United States. Congress. Senate. Committee on the.

The Industrial reorganization act. Hearings, Ninety-third Congress, first session [-Ninety-fourth Congress, first session], on S. 1167 (Volume pt. 7) online

. (page 99 of 140)
Online LibraryUnited States. Congress. Senate. Committee on theThe Industrial reorganization act. Hearings, Ninety-third Congress, first session [-Ninety-fourth Congress, first session], on S. 1167 (Volume pt. 7) → online text (page 99 of 140)
Font size
QR-code for this ebook


"Barriers to New Competition" (Cambridge: Harvard University Press, 1956).



5683

The second problem is that of data. The data necessary is closely held by the
coi-porations involved. Most of the studies that have been done have used Census
Bureau data which is aggregated and classified in such a way as to protect the
confidentiality of the companies. However, the aggregation also clouds the
economic significance of the statistics.

The third problem is that the structure-conduct-performance links can only
be expected to be true in a statisitical sense and not in every individual case.
Structure is an important determinant of pricing policy but the industry'^
history and product peculiarities also play a role. Consequently, even with per-
fect quantification of the variables and perfect data, we should still expect a
substantial amount of unexplained variance in statistical studies of structure-
conduct-performance links.

Because of data and classification problems, some of the best research has
used simple classification schemes rather than formal statistical methods to
illustrate the relationships. Very significant results were found in Michael
Mann's study of the average rates of return after taxes on the stockholder's
equity for thirty industries in the years 1950-1960. The result are shown in
Table 1.

TABLE 1— PROFIT RATES BY CONCENTRATION AND BARRIERS TO ENTRY

Very high Substantial Moderate

barriers barriers to low barriers

High concentration 16-4(8) 11-1(8) 11-9(5)

Moderate concentration (0) 12-2(1) 8-6(8)

Note: The number in parentheses is the number of industries in each category.

Source: H. Michael Mann, "Seller, Concentration, Barriers to Entry, and Rates of Return in 30 Industries, 1950-60,"
Review of Economics and Statistics. XLVIII (August 1966), 296-307.

A measure of performance for the industries is the deviation of the profit rates
from the average profit rates in the economy. Short term fluctuations in profit
rates are signals to investors to increase or decrease investment in certain in-
dustries, but long lasting deviations from normal profit levels are indications of
poor perfoi-mance. Prices in industries with abnormally high profit levels are
higher than the price necessary to bring forth the needed production, and con-
sequently the economy is not allocating scarce resources among competing proj-
ects as efficiently as possible. Also, a sustained high profit rate causes a redis-
tribution of income from consumers to stockholders compared to what would
occur under competitive conditions.

The average profit rate after tax on stockholder equity for all U.S. manufactur-
ing corporations for the years 1946-1965 was 9.0%."^ Using this figure as a cost of
capital, the Mann figures show that industries with high concentration and very
high barriers to entry bad returns 82% above what was required, an indicator
of poor performance. Those with substantial entry barriers had profit rates 24%
above the normal, and those with moderate to low entry barriers had profit rates
24% above the normal, and those with moderate to low entry barriers had profit
rates approximately equal to what was required. Although the Mann study shows
the general magnitude of profit variations with concentration and barriers to
entry, it could not identify the separate effects of concentration and barriers to
entry because no industries were found with very high barriers to entry and
moderate or low concentration.

Many studies have been done using multiple regression techniques to relate
various measures of strructure to profit rates. Among the best of these is one
done by Comanor and Wilson using forty-one consumer goods industries.* Struc-
tural measures used for explanatory variables included the advertising/sales
ratio and advertising per firm (representing product differentiation or brand
loyalty as a barrier to entry ) , economies of scale, capital requirements, and two
measures of concentration. The most significant explanatory variables (statisti-
cally significant at the 1% confidence level) were the advertising/sales ratio
and capital requirements, both measures of barriers to entry. Although tlie
Comanor and Wilson equations showed an increase in profit rates from an in-

3 Lawrence J. White, "The Automobile Industry Since 1945" (Cambridge: Harvard
University Press. 1971). p. 251.

•'William S. Comanor and Thomas A. Wilson. "Advertising: Market Strnctiire and Per-
formance, Review of Economics and Statistics XLIX (Nov. 1967), pp. 423-440.



