and this falls back
19	to a bit more traditional perspective -- if you are
20	familiar with, for example, the patent misuse law,
21	patent misuse occurs when a firm has arguably expanded
22	the temporal or the product aspect of what they are
23	trying to enforce beyond the four square corners of the
24	patent. So, sometimes firms will actually ask for, when
25	they are licensing their IP, they will ask for long-term

32

1	contracts that exceed the length of the patent life, and
2	so that arguably is a performance indication that maybe
3	this firm does have some substantial market or monopoly
4	power.
5	So, thank you very much.
6	(Applause.)
7	DR. WERDEN: Andrew Chin.
8	DR. CHIN: Thank you. Here is a picture from
9	the last time I saw Dean Schmalensee in the Microsoft
10	case.
11	My name is Andrew Chin. My web site is
12	andrewchin.com. You can get two of my recent articles I
13	will be talking about on that web site, recently
14	published, and the title of my talk is Defining Software
15	Product Markets.
16	There is time for just one main point, and that
17	is that relevant software product markets can be
18	correctly delineated using the existing techniques that
19	are described in the Merger Guidelines. By "correctly,"
20	I mean that the resulting market that you find is
21	appropriate, is an appropriate subject for antitrust
22	concern.
23	There is one tricky aspect to this, and that is
24	what I am focusing on today, is that the key to doing
25	this correctly is describing software products

33

1	accurately and at the right level of abstraction to
2	perform the analysis, because here is what can happen if
3	you get it wrong.
4	The conclusions of law of the District Court in
5	Microsoft grounded the liability for attempted
6	monopolization in a market for "platform level browsing
7	software for Windows." On appeal, the D.C. Circuit
8	found this description of the market to be varying and
9	imprecise and as a consequence reversed the attempted
10	monopolization liability and remanded the tying claim
11	for a rule of reason analysis under which the plaintiff
12	would have one hand tied behind their back. They would
13	be barred from more careful approaches to market
14	definition.
15	The approach of defining the browser software
16	product market in this way, though, was doomed to
17	failure because it defined the software product as "code
18	and nothing else," as essentially adopting the position
19	taken by Microsoft throughout the trial, that a software
20	product consists of code and nothing else.
21	Consider whether Microsoft would have taken the
22	same litigation position in a copyright infringement
23	suit. Had I purchased Office XP and made several copies
24	and sold those, put them on eBay, I doubt that a defense
25	that I had bought the code and therefore could do

34

1	anything I wanted with it would avail me very much in a
2	copyright infringement suit. So, the absurdity of that
3	position percolates throughout the D.C. Circuit's tying
4	analyses, both in the consent decree case and in the
5	appeals decision. I have argued in my Wake Forest Law
6	Review piece that throughout the D.C. Circuit's
7	analysis, it relies on this fallacy, and then go into
8	some of the consequences of relying on that fallacy in
9	that article.
10	Well, another approach was available to the D.C.
11	Circuit and to the District Court in the conclusions of
12	law, and that was kind of buried in the findings of
13	fact, but there was a discussion of a "market for web
14	browsing functionality," essentially defining the web
15	browser software product in terms of what it does. It
16	enables a user to browse the web; in short, to select,
17	retrieve and perceive web resources.
18	The conclusions of law did not cite this
19	finding. The D.C. Circuit followed suit and did not
20	cite it either but said as to the combined opinions of
21	the District Court that it failed to enter "detailed
22	findings defining what a browser is or what products
23	might constitute substitutes."
24	From that I take two points: One, that
25	antitrust analysis requires description in detailed

