Discontinuities and Earnings Management: Evidence from Restatements Related to Securities Litigation.

Discontinuities and Earnings Management: Evidence from Restatements Related to Securities Litigation.
Rev Account Stud (2014) 19:1191–1224
DOI 10.1007/s11142-014-9286-7
Inter-industry network structure and the crosspredictability of earnings and stock returns
Daniel Aobdia • Judson Caskey • N. Bugra Ozel
Published online: 29 April 2014
Ó Springer Science+Business Media New York 2014
Abstract We examine how the patterns of inter-industry trade flows impact the
transfer of information and economic shocks. We provide evidence that the intensity
of transfers depends on industries’ positions within the economy. In particular, some
industries occupy central positions in the flow of trade, serving as hubs. Consistent
with a diversification effect, we find that these industries’ returns depend relatively
more on aggregate risks than do returns of noncentral industries. Analogously, we
find that the accounting performance of central industries associates more strongly
with macroeconomic measures than does the accounting performance of noncentral
industries. Comparing central industries to noncentral ones, we find that the stock
returns and accounting performance of central industries better predict the performance of industries linked to them. This suggests that shocks to central industries
propagate more strongly than shocks to other industries. Our results highlight how
industries’ positions within the economy affect the transfer of information and
economic shocks.
Keywords Information transfer Inter-industry networks Aggregate
risk Earnings Stock returns
JEL Classification
D57 G14 M41
D. Aobdia
Kellogg School of Management, Northwestern University, Evanston, IL, USA
J. Caskey (&)
McCombs School of Business, University of Texas at Austin, Austin, TX, USA
e-mail: [email protected]
N. B. Ozel
UCLA, Anderson School of Management, Los Angeles, CA, USA
123
1192
D. Aobdia et al.
1 Introduction
The accounting literature has long recognized the role of information transfers
among firms in forming expectations about earnings and returns. Early studies in
this area document within-industry information transfers, consistent with the notion
that similar companies face similar economic shocks (e.g., Foster 1981; Clinch and
Sinclair 1987). More recent studies document information transfers via customer/
supplier links (e.g., Pandit et al. 2011). We focus on the effects of industries’
positions in the economy-wide flow of trade and examine how inter-industry trade
affects the transmission of economic shocks as well as the transmission of
information about those shocks. Our evidence suggests that transfers of economic
shocks and information depend not only on industries’ immediate trading partners
but also on their positions within the overall economy.1
Our analysis views industries through the lens of the economy-wide trade
between customers and suppliers. An industry’s role in propagating economic
shocks depends not only on its size but also on the extent to which it interacts with
different economic sectors as both a customer and a supplier (Acemoglu et al.
2012). We place particular attention on industries that form hubs in the flow of trade
(‘‘central industries’’). An industry’s role as a hub reflects both its direct trading
relationships—its immediate customers and suppliers—and its indirect trading
relationships with customers-of-customers, suppliers-of-suppliers, and so forth. Our
objective is to document whether and how an industry’s position in trade flows
influences (1) the relative importance of factors that affect the industry’s financial
performance, and (2) the extent to which an industry’s performance is associated
with the performance of its trading partners.
We predict that an industry’s position in the flow of trade impacts that industry’s
exposure to economic shocks, and hence its financial performance. We also predict
that an industry’s position influences its role in the transmission of shocks to other
industries, which, in turn, affects the cross-predictability of financial performance.
We provide evidence that the performance of central industries depends relatively
more on aggregate risks than does the performance of noncentral industries. We also
provide evidence that shocks to central industries have a stronger association with
shocks to a portfolio of their trading partners. This second set of tests is consistent
with either shocks originating in central industries propagating outward, or with
shocks originating outside of central industries affecting the central industries’
performance only if the shock has a broad impact on its trading partners.
As an example of how an industry’s position can affect its impact on the
economy, consider the automobile industry. Based on the Bureau of Economic
Analysis’ (BEA) data on U.S. gross domestic product (GDP), consumer expenditures on ‘‘vehicles and parts’’ comprise around 3 % of GDP, which is slightly more
than, but comparable to, the contribution of ‘‘clothing and footwear’’, and
consistently smaller than ‘‘food services and accommodations.’’ As evidence that
the auto industry occupies a relatively important position in the economy, consider
1
Later in the introduction, we discuss why we conduct our analysis at the industry-level rather than the
firm-level.
123
Earnings and stock returns
1193
that a strike at a General Motors plant in Flint, Michigan—an idiosyncratic shock to
the auto industry—led to a measureable impact on GDP (Montgomery and Vames
1998). Following the 2007–2008 financial crisis, the auto industry received a federal
aid package largely because ‘‘the government believed it could not afford to let the
[auto] industry fail’’ (Shepardson 2009). While the economic shock of the financial
crisis did not originate in the auto industry, politicians acted on the belief that the
industry could significantly amplify the crisis.
Our first set of findings indicates that central industries primarily face systematic,
rather than idiosyncratic, risk. This could stem from either central industries
obtaining some diversification by virtue of their exposure to wide swaths of the
economy, or, as predicted by Acemoglu et al. (2012) and examined in a
contemporaneous paper by Ahern and Harford (2013), because central industries
originate systematic risks. We estimate the R2s from one-factor (CAPM) and threefactor (Fama–French) regressions as gauges of the extent to which aggregate risks
explain stock returns. We then regress these R2s on explanatory factors, including
the industry’s centrality, market capitalization, average analyst coverage, and
average trading volume. The R2 tests indicate that aggregate risks explain a greater
portion of central industries’ returns than of noncentral industries’ returns,
consistent with central industries primarily facing systematic risks. We note that
central industries’ factor loadings resemble those of noncentral industries which,
combined with the central industries’ higher R2s, implies that central industries’
returns are less volatile due to having lower idiosyncratic risk.
We also conduct analogous tests where we first regress seasonal change in
industry-level return on assets (ROA) on a vector of macroeconomic statistics and
then examine the relation between centrality and the resulting R2s. Similar to the
returns tests, we find that central industries have higher R2s in regressions of ROA
on industrial production and interest rate variables. In regressions that include a
broader set of indicators, such as unemployment and housing starts, the association
between centrality and the R2s remains positive, but the statistical significance
depends on how we aggregate ROA to the industry level. The weaker association
between centrality and the R2s in the latter analysis is consistent with the tradebased centrality measure primarily relating to inter-corporate activity, rather than
overall economic activity.
Our second set of analyses examines how centrality impacts transfers of
information and economic shocks. Menzly and Ozbas (2010) find that industries’
ROA and stock returns can forecast the ROA and returns of the industries that they
trade with. Our tests show how centrality impacts these relations.
For each industry, we examine the association between its seasonal change in
quarterly ROA and the concurrent and one-quarter-ahead ROA of the industries it
trades with (linked industries). We find that the association between central
industries’ ROA changes and ROA changes of the industries they trade with is over
two times greater than that of noncentral industries. The concurrent ROA results are
consistent with both central industries initiating shocks that have a greater tendency
to propagate and with central industries being subject to correlated shocks to their
trading partners. The one-quarter-ahead ROA results are more suggestive of shocks
123
1194
D. Aobdia et al.
to central industries propagating outward; however, our empirical design cannot
isolate the source of the shocks.
We also test the relation between industries’ monthly stock returns and their
linked industries’ concurrent and 1-month-ahead stock returns. Central industries’
returns have a stronger relation with their linked industries’ concurrent returns than
do noncentral industries. This is consistent with the central industries and their
linked industries experiencing similar shocks, whether these originate with the
central industry or represent correlated shocks to their linked industries. The
1-month-ahead returns show that central industries’ returns do not provide an
incremental forecasting effect over and above the general link between trading
partners’ returns documented by Menzly and Ozbas (2010). In conjunction with the
ROA tests, the returns tests suggest that investors impound the information about
central industries’ ROA to a similar extent that they impound trading partners’
ROAs, in general.
Our study relates to the literature on the interrelationship between market
fundamentals and portfolio performance. Hong et al. (2007) show that there is a
positive association between an industry’s ability to forecast macroeconomic
fundamentals and the industry’s ability to forecast future market returns. Along the
same lines, Anilowski et al. (2007) and Bonsall et al. (2013) find that bellwether
firms’ earnings guidance predicts market-wide returns, where they identify
bellwether firms based on size and the past relation between earnings and
macroeconomic variables, respectively. Shivakumar’s (2007, 2010) discussions of
Anilowski et al. (2007) and Cready and Gurun (2010) call for research on the
relation between earnings and macroeconomic conditions and on inter-industry
information transfers in the context of financial reporting. Our study adds to this
literature by identifying a specific mechanism—inter-industry trade flows—that
moderates the relation between macroeconomic fundamentals and industry
performance.
Our trade-based notion of centrality conceptually and empirically differs from the
information-based measures used to identify bellwether firms whose earnings
guidance provides macroeconomic information (e.g., Hong et al. 2007; Anilowski
et al. 2007; Bonsall et al. 2013).2 The bellwether effect focuses on how much
information a manager’s forecasts provide about macro variables which, in turn,
depends on the firm’s sensitivity to those variables. The sensitivity need not overlap
with centrality, which is based on inter-industry trade. For example, Bonsall et al.
include Volcom Inc. as a bellwether firm because its forecasts are informative about
the macroeconomy. Our trade-based data classify Volcom in one of the least central
industries—apparel manufacturing. While apparel may have low trade volume with
other industries, Volcom’s forecasts can be informative about the overall economy if
its sales of premium clothing are highly sensitive to macroeconomic conditions.3 In
untabulated analyses, we examine the industries of bellwether firms as identified in
2
To give an analogy, miners conduct most of the activity in a coal mine, but what happens to the canary
can be quite informative about conditions in the mine.
3
Software publishing provides another example that the notion of trade-based centrality is distinct from
the bellwether effect. Based on trade data, we classify software publishing as having very low centrality.
This classification may stem from the software publishers relying on labor as its primary input, as
123
Earnings and stock returns
1195
Bonsall et al. (2013) and find that the correlation between an indicator variable for
bellwether firms and our measure of centrality is insignificant.
Our study also complements scholarship that examines potential information
transfers between trading partners. Cohen and Frazzini (2008) and Pandit et al.
