STRATEGIC ORGANIZATION Vol 1(1): 79–108

1476-1270[200302];1:1;79–108;031219
Copyright ©2003 Sage Publications (London,Thousand Oaks, CA and New Delhi)

The emergence and sustainability

of abnormal profits
Anita M. McGahan Boston University
Michael E. Porter Harvard University

Abstract
In this paper, we examine the emergence and the sustainability of abnormal profits among
businesses that were part of US public corporations between 1981 and 1994 and that
reported financial results for at least six years. Our results reveal strong asymmetries
between high and low performers. Overall, high performance is more stable than low per-
formance. High performers show profits above the average a decade earlier. In contrast,
low performers show profits that are slightly above average a decade earlier. Industry and
corporate-parent effects influence high performance to a far greater degree than low per-
formance. Low performance is dominated by business-specific effects.

Key words • competition • entrepreneurship • industry • performance • sustainability

The fields of strategy and organization deal with fundamental questions about
how company profits emerge and persist over time. Strategy researchers
have generated extensive theory on the sources of profitability, including the
industrial-organization, resource-based, property rights, organizational ecology,
and transaction-cost views. For example, resource-based theorists tend to
emphasize the importance of corporate and business-specific effects in the emer-
gence of high performance, and organizational ecologists emphasize the persis-
tence of both high and low performance due to inertia. While there is no
shortage of theories to explain the sources of profitability, there is little empiri-
cal evidence about the trajectory of profitability over extended periods of
time. Mueller (1986), Ghemawat (1991), and McGahan and Porter (1999) are
exceptions.
In this paper, we report the results of a panel study that begins to fill this
gap. Our purpose is to report empirical regularities in the different trajectories
of profitability for those businesses that retained their identities and reported
financial results for at least six years. The analysis distinguishes the influences on
the emergence and sustainability of abnormal profitability of year, industry,
corporate-parent, and business-specific effects. The approach reflects the idea
that industry, corporate-parent, and business-specific levels are all important for
79
80 S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )

understanding the role of both strategy and organization. Data are drawn from
the Compustat Business-Segment Reports for 1981 through 1994, which cover
the activities of all corporations publicly traded on US equity markets.
Industry effects on the emergence of high and low profitability arise from
structural impediments to competition that similarly affect incumbents. For
example, after its initial public offering in 1986, the Lotus Development
Corporation was a superior performer in the sense that its return on assets in
each year was greater than average. Its performance largely reflected its partici-
pation in the profitable PC applications software industry. The structural
impediments that generated industry effects in the software industry arose
through both strategy and organization. In this case, the decisions and actions of
particular individuals, such as those of Bill Gates, were instrumental in shaping
the industry structure through strategic and organizational choices that gave
rise to shared development tools, licensing policies, and barriers to entry.
Broadly, this paper provides evidence that suggests relationships between
various theoretical explanations for why industry effects arise.
Corporate-parent effects arise when a company’s return in a specific business
is uniquely affected by its pattern of diversification. A negative corporate-parent
effect arises when a diversified firm’s member businesses underperform their
industries on average. For example, the tendency of corporate parents to deploy
resources into related industries may partly explain why poor performance per-
sists in some businesses. This explanation for a negative corporate-parent effect
draws on both the industrial-organization and resource-based views. Like indus-
try effects, corporate-parent effects can be explained theoretically by the
industrial-organization, resource-based, and organizational views (among
others). In general, we allow for the fact that different theoretical explanations
may be relevant for understanding the sources and persistence of the effects.
This paper identifies the effects but does not discriminate among the theoretical
explanations.
A positive corporate-parent effect arises when a diversified firm’s member
businesses outperform their industries. For example, Taco Bell’s profitability
improved during the 1980s after it was acquired by Pepsico, a beverage and
food company with operations in other fast-food businesses. In this case, mem-
bership in the corporate parent improved the profitability of the member busi-
ness in ways that were not available to the average fast-food company, and in
ways that supplemented the improvements in the business-specific effect associ-
ated with a 1984 restructuring.
Business-specific effects arise from differences in the competitive positions
of an industry’s incumbents, and also are rooted in both strategy and organiza-
tion. Throughout this paper, the term ‘competitive positioning’ refers to the
broad variety of factors that influence the relative performance of a specific firm
within an industry compared to its direct rivals within the industry. Southwest
Airlines’ performance during the 1980s was largely due to business-specific
effects based on its particular competitive strategy in the airline industry: it
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 81

offered low price and convenience to business travelers on short-haul flights

between medium-sized cities; it achieved extremely low operating costs due to
route selection, few on-board services, efficiencies in aircraft utilization, and
highly motivated employees. Southwest substantially outperformed the in-
dustry as a whole, where limited barriers to entry and vigorous competition
between incumbents depressed average profitability. Case studies on Southwest
demonstrate that its no-frills approach, better segmentation, superior leader-
ship, and extraordinary execution all contributed to the business-specific effect.
In general, business-specific effects may arise from entrepreneurial insight, effec-
tive management of critical resources, superior organization design, effective
leadership, and other factors that influence strategy and organization.
Industry, corporate-parent, and business-specific effects also influence the
sustainability of profits, which we define as the tendency of abnormally high or
low profits to continue in subsequent periods. Theory gives rise to several
potential reasons why high profits may be sustained. Organizational ecology
focuses on impediments to imitation. Resource-based theorists suggest that
long-lived assets with appropriable value explain persistently high profitability.
Recent work in industrial organization emphasizes institutional structure as a
potential source of persistence. This paper presents evidence on actual patterns
in the persistence of high performance to shed light on the causal mechanisms
suggested by each of these theories.
Superior performance may deteriorate because of eroding industry struc-
ture, such as occurred with the deregulation of long-haul trucking during the
1980s. Profits also may decline because of the detrimental influence of a corpo-
rate parent. For example, during the 1980s, Ross Perot claimed that the com-
pany he founded, EDS, had suffered as a result of its acquisition by General
Motors. Performance in a particular business also may decline because of dam-
age to business-specific effects. The Heileman Brewing Company was bankrupt
in 1987 because it could not sustain a profitable position after competitors cut
prices in its markets.
To investigate these effects, we examine the profitability of businesses
covered in the Compustat Business-Segment Reports for 1981 through 1994.
We excluded data on subsequent years because of large distortions in reported
performance during the late 1990s. The data report the profitability of publicly
traded firms by ‘business segment’, i.e., by particular industries defined by 4-
digit SIC codes. In our statistical analysis, we decompose the profitability of
business segments into effects associated with the year, the industry, the corpo-
rate parent, and the segment itself. We include business segments in our esti-
mates even if the Compustat Business-Segment Reports do not contain a
complete data series for 1981 through 1994. In some previous studies, busi-
nesses were included only if a full time series was available. We believe that the
prior approach may have introduced selection bias in the calculation of effects
and hence in estimates of persistence. Our objective is to include all useful infor-
mation to ensure better estimates.
82 S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )

