Firm Size and Profitability

Author(s): Marshall Hall and Leonard Weiss

Source: The Review of Economics and Statistics, Vol. 49, No. 3 (Aug., 1967), pp. 319-331
Published by: The MIT Press
Stable URL: http://www.jstor.org/stable/1926642 .
Accessed: 28/06/2014 11:24

Your use of the JSTOR archive indicates your acceptance of the Terms & Conditions of Use, available at .
http://www.jstor.org/page/info/about/policies/terms.jsp

.
JSTOR is a not-for-profit service that helps scholars, researchers, and students discover, use, and build upon a wide range of
content in a trusted digital archive. We use information technology and tools to increase productivity and facilitate new forms
of scholarship. For more information about JSTOR, please contact support@jstor.org.

The MIT Press is collaborating with JSTOR to digitize, preserve and extend access to The Review of
Economics and Statistics.

http://www.jstor.org

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
 FIRM SIZE AND PROFITABILITY
Marshall Hall and Leonard Weiss *
A BASICpropositionof economictheoryis I
that, under competition, profit rates will Previous studies of the effect of size on
tend toward equality. Baumol, on the other profitability [7, 20, 21, and 22] have provided
hand, has put forward the hypothesis that only very imperfect information on the sub-
increased money capital will not only ject. They have been based on Internal Rev-
increase the total profits of the firm, but be- enue Service data for all corporations with
cause it puts the firm in a higher echelon of balance sheets and have therefore included
imperfectly competing capital groups, it may many firms that were suboptimal within their
very well also increase its earnings per dollar chosen industries. The profit rates for such
of investment" [5, p. 33]. firms do not belong on the long-run size-profits
The logic of Baumol's proposition is that curve.' Baumol's hypothesis would imply that
large firms have all of the options of small optimally-scaled firms in industries where the
firms, and, in addition, they can invest in lines minimum efficient scale is small would earn
requiring such scale that small firms are ex- lower rates of return than those in industries
where it is large.
cluded. It follows that ". . . so long as any
In addition, because of the aggregative na-
industries are peculiarly well suited to large
ture of the Internal Revenue Service data,
investments, and so yield disproportionate re- these studies have not been able to set up
turns to sizeable funds, then, provided capital controls for characteristics of firms other than
is prepared to move in response to profit dif- their scale and industry. Moreover, the In-
ferences, this will tend to be true of all other ternal Revenue Service lumps all firms of more
industries in which large firms operate" [5, than $250 million assets into a single size class,
p. 37]. If this hypothesis is correct, we should though some of the most interesting differences
find higher rates of return in large enterprises occur above that size.
even in the long run and even in the absence of The typical result of the Internal Revenue
barriers to entry other than those directly asso- Service data studies has been that beyond
ciated with availability of capital. An empir- some fairly small scale (five to ten million dol-
ical test of this proposition should give a basis lars in assets) size has little effect on profits
for judging the "height" of the "capital re- [7, pp. 53-65, and 20, pp. 35-37]. With other
quirements" barrier [2, p. 156], a potentially pertinent variables considered, and with scale
important element in the explanation of indus- differences above $250 million in assets dis-
try performance. tinguished, however, the results might be quite
The primary goal of this paper is to test different.
The present study attempts to circumvent
Baumol's hypothesis. It is divided into three
these previous difficulties by using the records
parts. In section I we discuss our data and
of individual firms as observations, by limiting
our model, in section II we present and an- itself to firms of optimal scale, and by intro-
alyze our results, and in section III we draw a ducing appropriate additional variables. The
few tentative conclusions. relationship of profit rates to these other vari-
* The authors are, respectively, Associate Professor of 1 Actually, Baumol's hypothesis does allow for long-run
Economics at Washington University and Professor of Eco- profit differences within an industry [5, p. 40] if the scale
nomics at the University of Wisconsin. Portions of the curves are properly shaped, since a small firm may find
computations for this study were financed by grants from suboptimal operations in an industry requiring large scale
the National Science Foundation. Computers at the Uni- to be as profitable as the best alternatives available for a
versity of Wisconsin and at Washington University were firm of its size. However, the IRS data provide no basis
used. The comments of Harold Watts, William Baranek, for distinguishing between such cases and those suboptimal
Lawrence Hexter, and Jan Kmenta were very useful in the firms which will, in the long run, either adjust their scales
preparation of this paper. or leave their industries.
[ 319 ]