(2011) conduct analyses at the firm level. On the one hand, the use of firm-level
disclosures of customers facilitates the examination of firm-level information
transfers. On the other hand, the firm-level data include a relatively small set of
large customers and do not have sufficient coverage to estimate firms’ roles in the
overall economy.4 The use of industry as the unit of measurement inherently
coarsens our data but allows us to incorporate the large amount of economic activity
that does not appear in the disclosures of sales to large customers. Both our study
and that of Menzly and Ozbas (2010) conduct industry-level analyses utilizing BEA
input/output tables to identify trading relationships. These data provide comprehensive coverage of the U.S. economy. Menzly and Ozbas (2010) provide evidence
of information transfers at the industry level but do not examine the role of industry
position, which is the focus of our study.
Our study is also related to the literature that examines the propagation of
economic shocks. Ahern and Harford (2013) examine how central industries drive
waves of mergers. While we show that the returns on the Fama–French portfolios
explain a relatively large portion of central industries’ returns, a contemporaneous
study by Ahern (2012) provides evidence that centrality, itself, represents a source
of systematic risk incremental to the risks captured by the Fama–French portfolios.
Acemoglu et al. (2012) examine a multi-sector setting to identify conditions under
which idiosyncratic shocks can lead to aggregate fluctuations. They show that
shocks to sectors that trade with a disproportionately large number of other sectors
can be amplified into aggregate fluctuations.5 While we do not differentiate our
analysis based on the origination point of the shocks, our finding that central
industries have stronger associations with aggregate fluctuations than do noncentral
ones is consistent with this theory.
The paper proceeds as follows. Section 2 discusses the concept of centrality as
well as our empirical predictions. Section 3 describes our data. Section 4 provides
Footnote 3 continued
opposed to trade with other industries. Bonsall et al. (2013) classify 27 software publishers as bellwether,
based on how informative their forecasts are about the overall economy.
4
For example, US firms are required to disclose only major customers that comprise at least 10 % of
revenues. This limits disclosures to a fairly small number of large customers and provides limited ability
to track the economy-wide flow of trade. According to customer and supplier links data compiled from
Compustat and available on Andrea Frazzini’s website, there were 454 firms with usable sales data in
1997, the vintage of BEA data we utilize. This is a small fraction of 12,000 firms on Compustat in that
year, and, according to Cohen and Frazzini (2008), the disclosed customers tend to be larger than their
suppliers.
5
As Lucas (1977) and others argue, in highly disaggregated economies, idiosyncratic shocks will remain
fairly confined. While Dupor (1999) and Horvath (1998, 2000) debate on whether sectoral shocks can
transfer into aggregate fluctuations, Acemoglu et al. (2012) provide a more complete answer to this
question by showing that sectoral shocks can lead to aggregate fluctuations only when there is
heterogeneity in the links between sectors. With heterogeneous links, shocks to hub sectors can lead to
aggregate shocks. Gabaix (2011) provides a similar model using firm-level shocks as a source of
aggregate fluctuations.
123
1196
D. Aobdia et al.
evidence that aggregate risks account for a relatively large portion of central
industries’ performance. Section 5 provides evidence on the propagation of shocks
between industries, and Sect. 6 concludes.
2 Measure of centrality and empirical predictions
2.1 Measure of centrality
Our empirical predictions pertain to industries that serve as hubs in the economywide flow of trade, so we must define what makes an industry a hub before
proceeding. We identify hub industries with respect to their role in inter-industry
trade flows. Following Ahern and Harford (2013) and Anjos and Fracassi (2012), we
use the BEA’s 1997 Input–Output tables to identify trade flows.6 The BEA reports
trade in two tables—the Make and Use tables. The Make table reports the value of
each commodity produced by each industry. The Use table reports the value of each
commodity purchased by each industry and plays a primary role in identifying interindustry trade.
The BEA uses a single set of IO industry codes to classify both industries and
commodities, where the largest producer of a given commodity has an industry code
that matches the commodity code. A single industry typically dominates the output
of a given commodity, and a single commodity typically dominates the output of a
given industry. For example, the ‘‘electric lighting equipment manufacturing’’
industry produces over 97 % of the ‘‘electric lighting equipment manufacturing’’
commodity, and that commodity accounts for over 97 % of that industry’s output.
Over 90 % of commodities are dominated by an industry that produces at least 75 %
of the output, and over 90 % of industries have at least 75 % of their output
concentrated in a single commodity.
The literature has used different approaches to identify inter-industry trade. Some
studies identify inter-industry trade solely with the Use table (e.g., Menzly and
Ozbas 2010), which implicitly assumes a one-to-one mapping between commodities
and industries. Others (e.g., Ahern and Harford 2013) utilize the Make table to
determine each industry’s market share of each commodity and then allocate the
purchases in the Use table based on these market shares. This process implicitly
assumes constant market shares within industries’ purchases of a given commodity.
For example, the ‘‘basic chemicals manufacturing’’ industry produces about 9 % of
the ‘‘resin, rubber, and artificial fibers manufacturing’’ commodity, and this
6
Like these studies, we focus on trade flows in 1997 since this year is the approximate midpoint of our
sample period. The BEA has been publishing input–output tables every five years until 2007, when the
BEA published the 2007 data in December 2013, after we completed our study. However, inasmuch as
the industry definitions vary in each version of these tables, it is not possible to form a consistent time
series based on different versions of input–output tables. Because there are relatively minor changes
between 1997 and 2002 industry classifications, we recompute our centrality measure using the 2002 data
and find that the correlation between the 1997 centrality and the 2002 centrality is high at 96 %. As a
robustness check we also replicate our baseline specifications using a combination of the 1997 and 2002
tables for the pre- and post-1999 periods, respectively. These replications yield qualitatively similar
results.
123
Earnings and stock returns
1197
approach assumes that all industries that purchase resin and rubber obtain about 9 %
of it from the ‘‘basic chemicals manufacturing’’ industry.
Because of the previously discussed high degree of overlap between industries
and commodities, there is not much of a practical difference between the two
approaches—the centrality measures we derive from the two approaches have a
correlation of 98.3 %. We adopt the latter approach that employs both the Make and
Use tables.7 In principle, combining the tables yields a more accurate gauge of trade
flows. One potential drawback is that the combination of the tables introduces
ancillary connections between industries that can behave as noise for purposes such
as identifying when one industry’s performance is most likely to impact another’s.
For example, an examination of the ‘‘resin, rubber, and artificial fibers manufacturing’’ industry may provide better information for gauging the economy-wide
effects of the corresponding commodity than combining that industry’s performance
with some fraction of the performance of the ‘‘basic chemicals manufacturing’’
industry.
To ensure a sufficient number of companies in each industry, we define industries
using the four-digit IO industry codes provided by the BEA. We exclude
government, special industries, value added, and final users (industry definitions
that start with the letters S, V, or F).8 Figure 1 plots the inter-industry links based on
the BEA data. The black circles denote industries that we classify as noncentral, and
the white squares represent central industries, which we define in Sect. 3.2. Because
every industry has some trade with nearly every other industry, the figure only plots
links with a strength measure, described below, greater than 3 %.
The inter-industry trade links can be represented in matrix form by a matrix A
with elements Aij = Aji = 1 if industries i and j trade with each other, and with
diagonal elements Aii = 0. Measures of industries’ influence utilize the matrix A.
Recognizing the heterogeneity in trading relationships, similar to Ahern and Harford
(2013) and Anjos and Fracassi (2012), we represent the links in A with measures of
the strength of trade between two industries. Specifically, we compute the elements
as:
1
s
s
s
s
P ij þ P ij þ P ji þ P ji
Aij ¼
;
ð1Þ
4
k sik
k skj
k sjk
k ski
where sij equals to the sales (in dollars) of goods and services by industry i to
industry j. The first (second) ratio measures the sales made by i to j as a percentage
of i’s total sales (j’s total purchases), gauging whether j is an important customer to
i (i is an important supplier to j). The last two ratios similarly measure whether i is
7
Prior versions of the paper relied solely on the Use table and obtained similar results.
8
The BEA’s ‘‘special industries’’ category includes noncomparable imports (industry code S003), scrap,
used, and secondhand goods (S004), rest of world adjustment (S006), and inventory valuation adjustment
(S007). We do not include government/special industry/value added/final uses industries in the
calculation of the strength measure because of the difficulties in the interpretation of certain associations
(e.g., negative sales values, changes in private inventories, etc.). In the calculation of the strength
measure, we include private consumption expenditures to calculate total sales numbers so that our
measures gauge importance relative to overall sales, rather than strictly business-to-business sales.
123
1198
D. Aobdia et al.
Fig. 1 Inter-industry network based on BEA input/output tables. The inter-industry network based on the
U.S. input–output matrix of 1997 provided by the BEA. Each node corresponds to an industry (based on
BEA’s four-digit industry code) and each edge corresponds to a link between industries with strength (Aij)
of 3 % or more. See Sect. 2.1 for the definition of strength of the link between two industries. The square
nodes indicate industries with a centrality score above the 0.109 threshold that we use to indicate central
industries, as we discuss in Sect. 3.2
an important customer to j, and j is an important supplier to i, respectively. A larger
Aij indicates a stronger link between industries i and j.
We measure an industry’s position in the economy using Bonacich’s (1972)
eigenvector centrality metric (hereafter ‘‘centrality’’).9 An industry rates as having
high centrality if it has strong trade links with industries that have high centrality.
Specifically, industry i’s centrality is:
1X
ci ¼
A c;
ð2Þ
j ij j
k
where k is a normalization factor. An industry’s centrality depends on both how
many other industries it trades with and those industries’ position within the overall
9
In the interest of maintaining the paper’s focus, we briefly describe the eigenvector centrality measure
here and refer interested readers to textbook treatments (e.g., Jackson 2008 or Newman 2010) or Borgatti
(2005) for discussions of centrality metrics. The term arises because the vector c of centralities can be
defined from (2) as kc = Ac, which is an eigenvalue equation for the matrix A. Other notions of
centrality, such as betweenness and closeness, pertain to shortest paths between industries and do not
seem particularly relevant to our setting (See, e.g., Borgatti 2005). For example, the iron and steel
industry’s economic importance likely stems more from its use in a variety of outputs than it having a
short route from production to final consumption.
123
Earnings and stock returns
1199
economy. Table 1 provides a list of the 21 most and 21 least central industries.