By examining the historical influence of each effect on segments that

become high and low performers, we explore patterns in the emergence of
abnormal profits. We examine patterns in the sustainability of abnormal profits
by studying the influence of each effect on the subsequent profitability of ini-
tially high and low performers. The results address the trajectory of performance
for only those business segments that maintain their identity for at least six
years. There could be systematic differences in the influence of industry, corpo-
rate parent and segment-specific characteristics for those segments that main-
tain their identity and those that are sold, acquired, closed, or redefined. Our
analysis cannot address these differences due to the absence of data, and the
findings should be interpreted accordingly. Although the analysis is fundamen-
tally descriptive, it provides an empirical foundation for exploring the nature of
the underlying strategic and organizational processes.
This study differs from prior research in a number of respects. First, it
brings together research in two separate lines. In one line, authors have studied
the influences of industry and firm-specific effects on profitability but have
neither analyzed how the effects change over time nor examined high and low
performance separately. Schmalensee (1985) found that industry effects account
for about 20% of cross-sectional variation in the profits of the business units
covered in the Federal Trade Commission’s (FTC) 1975 Reports on American
manufacturers. Rumelt (1991) showed that business-specific effects have an
important influence on the profits of manufacturing firms covered in FTC
Reports for 1974 to 1977. Using data on a broader variety of economic sectors,
McGahan and Porter (1997) showed that year, industry, corporate-parent, and
business-specific effects all influenced abnormal profits among Compustat firms
over the period 1981 to 1994. The effects may differ for high and low per-
formers, however, and thus endure at different rates.
In the second line of research, authors have studied the persistence of profits
over time, measuring persistence in several particular ways. Such research either
did not examine differences in the emergence and sustainability of profits, or did
not study the influence of industry, corporate-parent, and business-specific
effects on emergence and sustainability. Several of the studies in this line
(Mueller, 1986; Waring, 1996; McGahan and Porter, 1999) employed a method
that requires careful interpretation because it defines persistence as the
endurance of the incremental components of effects. Mueller (1986) used FTC
survey results that spanned a 23-year period to demonstrate significant persis-
tence in the ranking and profitability of very high and very low performers.
Using this method, Waring (1996) demonstrated persistence in the corporate
profits of US firms. McGahan and Porter (1999) used the Compustat Business-
Segment Reports to show that industry effects persist longer than corporate-
parent effects, which in turn persist longer than business-specific effects.
McGahan and Porter (1999) compared differences among industry, corporate-
parent and business-specific effects in the persistence of their shocks – that is, of
incremental changes that could be identified with a specifc year under study. This
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 83

paper differs in that it deals with persistence in the entire influence of the indus-
try, the corporate parent, and the specific business, regardless of whether the
influence arose from a fixed effect or from a shock. As a result, the analysis in
this paper deals with a broader set of issues that arise in the development of
strategy and in the effectiveness of organization.
Cubbin and Geroski (1987) used different methods to show endurance in
the firm-specific effects of 217 British manufacturers. Ghemawat (1991)
showed sustainability in the profits of firms covered in the PIMS database. Here,
we separately examine how industry, corporate-parent, and business-specific
effects emerge and how they are sustained. By bringing together research in the
two lines, we aim to shed light on the underlying competitive processes that
shape strategy and organization.
Our research also differs from previous studies because it is based on a
dataset that covers a wide variety of economic sectors over a period that spans
several business cycles. The screened Compustat reports contain information on
agriculture and mining, construction, manufacturing, wholesale and retail
trade, transportation, entertainment and lodging, and the service sector. Many
prior studies, including Schmalensee (1985); Rumelt (1991); Mueller (1977,
1986), and Cubbin and Geroski (1987), covered only the manufacturing sector.
As a result, our findings are more general than those of previous authors.
Finally, our research differs from previous studies in that it includes infor-
mation on segments even if we do not have a series of data on them in every year
from 1981 to 1994. Unlike Mueller (1986) and Waring (1996), we include
businesses on which we do not have a complete series because they prove to
account for a significant portion of activity in the economy. Entries and exits in
the Compustat Business-Segment Reports include acquisitions, divestitures,
and reclassifications. Thus, they often do not represent actual entries and exits
by businesses and should not be interpreted as economic entry and exit. In
McGahan and Porter (1999), we present the results of a number of tests on
selection bias with regard to exits and entries in the Compustat data. Although
we find no evidence of differences in rates of entry by high and low performers,
we do find differences in rates of survivorship for high and low performers. This
difference may be attributable to the viability of operating units or to reporting
bias. Our findings apply only to publicly traded firms, not to all businesses in
the economy. As a consequence of including businesses without a complete
series, our dataset includes information on businesses that are more varied in age
than those examined by previous authors. Our results are also less subject to sur-
vivorship bias than those of previous studies.

Methods

Our analysis involves several steps. First, we calculate the mean profitability and
the year, industry, corporate-parent, and segment-specific effects for each
84 S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )

business segment in each year. Following McGahan and Porter (1999), we use
sequential, weighted-least-squares techniques to partition profits according to
the following model:
Ai,k,t = m + gt + ai,t + bk,t + fi,k,t (1)
In this equation, the dependent variable is the ratio in percent of operating
income to identifiable assets of the business segment in industry i and corpora-
tion k at time t. The terms m, gt, ai,t, bk,t, and fi,k,t represent the mean, the
year effect, the industry effect, the corporate-parent effect, and the segment-
specific effect, respectively. Industry, corporate-parent, and segment-specific
effects differ by year. There is no corporate-parent effect unless a corporation has
more than one segment.
In the sequence of the estimation, we introduce effects in the following
order: mean, year, industry, corporate-parent, and segment-specific. Thus, the
segment-specific effects are defined as the residual after introduction of the year,
industry, and corporate-parent effects. As a result, any reporting error that arises
systematically at the year, industry and corporate-parent levels also affects the
measured segment-specific effects. Measurement error in accounting profits and
anomalies in accounting conventions may introduce bias in our assessment of
any of the effects. For example, a change in statutory requirements on pension
plans in a particular year may influence all businesses in the economy and create
a year effect. Similarly, a change in the inventory method of a diversified firm in
a specific year may generate a corporate-parent effect in the year. We have no
hypothesis about the source or direction of bias, and therefore cannot assess its
impact.
Observations are weighted by assets in the calculation of each effect to
assure that effects are not skewed by the influence of small segments.
Rather than report the thousands of estimates, we report a nested ANOVA
on equation (1).1 Abnormal profits are defined for each segment in each year as
the difference between profit and the grand mean; that is, as r i,k,t = Ai,k,t –m =
gt + ai,t + bk,t + fi,k,t.
Second, we present a set of charts that show the path of average abnormal
profits in the emergence and sustainability of high and low performance. The
charts partition abnormal profits by type of effect for segments that eventually
become high and low performers, and for segments that once were high and low
performers. For the emergence analysis, a business segment is classified as a high
performer if the sum of its industry, corporate-parent, and segment-specific
effects (i.e., ai,t+bk,t+fi,k,t) is greater than the median for all segments in the
last year for which we have data on the segment, and as a low performer if
the sum (i.e., ai,t+bk,t+fi,k,t) is less than the median in the last year for which
we have data. The partitioning is based on the median rather than the mean
because the median identifies half of the segments as high performers, and half
as low performers. If we were to break the sample by using the mean, then
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 85

skewness in the distribution of performance would lead to different numbers of

business segments in the high and low cohorts.
For the sustainability analysis, a segment is classified as a high performer if
the sum of the three effects is greater than the median in the first year for which
we have a complete record on the segment, and as a low performer if the sum is
less than the median in the first year for which we have a complete record. (A
complete record contains lagged information; thus, the classification for the sus-
tainability analysis is based on profits in the second calendar year for which we
have data on the segment.)
Third, we examine statistically the importance of industry, corporate-
parent, and business-specific effects over the period from 1982 to 1994 (i.e.,
over the same period covered in the charts). The importance of each type of
effect is assessed from its average in the emergence and sustainability of abnor-
mal profits. The statistical assessment is based on the mean rather than the
median. This allows reports on whether differences are statistically significant.
Finally, we examine persistence as it has been defined in the literature using
a statistical approach developed initially by Mueller (1986). The approach is
based on the idea that estimates of persistence should not be overly influenced
by abnormal profits in the first observation on a business segment. If persistence
were estimated as the percentage of profits in a year that had arisen in the prior
year, then the estimate would be unduly affected by the arbitrary initial starting
point. Under Mueller’s method, ‘persistence’ is defined as the tendency of incre-
mental components of abnormal profit to endure rather than the total amount by
which abnormal profits endure between periods. The incremental component is
the portion of abnormal profit that is unique in a particular year. Because persis-
tence applies only to incremental components, the statistical approach generates
results that must be interpreted carefully. For example, a high persistence rate
may be associated with a small average incremental component to profitability.
Thus, the persistence of the incremental component may be largely irrelevant to
the tendency of profits to last between periods. The results of some previous
studies, especially Waring (1996), have been interpreted without sufficient
attention to this definition of persistence. The consequence is a mistaken infer-
ence about the importance of a persistent effect to the continuing performance
of a firm. We use this statistical approach because it has become the standard in
the literature and because it makes our results comparable to those of previous
authors. In our discussion of results, we attend to the idiosyncrasies of the per-
sistence definition. Overall, we rely on the charts to convey information about
the broader patterns by which abnormal profits last between periods.