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
320 THE REVIEW OF ECONOMICS AND STATISTICS
ables is of interest by itself and will appear as tribution or construction. Many of the rest
a by-product of our study. had narrowlydefinedproducts such as linoleum,
explosives, shoe machinery, coal mining ma-
The Sample chinery, bearings, ships, photographic equip-
The firms studied were selected from the ment, or watches where no detailed production
Fortune "Directories of 500 Largest Industrial index exists and where two- or three-digit in-
Corporations"[10] for the years 1956 to 1962. dexes would have been quite unrepresentative.
Only firms that were listed in the top four hun- While the third group of exclusions tends to
dred in at least one of the seven years covered bias our sample against diversification, this
were included in the sample. There were 467 last group biases it in the opposite direction, if
such firms, but 126 of these were excluded from anything.
the sample for a variety of reasons. Finally, seven firms were excluded by error.
First, 21 firms were excluded because they In most cases they appeared in the top 400
were smaller than the minimum efficient plant only once during the years 1956 to 1962.2
scales as estimated by Bain for their industries Each firm in each year is treated as a sep-
[2, p. 72 and Appendix C]. This included all arate observation. Less than seven observa-
steel firms as small as Colorado Fuel and Iron, tions appear for a number of firms because of
all petroleum refineries as small as Skelly, and movements into or out of the Fortune directory
one automobile firm, Studebaker Packard. due to mergers or differential rates of growth.
Some of these firms were undoubtedly viable This introduceslittle bias, however, since grow-
within their selected geographical or special- ing and declining firms both receive such treat-
ized product sub-markets, but their exclusion ment.
does no harm to our study, while the inclusion The limited sample does not permit us to
of suboptimal firms would make the test of the observe the low end of the size-profit curve,
size-profit hypothesis equivocal. All of the but our range of observation is still quite wide.
other included firms in industries where esti- Firms within the sample range in assets from
mates could be made seemed to be larger than $33 million to $11 billion. Only 3.6 per cent
the minimum efficient scale. A review of the of the observations appear above two billion
included firms from industries requiring large- dollars and only 4.1 per cent appear below $50
scale national advertising or dealer systems million, so interpretationsof our results should
suggests that few if any were suboptimal from probably be restricted to that range.
a promotional point of view either. In most
industries, only a handful of industry leaders The Profit Concept
were large enough to appear in our sample. Our dependent variables is Ii,/Ei,, the rate
Second, we excluded four firms because their of return after tax on year-end equity ("in-
profit rates were in some way subject to public vested capital") as reported in Fortune plus
regulation. Western Electric, General Tele- and minus windfall losses and gains reported
phone, Pullman, and International Telephone in that source. (Throughout the paper i refers
and Telegraph were so treated. These would to the firm, j refers to the industry, and t refers
all have been excluded for lack of production to the year.) It is held that profits after taxes
indexes anyway. are the most appropriatevariable. Profit rates
Third, 38 firms were excluded because they before taxes might differ because of differing
were too diversifiedto permit us to assign them tax treatments of different industries, but entry
to a particular industry or weighted group of should bring profit rates after taxes toward
industries. These firms were distributed fairly
equality under competition.
evenly over the observed range of asset sizes. We prefer the rate of return on equity to
Fourth, 63 firms had to be excluded because
no index of industrial production or, in a few 2 A complete list of firms in the sample and another of

cases, no index of concentration, was available firms among the top 400 which were excluded appears in
an appendix available on request from Leonard W. Weiss,
for the industry to which they were assigned. Economics Department, University of Wisconsin, Madison,
A few of these were heavily committed to dis- Wisconsin, 53706.

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
 FIRM SIZE AND PROFITABILITY 321
that on total capital, partly because this is the economic profits are well known. We use the
profit rate reported in Fortune, but also be- reported accounting profits (excluding wind-
cause it seems theoretically correct. It is what falls). The undervaluationof assets and equity
managers acting in the owners' best interests due to inflation is limited because the dates
would seek to maximize. We argue that capital used (1956-1962) are well after the main post-
structure is an element of input mix. Either war inflation. Moreover, by introducing indus-
profit maximization or sales maximization try growth as a variable, we should correct
would require some optimal rate of borrowing, much of the tendency for slowly growing in-
which differs from industry to industry de- dustries to understate, relatively, their assets
pending on such things as stability and growth and equity and therefore overstate their profit
prospects. As a result, rates of return on assets rates due to large percentages of old plants.
should differ between industries, even in per- The growth variables should also remove much
fectly competitive long-run equilibrium, but of the difference in effect of accelerated depre-
rates of return on equity should tend toward ciation among industries with different growth
equality between industries. This becomes rates.
readily apparent when the implications of pure There is a variety of reasons to believe that
competition for profit rates are applied to such the remainingerrors in the profit data will tend
industries as banking where debt ratios com- to understate, if anything, profit rate differ-
monly exceed 90 per cent! Equalization of ences. (1) Profitable, and particularly large
rates of return on assets for such industries and profitable firms would have the greatest
with those typical of much of manufacturing incentive to adopt accounting practices that
would result in obviously unrealistic rates of understate their profits for tax purposes. (2)
return on equity. The same group of firms might generally be
Because there is still room for controversy expected to feel most vulnerable to public
over the appropriate rate of return, [22, pp. opinion and to understateprofit for public rela-
123-124] we also ran our regressions with tions reasons as well. (3) To the extent that
litl/Ait, profits after taxes as a percentage of managers seek to retain profits rather than pay
year-end assets, as an alternative dependent dividends, profitable firms might also under-
variable. Baumol, himself, presents his size- state profits for reasons of stockholder rela-
profit hypothesis both in terms of Iit/Eit [5, tions. At the same time, managers of unprofit-
p. 43] and IitlAit [5, p. 35-39]. As a by- able firms, if concerned with retaining control,
product of the present study, we should be able might well overstate profits. (4) It has been
to judge which rate of return is more nearly suggested that exceptionally profitable firms
equalized between industries. and particularly profitable firms in the public
The year-end ratio of equity to assets spotlight may take some of their earnings in
(Eit/Ait) will be introduced as an explanatory the form of congenial and/or socially attrac-
variable. According to the argument just pre- tive personnel [1] or in various other man-
sented, there should be a positive relationship agerial prerogatives [17], neither of which will
between EitlAit and lli,/Ai,. This will be re- appear as financial profits. Where recorded
enforced in the present study because our data profits are closer to zero, one would expect to
do not include interest payments, though these find fewer such non-pecuniary returns. (5)
are certainly part of the cost of total capital. Where the assets of business firms have
Ei,/Ai, is also introduced in the litl/Ei, model. changed hands at prices reflecting profit pros-
Although we expect inter-industry differences pects, equity will approximate the capitalized
in lli,/Ei, to be independent of Eit/Ai,, the value of expected profits so that reported
firms within particular industries patently do Htl/Eit will tend toward equality even where
have differing capital structures. Since large economic returns on capital at replacementcost
amounts of leverage (low Eit/Ai,) imply high are not equalized.
risks, one would expect a negative relationship Since virtually all of these considerations
between Hl1/Eit and Eit/Ait. point toward an understatement of profit dif-
The inconsistencies between accounting and ferences, we can probably expect that any de-