Appendix 2 briefly elaborates on the centrality measure.10
2.2 Empirical predictions
In any economy, the patterns of inter-industry trade have an impact on the
importance of an industry and on the factors that affect that industry’s performance.
Some industries, such as wholesale trade, have strong customer/supplier interactions
with several other industries and therefore are in more central positions. Others,
such as tobacco products manufacturing, are relatively isolated.
An industry’s centrality, as we define it in Sect. 2.1, reflects its influence on and
exposure to the overall economy. Recall that industry i’s centrality, ci, is defined in
P
(2) as ci ¼ 1k j Aij cj . An industry’s centrality increases with the number of
industries it trades with (having more cj’s in the sum), the importance of the
industries it trades with (having higher cj’s in the sum), and the strength of those
trading relationships (having higher Aij’s in the sum).
By their nature, shocks to central industries will tend to be associated with
macroeconomic factors. Acemoglu et al. (2012) predict that a shock originating in a
central industry will have a greater tendency to result in a macroeconomic shock
because of the trade that passes through, and is affected by, the central industry. We
reason that a shock to a central industry’s trading partners must be macroeconomic
for it to affect the central industry. First, if the industry is central by virtue of having
many trading partners (having more cj’s), then the shock must affect many of those
trading partners for it to impact the central industry. Second, if the industry is
central by virtue of trading with industries that are themselves central (having high
cj’s), then Acemoglu et al.’s (2012) arguments imply that shocks to those trading
partners will tend to be macroeconomic. Empirically, we cannot determine the
origin of shocks, but the nature of centrality implies that shocks experienced by
central industries will tend to be macroeconomic. Based on the preceding
discussion, we predict that, compared to the performance of the noncentral
industries, the performance of the central industries associates more strongly with
aggregate fluctuations and risks.
Because real economic shocks can transmit via trade linkages, information about
those shocks may similarly transfer and lead to cross-predictability of industry
performance. The predictability depends on the extent and nature of trade between
the industries. For example, if a given industry’s trading partners maintain large
inventory buffers, real shocks to that industry may transfer to its partners with a
delay, if at all. If the given industry represents only a small portion of its trading
10
We obtain similar results using an alternative measure, degree centrality, which is a strength-weighted
count of an industry’s neighbors, ((RjAij)/(# industries - 1)). We report results for the eigenvector
measure because it is more comprehensive, reflecting higher order links such as neighbors of neighbors,
and because of its common use in the literature on networks. Table 1, Panel A, omits the 6th ranked
industry, ‘‘monetary authorities, credit intermediation and related activities,’’ because we exclude
financial institutions from our analyses. In other words, the 22 most central industries are those listed in
Table 1, Panel A, plus ‘‘monetary authorities/credit intermediation.’’ For the same reason, Panel B omits
‘‘funds/trusts’’ and ‘‘insurance carriers.’’
123
1200
D. Aobdia et al.
Table 1 List of the most and least central industries
Rank
Name of the Industry
Panel A: Most central industries
1 (most)
Wholesale trade
2
Construction
3
Management of companies and enterprises
4
Real estate
5
Retail trade
6
All other administrative and support services
7
Plastics and rubber products manufacturing
8
Power generation and supply
9
Motor vehicle body, trailer, and parts manufacturing
10
Petroleum and coal products manufacturing
11
Other fabricated metal product manufacturing
12
Food manufacturing
13
Oil and gas extraction
14
Printing and related support activities
15
Employment services
16
Iron and steel mills and manufacturing from purchased steel
17
Semiconductor and electronic equipment manufacturing
18
Basic chemical manufacturing
19
Telecommunications
20
Architectural and engineering services
21
Management and technical consulting services
Panel B: Least central industries
1 (Least)
Cable networks and program distribution
2
Leather and allied product manufacturing
3
Ordnance and accessories manufacturing
4
Tobacco manufacturing
5
Apparel manufacturing
6
Animal production
7
Motion picture and sound recording industries
8
Nursing and residential care facilities
9
Social assistance
10
Pharmaceutical and medicine manufacturing
11
Textile product mills
12
Fishing, hunting, and trapping
13
Other transportation equipment manufacturing
14
Performing arts, spectator sports, museums, zoos, and parks
15
Religious, grantmaking and giving, and social advocacy organizations
16
Software publishers
17
Transit and ground passenger transportation
18
Household appliance manufacturing
19
Amusements, gambling, and recreation
123
Earnings and stock returns
1201
Table 1 continued
Rank
Name of the Industry
20
Water, sewage, and other systems
21
Water transportation
partners’ trade ties, then shocks to that industry will have a small impact on its
trading partners.
We hypothesize that a central industry’s performance has greater predictive
ability for its trading partners than does a noncentral industry’s performance.
Specifically, shocks to the performance of a central industry are more strongly
associated with the concurrent or future performance of industries to which the
central industry is linked, than shocks to a noncentral industry are associated with its
linked industries. Because centrality stems from strong trade exposure to the overall
economy, shocks to central industries should be experienced by those industries’
trading partners. In contrast, a shock to a single trading partner of a central industry
should have relatively little effect on the central industry because the central
industry’s performance depends more on the overall economy than the performance
of a single trading partner.
3 Data and descriptive statistics
3.1 Data
We obtain accounting and stock returns data from Compustat and CRSP,
respectively. Analyst-related data come from I/B/E/S. In all specifications, except
in the earnings tests in Sect. 4.2, we require the availability of Compustat quarterly
data and CRSP monthly returns.
We require that the historical NAICS industry code be available in Compustat to
merge our firm-related data with our centrality measures, available at the BEA fourdigit industry level. This restricts our sample period to the fiscal years beginning in
1985. We end our sample in 2011. We exclude financial institutions (BEA industries
beginning with 52) from the sample because their earnings and returns depend
relatively more on the regulatory environment than those of other firms.
3.2 Descriptive statistics
Figure 2 plots the centrality measure in descending order. The figure exhibits the
common phenomenon that a small number of nodes in the network, industries in our
case, exhibit high connectivity (e.g., Gabaix 2009). As the figure demonstrates,
while most industries have some connections with the other industries, there are
relatively few hub industries. Four industries—wholesale trade, construction,
management of companies, and real estate—exhibit the highest degree of centrality,
123
1202
D. Aobdia et al.
followed by a number of other industries with still relatively high centrality. We
confirm these observations in Table 2. The centrality measure quickly decreases
from its maximum of 0.274–0.095 at the 75th percentile—a difference of 0.179. The
distribution then levels out with 75 % of the observations clustered between the
0.030 values and the 0.095 values—a difference of 0.065.
There are two major discontinuities in the centrality values not counting several
back-to-back cliffs at the very beginning of the distribution: the first is at 0.122, and
the second is at 0.109. The distribution of centrality values becomes visibly smooth
beyond 0.109 (the dotted line in Fig. 2), suggesting that industries below this level
are relatively more comparable in terms of centrality. Accordingly, we choose to
define central industries as those that have centrality values greater than 0.109,
which accounts for the top 18 % of the industries. Our results remain similar when
we set the cutoff at 0.122 or when we use different cutoffs around the baseline
level.11 The industries with a centrality measure above the 0.109 threshold are
indicated by square nodes in the network plot in Fig. 1.
4 Are central industries more exposed to aggregate risks?
4.1 Returns tests
4.1.1 Research design
As we explain in Sect. 2.2, our first prediction is that aggregate risks explain a
relatively higher proportion of central industries’ returns, compared to noncentral
industries’ returns, for at least two reasons. First, some industries are central by
virtue of trading with many different parts of the economy, which provides some
diversification of idiosyncratic shocks. Second, shocks to central industries may
initiate systematic fluctuations (Acemoglu et al. 2012). To test this prediction, we
estimate the ability of CAPM and Fama–French three-factor models to explain
industries’ returns.
Specifically, we regress the excess returns rit for industry i in period t on a vector
ft of explanatory returns—either the excess return on the market portfolio or the
Fama–French factor-mimicking portfolios:
rit ¼ ai þ b0i f t þ eit :
ð3Þ
As in Roll (1988) and the literature on stock return synchronicity (e.g., Morck et al.
2000), the R2 from regression (3) measures the extent of industry-specific variation
in returns. In particular, industry i’s R2 provides an estimate of:
11
In particular, our analyses remain qualitatively similar when using a cutoff of 0.12, or 11 % of the
distribution of industries, when using a cutoff of 0.10, or 20 % of the industries. The 0.109 cutoff
classifies 22 industries as central—the 21 listed in Table 1, Panel A, plus ‘‘monetary authorities, credit
intermediation and related activities,’’ which is ranked 6th in the trade data used to compute centrality,
but we omit it from the table because we exclude financial institutions.
123
1203
0.20
0.10
0.00
Centrality
0.30
Earnings and stock returns
0
20
40
60
80
100
120
Centrality Rank
Fig. 2 Distribution of centrality measure. The distribution of the eigenvector centrality measure.
Eigenvector centrality (defined formally in Sect. 2.1) of an industry is as function of that industry’s direct
or indirect ties to the other industries in the economy as well as the strength of these ties. Larger values
indicate more central industries. The dotted reference line in the figure corresponds to 0.109, the cutoff
value we use to differentiate central industries from the noncentral industries
Table 2 Distribution of the
centrality variable
The distribution of the
eigenvector centrality measure
for the industries in the U.S. The
measure is calculated using
make and use tables of 1997
published by the BEA.
Industries are defined using the
BEA’s four-digit industry codes
and the total number of
industries is 123
R2i
Variable
Centrality
Minimum
0.030
5th percentile
0.041
10th percentile
0.046
25th percentile
0.056
50th percentile
0.072
75th percentile
0.095
90th percentile
0.125
95th percentile
0.147
Maximum
0.274
Mean
0.081
StDev
0.040
N
123
var b0i f t
var b0i f t
1
0 ¼
¼
;
¼
var
ðeit Þ
varðrit Þ
var bi f t þ varðeit Þ 1 þ
varðb0i f t Þ
ð4Þ
where var b0i f t measures overall variation attributable to the explanatory returns,
and var(eit) measures the industry-specific variation in returns.