Data

The complete Compustat Business-Segment Reports for 1981 to 1994 contain

151,929 records, each on a single business segment in a particular year. Since 15
86 S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )

December 1977 the Securities and Exchange Commission (SEC) has required
publicly-traded firms to report operating income, identifiable assets, and sales
by 4-digit SIC (i.e., by business segment). We exclude data prior to 1981
because of the high proportion of missing records in the reports for 1977–1980.
From the Compustat reports for 1981 through 1994, we eliminate records
without a primary SIC designation, and segments in SICs that are not identified
with actual industries (i.e., those described as ‘not elsewhere classified’, ‘non-
classifiable establishments’, or ‘government, excluding finance’). In addition, we
eliminate financial businesses designated as ‘depository institutions’ with SICs
in the 6000s, and segments that are the only organizations in their primary SIC
classifications in a particular year. Following previous authors, we exclude small
segments with sales less than $10 million or with assets less than $10 million
because their financial results are volatile and because they are often created for
the disposition of assets prior to exit, for example.
After screening and calculating lagged effects, our dataset contains 58,340
observations. These screens and characteristics of the dataset are the same as in
McGahan and Porter (1999). The screened dataset covers a substantial portion
of US economic activity, accounting for about half of non-financial corporate
sales and nearly a quarter of non-financial corporate assets reported to the
Internal Revenue Service (IRS) from 1985 to 1992. We have information on an
average of 4,488 business segments in each year, with 5.7 years of information
on average for each segment. The dataset covers 638 industries, 12,304 business
segments, and 7,005 corporations, of which 1,886 are diversified (i.e., report
information on more than one included segment). Diversified corporations in
our dataset participate in 2.6 business segments on average. The mean ratio of
operating income to identifiable assets among the business segments in the
screened sample is 9.94% with a variance of 248%. The average business seg-
ment has $901.2 million in identifiable assets. There is no statistically signifi-
cant difference in the average size or parent diversification level among
businesses with performance above and below the norm.
Because the average reported business segment is large, and because there
are just 2.6 measured segments per diversified parent, we believe that the aver-
age business segment is larger than a true operating business unit. There is
evidence that firms may aggregate activity beyond the true 4-digit level in their
reports on segments to limit disclosure and because a maximum of only 10 seg-
ments must be reported. The implied aggregation in reported business
segments suggests the pooling of activity that would otherwise be categorized
in more than one 4-digit SIC. Such pooling would tend to dampen industry
effects and enhance business-specific effects because we infer industry profits
from the overly diverse activities of segments categorized in the same 4-digit
SIC and assess the business-specific effects as the residual. If the business units of
diversified firms are related, the pooling also may dampen corporate-parent
effects because we infer the influence of the corporate parent from the profits of
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 87

member segments. Segment-specific effects may be diminished if the effects

of the different positions of a parent within true 4-digit industries are blurred
through aggregation. We interpret our results with these tendencies in mind.
We have at least six years of data on a total of 4,046 business segments,
which account for 65.2% of the total observations. McGahan and Porter (1999)
shows that relationships between aggregate measures of persistence are not sen-
sitive to the six-year exclusion rule. We confine our estimates of persistence to
these segments to exclude segments for which persistence estimates may be spu-
rious because of fluctuations in the business cycle, for example, and because our
methods require at least three years on a segment for calculation of a full set of
statistics. The average number of years per segment in our analysis of persistence
is 9.4.
The analysis reported in this paper deals with mobility between high and
low performance over time. Of the business segments, 36% are high performers
(i.e., rank in the top half of segments by profitability) at both the beginning and
the end of the period for which they are tracked; 14% begin as high performers
and end as low performers, 10% begin as low performers and end as high per-
formers, and 39% begin as low performers and end as low performers. The
remainder of the analysis investigates this pattern in greater depth.

Empirical results

In this section, we first present the results of a nested ANOVA to verify the
presence of year, industry, corporate-parent, and business-specific effects on high
and low performance. We then introduce and discuss charts that show the emer-
gence and sustainability of high and low performers. Finally, we present the
results of statistical analyses to confirm and expand on the regularities identified
in the ANOVA and in the charts.

The existence of effects

Table 1 shows the results of the nested ANOVA on equation (1), which we use
to verify the existence of the various effects. The analysis for all segments shows
that all types of effects are significant at the 1% level except corporate-
parent effects. Industry effects, while significant, may be understated because of
reporting error. We preserve corporate-parent effects in the analysis despite their
lack of significance in the aggregate because of their theoretical importance and
because McGahan and Porter (1997) and McGahan (1999) show a strong rela-
tionship between corporate-parent and industry effects. In a nested ANOVA,
McGahan and Porter (1997) show that the time-invariant portions of year,
industry, corporate-parent, and segment-specific effects add to incremental R2 in
nearly the same proportions as in Table 1.
 88
S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )
Table 1: Sequential analysis of variance of equation (1)a

All segments High performersb Low performersb

Source d.f. Incr. R2 F-Statc d.f. Incr. R2 F-Statc d.f. Incr. R2 F-Statc

Year 12 0.0044 21.24 12 0.0170 41.50 12 0.0037 8.88

Industry-year 6402 0.2130 1.73 5783 0.2959 1.22 5403 0.2254 0.99
Corporate-parent-year 10847 0.1883 0.91 8796 0.2997 0.73 7714 0.2278 0.61
Segment-specific-year 41078 0.5944 inf 14657 0.3873 inf 15961 0.5431 inf
Model 58340 1.0000 29249 1.0000 29091 1.0000
a Order of effects: year, industry-year, corporate-parent-year, segment-specific-year
b For this analysis, high performers and low performers are identified as segments with profits above and belowthe mean by year; segments with profits at the mean are identified as
high performers
c All F-statistics are significant at the 1% level except for the corporate-parent-year effects
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 89

Charting the emergence and sustainability of abnormal profits

Figure 1 shows the emergence of high and low performance in the sample. The
vertical axis represents the sizes of effects (as the contribution to the total ratio of
operating income to identifiable assets). The horizontal axis represents the num-
ber of years prior to the final year for which we have data on the segment. For
each segment, we lose the first observation in our statistical analysis of persis-
tence. Thus, we have a maximum of 13 years of data on each segment. Segments
are included in the calculations for the charts even if we do not have a complete
series of data on them. Thus, the plotted averages at the right side of the figure
(above year 0) represent the average of the effects in the last years for which data
is available. The information above the horizontal axis at the year marked 10, for
example, describes the averages for segments 10 years prior to the date (year 0)
at which they are classified as high performers. As a result of this approach,
the averages at the left side of the figure are calculated from fewer observations
than the averages at the right side of the figure. This occurs because segments
enter into the dataset continuously. At the bottom of the chart, we indicate the
number of observations for each year.
In the figure, the average abnormal profits among high and low performers
at year 0 are 9.3% and –11.8%, respectively. These numbers differ from each

