cross-industry momentum - semantic scholar · 2015-07-29 · cross-industry momentum ∗ lior...

Cross-Industry Momentum∗

Lior Menzly†

University of Southern California

Oguzhan Ozbas‡

University of Southern California

December 2004

Abstract

Industries related to each other through the supply chain (upstream or downstream) exhibit

strong cross-momentum. Cross-industry momentum is distinct from previously documented

firm and industry-level momentum. Trading strategies that consist of buying (selling) industries

with large positive (negative) returns to their related industries over the previous month yield

significant profits.

∗We wish to thank an anonymous referee and Mark Grinblatt for helpful comments and suggestions. Thanks also

to seminar participants at UCLA Anderson School of Management and USC Center for Investment Studies. Financial

support from the Marshall General Research Fund is gratefully acknowledged.

†Marshall School of Business, Hoffman Hall 701, Los Angeles, CA 90089; home.uchicago.edu/~liorm

‡Marshall School of Business, Hoffman Hall 701, Los Angeles, CA 90089; www-rcf.usc.edu/~ozbas

1 Introduction

A large body of empirical research now documents strong short to medium-term persistence in

stock returns. Jegadeesh and Titman (1993) find that buying past winners and selling past losers

can generate significant abnormal profits. A sizeable portion of these so-called momentum profits

appears to be based on persistence in industry returns (Moskowitz and Grinblatt, 1999).

The present paper adds to this line of work by documenting a new kind of return persistence —

positive cross-correlation among industries that are related to each other through the supply chain

— which we call cross-industry momentum to distinguish it from prior works on own-momentum

(hereafter momentum) at either the stock or industry level. Based on inter-industry flow of goods

and services from the Input-Output Benchmark Survey of the Bureau of Economic Analysis, we

find that related industry returns (upstream and downstream) lead industry returns. We also find

cross-industry momentum effects to be distinct from previously documented firm and industry level

momentum effects.

In addition, we explore the economic significance of our results and find that trading strategies

based on cross-industry effects produce economically significant profits. For example, a trading

strategy that consists of buying (selling) industries with large positive (negative) returns to their

upstream industries over the previous month yields an annual premium of more than 6 percent and

a Sharpe ratio of 0.7.

To motivate our empirical analysis of cross-industry momentum, we propose a simple limited

information model in which only a certain fraction of investors receive informative signals about

risky asset payoffs. Since then investors hold heterogenous beliefs, information gets impounded into

prices partially and gradually over time.1 More relevant to the analysis of this paper, we show that

cross-momentum arises naturally among risky assets with positively correlated fundamentals.

1Holden and Subrahmanyam (2002) provide a similar rationale for momentum.

1

Using firm-level data from Compustat, we find supportive evidence that related industries as

defined in our analysis indeed have positively correlated fundamentals. This result may appear

surprising at first if one’s prior is that related industries play what ultimately is a strategic zero-sum

game and that what is good news for a supplying industry must be bad news for the buying industry.

While in reality such strategic interaction undoubtedly takes place among related industries, it

would appear that an equally plausible alternative effect — the fortunes of related industries rising

and falling together due to demand or technology shocks — dominates empirically.

In a number of robustness tests, we find that large stocks (market capitalization above the

median NYSE cutoff) do not exhibit cross-industry momentum. That prices of large stocks are

efficient with respect to stock prices in related industries is supportive of our limited information

motivation for studying cross-industry momentum because both momentum and cross-momentum

effects arise in the model when some investors are more informed than others. One would expect

such differences in information among investors to be less prevalent for large firms than small firms.

For example, if information intermediaries focus most of their research efforts on large firms (Hong,

Lim and Stein (2000) show that large stocks have better analyst coverage) and as a result even out

informational differences among investors, then one would expect stock prices of large firms to be

efficient with respect to developments in related industries.

It is important to note that our findings of cross-industry momentum are robust to the exclusion

of stocks with market capitalization below the 20th NYSE percentile, and so the cross-industry

effects that we document in this paper are distinct from findings of delayed price response among

micro-cap stocks. (Also see Lo and MacKinlay (1990), Brennan, Jegadeesh and Swaminathan

(1993), Mech (1993), Hou (2002), Hou and Moskowitz (2002).) Furthermore, in contrast to delayed

price response that appears to be a predominantly intra-industry effect with large firms leading

small firms (Hou, 2002), cross-industry momentum is an inter-industry effect.