A higher R2 indicates that the explanatory returns account for a relatively greater
portion of the industry’s return variation. Note that this is a relative measure—a
123
1204
D. Aobdia et al.
stable
might have low fluctuations from macroeconomic risks (low
industry
var b0i f t ) but a high R2 due to also having low idiosyncratic risk (low var (eit)). We
expect that the R2s from these models will be higher for central industries than for
noncentral ones.12
We form value-weighted industry portfolios and estimate CAPM and Fama–
French three-factor models with both daily and monthly returns data using rolling
five-year windows beginning in 1981 and ending in 2011.13 Upon estimating the
CAPM and the three-factor models we test the association between centrality and
the R2s from these models using the following industry-level regression:
R2it ¼ a þ b1 Centralityi þ b2 Avg: Analyst Coverageit
þ b3 Avg: Proportion Tradedit þ b4 Log Avg: Market Valueit þ eit
ð5Þ
Because R2s are bounded by zero and one, we use a fractional logit model (Papke
and Wooldridge 1996) to estimate (5). A positive coefficient on Centrality, (b1)
indicates that aggregate risks explain a greater proportion of central industries’
returns, compared to noncentral industries’ returns. Because the R2s in the
regressions could also depend on the information environment and the liquidity of
the stocks traded, we add three control variables following Piotroski and Roulstone
(2004) and Kelly (2005): Average Analyst Coverage is the value-weighted average
of an indicator variable that equals one for any stock in the portfolio that is covered
by analysts at the end of the estimation period; Average Proportion Traded is the
average monthly trading volume as a percentage of shares outstanding during the
period (i.e., we compute the portfolio’s shares traded/shares outstanding each period
and take the average over the periods).14 Log Average Market Value is the logarithm
of the estimation-period average of the portfolio’s market capitalization, where we
compute the market values of the portfolios’ constituents as the CRSP shares
outstanding times market price at the beginning of each period in the regression
(monthly for monthly returns, and daily for daily returns). Based on Kelly’s (2005)
finding that R2s from market models increase with information availability and
liquidity, we expect positive coefficients on the control variables. Appendix 1
provides additional information on the computation of control variables. Both the
12
In some settings, such as in per share regressions, R2 can be problematic due to scale effects (e.g.,
Brown et al. 1999, Liu and Ohlson 2000). Scale impacts the R2 for reasons having nothing to do with the
model being estimated. Our approach of using R2s from returns regressions as the dependent variable
parallels its use in studies of stock return synchronicity (e.g., Morck et al. 2000; Durnev et al. 2003;
Piotroski and Roulstone 2004) and earlier work such as Roll (1988). In the synchronicity tests, as in ours,
the R2 has the clear interpretation as simply measuring the amount of return variation accounted for by
explanatory factors.
13
We use industry definitions as of the end of each estimation window. Because NAICS codes became
available in 1985, our estimation starts with the 1981–1985 window, and we estimate regressions for
1981–1985, 1982–1986, and so on until 2007–2011. Table 3 reports double-clustered standard errors on
year and industry to control for the within-industry correlation of residuals caused by overlapping periods,
and for the within-time correlation caused by the regression potentially having a better fit in some years
than in others (Thompson 2011).
14
This controls for the fact that the market price of a stock will remain the same if the stock is not traded
during a given period, which would result in a low R2.
123
Earnings and stock returns
1205
dependent variable and control variables are winsorized at the 1st and 99th
percentiles to reduce the impact of outliers in the specifications.
4.1.2 Results
We present our results from the estimation of (5) in Table 3. From the 123
industries in Table 2, this table excludes financial institutions (four industries with
BEA industry codes beginning with 52) and industries that have insufficient data in
CRSP and Compustat (‘‘fishing, hunting and trapping,’’ and ‘‘religious organizations’’; BEA industries 1140 and 813A, respectively). This leaves 117 industries for
which we conduct our analysis. Panel A presents descriptive statistics for central
and noncentral industries. We test for the difference of the means (distributions)
using a t test (Wilcoxon–Mann–Whitney test). All of the R2 measures indicate that
central industries have higher R2s than noncentral industries. The means test shows
no difference in analyst coverage between central and noncentral industries, while
the rank-sum test suggests that noncentral industries have somewhat greater analyst
coverage than central industries. Noncentral industries also have lower market
values, indicating the need to control for this variable.
Table 3, Panel A, also shows that the mean and the distributions of factor
loadings are similar between the central and noncentral industries. This suggests
that the central firms’ higher R2s are driven more by less exposure to idiosyncratic
risk thanby greater exposure
to systematic risk. Recall that the industry’s R2 is
varðeit Þ and that the term varb0 x represents the total exposure to
R2i ¼ 1= 1 þ var
i t
ðb0i xt Þ
systematic risk. The similar factor loadings in Table 3, Panel A, indicate that the
total exposure to systematic risk differs little between central and noncentral
industries. The variation in R2 must therefore be driven by the idiosyncratic portion
of returns.15
We present the results of the fractional logit estimations in Table 3, Panel B. The
first two columns show the results using daily returns data. The coefficient on
Centrality is positive in both CAPM and Fama–French specifications at the 1 %
level. Untabulated results indicate that, across the different specifications and at the
mean of the control variables, the marginal effect of the centrality variable, at the
mean of the other explanatory variables, ranges from 0.469 to 0.935. With the
control variables set to their means, Panel C reports that an increase in centrality
from 0.072 to 0.147, corresponding to the median and 95th percentile of centrality
in Table 2, would increase the average R2 by between 3.6 and 6.9 percentage points,
representing from 11.3 to 15.2 % increases in the predicted R2s, depending on the
specification. Consistent with Piotroski and Roulstone’s (2004) and Kelly’s (2005)
findings, our results in Panel B indicate that industries with higher analyst coverage
and industries with a higher market capitalization have higher R2s, indicating that
larger industries are more exposed to aggregate risks than other industries. Our
15
Table 3 compares means and the overall distributions, while untabulated analysis also indicates that
the percentiles of the factor loadings are roughly similar for central and noncentral industries.
123
1206
D. Aobdia et al.
Table 3 Analysis of the industry R2s from CAPM and Fama–French three-factor models
Variable
Mean
Central
Median
Noncentral
Central
Test of Equality
(Central–noncentral)
Noncentral
Mean
Distribution
(Rank-sum)
Panel A: Descriptive statistics
Regression variables
Centrality
0.141
0.068
0.127
0.067
7.42***
6.77***
Daily CAPM R2
0.408
0.319
0.389
0.307
3.97***
4.00***
Daily FF R2
0.467
0.362
0.457
0.352
5.23***
Daily Average Analyst
Coverage
0.837
0.820
0.886
0.936
0.60
-1.75*
Daily Average Proportion
Traded
4.557
5.041
3.289
3.772
-1.38
-1.65*
Log Average Daily Market
Value
17.460
15.994
17.364
16.189
4.05***
3.86***
Monthly CAPM R2
0.527
0.417
0.542
0.425
3.63***
3.37***
Monthly FF R2
0.629
0.506
0.641
0.520
4.71***
Monthly Average Analyst
Coverage
0.837
0.820
0.886
0.936
0.60
-1.75*
Monthly Average
Proportion Traded
0.950
1.050
0.687
0.786
-1.37
-1.65*
17.457
15.990
17.359
16.182
Log Average Monthly
Market Value
4.05***
4.84***
3.91***
3.86***
Factor loadings
CAPM beta daily
0.904
0.910
0.893
0.920
-0.13
FF Mkt factor daily
1.073
1.065
1.082
1.077
0.21
-0.49
0.16
SMB daily
0.366
0.463
0.398
0.454
-1.12
-1.00
HML daily
0.274
0.173
0.284
0.190
1.20
1.37
CAPM beta monthly
1.021
1.028
1.032
1.042
-0.12
-0.16
FF Mkt factor monthly
1.060
1.028
1.050
1.031
0.67
0.43
SMB monthly
0.300
0.421
0.353
0.377
-1.40
-1.16
HML monthly
0.172
0.116
0.169
0.113
0.70
0.79
2
2
Dependent Variable:
Daily R s
R2
CAPM
Fama–French
CAPM
Fama–French
3.752***
Monthly R s
Panel B: Fractional logit regressions
Centrality
Average Analyst Coverage
Average Proportion Traded
Log Average Market Value
123
2.148***
2.641***
3.508***
(3.045)
(3.761)
(3.463)
(3.678)
0.495**
0.397**
0.162
0.176
(2.187)
(1.974)
(0.654)
(0.818)
-0.002
0.058***
0.053***
0.024
(3.146)
(3.132)
(0.225)
(-0.021)
0.193***
0.175***
0.124***
0.117***
(7.081)
(6.755)
(3.639)
(3.656)
Earnings and stock returns
1207
Table 3 continued
Dependent Variable:
Daily R2s
R2
CAPM
Fama–French
CAPM
Fama–French
-4.744***
-4.167***
-2.716***
-2.240***
(-11.412)
(-10.705)
(-5.134)
(-4.891)
Number of observations
3,077
3,077
3,077
3,077
Clustering
Industry/year
Industry/year
Industry/year
Industry/year
Number of clusters
117/27
117/27
117/27
117/27
Chi square
127.33***
134.77***
34.23***
46.83***
R2
0.387
0.387
0.119
0.147
Constant
Monthly R2s
Estimated effect of increase in centrality from median (0.072) to the 95th percentile (0.147)
Daily R2s
CAPM
Monthly R2s
Fama–French
CAPM
Fama = French
Panel C: Estimated effect of centrality on R2
Predicted R2, centrality = 0.147
0.354
0.414
0.492
Predicted R2, centrality = 0.072
0.318
0.367
0.427
0.520
Difference
0.036
0.047
0.065
0.069
% Increase
11.3
12.8
15.2
0.589
13.3
The industry-level analysis of the association between centrality and exposure to aggregate risk using a
fractional logit model. The dependent variable is the R2 from regressing value-weighted industry portfolio
excess returns on either the excess return of the CRSP value-weighted market portfolio (CAPM), or the
Fama–French market, SMB, and HML portfolio returns. Estimations are conducted using 5-year windows. Centrality is defined in Sect. 2.1. Average Analyst Coverage is a value weighted measure of an
indicator for whether there is a quarterly consensus EPS forecast provided in I/B/E/S for the last quarter
of the analysis period. Average Proportion Traded equals the average over a 5 year period of the value
weighted percentage of shares traded. Log Average Market Value equals the logarithm of the average over
the 5-year period of the industry portfolio’s market value. Except for centrality, the variables are
winsorized at 1 and 99 %. See Appendix 1 for detailed variable definitions. Panel A presents descriptive
statistics for the dependent and control variables. There are 3,077 observations: 2,533 for noncentral
industries and 544 for central industries. Panel B presents the results of the fractional logit model. The
predicted R2s for the incremental effects in Panel C set the control variables at their sample means.