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
322 THE REVIEW OF ECONOMICS AND STATISTICS
tected size-profit relationship does, in fact, FIGURE 1. - EXPECTED RELATIONSHIP BETWEEN SIZE
exist. On the other hand, the failure of an AND RATE OF RETURN

expected relationshipbetween size and account-

ing profits to appear would not permit us to E
reject unequivocably the size-profit hypothesis.
The Size Concept
The firm size variable used in our study is
1/(log A,,), the reciprocal of the logarithm to
the base ten of total year-end assets expressed
in thousands of dollars. Baumol's own state-
ment of the size-profit hypothesis defined size
Log A
as the "amount of owned and borrowed money
capital" [5, p. 38] to which total assets is a mol [5, p. 381. Of course, our formulation of
good approximation. We feel that this concept the size-profit relationship is compatible with
of size is correct. A definition expressed in a nearly linear relationship over the observed
asset terms is superior to a sales or employment range if such appears appropriate.
concept of size because it is the difficulty of
financing large lumps of assets that limits en- Concentration
try to certain fields. Assets are superior to Ideally, we should also control for the
equity, even in an explanation of li,l/E&1,be- various elements of competitive structure such
cause it is the size of the total lump of capital, as buyer and seller concentration, other bar-
however financed, that determines the oppor- riers to entry besides capital cost, and product
tunities available to the firm. differentiation. We have had to limit ourselves
Baumol did not formally specify the func- to seller concentration (Cj) and assume that
tional relationship between size and profits. other elements of competitive structure were
We used a logarithmic form of asset size for independent of firm size. Weighted averages
the argument that the difficulty of raising an- of 1958 four-firmconcentration ratios for com-
other one per cent in assets is more nearly com- ponent four-digit products were computed for
parable between General Motors and, say, the industry of each firm [based on 18, Table
American Motors, than is the difficulty of rais- 4]. Corrections were made in many of the
ing another million dollars for each firm. We basic concentration ratios for noncompeting
used the reciprocalform because we anticipated subproducts, inter-product competition, and
that another percentage addition to assets the local and regional character of industry
might, in fact, be easier for General Motors to following a technique used previously by one
raise than for a smaller firm. In other words, of the writers [24 and 25].
a 100-million-dollar firm might have greater Concentration is, of course, a characteristic
difficulty entering an industry with a 50-mil- of the industry to which a firm belongs. In
lion-dollar capital requirement than a bil- many cases, the firms in our sample were as-
lion dollar firm would have in entering signed to three-digit industries according to
a field which requires a $500-million-dol- the list in Mergers and Super Concentration
lar initial investment. A second reason that [13, Appendix] and the weights used in com-
one might expect IlI,/Ei, to rise at a de- puting average concentration ratios were then
creasing rate with proportional increases in the 1958 product shipments reported by the
size is the increased possibility of public criti- Census. This listing was compared with the
cism of "excessive profits" as the firm becomes industry assignment of plants in the Fortune
larger. We expected to find a negative rela- Plant and Product Directory [11]. In 109
tionship between IHi/Eit and 1/log Ait follow- cases it was possible to assign firms to specific
ing the pattern shown in figure 1. This seems four-digit product categories so that no aver-
to correspond to the diagrammatic suggestion aging was necessary. In 19 cases, rough em-
of the size-profit relationship shown by Bau- ployment weights for industry mix from the

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
 FIRM SIZE AND PROFITABILITY 323
Fortune Plant and Product Directory were seems adequate to dispose of the possibility
used to combine concentration ratios of two that we have the line of causation reversed. If
or more industries.3 large firms grow no faster than small ones,
The usual expectation, of course, is that high there is little danger that high profits are the
concentrationwill be associated with high prof- cause of size rather than size being the cause
its because such concentration permits en- of high profits. Since past work on this ques-
hanced control over market prices. tion has covered both relative growth rates
within industries [14, 16] and in all manufac-
Growth and Stability turing [6] and since some of the studies have
An attempt is made to allow for changes in covered almost the same sample of firms as
demand and/or costs of production by intro- ours [6 and 14] we feel fairly confident that
ducing the previous five years' percentage firm growth can be taken as independent of
changesin output (Qjt/Qjt-1, Qjt-1/Qjt-2 . . firm size in our study.
Qjt-4/Qjt_5). The Federal Reserve index of Time dummies are introduced for each year
industrial production [9] for the industry to observed after 1956. These should show the
which the firm was assigned is represented by impact of economy-wide growth and fluctua-
Q. In some cases indexes for a number of in- tions on profit rates, while the industry growth
dustries are combined using the 1957-1959 variables show the effect of special fluctuations
weights given by the Federal Reserve or the or trends in the industry in question.
rough employment weights derived from the To summarize, then, our model is:
Fortune Plant and Product Directory.4 IIit/Eit = (a + b1) (1/log Ait)
Changes in industry output may reflect + b2Cj+ b3 QjtlQjt-1
either changes in demand or changes in cost + b4 Qjt-l/Qjt-2 -2+ be Qjt-2/Qjt-.3
that result in movements along the demand + b6 Qjt-3/Qjt-4 + b7 Qjt-4/Qjt-.,
curve. Although we cannot distinguish between + time dummiesfor 1957-1962
these, we expect them to have similar effects + b13Eit/Ait + Eit.
on profitability. Either an increase in demand We expected b1 and b13 to have negative signs
or a decrease in costs would result in increased and b2 through b7 to have positive signs.
profits unless it is fully anticipated. Industry
output was used instead of firm output because Correction for Heteroskedasticity
deflated data covering many previous years The variance of profit rates for large firms is
were available. Sales receipts of individual less than for small firms, [21 and 22, p. 48]
firms would be affected by price and are often presumably because of the greater diversifica-
not available for the whole 1951-1962 period. tion of large firms. If the large firms were
We have therefore controlled for industry equivalent to randomly selected combinations
growth and instability, but profit differences of N small firms, then the means of litlEit for
due to differential growth rates or differential the large firms would have a variance of 1/N
instability within industries have not been con- times the variance of IitlEit for the small
trolled. In effect, we assume that relative firms. This suggests that the sum of the squares
growth and instability within industries are of the residuals from a simple least squares fit
independent of firm size beyond minimal effi- of our model would be proportional to l/Ait.
cient scale. Figure 2 shows a plot of the sums of the
There is considerable evidence that average squared residuals from a simple least squares
growth rates (though not the variances of fit of the above model for successive groups of
growth rates) are in fact independent of size 30 observations for 1962. The curve shown in
[6, 14, 16, and 19]. Incidentally, this evidence figure 2 fits the equation Sum of Squares=
k ( 1/Ai62) where k is an arbitraryconstant given
'A full description of the corrections used and a list of
the resulting concentration indexes appears in the appendix the value of 50 billion to roughly fit the ob-
available on request from Leonard Weiss. The particular served values. It correspondsto the theoretical
industries to which firms were assigned are also shown.
'A full description of the output variables appears in
pattern of heteroskedasticity. It is plain from
the appendix available from Leonard Weiss on request. figure 2 that variance falls off less rapidly in