2

Finally, we investigate whether there is any systematic difference between economically large

and small supply chain relationships. To address this question, we allocate an industry’s upstream

and downstream industries to one of six portfolios (three upstream and three downstream) based on

the amount of inter-industry flow of goods and services reported in the Input-Output Benchmark

Survey and then explore the predictive power of these portfolio returns for industry returns.2 For

the most part, we find that cross-industry momentum is due to small and medium supply chain

relationships. Related industry portfolios consisting of the largest supply chain relationships have

no predictive power for industry returns, indicating that stock prices are efficient with respect to

developments in upstream and downstream industries that matter the most economically. As for

the model, there is still the question of why readily available past returns are not fully reflected in

prices. One possibility is that some investors find it costly to gather that information. Or it could

be that some investors are not fully aware of every supply chain relationship.

To our knowledge, the idea that limited information could generate persistent returns across

fundamentally related risky assets is largely unexplored. Perhaps most related to the present paper

is work by Moskowitz and Grinblatt (1999) who document autocorrelation in industry returns

and conjecture that this could be due to cross-correlation among stocks within the same industry,

though they do not provide direct evidence of these conjectured cross links like we do here for

transacting industries. Hong, Lim and Stein (2000) focus on autocorrelation in stock returns and

find that stocks with more analyst coverage exhibit shorter momentum which they interpret as

2We construct three upstream portfolios such that they represent an equal amount of inter-industry trade. To

be more concrete, the first upstream portfolio for an industry contains those upstream industries that supply it the

least yet together provide one-third of its purchases in total whereas the third upstream portfolio contains upstream

industries that supply it the most and also provide one third of its purchases (with the remaining upstream industries

providing one-third of its supplies assigned to the second upstream portfolio). We form three downstream portfolios

in the same way.

3

faster diffusion of dispersed information. Hong, Torous and Valkanov (2003) show that a number of

industries lead the stock market by up to two months, which is a consistent aggregate implication

of cross-industry momentum.

The paper proceeds as follows. Section 2 outlines a simple limited information model to motivate

our study of cross-industry momentum. Section 3 briefly describes the data sources and return series

used throughout the paper. In Section 4, we discuss our empirical approach and present our basic

findings about cross-industry persistence in returns. Section 5 explores trading strategies based on

cross-industry momentum. We present some evidence of systematic difference between large and

small supply chain relationships in Section 6 and provide concluding remarks in Section 7.

2 The Model

In this section, we outline a limited information model that emphasizes differences in information

and consequently differences in beliefs among investors, which together with limited risk bearing

capacity lead to momentum and cross-momentum in prices. To keep the analysis simple, we use

the standard CARA-normal setup and study competitive Walrasian equilibria.3 We provide formal

proofs of our propositions in the Appendix.

We first show that prices do not exhibit momentum when investors have homogeneous beliefs.

That is, we show that prices do not exhibit momentum when every investor receives the same

informative signal and as a result has the same posterior belief about future asset payoffs. This

example serves as an useful benchmark to the case in which only a certain fraction of the population

3Alternatively, one could use the noisy rational expectations equilibrium concept in which prices serve the dual

role of clearing markets and updating beliefs. Since prices would still not be fully revealing as in Grossman and

Stiglitz (1982), informational differences would remain and investors would hold heterogeneous beliefs, which would

then lead to momentum (Holden and Subrahmanyam, 2002) and cross-momentum in prices.

4

receives the informative signal. When investors differ in their information sets by having or not

having the informative signal, differences in posterior beliefs arise naturally. Once investors have

different posterior beliefs, we show that prices incorporate information partly and gradually, and

that prices exhibit momentum.4

Finally, we show that the above result about price momentum in one risky asset extends to

price momentum across risky assets with positively correlated payoffs. Specifically, we show that

equilibrium prices of two risky assets with positively correlated payoffs exhibit cross-momentum

when a fraction of the population receives an informative signal about one of the assets with further

implications for the other asset.

2.1 Momentum

Suppose that there are three dates t ∈ {0, 1, 2}, a single risky asset in positive supply z that pays

a liquidating dividend d at t = 2, and a riskless asset whose gross payoff is normalized to 1 and

hence is the numareire. There are n investors in the economy with constant absolute risk aversion

parameter a. Investors trade the risky asset at t = 0 and t = 1, and then consume the liquidating

dividend at t = 2. Their common prior belief at t = 0 is that d ∼ N¡d̄, σ2d

¢. At t = 1, investors

receive an informative but noisy signal s about d where s = d+ε and ε ∼ N¡0, σ2ε

¢. The informative

signal allows investors to update their beliefs about d and adjust their demands for the risky asset

at t = 1.

Proposition 1 When every investor receives the informative signal s about d, equilibrium prices

do not exhibit momentum.