Figure 1 The emergence of abnormal profits

90 S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )

other in absolute value for three reasons. First, high and low performers are iden-
tified by reference to the median profitability, not to mean profitability. Second,
various calendar years are represented in the calculation of averages at year 0.
Third, differences in size by assets can generate differences in the average
abnormal profits for high and low performers because the averages are weighted
by assets. One notable feature of the results in Figure 1 is that there is as much
net entry into the low-performing cohort as into the high-performing cohort
during the period, which suggests no systematic distortion associated with
entry. The total number of segments is about 4,000 in each of the final six years
because we restrict the dataset to include only those segments for which we have
six years of data. Most of the entry into the dataset occurs in years 6–12. The
number varies across the last six years because some segments are not observed
in the dataset for six continuous years. Entry can occur for a number of reasons,
including acquisition, the formation of a new company, the entry by an estab-
lished company into a new business, and changes in the conventions for report-
ing by an established company.
Figure 1 reveals several important empirical regularities. On average, a seg-
ment which is ultimately a high performer shows profits significantly greater
than the norm over the entire period for which data are available. Highly profit-
able segments have a history of high performance. This suggests that the causes
of high profit tend to endure and perhaps even accumulate. This result contra-
dicts the theory that high accounting profits occur subsequent to a period of low
or moderate accounting profits during which companies invest for later reward.
Instead, high performance among Compustat segments appears to result from
product- or factor-market imperfections that endure. The figure shows that the
average low performer shows profit slightly above the norm 12 years earlier.
Previously satisfactory performance is typical of segments that subsequently
become low performers. The Appendix contains an analysis of segments for
which we have a complete data series over the period from 1981 to 1994.
Survivorship is associated with higher profits for both high and low performers.
Figure 1 also shows that the influence of industry, corporate-parent, and
segment-specific effects differs markedly for high and low performers. The
influence of year effects is modest and nearly the same for high and low per-
formers, which suggests that the business cycle has little enduring effect on the
emergence of abnormal profits.
For high performers, participation in an attractive industry strongly con-
tributes to abnormal returns. This result supports theoretical explanations that
emphasize the interaction between industry structure and firm positioning as
crucial to the emergence of superior performance. Affiliation with a high-
performing corporate parent is also associated with the emergence of abnormal
returns, although to a somewhat lesser extent. This finding supports the idea
that a corporation’s ability to achieve superior performance through diversifi-
cation may be related to the attractiveness of the other industries in which it
competes. Empirically, industry effects are nearly as great as segment-specific
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 91

effects in the early development of high performance. Over time, segment-

specific effects become somewhat more important, which is consistent with the
idea that robust competitive positions build incrementally over time.
In contrast, low performance typically arises from business-specific effects
rather than from participation in an unattractive industry or from ownership by
a poorly performing corporate parent. Recall that the business-specific effects
may arise from any condition that affects performance relative to direct rivals
within a specific industry, and may be associated with superior insight,
changing external institutions, deteriorating resources, poor leadership, or poor
execution. This result means that the adverse effects on profitability associated
with industry events such as the deregulation of long-haul trucking or with cor-
porate events such as the acquisition of EDS by GM are not characteristic of the
average surviving low performer. Heileman’s experience in the brewing industry
is typical of low performers in the sense that its emerging poor performance
could be traced to its own loss of competitive position – hence a negative seg-
ment-specific effect. The importance of industry effects to high but not low per-
formers raises a question about how low-performing industries affect
profitability. The answer is that low performers tend to participate in both high-
profit and low-profit industries, but high performers tend to participate in
high-profit industries. Thus, the influence of the low-profit industries is offset
by the influence of the high-profit industries in the reported industry effect for
low performers. The strong influence of business-specific effects for low per-
formers in Figure 1 suggests that low performance may result from commit-
ment to untenable positions that are hard to reverse.
Figure 2 charts the subsequent sustainability of initially abnormal profits.
For this analysis, segments are classified as high or low performers based on their
profits in the earliest year for which we have data. Each successive point in the
figure represents the average performance of the high and low cohorts in a sub-
sequent year. Note that the averages represented at the left side of Figure 2 are
developed from more observations than those at the right side of the figure.
Thus, the sustainability chart is not symmetric to the emergence chart because
Figure 2 groups segments at year 0 based on their performance in the earliest
year, whereas Figure 1 groups segments at year 0 based on their performance in
the latest year. As a result, some of the segments that are classified as low per-
formers in the emergence analysis may be classified as high performers in the
sustainability analysis, and vice versa.
The change in number of observations for each year arises because of attri-
tion among segments in the dataset. Attrition may occur for a number of
reasons: bankruptcy or corporate merger may occur, companies may combine
segments in reporting, or a corporation may sell a business to another corpora-
tion that is represented in the dataset. The rates of attrition are greatest for years
6–12 because the data are constrained to include only those segments for which
we have more than six years of data. Note that attrition is just as great for high
performers as low performers, which may indicate that high performers are as
92 S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )

Figure 2 The sustainability of abnormal profits

likely to be sold as low performers. The results must be interpreted carefully to

apply only to those segments with continuity in identity over the period.
Figure 2 reveals several important empirical regularities about sustain-
ability. High performers show profits significantly greater than the norm 12
years later. A comparison of Figures 1 and 2 reveals that in both cases, high
profits are relatively enduring. In the Compustat sample, segments that are ini-
tially high performers report profitability that erodes somewhat faster than
profits emerge among segments that end up as high performers. This occurs
both because of exit and entry into the Compustat dataset and because of
changes in the profits of segments.
Figure 2 also indicates that initially low performing segments that continue
to report for 12 years eventually show profits that are near the norm. A compar-
ison with the emergence analysis indicates that the recovery from initial low
performance occurs more slowly than low performance develops. On average, a
business that is initially a low performer continues to post below-average
returns. Thus, although low performers often start as average performers, low
performance is difficult to reverse. Different causal mechanisms may be at work:
low performance may emerge from commitments to specialized assets that do
not create value for buyers, while low performance may endure because of
organizational inertia that prevents the redeployment of resources to more
productive uses.
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 93

In Figure 2, industry and corporate-parent effects are again significantly

larger for high performers than for low performers. Thus, the influence of indus-
try, corporate-parent, and segment-specific effects in Figure 2 is broadly consis-
tent with Figure 1, with one exception. In Figure 2, segment-specific effects are
initially large for high performers, but erode relative to industry and corporate-
parent effects. By contrast, Figure 1 shows that segment-specific effects remain
important to high performance in every year. On average, high performers
regress toward the mean almost exclusively because of diminished segment-
specific effects. Industry and corporate-parent effects are remarkably stable (see
also McGahan and Porter, 1999). In other words, the erosion of an advantageous
competitive position is the most significant threat to the sustainability of abnor-
mally high profits.
On average, low performance is corrected as segment-specific effects dimin-
ish. Thus, low performers tend to reduce the disadvantages in their competitive
positions over time, which is not surprising given that segment-specific effects
account for nearly all of low performance. A comparison of the results for high
and low performers indicates an asymmetry in how business-specific effects
change. This suggests fruitful avenues for further research into theoretical expla-
nations for why businesses can build competitive positions more easily than they
can dismantle positions.