35

1	terms as to what a software product is and in explicit
2	terms. Tell us what it is, not what it does. Well, at
3	one level of abstraction, a fairly high level, you can
4	just define what it is as the set of legal rights and
5	technological capabilities that enable a user to select,
6	retrieve and perceive web resources. You get two clues
7	as to what those rights and capabilities are, and they
8	come in the box.
9	They come in the box in the form of software
10	code on some tangible medium, such as a CD-ROM, and
11	accompanying documentation. Microsoft holds the
12	copyright on both the code on the medium and on the
13	documentation, so you do not own those, but the legal
14	rights and technological capabilities are defined by
15	reference to those accompaniments.
16	More detail is available but entirely
17	unnecessary; however, they are available. I describe
18	them fully in my Harvard Journal on Technology piece to
19	give comfort to those who may not be convinced that
20	these are well-defined concepts, and also, to address
21	the misconception that arises from viewing these
22	products as code that, for example, these are integrated
23	by virtue of being supported by the same body of code.
24	So, this addresses the product integration rhetoric that
25	came throughout the case.

36

1	Now, so, why do we not need that level of
2	detail? Because all that antitrust analysis requires is
3	in the language of Dupont, is first to identify
4	reasonably interchangeable software products from the
5	user perspective for performing the same purposes or
6	supporting the same user purposes. So, here is an
7	example. Here is an example of two products that
8	support the same user purpose at some level of
9	abstraction.
10	Converting binary to BCD. For those of you with
11	patent law backgrounds, this is the algorithm that was
12	found to be non-patentable in Gotshall versus Benson by
13	the Supreme Court. So, it is an historically
14	interesting example. You do not need to know what BCD
15	is, but this is a DOS program that will take a base 2
16	number and convert it to BCD.
17	Another way of doing this is create a Windows
18	application, a calculator with a bin-to-BCD button on
19	it. You type in the number, you click the button, and
20	it performs the same calculation. At some level we know
21	that these two applications serve the same user purpose.
22	So, if we run through the Merger Guidelines
23	analysis, we can look on the demand substitution side,
24	we see they are functionally interchangeable insofar as
25	they support the same user purpose; however, if we dig

37

1	deeper, they run on different code. How important is
2	that? Well, maybe if the user notices that one set of
3	code runs more slowly than the other, that might factor
4	into their preferences. The different user interfaces,
5	one might appeal more to some sets of consumers than
6	others. They run on different operating systems. So,
7	there is different platform preconditions for both
8	pieces of software, both software programs to operate,
9	but there is high overlap. Basically all modern Windows
10	applications have a DOS shell that you can go out to and
11	run the DOS program with. So, there is a high overlap,
12	but all of these can factor into the reasonable
13	substitutability or reasonable interchangeability
14	calculus.
15	Then on the supply side, you can identify
16	structural barriers to entry. For example, if a firm
17	with market power controls some of the preconditions for
18	either of these programs to operate.
19	But what we might need more structure on -- all
20	of these inquiries are fairly familiar, and whether you
21	are analyzing flexible wrapping materials or software
22	products, these are familiar modes of analysis to us
23	except possibly for the user purpose. How do you define
24	the user purpose for which a software product is used?
25	What is the appropriate level of abstraction?

38

1	Well, software engineering provides us a tool
2	for identifying the user purpose for a software product
3	at what I believe is the right level of abstraction.
4	So, if you look at this, this is called the essential
5	use case, and this is a way of describing the
6	functionality of a software product in terms of what the
7	user intends the system to do and how the system
8	responds to that intention. Does it meet its
9	responsibilities?
10	So, there are many ways of describing a web
11	browser. You could operate it, you could select items
12	with a mouse, you could use a trackball, you could use
13	voice. At this level of abstraction, those design
14	choices do not matter. The code that supports those
15	designs and implementations do not matter. All that
16	matters is what from the user's point of view is the
17	purpose supported. The precondition matters, and the
18	user intention system responsibilities matter. So, that
19	is the appropriate level of abstraction.
20	So, what I argue is that the box containing the
21	software and documentation, this Windows 98 item that
22	Microsoft markets, competes in at least two relevant
23	product markets, and both of the relevant product
24	markets that were described in the tying analysis, and
25	those are technically end use segments, one of which is

39

1	providing platform software that can be pre-installed to