4Skill driven differences in information acquisition costs among investors could plausibly lead to such equilibria —

in which only investors with information processing skills above a certain threshold (or equivalently, acquisition costs

below a certain threshold) choose to acquire informative signals.

5

However, when only a fraction α ∈ (0, 1) of the population receives the informative signal,

equilibrium prices exhibit momentum because the informative signal gets incorporated less than

fully at t = 1 due to informed investors’ limited risk bearing capacity.

Proposition 2 When only a fraction α ∈ (0, 1) of the population receives the informative signal s

about d, equilibrium prices exhibit momentum.

2.2 Cross-Momentum

Suppose now that there are two risky assets k ∈ {1, 2} both in positive supply z with correlated

liquidating dividends d1 and d2. The common prior belief at t = 0 is that (d1, d2) ∼ N¡d̄1,Σ

¢where

Σ =

⎡⎢⎢⎣ σ2d ρσ2d

ρσ2d σ2d

⎤⎥⎥⎦ . (1)

At t = 1, some investors receive an informative but noisy signal s1 where s1 = d1 + ε and

ε ∼ N¡0, σ2ε

¢.

Proposition 3 When only a fraction α ∈ (0, 1) of the population receives the informative signal

s1 about d1 and ρ > 0, equilibrium prices exhibit cross-momentum.

When only a fraction α ∈ (0, 1) of the population receives the informative signal about one of

the assets with implications for the other asset (ρ > 0), equilibrium prices exhibit cross-momentum.

3 Data

CRSP monthly return files (January 1963 - December 2002) and the Input-Output Benchmark

Survey of the Bureau of Economic Analysis (the BEA Survey, hereafter) constitute our main data

sources. The BEA Survey assigns overall economic activity in the U.S. into one of 85 industry

accounts, and reports the extent of inter-industry flow of goods and services among these industries.

6

Using the industry account-SIC code dictionary provided as part of the BEA Survey, we first

assign firms to their respective industries based on their reported SIC codes in CRSP.5 We then

calculate value-weighted monthly industry returns. Finally, we transform the macroeconomic data

from the BEA Survey into portfolio weights. Using these weights, we construct the return on the

representative supplier (upstream portfolio) and customer (downstream portfolio) for each industry.

Formally, our return series are calculated as follows:

Ri,t =Xj∈i

Mj,t−1Pj∈i

Mj,t−1Rj,t (2)

Rusi,t =

Xk 6=i

Ck,iPk 6=i

Ck,iRk,t (3)

Rdsi,t =

Xk 6=i

Ci,kPk 6=i

Ci,kRk,t (4)

where Ri,t is the value-weighted return of industry i in month t, Rj,t is the stock return of firm j

(in industry i) in month t, Mj,t−1 is the market capitalization of firm j at the end of month t− 1,

Rusi,t

³Rdsi,t

´is the upstream (downstream) return of industry i in month t, and Ck,i (Ci,k) is the

flow of goods and services from industry k (i) to industry i (k).

Table I presents summary statistics for the ultimate 65 industries used in this paper — out of

the original 85 industry accounts, we drop eight catch-all accounts (mainly related to government,

import and inventory adjustments) to avoid potential measurement error, another 11 accounts due

to a lack of sufficient number of firms to form meaningful industry returns and the wholesale/retail

industry account.6 Compared to widely used 48 Fama-French industries (Fama and French, 1997),

mean and standard deviation characteristics appear reasonable. Also Table A-I presents evidence

5 In a small number of instances where a SIC industry is assigned to more than one industry account, we keep

the first entry in the dictionary and drop the remaing entries to prevent any chance of a hard-wired cross-industry

momentum result based on own industry momentum (Moskowitz and Grinblatt, 1999).

6The problem with the wholesale/retail industry account is that the BEA Survey lumps all wholesale and retail

activity in the economy into one industry account. As a result, the industry account-SIC code dictionary, which we

7

that related industries as defined in our analysis have positively correlated fundamentals as mea-

sured by return on assets.

We should mention one important caveat before proceeding with our analysis — the volume of

inter-industry trade used in our construction of upstream and downstream returns comes from the

1987 BEA Survey. Obviously, we would like to avoid using data that only became available after

the fact for some of the years in our predictive regressions and trading strategies. However, we are

constrained by the fact that the BEA first published its survey in 1982. Since then, there have

been just three more surveys (1987, 1992 and 1997) as these surveys are published once in every

five years using comprehensive Census data. Moreover, out of the already limited four options that

we have, we cannot use the 1997 survey because the BEA switched the way it classified industries

in 1997 from SIC to NAICS, which makes the link to CRSP problematic.