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
324 THE REVIEW OF ECONOMICS AND STATISTICS
FIGURE 2. OBSERVED AND EXPECTED these constants are significantly greater than
HETEROSKEDASTICITY IN RATES OF RETURN
zero. In view of the weights used, this constant
Sum of
Residuals term is, in effect, the coefficient for an addi-
Squared

700\ tional variable, 1/VA\Ait It is shown as the

600 - coefficient for that variable at the ends of col-
500 \ umns (1) and (4). Since this coefficienthas a
400 -
positive sign, it offsets to some extent the ef-
300 -
fect of the negative coefficient for 1/log A1t.
200 _
Equations (2) and (5), in accordancewith our
100 _
correction for heteroskedasticity, are estimated
20 50 100 200 400
to pass through the origin, thus ensuring that
Median Asset Slze of Class In S
1,000
Milhons
2,000 5,000

the constant term (i.e., the coehficient for

practice than in theory, meaning that large 1/'\/Ajt), is zero.
firms are not perfectly diversified groups of
randomly selected smaller firms. This, of Size and Profits
course, is commonly observed [19]. Values for HitlE1tand 1it/AAitare plotted in
The theoretical relationship between size figure 3 assuming mean values for Cj (the in-
and variance is sufficiently close to the actual dustry growth variables) and Ei,lAit, and us-
one, however, that we feel justified in correct- ing the mean of the coefficients for the time
ing for heteroskedasticity by multiplying all dummies (the coefficient of zero for 1956 was
observationsof all variables by VAi, (V/assets) averaged in with the six time coefficients from
and introducing \/A as a further variable [ 12, table 1). The curves bear the same labels as
p. 235 and 15, pp. 207-211]. The decline in the equations reported in table 1. The initial
variance with firm size is most marked for ob- downward slope of curves (1) and (4) results
servations where assets are less than $200,- from our imperfect correction for heteroskedas-
000,000. As a check on our results, unweighted ticity and is of doubtful meaning. Equations
least squares regressions are also run for ob- (3) and (6) were not fitted to observations
servations above the $200,000,000 size. with assets below $200,000,000 so their ex-
trapolation backward to smaller sizes is also
questionable. Equations (2) and (5), where
II the error in our weighting system is suppressed,
The coefficients and standard errors esti- is probably our safest representation of the
mated from our data are shown in table 1. Six size-profit relationship.
equations appear there. In equations (1) We feel reasonably certain that the differ-
through (3) the dependent variable is Hit/Eit ence in size between $200,000,000 and $2 bil-
and in equations (4) through (6) it is HitlAit. lion raises reported HitlEit from 9.8 points to
Equations (1), (2), (4), and (5) were com- 10.9 to 11.3 points or by 11 to 15 per cent.
puted using VAit as weights. Equations (3) Over the same size range, IJt/Ait increases
and (6) are unweighted regressions fitted to from 6.3 - 6.9 to 7.2 points or by ten to 14
the observations where Ait exceeded $200,000,- per cent. The possibly questionable extension
000. of equations (2) and (5) back to $50,000,000
The coefficients relating 1/log Ait to profit suggests that the difference in assets from
rates are of the expected signs and significantly $50,000,000 to $2,000,000,000 involves a dif-
different from zero in all six equations. The ference in ltl/Eit from 8.6 points to 11.3 points
large values for this coefficient in equations and a difference in llit/Ait from 5.6 points to
(1) and (4) are misleading, however, because 7.3 points -increases of 30 per cent in both
our correction for heteroskedasticity is imper- cases.
fect. The constant terms in these equations We feel that the Baumol hypothesis has been
would be zero if the residual sum of squares strongly supported by the data. As we argued
were in fact proportional to 1/Ait. Actually earlier, there are many reasons for supposing

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
 FIRM SIZE AND PROFITABILITY 325

TABLE 1. -REGRESSION EQUATIONS WITH THREE ALTERNATIVE TREATMENTS OF HETEROSKEDASTICITY

Regression Coefficients (Standard Errors in Parentheses)

ll t/Est Dependent Variable isst /A i t Dependent Variable

Variable (1) (2) (3) (4) (5) (6)