Statistical analysis on the importance of industry, corporate-parent, and

segment-specific effects

We next use statistical methods to develop more precise findings. The analysis
confirms our previous results, and reveals additional regularities. To assess rela-
tive importance, we examined the average and standard deviation by type of
effect for high and low performers in the emergence and sustainability of abnor-
mal profits. Table 2 shows the results. The table indicates that:
• Large low performers show profitability closer to the norm than large high performers.
As a result, abnormal profitability is greater for high performers than low perform-
ers. The average abnormal profitability for high performers is greater in
absolute value than for low performers. This asymmetry is possible because
the effects are calculated by weighting each observation by the size of the
segment.
• Industry effects are especially important to high performance. Industry effects are
greater in size than any other type of effect in the sustainability of high per-
formance, and are nearly as large as segment-specific effects in the emer-
gence of high performance. Emerging high performers tend to participate
in structurally attractive industries, and sustained high performers benefit
from participation in attractive industries that remain attractive. Among
low performers, industry effects are considerably smaller in absolute value
than segment-specific effects in both cases.
 94
S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )
Table 2 Importance of industry, corporate-parent, and segment-specific effects on abnormal profitabilitya

Emergence Sustainability

High Performers Low Performers High Performers Low Performers

Effect Avg.b Std.D.b Avg.b Std.D.b Avg.b Std.D.b Avg.b Std.D.b

Abnormal Profitc 5.733 14.153 –3.850 12.268 4.041 15.316 –2.727 11.538
Year 0.121 1.224 0.182 1.214 0.128 1.219 0.179 1.218
Industry 2.099 6.974 –0.486 5.947 1.767 7.108 –0.320 5.787
Corporate-parentd 1.032 6.538 –0.090 5.642 0.765 6.522 0.118 5.627
Segment-specific 2.481 12.546 –3.456 11.743 1.381 13.210 –2.704 11.283
a Includes only segments for which we have at least six years of data
b Average and standard deviation over all years for which we have data
c Sum of year, industry, corporate-parent, and segment-specific effects
d Corporate-parent effects are reported as averages over all corporations, not just diversified corporations
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 95

• Although corporate-parent effects are consistently smaller in absolute value than seg-
ment-specific and industry effects, they contribute positively to high performance.
Among low performers, corporate-parent effects slightly offset other effects. Positive
corporate-parent effects contribute to both the emergence and sustainability
of high performance. High performers tend to belong to corporate parents
that positively contribute to profitability. On average, low performance is
slightly offset by positive corporate-parent effects, although corporate-par-
ent effects on low performers are so small as to be negligible.
• Segment-specific effects are important to both high and low performance, especially to
low performance. In the emergence and sustainability of high performance,
segment-specific effects account for 43% and 34%, respectively, of abnor-
mal profits on average. In the emergence and sustainability of low perfor-
mance, segment-specific effects account for 90% and 99%, respectively, of
abnormal profits on average. Thus, segment-specific effects are important to
high performance, and dominate low performance.
• The effects of the business cycle are not important in the emergence and sustainability
of high and low performance. High and low performance is not materially
influenced by year effects. On average, year effects influence both high and
low performers positively in our sample. (This result is possible because we
have more observations for years at the peaks of business cycles than at the
troughs of business cycles.)
See McGahan and Porter (1997) for an analysis that breaks down the contribu-
tion of industry, corporate-parent and business-specific effects for agriculture &
mining, manufacturing, transportation, wholesale & retail trade, lodging
& entertainment, and services. While the decomposition of variance in Table 2
confirms the results in Figures 1 and 2, it does not provide direct evidence on
the rates at which abnormal profits emerge and are sustained. The next section
provides a direct assessment of the relative rates of change in each type of effect.

Statistical analysis of the persistence in incremental components of effects

Figures 1 and 2 reveal that some types of effects appear to diminish over time
while others appear to accumulate. To explore these regularities, we use a widely
adopted statistical approach to measure the persistence in the incremental com-
ponents of effects. Specifically, we stipulate that for each observation, the year,
industry, corporate-parent, and segment-specific effects each consist of a fixed
component and an incremental component. We also stipulate that the sum of
these effects (which equals the abnormal profit) consists of a fixed component
and an incremental component. For example, the segment-specific effect, fi,k,t,
is assumed to equal fi,k+si,k,t. The fixed component, fi,k, is the average amount by
which an effect differs from zero in every year for which we have data on the
segment. The incremental component, si,k,t, is the additional amount (above or
below the fixed component) that arises in year t.
96 S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )

Following Mueller (1986), Waring (1996), and McGahan and Porter

(1999), we allow the incremental component of each effect to follow the first-
order autoregressive process si,k,t=rSS,i,ksi,k,t–1+ki,k,t where ki,k,t is a random shock
at time t. McGahan and Porter (1999) use the same specification to study per-
sistence without distinguishing the emergence of abnormal profits from the sus-
tainability of abnormal profits. This expression defines rSS,i,k, which is
the rate of persistence in the incremental components for segment i,k. The rate
of persistence may take on positive or negative values. Some algebraic substitu-
tion yields the following expression in fi,k,t, fi,k,t–1, and rSS,i,k, which we use to
estimate rSS,i,k:
fi,k,t = di,k + rSS,i,k f,k,t–1 + ki,k,t (2) 2
Equation (2) stipulates that fi,k,t follows a first-order autoregressive process3
with a drift captured by di,k. The fixed component of the segment-specific effect,
denoted fi,k, is equal to di,k/(1–rSS,i,k). Equations (3) through (6) show similar
relationships for the year, industry, and corporate-parent effects, and for the sum
of effects. For each equation, we assume that the error is independently drawn
from a distribution with mean zero and constant but unknown variance.
gt = a +rYR,i,k gt–1 + li,t (3)
ai,t = bi +rIN,i,k ai,t–1 + qi,t (4)
bk,t = ck +rCP,i,k bk,t–1 + zk,t (5)
ri,k,t = ti,k + ri,k ri,k,t–1 + «i,k,t (6)
Ordinary least squares (OLS) estimation of equations (2) through (6) yields
biased estimates of ri,k, rYR,i,k, rIN,i,k, rCP,i,k, and rSS,i,k because the errors associ-
ated with some years are correlated with dependent variables in other years.4 We
correct for this bias using a formula developed by Nickell (1981).5 With
Nickell’s correction, the OLS procedure generates an estimate of persistence by
type of effect for each of the business segments in our dataset. We obtain an
average by weighting each estimate by the inverse of its variance.6 The intuition
is that the average should reflect the precision with which the constituent esti-
mates are calculated. The inverse of the variance of each estimate accounts for
both the number of observations on the segment and the degree of fluctuation in
the incremental components for the segment. From the persistence estimates,
we impute the fixed and incremental components of each effect. A summary
measure of goodness-of-fit for each of the average persistence rates is obtained
from a weighted sum of the variances of the residuals in each of the regressions,
and from a weighted sum of the dependent variables in each of the regressions.7
In the statistical analysis, we eliminate spurious estimates by assessing per-
sistence only when we have at least six years of data on a segment. McGahan and
Porter (1999) contains a specification test to verify that estimates of persistence
are not sensitive to this criterion. Changes in the corporate ownership of a seg-
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 97

ment are treated as exits and subsequent entries. We do not assess the persis-
tence of profitability between years for any segment unless we have data on both
the current and previous year for the segment; as a consequence, we omit infor-
mation on a segment for the year of a change in corporate parent. Our results
should be interpreted carefully as applying only to those segments that retain
their identities for at least six years, and not to all segments in a particular year.
Table 3 shows the results. Panel (a) shows statistics on the emergence of
abnormal profits, and panel (b) shows statistics on the sustainability of abnormal
profits. The first section of each panel shows the average estimates of the fixed
(i.e., fi,k) and incremental components of each effect (si,k,t). The second section of
each panel shows estimated rates of persistence in the incremental components
and the third section shows goodness-of-fit of the persistence estimates.
In several instances, the estimated rate of persistence in the sum of effects is
greater than the estimated rates for each of the constituent effects. For example,
in panel (b), the estimated rate of persistence among high performers in the sum
of effects is 80.6%, which is greater than each of the estimates for the year,
industry, corporate-parent, and segment-specific effects of 65.8%, 76.6%,
71.5%, and 70.8%, respectively. This outcome is possible because the fixed and
incremental components of effects vary asynchronously, so that aggregation
across effects can yield greater stability than in each of the constituent effects.
Mann–Whitney tests indicate that the means are significantly different at the
99% level for high and low performers in every case except for the following: the
year effects, and the incremental component of the corporate-parent effects.
Note that the fixed component of the corporate-parent effects is significantly
different for high and low performers, and that the rates of persistence in the
corporate-parent effects are significantly different for high and low performers.
Thus, we draw our conclusions about the year, industry, corporate-parent, seg-
ment-specific and overall abnormal profits with confidence of their statistical
robustness. In supplementary analyses on both the emergence and the sustain-
ability of abnormal profits, we found no significant correlation between the
fixed components, the persistence rates, the sizes of businesses, and the amount
of diversification.
The results in panel (a) suggest the following regularities in the emergence of
abnormal profits for businesses that maintain their identities for at least six
years:
• On average among businesses that retain their identities for six or more years, low per-
formance accumulates faster than high performance. Panel (a) shows a persistence
rate in abnormal profits for low performers that is greater than for high per-
formers (82.1% vs. 77.4%). This occurs because the incremental compo-
nent of high performance is small on average; high performance is largely
fixed. Persistence in the incremental component is not as relevant to the
emergence of high performance as low performance. The incremental com-
ponents of low performance accumulate over time on average.
Table 3 Estimates of persistence in the incremental components of effects