As a robustness check, we have run our programs using the 1982 and 1992 surveys. The results

are essentially unchanged and so for brevity we report results using the 1987 survey only. We obtain

almost identical results because the correlation among the surveys is relatively high. Specifically, the

correlation of upstream portfolio weights, Ck,iPk 6=i

Ck,i, is 0.9647 between 1987 and 1982, 0.9628 between

1987 and 1992, and 0.9456 between 1982 and 1992. The corresponding statics for downstream

portfolio weights, Ci,kPk 6=i

Ci,k, are 0.9474, 0.9466, and 0.9382. It is clear from these numbers that the

way industries interact with each other has not changed much over time, and that there is a high

degree of structural stability in the upstream and downstream relationships. So even if there were

surveys before the beginning of our sample period in 1963, it seems unlikely that they would have

differed significantly from the 1987 survey.

use to translate industry accounts into SIC codes, is coarse and has the wholesale/retail industry account linked with

every wholesale and retail SIC code. Because of this lack of precision, we exclude the wholesale/retail account to

prevent measurement error in computing related industry (upstream and downstream) returns.

8

4 Empirical Methodology and Basic Findings

Our main empirical approach entails estimation of panel and Fama-MacBeth regressions to predict

industry returns using lagged returns in related upstream and downstream industries. We use the

same standard techniques that have been used quite extensively in previous work on momentum.

4.1 Panel Regressions

To explore cross-industry persistence in industry returns, we estimate variants of the following

panel regression:

Ri,t = at + b1 ∗Rusi,t + b2 ∗Rds

i,t + c1 ∗Ri,t−1 + c2 ∗Rusi,t−1 + c3 ∗Rds

i,t−1 (5)

where Ri,t, Rusi,t and R

dsi,t are contemporaneous own, upstream and downstream returns of industry

i in month t, respectively. We include month fixed effects at to absorb systematic market return in

month t and compute standard errors that are robust to clustering (dependence of error terms) at

the monthly level.7 Naturally, we are interested in the predictive power of Ri,t−1, Rusi,t−1 and R

dsi,t−1

which are lagged own, upstream and downstream returns of industry i in month t− 1, respectively.

Table II reports our results in five columns. The first column shows that Ri,t is strongly

and contemporaneously related to returns in upstream and downstream industries as evidenced

by statistically significant positive coefficients on Rusi,t (0.335) and Rds

i,t (0.208). The strength and

magnitude of these coefficients shows that the BEA Survey is providing us with economically

meaningful industry relationships. It is quite striking that we can identify a group of related

industries and explain industry returns beyond the systematic market return captured with the

month fixed effect at.

We next investigate own industry momentum (Moskowitz and Grinblatt, 1999) in column 2.

7See Froot (1989), Rogers (1993) and Wooldridge (2002, section 13.8.2).

9

Consistent with earlier findings of industry momentum, we find that high past industry returns

predict high future industry returns as evidenced by a statistically significant positive coefficient

on Ri,t−1 (0.030). While the industry definitions used in this paper are somewhat different from

those used in the literature, it is reassuring to see that the BEA Survey industry accounts exhibit

similar return patterns that have been found to be significant in previous work.

Having established these basic results, we explore the predictive power of returns in upstream

and downstream industries in column 3. As evidenced by statistically significant positive coefficients

on Rusi,t−1 (0.058) and Rds

i,t−1 (0.038), we find strong cross-industry momentum from upstream and

downstream industries. Compared to own industry momentum, cross-industry momentum from

upstream industries appears to be more than twice as large, and almost 50 percent larger from

downstream industries. Combined together, cross-industry momentum from both upstream and

downstream industries presents a magnitude of predictability that is more than three times that of

own industry momentum.

Finally in columns 4 and 5, we test whether these different sources of return predictability

interact with each other in any meaningful way. It appears that cross-momentum from upstream

and downstream industries and own industry momentum are pair-wise independent from each

other as evidenced by statistically insignificant coefficients on interaction terms Rusi,t−1 ∗ Rds

i,t−1,

Rusi,t−1 ∗Ri,t−1 and Rds

i,t−1∗ Ri,t−1 at the one-percent level.

4.2 Fama-MacBeth Regressions

To further explore the statistical robustness and economic significance of our results so far about

strong cross-industry momentum from upstream and downstream industries, we estimate Fama-

MacBeth type regressions in this subsection. Essentially, we run regressions of the following form

10

for each month:

Ri,t = a+ λmarket ∗ bβmarket

i,t + λHML ∗ bβHML

i,t + λSMB ∗ bβSMB

i,t (6)

+b1 ∗Ri,t−1 + b2 ∗Rusi,t−1 + b3 ∗Rds

i,t−1

where bβmarket

i,t , bβHML

i,t , and bβSMB

i,t are return betas for industry i in month t (estimated using

industry returns in months t − 60 through t − 1) to control for known patterns in stock returns.