Constant -19.55 -39.82 -30.42 -17.98 -31.78 -24.06
[VAft in (1) (5.126) (3.988) (6.577) (2.861) (2.235) (3.586)
and (4) ]
1/log ALt -183.4 -48.51 -36.76 -123.5 -31.65 -20.52
(22.6) (6.309) (16.17) (12.61) (3.536) (8.819)
Cj .0283 .0253 .0166 .0153 .0133 .0062
(.0073) (.0074) (.0108) (.0041) (.0041) (.0059)
Qjt/Qj t-l 17.59 17.63 15.56 10.68 10.70 7.954
(1.478) (1.491) (2.630) (.8248) (.8354) (1.434)
Qjtgl/Qjt 2 12.94 13.19 10.91 7.902 8.069 6.225
(1.356) (1.367) (2.391) (.7569) (.7663) (1.304)
Qjt-2/QJts-3 10.12 10.48 4.454 7.017 7.259 3.686
(1.303) (1.313) (2.288) (.7271) (.7358) (1.248)
Qjt s/Qtj 4 8.009 8.101 3.937 5.819 5.881 3.193
(1.251) (1.261) (2.132) (.6981) (.7070) (1.163)

Qj t 4lQ. t-5 5.415 5.267 8.380 3.429 3.328 4.785

(1.222) (1.232) (1.954) (.6822) (.6908) (1.065)
Dummy with
value of 1 if:
1957 -.5291 -.5618 -.1965 -.3906 -.4129 -.3431
(.4968) (.5011) (.8152) (.2773) (.2809) (.4445)
1958 -1.330 -1.359 .03185 -.9843 -1.004 -.2792
(.4860) (.4902) (.7960) (.2713) (.2747) (.4340)
1959 -2.346 -2.321 -1.501 -1.337 -1.320 -.7236
(.4908) (.4950) (.7954) (.2740) (.2775) (.4338)
1960 -2.609 -2.716 -2.629 -1.298 -1.370 -1.201
(.4141) (.4173) (.6831) (.2311) (.2339) (.3725)
1961 -2.725 -2.773 -1.861 -1.572 -1.605 -1.161
(.4834) (.4875) (.7984) (.2698) (.2732) (.4354)
1962 -2.821 -2.852 -1.514 -1.755 -1.776 -.9740
(.4643) (.4683) (.7608) (.2592) (.2625) (.4149)
EittA gt .0332 .0419 .0411 .1182 .1241 .1224
(.0091) (.0091) (.0130) (.0051) (.0051) (.0071)
1/V/Ait 2718. .... .... 1851. .... ....
[constant in (437.6) (244.2)
and (4)]
-.2906.
R2 .7041a 8436 .1202 .7824a .8841 a .2911
degrees of freedom 2179 2180 966 2179 2180 966
a R2 is biased upward in equations (1), (2), (4), and (5) due to the nature of the correction for heteroskedasticity. The R2's in equa-
tions (2) and (5) are not comparable to those for equations (1) and (4)

that we have understated, if anything, the rayon [2, p. 170]. His capital requirement
relationship between size and profits. barrier reached its maximum in industries such
We can make some judgment about the as steel where minimum efficient scale may re-
height of the "capital requirements"barrier to quire 500 million dollars. Using this range
entry from the slopes of the curves in figure 3. only, equations (2) and (5) suggest that firms
Several of the industries in which Bain found in industries with such high entry fees could
only "moderate" barriers are typified by elevate HI1/Ei,from 8.8 to 10.4 or by about 18
"small"firms in the $50,000,000 neighborhood. per cent and could elevate IIit/Ait from 5.7 to
This is true of tires, metal containers, and 6.8 or by about 19 per cent without attracting

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
326 THE REVIEW OF ECONOMICS AND STATISTICS

FIGURE 3. -RELATIONSHIP BETWEEN ASSET SIZE AND RATES OF RETURN IN REGRESSIONS

(1) -(6) WHEN OTHER VARIABLES ARE ASSIGNED THEIR MEAN VALUES

Rate of
Return
14

'3 -0(

2 ?
11

I0
@~~~~~~ it

O IlJI II * I I 1I I .1
20 Mn. 5OMn. IOOMn. 200Mn. 500 Mn. I Bn. 2Bn. 5Bn. IOBn.

entry from those capable of breaking into in- from firm size as a determinantof profits. This
dustries with moderate barriers. While these is an important conclusion because concentra-
figures are based on a slightly dubious regres- tion and size are correlated,5and without hold-
sion, any possible upward bias in them is ing concentration constant it would be impos-
probably offset by the very probable under- sible to distinguish the effects of size. As we
reporting of high profits. have measured it in table 1, the effect of con-
The expected tendency for the size-profit centration is significant but small. A 40-point
curve to flatten out proved very weak. This increase in Cj (from 30 to 70 which includes 77
suggests that capital is rationed approximately per cent of our observations) would increase
in proportion to assets over the whole range of Hj,1Ej,by only 0.7 to 1.1 points and would in-
large-scale firms observed and that the capital crease II1/Aj, by only 0.3 to 0.6 points. At the
cost barrier to entry is approximately propor- 5 The simple (unweighted) correlation coefficient relat-
tional to the log of assets. ing Cj to Att is .1569 and that relating C; to 1/log Ait is
Market concentrationturns out to be distinct -.2906.