Standard errors are clustered at the industry and year levels (via bootstrapping for the Wilcoxon rank-sum
test in Panel A). The R2 of the regressions in Panel B is computed as in OLS (1 - SSR/SST) following
Papke and Wooldridge (1996)
*, **, *** Significance at two-sided 10, 5, and 1 % levels, respectively
analysis using monthly returns yields results that resemble those using daily returns.
For both CAPM and Fama–French three factor models, we find that central
industries have higher R2s. Overall these findings are consistent with our prediction
that central industries’ returns depend relatively more on aggregate risks.16
16
In untabulated analyses, we confirm that our results are qualitatively unchanged when using OLS
regressions, regressions where continuous variables are replaced with their ranks, and in an OLS
regression of the log-odds ratio log(R2/(1 - R2)), similar to that used by Piotroski and Roulstone (2004).
Prior work has shown that analysts tend to be industry-focused (Boni and Womack 2006; Kadan et al.
2012). If analysts find it relatively more attractive to cover central industries, this could cause the analyst
123
1208
D. Aobdia et al.
4.2 Earnings tests
4.2.1 Research design
We now conduct tests to measure the extent to which summary measures of
macroeconomic activity explain industries’ earnings. The test design resembles the
returns tests in Sect. 4.1. We first estimate industry-level regressions of seasonal
changes in quarterly ROA on a vector of macroeconomic statistics from Bonsall
et al. (2013). We then examine the association between centrality and the R2s from
these regressions. The first stage regressions have the form:
ROAiq ROAi;q4 ¼ ai þ b0i ðxq xq4 Þ þ eiq ;
ð6Þ
where ROAiq denotes industry i’s ROA in quarter q, and xq denotes the quarter
q vector of macroeconomic statistics taken from the Federal Reserve’s FRED
database—the AAA corporate bond yield (AAA), three-month Treasury yield
(TB3MS), 10-year Treasury yield (GS10), unemployment (UNRATE), consumer
price index for urban consumers (CPIAUCNS), log housing starts (HOUST), log
industrial production (INDPRO), and log seasonally adjusted real GDP (GDPC1).
We report regressions based on two methods for aggregating firm-level ROA
(earnings from continuing operations—Compustat IBQ—divided by average total
assets) to the industry level. First, we weight ROA by beginning-of-quarter market
value, analogous to the value-weighting of the portfolios for the returns tests.
Second, we weight by total assets so that the ROA is total industry earnings from
continuing operations divided by the average of industry total assets.
4.2.2 Results
Table 4 reports the results of the fractional logit regressions. The 117 observations
are the 117 industries from Table 3. We include a control for log of the industries’
average beginning-of-quarter market values over the estimation period. We do not
include the analyst coverage and trading volume controls from Table 3 because
these do not seem relevant to the ROA regressions. Panel A reports the regressions
where the first stage includes the full set of macroeconomic statistics. The R2s
from regressions with value-weighted ROA show a positive association with
centrality; however, the association is statistically insignificant for the regressions
of asset-weighted ROA (p values of 0.13 and 0.11, respectively). Also, in the
regression with value-weighted ROA that controls for market value, the model’s
summary Chi square statistic of 4.4 is insignificant at conventional levels (p value
of 0.108).
As we discuss in the introduction, the trade-based centrality measure primarily
captures the corporate sector of the economy. We accordingly run a second set of
Footnote 16 continued
coverage control variable to absorb some of the effect of centrality; however, we have been unable to
devise a way to distinguish how much, if any, of the relation between analyst coverage and R2 stems from
an indirect impact of centrality. Untabulated analysis shows that the coefficient on centrality is similar but
slightly lower in regressions that exclude the analyst coverage control.
123
Value-weighted ROA
5.0**
0.052
Chi square
R2
(-5.772)
0.052
5.4*
117
(-13.753)
117
0.059
6.6**
117
(-14.588)
0.060
6.7**
117
(-5.445)
-2.367***
(0.416)
-2.343***
-2.245***
-2.197***
(2.301)
(0.246)
(2.569)
4.103**
0.041
2.6
117
(-2.777)
0.018
(2.011)
(2.234)
4.266**
Asset-weighted ROA
0.035
2.3
117
(-8.958)
0.011
4.057**
0.062
4.4
4.152**
Number of observations
Constant
Log Average Market Value
Centrality
Panel B: First stage includes subset of statistics (AAA, T3, T10, and IP)
Value-weighted ROA
0.054
R2
Dependent variable: R2
3.9**
Chi Square
(-2.956)
117
(-9.921)
117
-1.356***
(-0.552)
-1.419***
-1.690***
-1.611***
(1.605)
3.488
-0.028
(1.504)
3.231
(-0.602)
(2.108)
Asset-weighted ROA
-0.030
4.203**
3.932**
(1.967)
Number of observations
Constant
Log Average Market Value
Centrality
Panel A: First stage includes full set of statistics (AAA, CPI, T3, T10, unemployment, housing, IP, and GDP)
Dependent variable: R2
Table 4 Analysis of industry R2s from regressions of ROA on macroeconomic statistics
Earnings and stock returns
1209
123
123
0.196
0.055
28.1
Difference
% Increase
22.4
0.045
0.201
0.246
29.6
0.037
0.125
0.162
30.5
0.040
0.131
0.171
Asset-wgt.
*, **, *** Significance at two-sided 10, 5, and 1 % levels, respectively
The industry-level analysis of the association between centrality and exposure to aggregate risk using a fractional logit model. The dependent variable is the R2 from
regressing seasonal changes (current vs. same quarter in prior year) in value-weighted or asset-weighted quarterly ROA on a vector of seasonal changes macroeconomic
statistics—the AAA corporate bond yield (AAA), three-month Treasury yield (T3), 10-year treasury yield (T10), unemployment, housing starts, industrial production (IP),
and gross domestic product (GDP). Log Average Market Value is the logarithm of the portfolio’s market value over the estimation period. Except for centrality, the
variables are winsorized at 1 and 99 %. See Appendix 1 for detailed variable definitions. Panel A presents regressions where the first stage includes the full vector of
statistics, and Panel B presents regressions where the first stage includes a subset of the statistics. The predicted R2s for the incremental effects in Panel C utilize the
regressions with the control for market value and set the control variable to its sample mean. The R2s of the fractional logit regressions are computed as in OLS (1 - SSR/
SST) following Papke and Wooldridge (1996)
0.251
Predicted R2, centrality = 0.072
Value-wgt.
Value-wgt.
Asset-wgt.
Subset of statistics
Full set of statistics
Predicted R2, centrality = 0.147
Panel C: Estimated effect of centrality on R2
Table 4 continued
1210
D. Aobdia et al.
Earnings and stock returns
1211
regressions that includes only industrial production and interest rate variables, as a
gauge for macroeconomic statistics that are more related to the corporate sector.
Table 4, Panel B, presents the results of regressing the R2s from these regressions
on centrality with a control for market value. The R2s have a positive association
with centrality in both the value- and asset-weighted ROA specifications. The
stronger results in Panel B are consistent with the trade-based centrality measure
reflecting inter-corporate activity, which is distinct from the bellwether effect of
Bonsall et al. (2013) that has greater sensitivity to the consumer side of the
economy.
Table 4, Panel C, reports that an increase in centrality from 0.072 to 0.147,
corresponding to the median and 95th percentile of centrality, would increase the
average R2 by between 0.037 and 0.040 percentage points in the regressions that use
a subset of the macroeconomic statistics and include the control for market value.
We compute the predicted R2s with the log average market value set to its sample
mean. The size of the effect on R2 is similar to that for returns reported in Table 3
but with a greater percent increase in R2 of about 30 %.
5 Do shocks to central industries propagate more strongly than shocks
to noncentral industries?
In this section, we test our second prediction that, compared to noncentral
industries, a central industry’s performance has a stronger association with its
trading partners’ performance shocks. This could reflect two scenarios. First,
aggregate shocks can originate in central industries (Acemoglu et al. 2012) and then
propagate to those industries’ trading partners. Second, central industries tend to
have trading partners in many other industries, so that many of their trading partners
must have a correlated shock (i.e., an aggregate shock) for it to impact the central
industry. Both scenarios lead to the prediction that a central industry’s performance
correlates relatively more strongly with the performance of a portfolio of the
industries it trades with. We test this prediction in terms of accounting performance
and returns performance.
5.1 Earnings predictability
5.1.1 Research design
In this subsection, we test our second prediction using accounting performance. In
particular, we test whether the accounting performance of an industry, which we
denote as ‘‘source industry,’’ predicts the current and future accounting performance
of the industries to which it is linked (hereafter ‘‘linked industries’’), and whether
the effects are enhanced when the source is a central industry. We measure
accounting performance using ROA, computed as earnings from continuing
operations divided by the average of the assets at the beginning and end of the
period. We compute industry-level ROA changes by weighting each firm’s ROA
123
1212
D. Aobdia et al.
change by its beginning-of-quarter market capitalization.17 We estimate the
following regression at the industry level:
Linked ROA Changeit ¼ a þ b1 Source ROA Changeit þ b2 Centrali
þ b3 Centrali Source ROA Changeit þ
4
X
cs Linked ROA Changei;ts þ eit ;
s¼1
Linked ROA Changei;tþ1 ¼ a þ b1 Source ROA Changeit þ b2 Centrali
þ b3 Centrali Source ROA Changeit þ
4
X
cs Linked ROA Changei;ts þ eit ; ð7Þ
s¼0
where Source ROA Change is an industry’s ROA change, and Linked ROA Change
is the weighted average ROA change of the source industry’s trading partners.
Central is an indicator variable that equals one if the centrality of the industry is
greater than 0.109 and zero otherwise. The interacted coefficient Central 9 Source
ROA Change measures the impact of centrality on the relation between the Source
and Linked industries’ ROA changes. To maintain a constant sample for both
regressions, we exclude the last quarter from the concurrent ROA regression.