5684

crease in concentration without clianging barriers to entry, the rate of increase
calcuhited was not statistically significant. The difficulty in measuring the sepa-
rate effects of concentration and barriers to entry arises because the two vari-
ables tend to rise and fall together, so that too few observations are available
on their separate influences to estimate them with confidence. The Comanor and
AN'ilson etiuations explained 46% of the variance in profit rates for the in-
dustries studied.

Although the separate influences of concentration and barriers to entry cannot
be estimated statistically, they can be identified through economic reasoning.
A highly concentrated industry can coordinate its pricing actions relatively
easily to raise prices above the competitive level. However, if barriers to entry
•are low or nonexistant, the high profits made will be eliminated as new firms
enter the industry. The potential comitetitors keep prices down almost as effective-
ly as actual competitors in the industry. The potential competitors are not totally
effective because some difficulties always exist for new entrants and there is a
time lag between the time a corporation decides to enter an industry and the
time it achieves full production. Consequently, we expect to see higher profits in
highly concentrated industries than non-concentrated ones of the same barriers to
entry, but the profits can never rise very far above the competitive level over a
long period of time in the absence of barriers to entry. An industry with high
barriers to entry and very low concentration also could not raise profits very
far above the competitive level because of the difficulties of coordinating prices
and avoiding price wars. However, no such industries exist or are likely to exist.
If entry is blockaded, there is much more incentive for the strongest firms to
expand their market shares than if entry is free, which eventually leads to con-
centrated industries.

The above conclusions suggest that barriers to entry rather than concentration
should be the primary concern of antitrust policy. If an industry's entry barriers
can be reduced, it will evolve toward competitive performance naturally without
the necessity for governmental intervention to reduce concentration. Conversely,
if firms are split to reduce concentration but entry barriers remain high, the
industry will naturally evolve back to monopolistic behavior. Of course, the best
way to restore competitive behavior is to reduce both concentration and barriers
to entry, but in many cases direct deconcentration could lead to disruption and
inefficiency for both the manufacturer and the customer.

Because the statistical studies do not aceoinit for all the variance in profit
rates, it is necessary to make a complete study of the computer industry to de-
termine how closely it conforms to the expected patterns in order to determine
what reorganization is necessary. It is also necessary to study the industry for
information on how structural change would affect technological progress. Sta-
tistical results on structure-progress links are extremely poor and sometimes
contradictory, a result of the pro))lems of quantifying progi-ess and the oppor-
tunities for progress. And finally, it is necessary to examine the industry to de-
termine the source and extent of barriers to entry, and to determine what could
be done to reduce those entry barriers.

II. COMPUTER INDUSTRY CHAKACTERISTICS

This section summarizes the structure, conduct, and performance of the com-
puter industry, emphasizing the aspects of the industry that are relevant to the
protilems of reorganization. It is based upon information developed in an extensive
study of the industry by the author .°

Before defining the market shares held by various firms in the industry, it is
necessary to define the industry. There is no definitive economic answer to the
industry definition question. An industry should include all products which are
close substitutes:! for each other and exclude products which are not good sub-
stitutes for others in the group. Because of tlie varying degrees of substitxition
amnng different products, it is necessary to choose definitions that relate to the
specific problem wtih which one is concerned. A meaningful definition of the
industry which is suitable for this study is general purpose computers and their
as.sociated perpheral equipment. This definition excludes analog computers, spe-
cial purpose computers, and data entry devices such as keypunch machines and
terminals. The logic behind this definition is that we should include computer

^ For expansion and clarification of the points mentioned here see Gerald Brock, "The
Fnited States Computer Industry 1956-1973," (unpublished Ph. D. dissertation, Harvard
University, 1073).



73.1 ....




21.3
20.8


98.3


71.2 .__.




97.3


70.7 ....




20.0


96.0


70.4 ....