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
 FIRM SIZE AND PROFITABILITY 327
point of means this is a six to 11 per cent centration dummies instead of a continuous
increase in IlI,lEit and a four to nine per cent in- concentration variable, the rate of return on
crease in Iitl/Ait. Concentration would seem equity was 1.921 points higher when the
to be less important than the capital require- (.5452)
ments barrier as a determinant of profitability. concentration ratio was 40 to 49 than when it
As an additional check, we attempted regres- was zero to 29, but it was only 0.2492 higher
sions similar to those in table 1 using dummy (0.5140)
variables for industries with concentration ra- if concentrationexceeded 59, and it was actual-
tios of 30 to 39, 40 to 49, 50 to 59, and of 60 ly lower if concentrationwas 30 to 39 or 50 to
or more. It was hoped that by this expedient 59, (-1.434 and -.4762, respectively). We
we could allow for the nonlinear relationship (.6103) (.5036)
between concentration and profits noted by have retained the linear relationship between
Bain [4]. The coefficients for 1/log A1t were concentrationand profits in the main argument
actually enhanced by this alternative formula- of our paper as more plausible than these rela-
tion. In equations correspondingto (1), (2), tionships.
(4), and (5) they were, respectively:
-2 34.4, -53.11 , - 152.5, Response of Profits to Changes Over Time
and -44.96.6
(21.56) (6.699) (13.21) (4.100) The coefficients of other variables are of in-
The coefficients for the concentration dummy terest in themselves. Past industrial output
variables had unexpected values in some cases, growth seems to have a very regular effect on
however. For instance, with an equation iden- profitability. Table 2 shows the percentages
tical to equation (2), except for the use of con- of total identified influence of past growth
TABLE 2. - DISTRIBUTION OF THE EFFECTS OF LAGGED CHANGES
IN INDUSTRIAL PRODUCTION ON RATES OF RETURN

Mean Values Effect on Profitability a

Assets
more than II/E Dependent H/A Dependent
All $200 million
Variable firms dollars (1) (2) (3) (4) (5) (6)

Qjt/Qjt-i 1.032 1.034 32.4 32.1 36.0 30.6 30.4 30.8

Qj t-I/Qj t-2 1.039 1.040 24.0 24.2 25.4 22.8 23.0 24.2
QJ ts2/QJ t-3 1.032 1.032 18.7 19.1 10.3 20.0 20.3 14.2
Qjt_3/Qjt-4 1.041 1.042 14.9 15.0 9.1 16.8 16.9 12.4
Qjt_4/QJt-5 1.031 1.028 10.0 9.6 19.2 9.8 9.4 18.4

a Percentages computed by (Vsk bk)/[ (Vk be) where Vk iS Qjt-k+l / Qj e- and be is the correspondingcoefficient from table 1.

6 Since the largest firm was also one of the most profit-
rates using the regression coefficients from
able firms (General Motors) we further tested our hypoth- table 1 and the mean values shown in the first
esis by running the same regressions with that firm ex-
cluded from the sample. The coefficients for 1/log Att were two columns of table 2.
reduced but were still significant. In equations correspond- The rapid drop in past influences sup-
ing to (1), (2), (4), and (5) but with concentration dum-
mies instead of the continuous concentration variable, the
ports Stigler's contension that market
coefficients for 1/log Att were: -123.7, -18.02, changes are rapidly translated into investment
(23.86) (7.079) changes and have only mild and temporary
-143.2, and -23.29, respectively. We would argue,
(16.27) (4.328)
effects on profitability [22, pp. 4-6, 18-19, 37,
however, that General Motors is perhaps the prime example and 69-70]. In his study the average correla-
of the Baumol hypothesis in the American economy today tion coefficient relating profit rates in year t
and that there is no good economic reason for excluding it
from the sample. We predict high profit rates in large firms
and in year (t + n) fell to nonsignificance by
and should not be surprised, therefore, when the largest the time n reached three years for the 1937-
firm yields one of the highest rates of return. Moreover, 1947 period and seven years for the 1947-
the rates of returns of the remaining large firms seem par-
ticularly likely to be understated because oil producers play
1957 period [22, pp. 70-71]. Our finding that
such a large role among them. changes in output five years back account for

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
328 THE REVIEW OF ECONOMICS AND STATISTICS
only ten per cent of the allocable influences of equation (1) and -.02141 in an expression
past changes on profit rates is similar to Stig- (.00822)
ler's peace-time results. Extrapolating back- correspondingto equation (2).
wards from the values in table 2, we would The relationship between litl/Ait and EitlAit
expect a near zero influence of output changes is bound to be positive because we left out
in periods more than six years earlier. The interest payments from our return on total
less consistent decline in the effect of past capital. Moreover, it can be shown that the
changes on the profitability of the largest firms elasticity of HII/Ajtwith respect to EitlAit is
(equations 3 and 6) may be related to the one greater than the elasticity of IIitlEit with
greater stability of profits discovered by Stig- respect to EIlAi,.' Where the latter is positive,
ler in concentrated industries [22, p. 70].7 the elasticity of lIj,/Aj, with respect to EitlAit
The coefficients for the time dummies quite must exceed one.
consistently show negative signs, meaning that, Figure 4 shows the computed relationship
as might have been expected, profits in the
other years of our sample were lower than in FIGURE 4. - RELATION OF RATES OF RETURN TO
EQUITY - ASSET RATIOS USING A CONTINUOUS CON-
1956. Profits decline, however, as the 1956- CENTRATION VARIABLE
1962 period progresses, even though the worst
Percent
year of that period in terms of economic activ-
ity was 1958. The lower profit rates in later 10
years may reflect, in part, a fuller adjustment
for past inflation as more and more of the 6
capital of these firms is recorded at original 4
costs close to current replacement costs. lrfrAt