98
(a) The Emergence of Abnormal Profitsa

Year Industry Corp.-Par.b Seg-Spec. Ab. Profits

High Low High Low High Low High Low High Low

S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )
Average effect c 0.121 0.182 2.099 –0.486 1.032 –0.090 2.481 –3.456 5.733 –3.850
Fixed component d 0.043 –2.249 8.180 0.128 7.673 –11.920 –18.365 –4.445 11.733 –8.794
Incr. component d 0.727 2.989 –5.724 –0.635 –6.735 12.573 20.136 0.487 –1.680 5.752
Estimated persistence rates
Average d 0.642 0.658 0.784 0.744 0.638 0.787 0.717 0.592 0.774 0.821
Standard deviation e 0.005 0.009 0.002 0.006 0.002 0.058 0.020 0.021 0.000 0.000
Goodness of fit f 0.479 0.475 0.880 0.680 0.698 0.362 0.795 0.911 0.936 0.459

(b) The Sustainability of Abnormal Profitsa

Year Industry Corp.-Par.b Seg-Spec. Ab. Profits

High Low High Low High Low High Low High Low

Average effect c 0.128 0.179 1.767 –0.320 0.765 0.118 1.381 –2.704 4.041 –2.727
Fixed component d –0.644 –1.966 1.745 7.369 -0.460 –4.697 –2.730 –25.662 –0.742 10.792
Incr. component d 1.377 2.739 0.444 –7.436 1.535 5.232 1.916 24.146 4.926 –10.868
Estimated persistence rates
Average d 0.661 0.643 0.760 0.771 0.643 0.781 0.620 0.712 0.808 0.753
Standard deviation e 0.004 0.006 0.002 0.003 0.028 0.024 0.029 0.015 0.000 0.000
Goodness of fit f 0.472 0.482 0.759 0.873 0.428 0.624 0.809 0.903 0.915 0.457

a Includes only segments for which we have at least six years of data
b Corporate-parent effects reported as averages over all corporations, but corporate-parent persistence rates are reported only for diversified corporations
c The averages of the fixed and incremental components do not sum to the average effect because the components are weighted to reflect the precision of persistence estimates
d Weighted by the inverse of the variance on each estimate
e Weighted by the inverse of the variance on the standard errors
f This measure is the weighted average of the R2 in the OLS regression on each segment; the weights are the inverses of the standard errors
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 99

• Industry effects accumulate over time for the average high performer but not for the aver-
age low performer. In panel (a), the average incremental component of the
industry effect is negative for both high and low performers. Persistence
in the negative incremental component for high performers indicates that
industry effects diverge from the economic mean on average. Persistence in
the negative incremental component for low performers indicates that indus-
try effects for these firms converge toward the economic mean on average.
Thus, industry effects contribute to the emergence of high performance but
not to the emergence of low performance. The incremental components of
industry effects are also more persistent for high performers than low per-
formers.
• Corporate-parent effects accumulate and contribute to divergence from the mean for
both high and low performers. In panel (a), the incremental components of the
corporate-parent effects take the same signs as the incremental components
of the sum of effects. Thus, corporate-parent effects contribute to the diver-
gence of abnormal profits from the mean. Corporate-parent effects on high
performers are about half as large as industry effects on average, and persist
at a lower rate than industry effects. Corporate-parent effects on low per-
formers are smaller than any other type of effect (including year), although
estimated rates of persistence in the corporate-parent effects on low per-
formers show greater variability than any other type of effect.
• The incremental components of the segment-specific effects persist at a greater rate for
high performers than low performers on average. In the emergence of high perfor-
mance, segment-specific effects diminish over time on average (given the
positive incremental component on average). Panel (a) also shows a positive
incremental component of the segment-specific effect for low performers,
which indicates that segment-specific effects diverge from the mean for the
average low performer. In panel (a), incremental segment-specific effects
persist at 71.7% for high performers and 59.2% for low performers.
Panel (b) shows persistence rates in the sustainability of high and low perfor-
mance for businesses that maintain their identities for at least six years. The
results indicate:
• On average among segments that retain their identities for six or more years, the incre-
mental components of abnormal profits deteriorate at about the same rate for high and
low performers on average. Panel (b) shows a persistence rate in the sum of
effects for high performers that is about the same as for low performers
(80.6% vs. 81.34%). In the sustainability of abnormal profits, the erosion
of the incremental components of high performance occurs at about the
same rate as the erosion of the incremental components of low performance.
The competitive processes that cause regression to the mean in high perfor-
mance appear to have similar effect to the corrective processes that cause
regression to the mean in low performance.
100 S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )

• In the sustainability of abnormal profits, industry effects contribute to convergence

toward the mean for the average segment. In panel (b), the average incremental
component of the industry effect is positive for high performers and nega-
tive for low performers. Thus, industry effects on both high performers and
low performers tend to diminish in absolute value over time, although the
gross impact of industry effects is considerably less for low performers than
for high performers. Average persistence in the incremental components of
industry effects is similar for high and low performers.
• Corporate-parent effects contribute to convergence toward the mean for high performers,
but do not contribute to convergence for low performers on average. In panel (b), the
average incremental component of the corporate-parent effect is positive for
both high and low performers. Thus, corporate-parent effects contribute to
convergence for high performers, but not for low performers.
• On average, segment-specific effects diminish over time for high performers. For the
average low performer, segment-specific effects accumulate rather than diminish.
Panel (b) shows positive average incremental components of the segment-
specific effect for both high and low performers. Thus, segment-specific
effects diminish over time on average for high-performing segments. For
the average low performer, however, the segment-specific effect accumulates
over time on average.
A supplementary analysis on results by sector does not generate information on
statistically significant differences in persistence rates because of high variance
on the persistence estimates. The estimated persistence rates by sector have high
variance both because there are small numbers of observations in some sectors
and because we incorporate into the variance estimates a correction that
accounts for the fact that the rates are themselves built from estimated para-
meters with non-zero variances. We did find significant differences in the effects
for businesses belonging to single-segment and diversified companies, however.
Significant differences for single-segment and multi-segment firms arise in the
fixed components of the industry effects and in the persistence rates of the busi-
ness-segment effects. Single-segment businesses tend to have higher fixed com-
ponents of industry effects but somewhat lower persistence rates in
business-specific effects. This may arise because single-segment businesses parti-
cipate in more attractive industries while diversified firms have greater scope for
enhancing favorable influences on positioning. These findings point to opportu-
nities for further research on the structural and behavioral differences across
focused and diversified firms that give rise to the emergence and sustainability
of abnormal profits.