We weight the portfolio of trading partners by the strength of link to the source
industry. Specifically, denoting by Aij the strength of the link between source
industry i and linked industry j [see expression (1)], we compute Linked ROA
Changeit as:
P
j Aij ROA Changejt
P
:
ð8Þ
Linked ROA Changeit ¼
j Aij
Strength-weighting the portfolios of linked industries controls for the trade ties
between the source and linked industries.18 Nearly every industry has at least some
trade with almost every other industry, and strength-weighting provides a means of
including primarily the small set of trading partners that comprise the bulk of the
industry’s business. Alternative measures, such as equal- or value-weighting the
ROA changes of every industry with which the source industry trades, would result
in Linked ROA Change variables that approximately equal the economy-wide ROA
change, regardless of the source industry. For example, the strength-weighting
yields a Linked ROA Change variable for the ‘‘animal production’’ industry comprised mostly of the five industries that constitute over 80 % of its trade. An equalor value-weighted Linked ROA Change would treat these five industries the same, in
the case of equal-weighting, or possibly give them less weight, in the case of valueweighting, as other industries such as construction.
17
Our results are similar when defining industry ROA as the industry’s total earnings divided by the
average of beginning and ending total industry assets, which is equivalent to using asset weights rather
than market value weights to compute industry ROA.
18
Normalizing the weights removes any mechanical relation between centrality and the dependent
variable (see expression 2).
123
Earnings and stock returns
Table 5
1213
Earnings predictability
Dependent variable: Linked ROAt change
(1)
(2)
Panel A: Concurrent industry ROAs
Source ROAt change
0.043***
(2.760)
Central
0.000
Central 9 Source ROAt change
0.066***
(0.028)
(3.535)
Linked ROAt-1 change
Linked ROAt-2 change
Linked ROAt-3 change
Linked ROAt-4 change
Constant
0.485***
0.460***
(9.641)
(10.068)
0.047
0.042
(1.036)
(0.957)
0.057
0.054
(1.442)
(1.382)
-0.256***
-0.248***
(-3.079)
(-3.201)
-0.001***
-0.001***
(-4.841)
(-4.878)
Number of observations
11,812
11,812
Clustering
Industry/Quarter
Industry/Quarter
Number of clusters
117/107
117/107
Adjusted R2
0.307
0.329
F-statistic
682.3***
448.1***
F-test: Source ROA Change ? Central 9 Source ROA change = 0
F-statistic
16.2
p value
0.000
Dependent variable: Linked ROAt?1 Change
(1)
(2)
Panel B: One-quarter-ahead industry ROAs
Source ROAt change
0.019***
(4.323)
Central
0.000
Central 9 Source ROAt change
0.021**
(-1.269)
(2.347)
Linked ROAt change
Linked ROAt-1 change
Linked ROAt-2 Change
Linked ROAt-3 change
0.515***
0.502***
(9.650)
(9.462)
0.039
0.035
(0.833)
(0.757)
0.057
0.056
(1.440)
(1.407)
-0.319***
-0.320***
(-3.275)
(-3.287)
123
1214
D. Aobdia et al.
Table 5 continued
Dependent variable: Linked ROAt?1 Change
Linked ROAt-4 change
(1)
(2)
0.131**
0.131**
(2.301)
(2.304)
Constant
-0.001***
-0.001***
(-3.936)
(-3.932)
Number of observations
11,812
11,812
Clustering
Industry/quarter
Industry/quarter
Number of clusters
117/107
117/107
Adjusted R2
0.312
0.316
F-statistic
535.4***
359.1***
F-test: Source ROA change ? Central 9 Source ROA change = 0
F-statistic
22.3
p value
0.000
The industry-level analysis of the association between centrality and the predictability of accounting
performance. The dependent variable is equal to the weighted seasonally differenced quarterly return-onassets of the linked industries of a given industry i (i.e., the source industry). The seasonally differenced
quarterly ROA of the linked industries is computed as the average ROA change of each linked industry
multiplied by the strength of the link summed over all linked industries and normalized by the sum of
strength of all links of the source industry. Each industry’s average ROA is computed as the market value
weighted average of the ROA changes of the firms comprising the industry. Panel A presents results for
the concurrent association between the ROA of the linked industries and the source, and Panel B presents
the results where the dependent variable is one-quarter-ahead ROA of the linked industries. See Appendix
1 for detailed variable definitions. Standard errors are double-clustered at the industry and quarter level
*, **, *** Significance at two sided 10, 5, and 1 % levels, respectively
We predict that the coefficient on b1 is positive if source industries’ ROA
changes are correlated with their trading partners’ ROA changes (Menzly and Ozbas
2010). We also predict that the coefficient on b3 is positive if central industries’
earnings are more strongly associated with the earnings of their trading partners than
are the earnings of noncentral industries. We include lagged values of Linked ROA
Change to control for future predictability of within industry earnings changes,
following Bernard and Thomas’s (1990) findings at the firm-level. In particular, we
predict that c1, c2, and c3 should be positive while c4 should be negative, due to
positive autocorrelation of earnings for the prior three quarters and negative
autocorrelation of earnings for the prior fourth quarter, as in Bernard and Thomas
(1990). For the purpose of consistency in our regressions, we restrict our sample to
firms having fiscal quarters ending in March, June, September, and December.
5.1.2 Results
We present our results from the estimation of (7) in Table 5. The results in Panel A
present specifications using concurrent earnings change as the dependent variable.
Column (1) presents specifications using the control variables only. Consistent with
123
Earnings and stock returns
1215
Bernard and Thomas (1990), we find that the first three lags of Linked ROA Change
are positively associated with the current value of this variable. One of the three
associations is significant. Also consistent with Bernard and Thomas (1990), the
coefficient on the fourth lag of Linked ROA Change is negative and significant.
We add Source ROA Change in Column (2), as well as its interaction with the
Central dummy variable. We find a significant positive association between source
and linked industries’ concurrent year-on-year changes in ROA, evidenced by a
positive coefficient on Source ROA Change. We also find that this positive
association is significantly enhanced when the source industry is central. In
particular, the coefficient more than doubles from 0.043 to 0.109 (=0.043 ? 0.066).
In Panel B, we present the results of the tests where the dependent variable is
one-quarter-ahead value of the Linked ROA Change. Results in Column (1),
including control variables only, are similar to those of Column (1) in Panel A. In
Column (2), consistent with our predictions, we find that Source ROA Change at
quarter t positively predicts Linked ROA Change at quarter t ? 1, with a significant
improvement of the predictive power when the source is a central industry, as
evidenced by a positive coefficient on the interaction Central 9 Source ROA
Change. The predictive coefficient increases from 0.019 to 0.040 (=0.019 ? 0.021),
comparable to the results in Panel A.
The evidence from Table 5 indicates that, while all industries’ earnings predict
their trading partners’ earnings, the effect is much stronger for central industries.
Untabulated tests show that we obtain similar results for alternative earnings-based
numerators such as EBITDA and EBIT. We also obtain similar results with
concurrent performance using two cash-flow based measures, cash from operations
and free cash flow; however, the centrality interaction has an insignificant relation
with one-quarter-ahead changes in the cash flow performance measures. Additionally, we ran untabulated regressions separately for portfolios of linked central and
linked noncentral industries. The association of source ROA changes and the
incremental effect of central industries is similar for both sets of linked industry
portfolios. Both the coefficients and t statistics are larger for portfolios of noncentral
linked industries, though with little statistical difference between them.19 These
additional tests indicate that the effect of centrality in Table 5 is not driven solely by
correlations among central industries.
The tests in Table 5 support our hypothesis that information transfers and the
propagation of shocks depend not only on trading links between industries but also
on industries’ positions in the flow of trade. By definition, central industries have
direct or indirect trading links to large portions of the economy. Shocks to these
industries therefore are more likely to be either the origins of or reflective of
aggregate risks that affect other segments of the economy.
19
We tested for a difference in the coefficients using a single regression where we interact all of the
explanatory variables with an indicator variable that equals one for the portfolios of central linked
industries. The test for the difference between the central and noncentral linked industries’
Central 9 Source ROA Change coefficients is not significant at conventional levels.
123
1216
D. Aobdia et al.
5.2 Returns predictability
5.2.1 Research design
We now test our second prediction using stock returns. Given the evidence in
Table 5 that central industries’ earnings have a relatively strong relation with their
trading partners’ earnings, we expect that stock returns, which reflect expectations
of earnings, exhibit a similar relation. Unlike the predictability of earnings, the
predictability of future stock returns requires some sort of friction or inefficiency.
Hence we expect that our findings for the cross-predictability of future returns will
be weaker than the findings for accounting performance and for concurrent returns.
Our analysis of the predictability of returns is similar to that in Hong et al.
(2007). In particular, for each industry i and month t we compute the value-weighted
return, based on the beginning-of-month market values of firms in the industry. We
subtract the value-weighted NYSE/AMEX/NASDAQ return to obtain the industry’s
abnormal return Rit. We estimate the association between the returns of an industry
and its trading partners by regressing the abnormal returns of the portfolio of trading
partners on the returns of the industry, as follows:
Returns linkedit ¼ a þ b1 Returns sourceit
þ b2 Centrali þ b3 Centrali Returns sourceit þ c Controlsit þ eit ;
Returns linkedi;tþ1 ¼ a þ b1 Returns linkedit þ b2 Returns sourceit
þ b3 Centrali þ b4 Centrali Returns sourceit þ c Controlsit þ ei;tþ1 ; ð9Þ
where Returns source is Rit, and Returns linked is the return on the portfolio of
trading partners. To maintain a constant sample for both regressions, we exclude the
last month from the concurrent returns regression. As in the ROA regressions, we
weight trading partners by the strength of trade as follows, where Aij denotes the
strength of the link between source industry i and linked industry j [see expression
(1)]:
P
j Aij Rjt
Returns linkedit ¼ P
:
ð10Þ
j Aij
Central is an indicator variable that equals one if the centrality of the industry is
[0.109 and zero otherwise. The interacted coefficient Central 9 Returns source
measures the impact of centrality on the relation between the Source and Linked
industries’ returns. We follow Hong et al. (2007) by including the following
variables as controls for time varying risk: BAA-AAA Spread, the default spread,
defined as the difference between the Moody’s BAA-rated and AAA-rated bond
yields; Inflation, the seasonally adjusted monthly inflation, measured as the growth
rate of the consumer price index; Stdev of Market Returns, the daily market
volatility estimated over one month; Market Dividend Rate, the one-year market
dividend yield, computed as the total dividend from the CRSP market portfolio
divided by the current market level.