11.6
8.3
9.U


87.5


74.5




89.4


72.5


75.9


88.5


66.7


74.8


9.2


85.2


69.7


74.2


7.8


87.4


74.3


74.3


6.3


88.9


74.6


73.3


5.6


88.2


70.6 ....




7.3
7.7


86.1


G7.4 ....




87.1



5685

systems which are generally in competition with each other and also the equip-
ment which must be used with a particular CPU design or set of interface
specifications.

TABLE 2.— MARKET SHARES
I Percent]

Next
Year IBM IBM (internal) largest firm Top 4 firms

1956

1958 -

I960..

1962

1963

1964

1965

1966

1967

1968

1970

1971

Source: IBM internal figures taken from ISM Quarterly Product Line Assessment, November 1968, p. 13, Other figures
calculated by author from data on computer installations in the Diebold Automatic Data Processing Newsletter and the
"Monthly Computer Census" of Computers and Autonution.

Table 2 shows the market shares of IBM, the next largest company, and the
top four companies together for various years. All figures are based on installed
machines during a given year, rather than new sales for that year. Because
computers are a capital good, the proper measure of market share is the stock of
capital installed rather than the changes to that stock. So long as market shares
remain approximately constant, the stock and sales market share figures will be
approximately the same. The major computer companies consider their installa-
tion figures proprietary ; consequently the calculations are based on estimates by
industry authorities of the installations of each model of computer in each year.
Confirmation for the general range of the market share figures is provided by
IBM's own estimates of its market share for "systems and peripherals" as shown
in the second column.

An analysis of the calculated market share figures shows that IBM's market
share generally rises with the introduction of a new product line and falls at the
end of the line, but has averaged around 707c. IBM's share hit its high points in
1963 (at the peak of second generation installations such as the 1-101 and 7690)
and 1968 (at the peak of third generation machines, primarily the System/.360).
Low points were reached in 1965 and 1971. The IBM internal figures confirm the
level of IBM's market share but not the pattern of market share changes. While
the calculated figures indicate a rise in IBM sliare from 72.5'/<: in 1964 to 74.6%
in 1968, the IBM internal figures show a drop in share from 75.9% to 73.3% over
the same time period.

Through 1968, the second largest firm by dollar value of installed machines was
the Univac division of Sperry Rand Corporation. Univac's market share dropped
steadily from 21.3% in 1956 to 5.6% in 1968. After 1962, no company other than
IBM held over 10% of the market in any year. The market ])attern became one
dominant firm (IBM with 70% of the market), .seven major competitors (Univac,
Honeywell, Control Data, G.E., Burroughs, N.C.R., and R.C.A. with two to eight
per cent apiece), and a number of minor firms with less than 1% of the market
each. Concentration in tlie industry as expressed Ijy the combined market share
of the largest four firms declined steadily as Univac lost ground and firms below
the top four gained share. Measured by the four firm concentration ratio, the
computer industry is less concentrated than several other industries such as
automobiles, primary aluminum, and flat glass, but no other non-regulated major
industry is so dominated by one firm as the computer industry.

Barriers to entry in the computer industry can be separated into economies of
scale, product differentiation, and capital costs. Economies of scale form a barrier
to entry because a new entrant must either suffer liigher costs than the estab-
lished companies or enter at a sufiiciently large scale to take advantage of the
economies of scale, with a consequent probability of causing a decline in industry
price either through excess production or retaliation by the other firms. To study
economies of scale, the production of computer systems can bo divided into three



5686

categories, mauufacturing, sales aud maiuteuauce, and design and software, each
with different cliaracteristics.