Our specifications leave out contemporaneous upstream and downstream returns so that we can

ultimately interpret our results as premiums that can be replicated with a tradeable portfolio

strategy.

After estimating variants of specification (6) for each month, we calculate means and standard

deviations of estimated coefficients (assuming independence of estimated coefficients across each

monthly regression), and report our results in Table III. Column 3 shows that previous findings of

own industry momentum are robust to controlling for systematic book-to-market and size effects

as evidenced by a statistically significant positive coefficient on Ri,t−1 (0.029).

We include our main predictive variables, namely lagged upstream and downstream returns, in

column 4 and find that cross-industry momentum is also robust to controlling for systematic book-

to-market and size effects as evidenced by statistically significant positive coefficients on Rusi,t−1

(0.079) and Rdsi,t−1 (0.050). Once again, the combined magnitude of cross-industry momentum from

both upstream and downstream industries appears to surpass that of own industry momentum —

the combined total is a quite remarkable annual premium of 12.9%.

In column 5, we further control for industry book-to-market and size directly, and find a similar

combined premium of 12.5%. Also in unreported firm-level Fama-MacBeth regressions that control

for firm-level short term reversal effects at the one-month horizon and medium term momentum

effects at the one-year horizon, coefficients on both Rusi,t−1 and R

dsi,t−1 continue to remain statistically

and economically significant.

11

5 Trading Strategies

Based on our findings of robust cross-industry momentum, we now turn our attention to trading

strategies that exploit this apparent return predictability. In addition to assessing the profitability

of these trading strategies, our goal is to explore and better pinpoint the source of cross-industry

momentum profits.

Our trading strategies consist of ranking industries into five bins on a monthly basis (presumably

at the beginning of a month) based on returns in their upstream or downstream industries in the

previous month. After ranking industries in this fashion, we construct zero-cost equal-weighted and

value-weighted (using industry market capitalization) portfolios that buy industries in the high bin

and sell industries in the low bin.

Table IV reports our results. In Panel A, trading strategies based on cross-industry momentum

from upstream industries yield annual premiums of more than 6%. The premium on the equal-

weighted strategy is slightly higher than the premium on the value-weighted strategy (6.8% versus

6.5%) whereas the Sharpe ratio of the equal-weighted strategy is significantly better than that of

the value-weighted strategy (0.852 versus 0.573). Figure 1 graphs potential cumulative trading

profits beyond the first month.

Interestingly, trading strategies based on cross-industry momentum from downstream industries

(reported in Panel B) yield annual premiums of 5.6% and 6.0% for equal- and value-weighted

strategies, respectively — which are slightly higher than the estimates in our panel and Fama-

MacBeth regressions, though it is important to note that our typical trading strategy is built on a

single sort and thus is not perfectly comparable to the multivariate regressions of Section 3.

12

5.1 Robustness Checks

We next perform a number of robustness checks. First, we look into whether there are any large

differences in the market capitalization of industries in the long and short legs of our trading

strategies and find nothing that may pose implementation problems (Panel C in Table IV). The

mean industry market capitalization in the long leg of the upstream strategy is $8,351 million,

which is only slightly more than the mean industry market capitalization of $7,722 million in the

short leg. Likewise, the difference between the long and short legs of the downstream strategy is

negligible ($6,180 million versus $6,129).

Second, we look at whether only a small number of industries enter into our trading strategies

and drive our results, and find that while there is some heterogeneity in inclusion probabilities (Table

V), the amount is not excessive to pose implementation problems. For example, if industries were

identical in their inclusion probabilities, they would enter our trading strategies roughly 1.5 percent

of the time. It turns out that the maximum inclusion probability of an industry rarely exceeds 3

percent in any of the leg-strategy combinations and, when it does, the industry in question also

appears in the opposite leg of the strategy with a high probability. This last fact is especially

important because it shows that the profitability of our trading strategies is not simply driven by

going long (short) in industries with historically high (low) returns.

Third, we regress monthly returns from our trading strategies on widely recognized return

factors such the Fama-French HML and SMB, and Carhart MOM, and find that cross-industry

momentum returns are essentially orthogonal to them (Table VI). The intercept is 0.5% for both

upstream and downstream strategies — which translates into an annualized Jensen’s alpha of 6%

in line with our previous results. We also try to alleviate a potential concern that our trading

strategies might be timing the return factors — increase and decrease exposure during positive

and negative realizations, respectively. In even-numbered specifications in Table VI, we include

13

interaction terms with a business-cycle indicator variable using NBER recession dates and find no

sign of factor timing.