123
Earnings and stock returns
1217
Table 6 Returns predictability
Dependent variable: Concurrent returns linked
(1)
(2)
Panel A: Concurrent industry returns
Returns source
0.102***
(9.393)
Central
0.000*
(1.781)
Central 9 Returns source
0.097***
(3.350)
BAA–AAA spread
Inflation
Stdev of Market Returns
0.012***
0.010***
(3.342)
(3.232)
0.110
0.111
(0.444)
(0.512)
-0.537**
-0.418**
(-2.563)
(-2.279)
Market Dividend Rate
-0.449***
-0.380***
(-3.529)
(-3.396)
Constant
0.002
0.001
(0.684)
(0.469)
Number of observations
37,453
37,453
Clustering
Industry/Month
Industry/Month
Number of clusters
118/335
118/335
Adjusted R2
0.029
0.141
F-statistic
228.7***
605.8***
F-test: Returns source ? Central 9 Returns source = 0
F-statistic
48.8
p value
0.000
Dependent variable: Returns linked month t ? 1
(1)
(2)
0.067*
0.059
Panel B: 1-month-ahead industry returns
Returns linked
(1.694)
Returns source
(1.537)
0.006**
(2.363)
Central
0.000***
(6.387)
Central 9 Returns source
0.008
(1.238)
BAA–AAA spread
0.008**
0.008**
(2.272)
(2.263)
Inflation
0.229
0.230
(0.879)
(0.882)
Stdev of Market Returns
-0.140
-0.137
(-0.615)
(-0.598)
123
1218
D. Aobdia et al.
Table 6 continued
Dependent variable: Concurrent returns linked
Market Dividend Rate
(1)
(2)
-0.338***
-0.337***
(-2.648)
(-2.638)
Constant
-0.001
-0.001
(-0.200)
(-0.220)
Number of observations
37,453
37,453
Clustering
Industry/Month
Industry/Month
Number of clusters
118/335
118/335
Adjusted R2
0.025
0.025
F-statistic
144.8***
91.8***
F-test: Returns source ? Central 9 Returns source = 0
F-statistic
4.9
p value
0.027
The industry-level analysis of the association between centrality and the predictability of stock returns.
The dependent variable is equal to the weighted monthly returns of the linked industries of a given
industry i (i.e., the source industry) less the market returns. To calculate the weighted monthly returns of
the linked industries, we first calculate value-weighted returns of all firms in each linked industry then
multiply these returns with the strength of the link of between the industry and the source industry and
normalize this value by the sum of strength of all links of the source industry. Panel A presents results for
the concurrent association between the returns of the linked industries and the source, and Panel B
presents the results where the dependent variable is 1-month-ahead return of the linked industries. See
Appendix 1 for detailed variable definitions. Standard errors are clustered at the industry and month level
*, **, *** Significance at two sided 10, 5, and 1 % levels, respectively
We predict a positive coefficient b1 on Returns source if source industries’
returns are positively correlated with their trading partners’ returns. A positive
coefficient would be consistent with Menzly and Ozbas’ (2010) finding that stocks
of economically related supplier and customer industries predict each other’s
returns. A positive coefficient on the interaction term, Central 9 Returns source,
indicates that, compared to noncentral industries, central industries’ returns have a
stronger association with their trading partners’ returns.
5.2.2 Results
We present our results from the estimation of (9) in Table 6. In Panel A, we
examine the concurrent associations. In Column (1), we regress current period
linked industries’ abnormal returns on the control variables. We find that higher
spreads between AAA and BAA bonds, lower standard deviation of market returns
and lower dividend rates are positively associated with concurrent abnormal
returns. In Column (2), we add the returns of the source industries and their
interactions with the central dummy. The positive coefficient of 0.102 on Returns
source indicates that the industries’ returns co-move with the returns of their
trading partners (Menzly and Ozbas 2010). The association nearly doubles when
the source is a central industry, as evidenced by the coefficient of 0.097 on the
123
Earnings and stock returns
1219
interaction Central 9 Returns source. The R2s in the specification go from 0.029
in Column (1) to 0.141 in Column (2), indicating that the returns of the source and
their interaction with the central dummy have strong incremental explanatory
power in the regression. Overall, these results provide evidence that central
industries’ returns have a stronger association with their trading partners’ returns
than do noncentral industries.
Table 6, Panel B, presents the results of the specifications using 1-month-ahead
returns for the linked industries. Column (1) presents the results for baseline
specifications with inclusion of the abnormal returns for the linked industries at
month t. Consistent with the literature on momentum (e.g., Moskowitz and
Grinblatt 1999), we find that the returns of the linked industries at month t are
positively associated with the returns of the linked industries at month t ? 1. We
introduce the returns of the source and their interactions with the central dummy
in Column (2). We find a positive coefficient on the returns of the source,
significant at the 5 % level, indicating that industries predict the combined returns
of their suppliers and customers, consistent with Menzly and Ozbas (2010). The
interaction Central 9 Returns source is insignificant, indicating that industry
position, as measured by centrality, does not alter the baseline delayed reaction
between customer and supplier industries documented by Menzly and Ozbas
(2010). While the tests in Table 5 provide evidence that central industries’
earnings have a relatively strong association with their trading partners’ future
earnings, the results in Table 6, Panel B, indicate that investors are no less
efficient in impounding this information into their trading partners’ prices than
they are for noncentral industries.20
6 Conclusion
We investigate the role of inter-industry trade flows as a source of transfers of
economic shocks and information among firms. We hypothesize that an industry’s
position in the inter-industry network is an important determinant of that
industry’s exposure to aggregate shocks and the extent to which those shocks
affect the industries to which it is linked. Building on both a diversification-based
argument and on prior theoretical findings on the propagation of industry shocks,
we predict and find that the performance of firms in central industries depends
relatively more on aggregate risks than does the performance of firms in
noncentral industries.
20
In the regressions for future returns, the significance of the interaction coefficient Central 9 Returns
source is sensitive to alternative treatments. For example it loads positive and significantly when using a
different cutoff for the centrality variable at 0.10 or when replacing our measure of eigenvector centrality
with weighted degree centrality. Similar to the ROA cross-predictability tests, we ran the returns tests
separating the linked industry portfolios into central and noncentral industries. As in Table 5, we find that
centrality only has an incremental effect on concurrent returns. Similar to the ROA tests that separate the
linked portfolios by centrality, the coefficient on centrality and the t statistic are larger for the portfolio of
noncentral linked industries, but the difference between the coefficients is not statistically significant at
conventional levels.
123
1220
D. Aobdia et al.
We also document that, compared to a noncentral industry, a central industry’s
performance is more strongly associated with the concurrent and future performance
of the industries to which it is linked. This is consistent with either shocks to a
central industry’s performance propagating more strongly than shocks to a
noncentral industry’s performance, or with shocks to central industries’ trading
partners impacting central industries when those shocks are correlated. This
suggests that central industries play a particularly important role in the transmission
of shocks.
Our study adds to the literature on the gradual diffusion of information across
asset markets via inter-industry trade flows. Prior research examines the crosspredictability of returns across economically linked firms, and our findings suggest
that such associations are stronger when economic shocks flow from a central firm
to a noncentral firm than vice versa. The strong relation between central industries
and their trading partners calls attention to their usefulness in forecasting the
performance of industries with which they trade.
Our paper also contributes to the literature on the propagation of idiosyncratic
shocks in the economy. Consistent with Acemoglu et al.’s (2012) prediction that
central industries play a key role in amplification of idiosyncratic shocks into
aggregate shocks, our empirical evidence indicates that the shocks to central
industries propagate more strongly than those to noncentral ones.
Acknowledgments We thank David Aboody, Sam Bonsall, Paul Fischer (editor), Robert Freeman,
Rebecca Hann (RAST discussant), Adrienna Huffman (AAA discussant), Jack Hughes, Gil Sadka, Brett
Trueman, an anonymous referee, and seminar participants at Columbia University, University of Texas
Austin, the 2013 Temple University Accounting Conference, the 2013 Review of Accounting Studies
Conference, and the American Accounting Association 2013 Annual Meeting for helpful comments. We
also thank Sam Bonsall for providing data that identify bellwether firms.
Appendix 1
See Table 7.
123
The value of eigenvector centrality. The calculation is described in detail in Sect. 2.1. The measure is computed based on the 1997 BEA ‘‘Make’’ and ‘‘Use’’
tables
Indicator variable that is set equal to one if the centrality of the industry is [0.109 and zero otherwise
Centrality
Central
Average over a five-year period of the monthly (daily) percentage of shares traded, defined as the total monthly (daily) trading volume of the stocks in the
portfolio divided by the total number of shares outstanding, per CRSP, for the stocks in the portfolio
Logarithm of the average over a 5-year period of the portfolio’s market capitalization, computed at the beginning of each period (monthly for monthly returns
and daily for daily returns) as shares outstanding times market price in CRSP
Average Proportion
Traded
Log Average Market
Value
Monthly change in the seasonally adjusted consumer price index (from FRED)
The standard deviation of daily market returns (CRSP value-weighted market portfolio) estimated over a period of 1 month
Total dividend from the CRSP market portfolio over the past 12 months divided by the current market level
Stdev of Market
Returns
Market Dividend Rate
The difference between the yield of BAA-rated and AAA-rated bonds (from FRED)
Market-capitalization-weighted average monthly returns of the firms in the source industry less value-weighted market return for the same month
Inflation
BAA-AAA Spread
Returns source
Returns linked
Strength-weighted sum of the monthly returns on linked industries normalized by the sum of strength of all links of the source industry, less the value
weighted market return. The weights are the strength of the link between the source and the linked industries and the monthly industry returns are equal to
value weighted firm-level returns where the weights are market capitalization at the beginning of the month
Market value weighted average of the change in ROA (IBQ/Average ATQ) between quarter t and quarter t - 4 of firms in the industry
Source ROA change
Table 6
Strength-weighted sum of year-on-year changes in ROA (IBQ/Average ATQ) of linked industries normalized by the sum of strength of all links of the source
industry. The weights are the strength of the link between source and linked industries. Industry changes in ROA equal the market value weighted average of
the firm-level change in ROA (quarter t less quarter t - 4)
Linked ROA change
Table 5
Log Average Market
Value
Logarithm of the average over the sample period of the portfolio’s market capitalization, computed at the beginning of each quarter as shares outstanding
times market price in Compustat (CSHOQ 9 PRCCQ)
The value-weighted average of an indicator variable that is set equal to one for any stock in the portfolio that has analyst coverage (quarterly consensus EPS
forecast provided in I/B/E/S) for the last quarter of the estimation period and zero otherwise
Average Analyst
Coverage
Table 4
Excess return on the value-weighted industry portfolio (value-weighted CRSP return less the risk-free rate from the Fama–French database)
Industry excess return
Table 3
Description
Variable
Table 7 Variable description
Earnings and stock returns
1221
123
1222
D. Aobdia et al.
Appendix 2: Additional details on centrality measure
This appendix provides additional discussion of the measurement of centrality.