Leverage and Profits 25 /1 /

/
I , r/, , . * ,

The coefficients for E.I Ai provide some

basis for evaluating the proposition that 0 10 20 30 40 50 60 70 80 90 loo
Hi,/Ei, tends to be equalized between industries E/A
because Ei Ai, is adjusted to the special con-
ditions of the industries in question. Our hy-
pothesis suggests a zero relationship between between Ej Ait and the profit variables using
Hit/E,it and Eit/A it in inter-industry compari- equation (1) (solid lines) and equation (4)
sons but a negative relationship for intra-indus- (broken lines). Figure 5 shows a similar com-
try comparisons, reflecting the greater risks parison when concentration is represented by
associated with greater leverage. In fact, the class interval dummies. Using equation (1)
relationship between Eit/Ait and IIitlEit turned HiI/Ait is derived by (IIjt/Eit) X (Eit/Ait). Us-
out to be significantly positive. A possible af- ing equation (4) IitlEit is derived by (lIit/Ait)
ter-the-fact explanationis that relatively profit- * (Eit/Ait). Rough estimates of the rate of
able firms take some of the exceptional returns return on total capital including interest pay-
in the form of reduced risks. It should be ments (Hit + r)/Ait are also included assum-
pointed out, however, that when concentration ing an interest rate of four per cent on all non-
was introduced by a series of class interval
8The elasticity of II/E is
dummies rather than a continuous linear vari- {E HA /I E'
able, the relationship between IIitlEit and A /E JXtE A3-
EitlAit became negative as had been initially The elasticity of I/A
expected. The regression coefficient was then
-.03053 in an expression corresponding to ()(
(E I II E)
A A A A
(.00814) E H E E /H E H/ E E
A E A A E A E A A
'These past output change variables have a slightly less
regular rate of decline when concentration dummies replace
{E 0 II EA
=t .- /}3 /D J+1
the continuous concentration variable.

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
 FIRM SIZE AND PROFITABILITY 329
FIGURE 5. - RELATION OF RATES OF RETURN TO in this case shows (lit + r)/Ait no more af-
EQUITY - ASSET RATIOS USING CONCENTRATION fected by Ai, than lIit/Ei,. The relationships
DUMMIES
derived from lItl/Ait support those shown in
Percent
figure 4.
15

-
10 8-I
III
To summarize,our main conclusions are that
size does tend to result in high profit rates as
Baumol proposed, that there is a significant
2 / though probably not enormous capital require-
ments barrier as a result, and that this barrier
/1 / very likely has a greater effect on profit rates
O 10 20 30 40 50 60 70 80 90 100 than concentration, the traditional index of
E/A market power. As by-product results of our
study we found support for Stigler's conten-
equity capital.9 The three sets of relationships tion that profits are almost completely equal-
are similar over the range of observations ized in about half a decade for firms of given
(.23 < Eit/Ait < .92) in both charts. The re- size and concentration,we detected a persistent
sulting values of (llt + r)/Ait still seem to be tendency toward lower profit rates as the 1956-
much more affected by the amount of leverage 1962 period progressed, and we found support
than is litl/Eit. Using equation (1) from table for the use of the ratio of profits to equity over
1 (i.e., figure 4), at the mean value of Eit/Ait, the ratio of profits plus interest to total assets
the elasticity of nitlEit with respect to Eit/Ait in inter-industry studies. These results have a
is .19, that of litl/Ait with respect to Eit/Ait is number of interesting implications.
1.19, and that of (iIlt + r)/Ait with respect to In Baumol, the size-profits hypothesis was
Eit/Ait is .72. It would take an interest rate of one of the bases for his contention that en-
10.5 per cent to make this elasticity as low as trepreneurs forego current profits to maximize
.19. Alternatively, it could equal .19 with a growth [5, p. 33]. He introduced various
mean interest rate of four per cent if the elas- other considerations in support of his sales
ticity of the interest rate with respect to Eit/Ait maximization hypothesis which we have not
were as great as -3.2. Since such high interest examined, but to the extent that his argument
rates and such high sensitivity of interest rates can be based on the size-profit relationship
to leverage seem quite unrealistic for firms of alone, we feel that we have given it some
the size discussed here, we conclude that the support.
rate of return on assets is, in fact, more sensi- The capital requirements barrier to entry,
tive to Eit/Ait than the rate of return on equity. about which Bain expressed uncertainty [2, p.
It follows that the latter-is, indeed, more nearly 156] and Stigler expressed skepticism [23, p.
equalized among industries and is therefore 27] would appear to be "substantial" though
the appropriateprofit measure to use for inter- probably not "high" unless accounting profits
industry comparisons. understate high profit rates by perhaps a half
The less reliable results shown in figure 5 or more. To go by Bain's final assignments of
largely confirm those of figure 4, though the overall barriers, this is apparently the evalua-
relationship derived from the lItl/Eit equation tion he gave to the barrier, also.
The classic criticism of "monopoly profits"
9 (Iltt+r)/Ait =Ilit/Att+.04 (1-Eut/Aut). The fig- has been that they imply product prices in ex-
ure four per cent is slightly lower than the Aaa bond rate for
1956-1962 (4.07) and the average short-term rates for loans cess of opportunity costs of factors employed
of more than $200,000 (4.59). Both figures were derived and therefore result in the misallocation of re-
from [8, pp. 199-200 and earlier issues]. Perhaps a differ- sources. To the extent that these profits are
ent interest rate could be justified since a substantial part
of (At - Eit) is accounts payable, reserve accounts, and
due to the capital requirements barrier rather
the like. than to concentration or other elements of