Fourth, we return to the multivariate panel regression approach of Section 3 to alleviate a

concern that our single-sort cross-industry momentum trading strategies might be benefiting from

own-industry momentum. In Table VII, we estimate variants of the following panel regression:

Ri,t = at + bLong ∗ 1hRi,t−1 ≥ pRk,t−1,80%

i+ bShort ∗ 1

hRi,t−1 ≤ pRk,t−1,20%

i(7)

+cLong ∗ 1hRusi,t−1 ≥ pRus

k,t−1,80%

i+ cShort ∗ 1

hRusi,t−1 ≤ pRus

k,t−1,20%

i+dLong ∗ 1

hRdsi,t−1 ≥ pRds

k,t−1,80%

i+ dShort ∗ 1

hRdsi,t−1 ≤ pRds

k,t−1,20%

iwhere the indicator variables proxy for our trading strategies by taking on a value of one for

observations that fall into either a high bin (above the 80th percentile) or a low bin (below the 20th

percentile) for the corresponding return variable (previous month own, upstream or downstream

industry return).8 Columns 1 through 3 include each trading strategy separately and thus represent

the profitability of single-sort strategies. Compared to the profitability of own industry-momentum

of about 6.5% annually (19 basis points on the long leg and 35 basis points on the short leg multiplied

by 12), upstream and downstream cross-momentum strategies yield 6.8% (29 basis points on the

long leg and 28 basis points on the short leg multiplied by 12) and 5.6% (17 basis points on the long

leg and 30 basis points on the short leg multiplied by 12), respectively.9 In column 4, we include

all three trading strategies simultaneously and find essentially the same results — it appears that

the returns from the three trading strategies are fairly orthogonal to each other. In column 5, we

exclude the month of January and find similar results. In columns 6 and 7, we split the sample

into two halves, 1963-1982 and 1983-2002, and find that while profitability has remained robust,

8We thank Mark Grinblatt for suggesting this specification as a robustness check.

9Note that these two upstream and downstream yields are numerically equivalent to the equal-weighted trading

results in Table IV because the univariate regression coefficients amount to equal-weighted sample average returns.

14

its composition has changed somewhat over time. Most notably, it appears that the profitability

of the upstream cross-momentum strategy has switched from the short leg in the first half of the

sample to the long leg in the second half, indicating a potentially easier source of profits. In column

8, we explore the joint profitability of the upstream and downstream cross-momentum strategies

and find an annual premium of roughly 9.3% (35 basis points on the long leg and 43 basis points

on the short leg multiplied by 12).

Finally, we plot the historical profitability of upstream and downstream strategies in Figure 2.

Except at the very end of the sample period, profits appear to be remarkably consistent over time.

In Figure 3, we provide some historical comparison with Fama-French HML and SMB, and Carhart

MOM returns.

5.2 Large versus Small Stocks

Our trading strategies produce insignificant results when we restrict attention to the universe of

large stocks (market capitalization above the median NYSE cutoff). Table VIII reports these

results. Both upstream and downstream strategies yield essentially no profits.

The absence of cross-industry momentum among large stocks may suggest that our findings are

purely driven by small stocks. Table IX shows that is not the case. When we restrict attention

to firms with market capitalizations above the 20th percentile NYSE cutoff, we obtain significant

results for both upstream and downstream strategies in the order of 5% and 6% as before. And so,

cross-industry momentum is not driven by micro stocks.

We can think of four reasons that may explain these results. First, it could be that a single

SIC code from CRSP does not do a good job of describing a large firm’s overall business. That

is, we may not have as precise a set of related upstream and downstream industries for large firms

as we do have for small firms and the insignificant results with large stocks may be due to this

15

measurement error. Second, it could be that prices of large stocks incorporate most information

from related industries, no matter how dispersed — say because large stocks are actively followed by

information intermediaries who specialize in collecting and analyzing such dispersed information.

We know from prior research that large stocks have better analyst coverage. Third, it could be

that we reduce the information content of upstream and downstream returns when we restrict

attention to large stocks — especially if it is information particularly in the prices of small stocks

that diffuses gradually to both large and small stocks over time. Fourth, it could be that most large

firms are integrated into upstream and downstream industries and that they do not have clear firm

boundaries that we can identify with a single SIC code from CRSP.