Conclusion

In this study, we examine the behavior of profitability among business segments

covered in the Compustat Business-Segment Reports for 1981 to 1994 and that
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 101

retain their identities for at least six years. Our purpose is to report on empirical
regularities in the influence of year, industry, corporate-parent and business-
specific effects on high and low performance over time. This study differs from
prior research by examining differences in the emergence and sustainability of
abnormal profits, by examining changes in the influence of effects over time,
and by distinguishing differences among high and low performers.
We chart the emergence and sustainability of high and low performance
among a large group of firms engaged in the US economy, and corroborate the
results using statistical methods. In analyzing changes over time, we distinguish
between the incremental and fixed components of abnormal profit and note that
the standard measure of persistence applies only to the incremental components.
We are unable to explore the differences in industry, corporate parent, and busi-
ness-specific factors between business segments that retain their identity for six
years and those that are sold, acquired, closed, or redefined. The persistence rates
reported in this paper apply only to the incremental component of the effects for
business segments with at least six years of longevity. Some previous studies
have reported persistence estimates that have been interpreted more generally.
Because the incremental components of abnormal profit may be quite small,
this misinterpretation can lead to incorrect inferences about the emergence and
sustainability of abnormal profits. To best convey our results, we interpret the
persistence estimates in context of the charts presented earlier.
The analysis reveals the following broad regularities about the process of
competition:
1 Industry effects are more important than business-specific and corporate-
parent effects in the sustainability of high performance. Business-specific
effects are more important than industry and corporate-parent effects in the
emergence and sustainability of low performance, and in the emergence of
high performance.
2 Industry and corporate-parent effects are more important on average to high
performance than to low performance. Business-specific effects are more
important on average to low performance than to high performance.
3 On average, high performance is preceded by high performance, whereas
low performance is preceded by average performance.
4 High and low performance erode at about the same rate.
While our analysis is descriptive, the results are consistent with fundamental
principles of strategy and organization. The emergence of low performance is
distinguished not by a history of poor performance but by the onset of an
adverse competitive positioning in a situation where industry and corporate-
parent effects are neutral on average. The absence of negative industry effects on
low performance suggests that low performance is not triggered by competitive
imitation. Instead, low performance apparently emerges when a competitor that
initially generated an average return suffers from an eroding position relative to
rivals.
102 S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )

On average, the emergence of high performance reflects a combination of

several circumstances: creating a business-specific position with benefits that are
preserved over time; participation in an attractive industry; and the beneficial
influence of a corporate parent. High performance is not normally associated
with mastering profit-dissipating rivalry in an unattractive industry.
On average, the industry and corporate-parent effects of high performers are
more sustainable than their business-specific advantages. This may mean that
positioning effects due to strategy and to organization are exposed to competi-
tive threats. Given the enduring nature of the positive industry effect on the
average high performer, it appears that high performers do not resort to profit-
dissipating rivalry to fend off imitation, perhaps in the interest of preserving
industry structure.
On average, low performance is sustained for over a decade. The longevity
and importance of segment-specific effects suggests that low performers adopt
losing strategies or adopt management practices that are hard to modify.
Candidate explanations include organizational inertia and sunk-cost effects that
lock a company into a losing strategy.
There is a stark contrast between sustaining high performance and the
emergence of low performance. Emergent low performers appear to be locked
into their adverse competitive positions and find it hard to recover their profit-
ability. Sustained high performers appear to preserve industry and corporate
benefits as their distinctiveness within a business erodes over time. When a high
performer can preserve its distinctive business-specific effect without harming
industry structure and without harming other corporate members, then the
high performance appears to accumulate. In sum, a distinctive idiosyncratic
business-specific effect appears to be necessary for the emergence of superior
profitability. Overall, however, sustained high profitability apparently follows
from the decision to preserve industry attractiveness and corporate structure
rather than to obtain a fully distinctive business-specific advantage without
regard for the industry and corporate consequences.
This paper also points to opportunities for further research. While the
descriptive analyses in this paper do not test theoretical hypotheses, the results
are broadly consistent with theory from a number of disciplines and perspec-
tives. One major avenue for research lies in identifying the causal mechanisms
that explain why superior performance tends to emerge quickly while inferior
performance emerges slowly and tends to persist. Sophisticated studies of
mobility are needed to identify structural features associated with the emergence
and sustainability of high and low performance (see Furman and McGahan,
2002 as an example of one study that takes on the question of whether perfor-
mance is predictable).
A second area for further research involves questions about strategic choice.
How can a firm make strategic decisions that enhance superior performance and
curtail inferior performance? Further research on these questions may require
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 103

enhanced methods that account for complex relationships between the esti-
mated effects (for example, see Hambrick and D’Aveni, 1988).
The results also indicate important relationships in the causal explanations
favored by theorists aligned with the industrial-organization view, the resource-
based view, property rights explanations, transaction-cost economics, and orga-
nizational ecology. Each theoretical lens may be particularly powerful at a
specific point in the evolution of a business. The greatest promise for further
theoretical study may be in integrating the various explanations and in demon-
strating how they can be related to a broader framework that links strategic and
organizational choices to the industry, corporate, and business-specific context.

Appendix: the emergence and sustainability of abnormal profits

among segments with a complete series

The central analyses in our paper raise several questions about patterns of emergence and
sustainability for newly tracked segments and segments on which we have a complete
series of data. In the emergence analysis, for example, the pattern of high performance
suggests that newly tracked segments may enter at the same high rate of profitability as
segments on which we have a complete series of data. In this Appendix, we report
the analyses for segments on which we have complete series and indicate differences in
the results. A principal finding is that the segments on which we have a complete series
have higher profits on average than those on which we do not have a complete
series. Otherwise, the broad patterns reported in the body of the paper for the entire
sample also hold for the segments that are analyzed in this Appendix.
In Figure 1, high performers show profits significantly greater than the norm 12
years prior to the date on which they are classified as superior. This regularity raises a
question about whether high performance is immediately evident among segments that
enter the Compustat dataset during the period under study. Figure A1 shows the emer-
gence of high and low performance for only those segments on which we have a com-
plete series. The pattern for high performers is strikingly similar to the pattern for the
full cross-section of high performers. For low performers, however, the pattern is some-
what different. The comparison suggests the following relationships for the average
segment as abnormal profits emerge:
• At the time that a segment is first tracked by Compustat, its abnormal profits are
about the same as the abnormal profits of mature performers in its cohort.
• The low performance that emerges among newly tracked businesses is considerably
more severe than the low performance of mature businesses.
Figure 2, which shows the sustainability of high and low performance, raises the possi-
bility that abnormal profits steadily erode for surviving segments. There is ambiguity in
Figure 2 because surviving segments are grouped with those no longer tracked. (Recall
that segments may be dropped from Compustat because of exit, privatization, merger of
the corporate parent, sale of the division, or reclassification of the division to another
104 S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )

Figure A1 The emergence of abnormal profits among segments with a complete series

SIC.) Figure A2 shows the sustainability of high and low performance for only those
segments on which we have a complete series. A comparison suggests the following rela-
tionships:
• ‘Exit’ among the high performers in the Compustat database occurs after a greater
than average decrease in abnormal returns.
• The high performance of businesses that do not exit is initially better than the aver-
age for all high performers.
• The low performers in the Compustat database that eventually exit show worse
profits than the average among low performers.
• The low performance of businesses that do not exit is initially better than the aver-
age for all low performers.
Survivorship is associated with higher than average profitability for both high and low
performers. The longevity of low performance in both cases suggests that low performers
are locked into their poor positions for several years, making exit difficult. Even long-
tracked low performers recover only to average profitability during the 12-year period
subsequent to their initial classification.
Among high performers, those that exit the Compustat database have seen a large
portion of their abnormally high profits erode prior to their exit. Thus, there is an
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 105

Figure A2 Sustainability of abnormal profits among segments with a complete series

asymmetry between patterns of entry and exit for high performers. Whereas newly
tracked high performers show profits comparable to those of mature high performers,
exiting high performers are less successful.
A comparison of Figures 2 and A2 reveals a higher segment-specific effect among
long-tracked high performers compared to` the average high performer. The favorable
segment-specific effects of exiting high performers diminish quicker than those of sur-
viving high performers. By contrast, the adverse segment-specific effects of exiting low
performers appear to be larger than those of surviving low performers. Figures 2 and A2
suggest the following hypotheses about the sustainability of abnormal profits for the
average business:
• Surviving high performers are distinguished by particularly favorable segment-
specific effects, suggesting an unusually distinctive position.
• Surviving low performers are distinguished by less adverse segment-specific effects
compared to the average low performer.
One interpretation that is consistent with these patterns is that exit from Compustat
occurs for high performers in attractive industries and attractive parents after their com-
petitive positions erode. While exit is a viable option for such segments, it may not be
viable for the typical low performer. Instead, low performers may be forced to reposition
within their industries over a period of several years.
106 S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )

Notes

We are grateful to Raffi Amit, Pankaj Ghemawat, Belen Villalonga, Geoff Waring, three anony-
mous referees, the editors, and to participants in seminars at the Academy of Management,
Harvard, INSEAD, Stanford, and the University of California at Berkeley for comments and dis-
cussions related to this paper. Special thanks to Arthur Schleifer for extensive discussions about
statistical methods. Thanks to Todd Eckler, Jan Rivkin, and Sarah Woolverton for help in extract-
ing and assimilating data. Boston University’s Systems Research Center, the Boston University
Institute for Leading in a Dynamic Economy (BUILDE) and the Division of Research at the
Harvard Business School provided generous financial support for this project.