Figure 3 replicates Fig. 1 from the main body, with the exception that we use shapes
other than squares to denote industries included in the discussion of this appendix.
The triangle denotes ‘‘petroleum and coal products,’’ the pentagon denotes
‘‘employment services,’’ and the hexagon denotes ‘‘wholesale trade.’’
A basic measure of an industry’s centrality is the raw number of industries it trades
with, which is called the industry’s degree. This metric ranks wholesale trade as the
most central industry, while petroleum and coal and employment services share the
rank of 11th most central based on Figure B where we omit weak linkages between
industries.21 Representing the network by the matrix A, with elements Aij = Aji = 1 if
industry i trades with industry j, and Aii = 0, an industry’s degree is di = RjAij.
The simple count of links fails to distinguish the relative importance of an
industry’s trading partners. For example, visual inspection of Figure B suggests that
employment services, the pentagon, has more highly connected trading partners
than petroleum and coal, the triangle. The eigenvector centrality metric addresses
this. Instead of counting links (di = RjAij), eigenvector centrality weights the links
Fig. 3 Inter-industry network based on BEA input/output tables. The inter-industry network based on the
U.S. input–output matrix of 1997 provided by the BEA. Each node corresponds to an industry (based on
BEA’s four-digit industry code), and each edge corresponds to a link between industries with strength
(Aij) of 3 % or more. See Sect. 2.1 for the definition of strength of the link between two industries. The
triangle, pentagon, and hexagon show the petroleum and coal products, employment services, and
wholesale trade industries, respectively, which we refer to in the discussion in this appendix. These three
nodes, along with the square nodes, indicate industries with a centrality score above the 0.109 threshold
that we use to indicate central industries, as we discuss in Sect. 3.2
21
All industries have degrees ranging from 119 to 122 if we do not exclude weak links because every
industry has some, albeit small, amount of trade with nearly every other industry.
123
Earnings and stock returns
1223
by their importance, ci = RjAijcj. With this metric, wholesale trade, the hexagon in
Figure B, maintains the highest centrality. Employment services has the eighth
highest rank, which exceeds the degree-based rank of 11 on account of Employment
Services interacting with relatively important sectors of the economy. Petroleum
and coal ranks as 36th in terms of eigenvector centrality. Because petroleum and
coal plays a relatively peripheral role in this example where we do not distinguish
links by strength of trade, its importance in terms of eigenvector centrality drops
significantly relative to its rank in terms of degree.
The eigenvector centrality metric that we discuss in the preceding examples
suffers from a shortcoming that we remedy in the measure used in our main
analysis. The examples fail to account for heterogeneity in the amount of trade
between industries. In our main analysis, we allow the matrix A of trade links to
reflect the strength of trade. This has substantial effects on industries’ rankings. For
example, petroleum and coals’ eigenvector centrality ranks as 11th, rather than 36th,
while employment services drops to 16th when we account for the amount of trade
they conduct with other industries.
References
Acemoglu, D., Carvalho, V. M., Ozdaglar, A., & Tahbaz-Salehi, A. (2012). The network origins of
aggregate fluctuations. Econometrica, 80(5), 1977–2016.
Ahern, K. (2012). Network centrality and the cross section of stock returns. Working paper, University of
Southern California.
Ahern, K., & Harford, J. (2013). The importance of industry links in merger waves. Journal of Finance,
69(2), 527–576.
Anilowski, C., Feng, M., & Skinner, D. J. (2007). Does earnings guidance affect market returns? The
nature and information content of aggregate earnings guidance. Journal of Accounting and
Economics, 44(1–2), 36–63.
Anjos, F., & Fracassi, C. (2012). Shopping for information? Diversification and the network of industries.
Working Paper, University of Texas.
Bernard, V. L., & Thomas, J. K. (1990). Evidence that stock prices do not fully reflect the implications of
current earnings for future earnings. Journal of Accounting and Economics, 13(4), 305–340.
Bonacich, P. B. (1972). Factoring and weighing approaches to status scores and clique identification.
Journal of Mathematical Sociology, 2(1), 113–120.
Boni, L., & Womack, K. (2006). Analysts, industries, and price momentum. Journal of Financial and
Quantitative Analysis, 41(1), 85–109.
Bonsall, S., Bozanic, Z., & Fischer, P. E. (2013). What do management earnings forecasts convey about
the macroeconomy? Journal of Accounting Research, 51(2), 225–266.
Borgatti, S. (2005). Centrality and network flow. Social Networks, 27(1), 55–71.
Brown, S., Lo, K., & Lys, T. (1999). Use of R2 in accounting research: Measuring changes in value
relevance over the last four decades. Journal of Accounting and Economics, 28(2), 83–115.
Clinch, G., & Sinclair, N. (1987). Intra-industry information releases: A recursive systems approach.
Journal of Accounting and Economics, 9(1), 89–106.
Cohen, L., & Frazzini, A. (2008). Economic links and predictable returns. Journal of Finance, 63(4),
1977–2011.
Cready, W., & Gurun, U. (2010). Aggregate market reaction to earnings announcements. Journal of
Accounting Research, 48(2), 289–334.
Dupor, B. (1999). Aggregation and irrelevance in multi-sector models. Journal of Monetary Economics,
43(2), 391–409.
123
1224
D. Aobdia et al.
Durnev, A., Morck, R., Yeung, B., & Zarowin, P. (2003). Does greater firm-specific return variation mean
more or less informed stock pricing? Journal of Accounting Research, 41(5), 797–836.
Foster, G. (1981). Intra-industry information transfers associated with earnings releases. Journal of
Accounting and Economics, 3(3), 201–232.
Gabaix, X. (2009). Power laws in economics and finance. Annual Review of Economics, 1, 255–293.
Gabaix, X. (2011). The granular origins of aggregate fluctuations. Econometrica, 79(3), 733–772.
Hong, H., Torous, W., & Valkanov, R. (2007). Do industries lead stock markets? Journal of Financial
Economics, 83(2), 367–396.
Horvath, M. (1998). Cyclicality and sectoral linkages: Aggregate fluctuations from sectoral shocks.
Review of Economic Dynamics, 1(4), 781–808.
Horvath, M. (2000). Sectoral shocks and aggregate fluctuations. Journal of Monetary Economics, 45(1),
69–106.
Jackson, M. (2008). Social and economic networks. Princeton, NJ: Princeton University Press.
Kadan, O., Madureira, L., Wang, R., & Zach, T. (2012). Analysts’ industry expertise. Journal of
Accounting and Economics, 54(2–3), 95–120.
Kelly, P. J. (2005). Information efficiency and firm-specific return variation. Working Paper, New
Economic School, Moscow.
Liu, J., & Ohlson, J. (2000). The Feltham–Ohlson (1995) model: Empirical implications. Journal of
Accounting, Auditing and Finance, 15(3), 321–331.
Lucas, R. E. (1977). Understanding business cycles. Carnegie-Rochester Conference Series on Public
Policy, 5, 7–29.
Menzly, L., & Ozbas, O. (2010). Market segmentation and cross-predictability of returns. Journal of
Finance, 65(4), 1555–1580.
Montgomery, J., & Vames, S. (1998, June 23). Strike at GM could weigh down second-quarter GDP in
U.S. The Wall Street Journal.
Morck, R., Yeung, B., & Yu, W. (2000). The information content of stock markets: Why do emerging
markets have synchronous stock price movements? Journal of Financial Economics, 58(1–2),
215–260.
Moskowitz, T. J., & Grinblatt, M. (1999). Do industries explain momentum? Journal of Finance, 54(4),
1249–1290.
Newman, M. (2010). Networks: An introduction. Oxford: Oxford University Press.
Pandit, S., Wasley, C. E., & Zach, T. (2011). Information externalities along the supply chain: The
economic determinants of suppliers’ stock price reaction to their customers’ earnings announcements. Contemporary Accounting Research, 28(4), 1304–1343.
Papke, L. E., & Wooldridge, J. M. (1996). Econometric methods for fractional response variables with an
application to 401(k) plan participation rates. Journal of Applied Econometrics, 11(6), 619–632.
Piotroski, J. D., & Roulstone, D. T. (2004). The influence of analysts, institutional investors, and insiders
on the incorporation of market, industry and firm-specific information into stock prices. The
Accounting Review, 79(4), 119–1151.
Roll, R. (1988). R2. Journal of Finance, 43(3), 541–566.
Shepardson, D. (2009, November 24). Fixing Detroit’s automakers one year later (first of two parts); The
inside story of the GM, Chrysler bailouts. The Detroit News.
Shivakumar, L. (2007). Aggregate earnings, stock market returns and macroeconomic activity: A
discussion of ‘Does earnings guidance affect market returns? The nature and information content of
aggregate earnings guidance’. Journal of Accounting and Economics, 44(1–2), 64–73.
Shivakumar, L. (2010). Discussion of aggregate market reaction to earnings announcements. Journal of
Accounting Research, 48(2), 335–342.
Thompson, S. (2011). Simple formulas for standard errors that cluster by both firm and time. Journal of
Financial Economics, 99(1), 1–10.
123
Was this manual useful for you? yes no
Thank you for your participation!

* Your assessment is very important for improving the work of artificial intelligence, which forms the content of this project

Download PDF

advertisement