These are all interesting possibilities and ultimately relate to alternative economic mechanisms

that could be behind cross-industry momentum. The investigation of these interesting possibilities

requires non-CRSP data sources such as IBES to explore the role of analyst coverage and Compustat

segment files to obtain detailed data about business composition. In related current work, we are

exploiting these databases to better understand cross-industry momentum.

6 Large versus Small Supply Chain Relationships

In this section, we investigate whether there is any systematic difference between economically large

and small supply chain relationships. To address this question, we allocate an industry’s upstream

and downstream industries to one of six portfolios (three upstream and three downstream) based on

the amount of inter-industry flow of goods and services reported in the Input-Output Benchmark

Survey and then explore the predictive power of these portfolio returns for industry returns.

We construct the three upstream portfolios so that they contain an equal amount of inter-

industry trade. Specifically, we first sort upstream industries according to their sales to the reference

industry in ascending order and then allocate them to one of three portfolios: 0-33%, 33-66% and

16

66-100%. The 0-33% portfolio is comprised of a set of upstream industries whose sales are the

smallest and add up to 33% of the total purchases of the reference industry. The 33-66% portfolio

is comprised of the next set of upstream industries whose sales constitute another 33% of the

total. The 66-100% portfolio is comprised of the remaining upstream industries. We form three

downstream portfolios in the same way and estimate regressions of the form:

Ri,t = at + b1 ∗Ri,t−1 + b2 ∗Rus:0−33%i,t−1 + b3 ∗Rus:33−66%

i,t−1 + b4 ∗Rus:66−100%i,t−1 (8)

+b5 ∗Rds:0−33%i,t−1 + b6 ∗Rds:33−66%

i,t−1 + b7 ∗Rds:66−100%i,t−1

Table X reports our results in three columns. The first column shows that a significant portion

of upstream cross-industry momentum comes from the smallest relationships as evidenced by a

statistically significant coefficient on Rus:0−33%i,t−1 (0.069). The coefficient on Rus:33−66%

i,t−1 (0.034) is

also significant but less than half the coefficient on Rus:0−33%i,t−1 . Interestingly, the economically

most important upstream relationships represented by Rus:66−100%i,t−1 do not appear to be behind our

findings of cross-industry momentum. Some descriptive statistics might help to give a sense of the

typical upstream relationships covered by the three portfolios. The mean upstream supply ratio,

Ck,iPk 6=i

Ck,i, is 1%, 11% and 18% in the 0-33%, 33-66% and 66-100% portfolios, respectively.

We next consider downstream relationships in column 2 and find that the entire downstream

cross-industry momentum comes from the smallest relationships as evidenced by a statistically

significant coefficient on Rds:0−33%i,t−1 (0.049) — neither the 33-66% nor the 66-100% portfolio appears

to be behind our findings of cross-industry momentum. Once again, some descriptive statistics

might help. The mean downstream purchase ratio, Ci,kPk 6=i

Ci,k, is 1%, 12% and 16% in the 0-33%,

33-66% and 66-100% portfolios, respectively. We consider upstream and downstream portfolios

simultaneously in column 3, and obtain similar results.

17

7 Conclusion

This paper documents strong cross-industry persistence in returns. Based on data about the flow

of goods and services between industries from the BEA Survey, we show that industry returns

lag returns in related industries (upstream and downstream). Trading strategies that consist of

buying (selling) industries with large positive (negative) returns to their upstream or downstream

industries over the previous month yield significant profits.

We interpret these findings as suggestive of partial and gradual diffusion of information across

fundamentally related risky assets. In ongoing work, we are investigating a similar phenomena in

the formation of analyst earnings expectations.

A potentially interesting extension could be to investigate whether the stock markets of related

economies exhibit cross-momentum. Analogous to the flow of goods and services in the Input-

Output Benchmark Survey, one could use import and export data to measure economic linkages

between countries.

More work lies ahead before we know whether cross-industry momentum is a widespread return

pattern in other stock markets. Fortunately, the Input-Output Benchmark Survey of the Bureau

of Economic Analysis is a fairly common type of analytical exercise carried out in most OECD

countries. In addition to exploring cross-industry momentum internationally, a cross-country study

can also shed some light on how financial sophistication (as measured by the speed with which

fundamental information from related industries is processed and priced) is related to the level of

financial development or the structure of the financial system.