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
330 THE REVIEW OF ECONOMICS AND STATISTICS
market power, this criticism seems open to tential competition in those size classes where
question. Capital in large lots would be a dif- exceptional profits can currently be earned.
ferent factor of production from capital in While some conglomerates may increase min-
small sums. A billion dollars in capital would imum efficient scale in a way that is of ques-
be more productive in its best employment tionable social value (e.g., by increasing pro-
than the total of one thousand lots of a million motion costs in differentiated goods produc-
dollars each. Given the imperfection of capital tion), the primary effect of most such inter-
markets, the high returns to large lots of capital industry mergers among large firms seems to
are social opportunity costs, and are properly be to produce very large firms. The formation
included in product price. There are still mis- of such giants by merger may well be the most
allocations, but they are due to imperfect cap- feasible way of creating lumps of capital cap-
ital markets. The misallocative effect of mon- able of threatening industries presently in-
opoly as it is usually understood is indicated sulated by a capital cost barrier. Conglomer-
only by profits in excess of those accounted for ate mergers, at least outside of the highly dif-
by size - for instance, the relatively mild ad- ferentiated goods fields, may well be the most
ditional profits attributable to concentration. promising means of reducing the present dif-
It seems to follow from this last point that, ference in capital cost between large and mod-
so long as we are faced with imperfect capital *eratescale enterprises.
markets, "normal profits" differ from size to
size. This would throw doubt upon the use of REFERENCES
open market interest rates as standards for
judging the existence of "excess" profits as an [1] Alchian, A. and R. A. Kessel, "Competition,
Monopoly and the Pursuit of Money," in Aspects
element of competitive performance [3, p. 367] of Labor Economics (Princeton: Princeton Uni-
in large firms since open market capital is sup- versity Press, 1962), 157-175.
plied in small lots and new money is limited [2] Bain, J. S., Barriers to New Competition (Cam-
by the amounts of capital already in the firm. bridge: Harvard University Press, 1956).
Capital in large enough lots to finance efficient [3] , Industrial Organization (New York: John
investments in industries requiring very large Wiley and Sons, 1959).
[4] , "Relation of Profit Rate to Industrial
scale can be expected to require higher returns Concentration, American Manufacturing, 1936-
than market rates of interest until we see some 1940," Quarterly Journal of Economics, XLV
revolutionary changes in the character of cap- (Aug. 1951).
ital markets. [5] Baumol, W. J., Business Behavior, Value, and
In terms of policy, our results point more Growth (New York: MacMillan, 1959).
[6] Collins, N. R., and L. E. Preston, "The Size
toward an attack on capital market imperfec- Structure of the Largest Firms," American Eco-
tions than to anti-trust programs oriented to- nomic Review, LI (Dec. 1961).
ward divestiture, dissolution, and merger. The [7] Crum, W. L., Corporate Size and Earning Power
anti-trust approach cannot be rejected on the (Cambridge: Harvard University Press, 1939).
basis of our findings- concentrationdoes have [8] Federal Reserve Board, Federal Reserve Bulletin
(Feb. 1964, Aug. 1961, and Dec. 1959).
a significant effect along with firm size. Yet [9] , Industrial Production (Washington,
the greater apparent effect of size than of con- 1960); and "Industrial Production" (monthly).
centration on profit rates suggests that capital [10] "Directory of the 500 Largest Industrial Corpo-
market policies might be given more emphasis rations," Fortune (annually in the July or August
than they now receive. Since the capital cost issues).
[11] Fortune, Plant and Product Directory, 1963-
barrier seems to extend to giant firms, efforts 1964.
to reduce restrictions imposed by capital ra- [12] Goldberger, A., Econometrics (New York: John
tioning might well be extended beyond the Wiley and Sons, 1964).
"small business" sphere where the main em- [13] House of Representatives, Select Committee on
phasis has been to date. Small Business, Mergers and Super-Concentra-
tion (Nov. 1962).
At the same time, we suspect that a whole- [14] Hymer, S., and P. Pashigian, "Firm Size and
sale policy of suppressing truly conglomerate Growth Rates," Journal of Political Economy
mergers may restrict an important form of po- (Dec. 1962).

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions
 FIRM SIZE AND PROFITABILITY 331
[15] Johnston, J., Econometric Methods (New York: [21] , "The Variability of Profitability with Size
McGraw-Hill, 1963). of Firms, 1947-1958," Journal of the American
[16] Mansfield, E., "Entry, Innovation and Growth Statistical Association (Dec. 1964), 1,183-1,193.
of Firms," American Economic Review, LII [22] Stigler, G., Capital and Rates of Return in Man-
(Dec. 1962). ufacturing Industries (Princeton: Princeton Uni-
[17] Scitovsky, T. de, "A Note on Profit Maximization versity Press, 1963).
and it's Implications," Review of Economic Stud- [23] , "Monopoly and Oligopoly by Merger,"
ies, II (Winter, 1943), 57-60. American Economic Review, Proceedings (May
[18] Senate Judiciary Committee, Subcommittee on 1950), reprinted in Readings in Industrial Organ-
Anti-Trust and Monopoly, Concentration in Man- ization and Price Policy (Homewood, Ill.: Rich-
ufacturing Industries, 1958 (Washington, 1962). ard D. Irwin, 1958).
[19] Simon, H., "Comment: Firm Size and Rate of [24] Weiss, L., "Average Concentration Ratios and
Growth," Journal of Political Economy (Feb. Industrial Performance," Journal of Industrial
1964). Economics (July, 1963).
[20] Stekler, H. 0., Profitability and Size of Firm [25] , "Concentration and Labor Earnings,"
(Berkeley: Institute of Business and Economic American Economic Review (March, 1966).
Research, University of California, 1963).

This content downloaded from 193.105.245.179 on Sat, 28 Jun 2014 11:24:10 AM

All use subject to JSTOR Terms and Conditions