Mauufacturing economies of scale are negligible. A study by IBM of its costs
of manufacturing peripheral equipment versus those of competitors concluded :
"OEM costs are strikingly more similar than dififerent from IB:M" and "OEM can
match our manufacturing costs''.^ The same study reported that IBM had a 15%
cost advantage as a result of its circuit production. However, the advantage did
not extend to all circuits. A report to IBM's Management Review Committee
stated : "in certain areas like T'L in which costs are relatively independent of
volume, competition is estimated to have as much as 30% cost advantage." ' On
the Aspen tape drive (IBM 2420-5), IBM's studies showed its direct manufactur-
ing costs as $4507 compared with $4950 for competitive versions of the product.
The $443 difference represents an advantage for IBM of 9.S% of the direct cost
or 1.8% of the $25,200 selling price. On the 2314 disc subsystem, IBM's direct
costs were given as $24,737 compared with competitor's $30,660, a $5923 difference
representing a 24% difference on the direct costs and 2.5% difference on the
$237,105 purchase price.'' Because the independents were producing at very small
scales and IBM was producing at a very large scale, the IBM cost advantages
indicate moderate economies of scale. However, part of the difference is accounted
for by the superior performance specifications of the independent product reduc-
ing the significance of economies of scale in manufacturing to an unimportant
baiTier to entry.

Sales and maintenance are classified together because they have similar econ-
omies of scale characteristics. Both activities are performed by labor with very
little capital equipment. Consequently, there are no economies of scale of the
traditional type arising from the indivisibility of high volume capital machinery.
However, there is some advantage to the larger firm because of a reduction in
travel time, increasing the productivity of salesmen and maintenance engineers
and reducing the customer wait time for service. Because of the critical nature
of computer equipment availability, a small company may find it necessary to
hire an inefficiently large number of maintenance men to be sure one is on duty
near enough to each installation to respond to a problem with the necessary speed.
The economies of scale due to travel time are minor in densely populated metro-
politan areas, but can become a significant problem for a new company trying to
expand into small and medium size cities and towns. As a result, sales of IBM's
smaller competitors have been largely limited to the major cities.

Design and software both are characterized by high cost for the first unit
produced and practically zero cost for additional units. Consequently, the unit
cost of production drops with each additional unit produced, making it difficult
for a new firm to compete with a large estahlished firm. Design and software to-
gether account for 20% of the total cost of producing a computer system according
to a survey of computer manufacturers." The effective marginal cost is sub-
stantially greater than zero in software and design because of the difficulty of
designing machines or writing software for a large market. The market is not
made up of homogeneous users causing software to incur a marginal cost from the
extra work required to make it general enough to serve a variety of needs. Gen-
eral software also requires more computer time in overhead than software
specialized for the individual user or similar group of users. In some cases this
overhead cost is great enough to justify the writing of modifications or additions
to the manufacturer supplied operating system to be paid for strictly by the
savings in computer time. Although data is not available to adequately evaluate
the net advantage to the largest firm from software economies of scale, it appears
that IBM's major competitors have been able to produce adequate software with-
out prohibitive costs, suggesting that the software advantages of the largest firm
are not insurmoimtable.

Two factors account for the existence of product differentiation or brand
loyalty as a barrier to entry : the difficulty of making rational computer selections
and the cost to the customer of switching among computer manufacturers. Selec-

Plahitlff's Exhibit 25. p. 25. in Telex vs. IBM. Case No. 72-C-18 and 72-C-89 in U.S.
District Court for the Northern District of Oklahoma, later reference cited as Teles vs.
IBM.

' IBM, "Manajrement Committoe to Management Review Committee Report," October 25,
1971. Plaintiff's Exhibit 391A-142 in Telex vs. IBM.

8 IBM, H. E. Cooley Presentation to G. B. Beitzel, June 24, 1970, Plaintlflf's Exhibit 40
in Telex vs. IBM.

» Software was reported as 10% of cost and engineering effort as 4%. See Mohammed K.
Hamid, "Price and Output Decisions in the Computer Industry," Ph. D. dissertation, Uni-
versity of Iowa, 1966. (Ann Arbor: University Microfilms, Inc.) p. 203.