1 Equation (1) is not estimable through a sequential ANOVA because the model is fully speci-
fied. Full specification is necessary to capture intertemporal changes in effects. See McGahan
and Porter (1997) for a nested ANOVA on a model in which year, corporate-parent, and
business-specific effects do vary by year.
2 Note that equation (2) can be derived from the following expressions: si,k,t=rSS,i,ksi,k,t–1+ki,k,t
and fi,k,t=fi,k + si,k,t. The derivation requires substitution of si,k,t–1=fi,k,t–1–fi,k into
si,k,t=rSS,i,ksi,k,t–1+ki,k,t, and substitution of the resulting expression for si,k,t into fi,k,t=
fi,k + si,k,t.
3 We adopt the notation that an underlined subscript indicates the average over the obser-
vations denoted by the subscript. There is no information in either the ratio
(fi,k+rSS,i,ksi,k,t–1)/(fi,k+si,k,t) or (di,k+rSS,i,k fi,k,t–1)/fi,k,t because si,k,t=rSS,i,ksi,k,t–1 and fi,k,t=
di,k+rSS,i,k fi,k,t–1; thus, (fi,k+rSS,i,ksi,k,t–1)/(fi,k+si,k,t)=(di,k+rSS,i,k fi,k,t–1)/fi,k,t=100%.
4 Consider, for example, equation (2) for business segment i,k in 1989. The equation stipu-
lates that fi,k,89 is a function of ki,k,89. Now consider equation (2) for the same segment in
1990. This equation indicates that fi,k,90 is a function of fi,k,89, but we cannot claim that the
two records are independent, because fi,k,89 is correlated with ki,k,89 in the prior record.
Thus, the estimate of the coefficient on fi,k,t is systematically biased.
5 See Waring (1996) and McGahan and Porter (1999) as well as Nickell (1981). We con-
ducted several simulations to verify Nickell’s approach, and found that it performed quite
well except for extreme OLS estimates. We therefore limited the Nickell corrections to
absolute values less than or equal to one. Thanks to Geoff Waring for discussions about this
approach.
6 The variance of each of the OLS estimates of persistence, which we here denote rx,i,k (where x
may signify year, industry, corporate-parent, segment-specific, or the sum of effects) is given
by var(rx,i,k)=
[var(xi,k,t)–est(rx,i,k)2var(xi,k,t–1)]/[(n–2)var(xi,k,t–1)] where n represents the number of years of
data for the segment.
7 In the calculation of goodness-of-fit, the weights are the inverses of the variances of the sam-
pling variances, which we obtain from Var(vx,i,k)=2Fx4/(n–1) and est(Fx2)=
(n Var(xi,k,t)–est(rx,i,k)2 n Var(xi,k,t–1))/(n–2), where Fx is the standard deviation of the popula-
tion parameter rx,i,k and vx,i,k is the sampling variance of the estimate of rx,i,k (see Hald,
1952: 203 and Kendall, 1952: 229). Thanks to Arthur Schleifer for this derivation.

References

Cubbin, J. and Geroski, P. (1987) ‘The Convergence of Profits in the Long Run: Inter-firm and
Inter-industry Comparisons’, Journal of Industrial Economics 35(4): 427–42.
 M C G A H A N & P O RT E R : A B N O R M A L P RO F I T S 107

Furman, Jeffrey L. and McGahan, Anita M. (2002) ‘Turnarounds’, Managerial and Decision
Economics 23(4–5): 283–300.
Ghemawat, Pankaj (1991) Commitment: The Dynamic of Strategy. New York: Free Press.
Hald, A. (1952) Statistical Theory with Engineering Applications. New York: Wiley & Sons.
Hambrick, Donald C. and D’Aveni, Richard (1988) ‘Corporate Failure as a Downward Spiral’,
Administrative Science Quarterly 33(April): 1–23.
Kendall, Maurice G. (1952) The Advanced Theory of Statistics, Vol. I, 5th edn. London: Charles
Griffen & Co.
McGahan, Anita M. (1999) ‘The Performance of US Corporations: 1981–1994’, Journal of
Industrial Economics 47(4): 373–98.
McGahan, Anita M. and Porter, Michael E. (1997) ‘How Much Does Industry Matter, Really?’,
Strategic Management Journal 18(4): 15–30.
McGahan, Anita M. and Porter, Michael E. (1999) ‘The Persistence of Shocks to Profitability:
Comparing the Market-Structure and Chicago Views’, Review of Economics and Statistics 81(1):
43–53.
Mueller, Dennis C. (1977) ‘The Persistence of Profits Above the Norm’, Economica 44 (November):
369–80.
Mueller, Dennis C. (1986) Profits in the Long Run. Cambridge: Cambridge University Press.
Nickell, Stephen (1981) ‘Biases in Dynamic Models with Fixed Effects’, Econometrica 49(6):
1417–26.
Rumelt, Richard (1991) ‘How Much Does Industry Matter?’, Strategic Management Journal 12(3):
167–85.
Schmalensee, Richard (1985) ‘Do Markets Differ Much?’, American Economic Review 75(3):
341–51.
Waring, Geoff (1996) ‘Industry Differences in the Persistence of Firm-Specific Returns’, American
Economic Review 86(5): 1253–1265.

Anita M. McGahan is Professor and Chairman of the Strategy & Policy group at the
Boston University School of Management and a Senior Institute Associate at the Harvard
Institute on Strategy and Competitiveness. She has experience at Morgan Stanley and
McKinsey & Company, and was recently named one of five international experts on the
strategic use of technology by CIO magazine. McGahan is the author of over 60 articles and
case studies on industry transformation and competition, and is currently writing a book
entitled How Industries Evolve. Recent publications include ‘What Do We Know About
Variance in Accounting Profitability?’ with Michael E. Porter in Management Science (July 2002)
and ‘Turnarounds’ with Jeffrey L. Furman in Managerial and Decision Economics (June–August,
2002). Address: 595 Commonwealth Avenue, Boston, MA 02215, USA. [email:
[email protected]]

Michael E. Porter is the Bishop William Lawrence University Professor at Harvard

University. He is a leading authority on competitive strategy and the economic development
of nations, states, and regions. The author of 16 books and over 85 articles, Porter’s ideas
have guided economic policy throughout the world. Competitive Strategy: Techniques for
Analyzing Industries and Competitors, published in 1980, is in its 58th printing and has been
translated into 17 languages.The 1990 book, The Competitive Advantage of Nations, motivated
by his appointment in 1983 to the President’s Commission on Industrial Competitiveness,
108 S T R AT E G I C O R G A N I Z AT I O N 1 ( 1 )

launched his second major body of work on competitiveness and economic development.
Porter’s third major body of work deals with the relationship between competition and
society.The holder of eight honorary doctorates, Porter has won numerous awards for his
books, articles, and public service in several fields. Address: Soldiers Field, Boston, MA 02163,
USA. [email: [email protected]]