5687

tion is difficult because of the multi-faceted nature of computer performance as
well as uncertainty about future enhancements or the performance of announced
but undelivered machines. Computer performance depends upon a variety of dif-
ferent types of internal CrU operations, the speed and characteristics of several
kinds of input-output equipment, and upon the performance of the operating
system and language compilers. The relative importance of the various factors
depends upon the actual job being considered. Consequently, there is no single
measure of computer performance that can serve as a valid ranking of the rela-
tive performance for all computers except in relation to a single well-defined
computing task. The only way to get an exact performance rating for potential
computers on a given customer's applications is to actually program all of his
applications for each computer and run them under production conditions. The
programming effort involved makes this prohibitively expensive for most cus-
tomers. It is impossible if at least one of the potential computers is not yet in
production status or is not available for test runs, a common situation. Decisions
for most customers are based upon guesses or estimates about the computer's
performance. The methods include rough estimates from the basic performance
parameters of the machine (memory access time, tape si)eed, etc.), published
comparisons using typical jobs, simulation studies of performance, and partial
benchmarking by programming one or more common jobs of the user.

Because of the impossibility of easily comparing the performance of various
computers, managers responsible for computer selection show considerable dis-
agreement as to the criteria thnt should be used. E. G. Schuster gathered recom-
mendations for computer selection in a hypothetical case from 498 computer
managers. Each person in the study was supplied with information about the
potential systems, including the price and a presumed certain figure for the
amount of time each system required to perform the user's jobs, eliminating the
performance uncertainty discussed above. Thus it was possible to make a purely
economic decision for this simplified case.

In analysing the results, Schuster found that the current manufacturer was
chosen 26 out of 27 times if other factors were equal. If a competitor offered a
20% price discount with equal performance, the current manufacturer was still
selected one out of three times. A second generation computer was selected over
a third generation one only 17% of the time even when it offered 30% better
performance than the current vendor and 10% better than the other competitors
with third generation systems. This indicates that computer managers consider
being up to date in technology to be significant for its own sake rather than just
because newer systems generally either have better performance or are less
expensive than older systems. The selections also indicated a preference for
extensive support personnel to be supplied by the manufacturer. An increase in
support covered by an additional price increase, with an effective price higher
than either the present vendor or another competitor, caused the high support
company to be chosen 38% of the time, the present vendor 58%, and other
competitor 4%.^°

The most significant finding of the Schuster study was that managers dis-
agree on the proper selection criteria even when confronted with all of the
relevant facts under presumed perfect certainty. This is a very unusual result
for tlie selection of a very expensive producer good in which a 10% saving in
monthly rental could be more than the salary of a full time selection expert.
Faced with uncertainty about the validity of competitors' claims, most managers
will remain with their current supplier unless they are extremely dissatisfied
with the service received, causing severe difficulties for a new firm trying to
attract customers.

The second major cause of brand loyalty is the cost of switching among
computer manufacturers. The customer makes a large investment in a data
base and programs to perform his applications. The programs and data are
specialized to a particular computer system and are often of equal or greater
value than the computer hardvvare. The degree of specialization of programs
and data is a measure of customer loyalty to his manufacturer. In the extreme
case where all programs and data were totally non-transferable to a competi-
tive machine, even a competitive price of zero could not entice the customer away
because the cost of reconstructing programs and data would be greater than
the savings in machine rental. The widespread use of higher level programming
languages such as Fortran and Cobol which are similar across machine types



10 Elmer G. Schuster, "Selective Demand Determinants in the Computer Acquisition
Process," Ph. D, dissertation, The American University, 1969 (Ann Arbor: University
Microfilms, Inc.), pp. 31-35.



5688

has reduced but not completely eliminated tlie problem. A potential competitor
must still design his system to exactly match the specifications of the supplier
whose customers he wishes to attract, or offer subtantially reduced prices to
induce the customer to pay the cost of converting. In practice, this means that
most manufacturers adopt the practices of IBM in order to remain competitive,
allowing computer standards to be manipulated for competitive advantage.^^

The uncertainty invitlved in computer selection combined with the cost of
switching to a new manufacturer causes brand loyalty to be very strong in the
computer systems market, which poses a formidable barrier to entry. How-
ever, two segments of the market have largely escaped the influence of brand
loyalty. The plug compatible peripheral manufacturers produce identical copies
of IBM input-output units except for occassionally improved performance. They



Online LibraryUnited States. Congress. Senate. Committee on theThe Industrial reorganization act. Hearings, Ninety-third Congress, first session [-Ninety-fourth Congress, first session], on S. 1167 (Volume pt. 7) → online text (page 99 of 140)