18

A Appendix

Proof of Proposition 1. Investor demand for the risky asset at t = 0 is

x0 =d̄− p0aσ2d

. (9)

Market clearing requires

nx0 = z. (10)

Substituting in investor demand yields

p0 = d̄− az

nσ2d (11)

After investors receive the informative signal s, their demand for the risky asset at t = 1 becomes

x1 =Ed|s − p1

aσ2d|s(12)

where

Ed|s = d̄+σ2d

σ2d + σ2ε| {z }βs

¡s− d̄

¢(13)

σ2d|s = σ2d

µ1− σ2d

σ2d + σ2ε

¶. (14)

Posterior beliefs about the liquidating dividend come from a normal projection of s on d

d = d̄+ βs¡s− d̄

¢+ ηs (15)

where the residual uncertainty about the liquidating dividend ηs is distributed N³0, σ2d|s

´. By the

optimality of the projection

ηs ⊥¡s− d̄

¢. (16)

Market clearing at t = 1 requires

nx1 = z. (17)

19

Substituting in investor demand yields

p1 = d̄+ βs¡s− d̄

¢− az

nσ2d|s.

Without loss of generality, define returns

r1 = p1 − p0 (18)

r2 = p2 − p1 (19)

and see that equilibrium prices do not exhibit momentum because the informative signal is incor-

porated fully at t = 1.

Cov (r1, r2) = Cov¡βs¡s− d̄

¢, d− βs

¡s− d̄

¢¢(20)

= Cov¡βs¡s− d̄

¢, ηs¢

(21)

= 0£ηs ⊥

¡s− d̄

¢¤(22)

Proof of Proposition 2. The equilibrium price at t = 0 is the same as in Proposition 1 and

so

p0 = d̄− az

nσ2d.

At t = 1, α (ni/n) fraction of the population receives the informative signal s and their demand

for the risky asset at t = 1 becomes

xi1 =Ed|s − p1

aσ2d|s. (23)

For (1− α) (nu/n) fraction of the population, demand for the risky asset at t = 1 is

xu1 =d̄− p1aσ2d

.


αnxi1 + (1− α)nxu1 = z. (24)

20

Substituting in investor demands XI1 and XU

1 yields

p1 = d̄+ασ2d

ασ2d + (1− α)σ2d|s| {z }0<γ<1

βs¡s− d̄

¢− az

n

σ2d|sσ2d

ασ2d + (1− α)σ2d|s

Because the informative signal is not incorporated fully at t = 1, equilibrium prices exhibit mo-

mentum.

Cov (r1, r2) = Cov¡γβs

¡s− d̄

¢, (1− γ)βs

¡s− d̄

¢+ ηs

¢(25)

= γ (1− γ)β2s¡σ2d + σ2ε

¢(26)

Proof of Proposition 3. Investor demand for the two risky asset at t = 0 is

X0 =1

aΣ−1

¡d̄1− P0

¢.


nX0 = z1.

Substituting investor demand yields

P0 = d̄1− az

nΣ1.

After a fraction α of the population receives the informative signal s1, their demand at t = 1

becomes

Xi1 =

1

aΣ−1s1

⎛⎜⎜⎝⎡⎢⎢⎣ Ed1|s1

Ed2|s1

⎤⎥⎥⎦− P1

⎞⎟⎟⎠ (27)

21

where

Ed1|s1 = d̄+σ2d

σ2d + σ2ε| {z }βd1,s1

¡s1 − d̄

¢(28)

Ed1|s1 = d̄+ρσ2d

σ2d + σ2ε| {z }βd2,s1

¡s1 − d̄

¢(29)

Σs1 = σ2d

⎡⎢⎢⎣³1− σ2d

σ2d+σ2ε

´ρ³1− σ2d

σ2d+σ2ε

´ρ³1− σ2d

σ2d+σ2ε

´ ³1− ρ2

σ2dσ2d+σ

2ε

´⎤⎥⎥⎦ . (30)

Posterior beliefs about the liquidating dividend come from a normal projection of s1 on d1 and d2.

d1 = d̄+ βd1,s1¡s1 − d̄

¢+ ηd1,s1 (31)

d2 = d̄+ βd2,s1¡s1 − d̄

¢+ ηd2,s1 (32)

where the residual uncertainty about the liquidating dividends is distributed N (0,Σs1) . By pro-

jection optimality

ηd1,s1 , ηd2,s1 ⊥¡s1 − d̄

¢. (33)

For (1− α) fraction of the population, demand for the risky asset at t = 1 is

Xu1 =

1

aΣ−1

¡d̄1− P1

¢.


αnXi1 + (1− α)nXu

1 = z1. (34)

Substituting in investor demands XI1 and XU

1 yields

p1,1 = d̄+ασ2d

ασ2d + σ2ε| {z }0<γd1<βd1,s1

¡s1 − d̄

¢− az

n(1 + ρ)

σ2dσ2ε

ασ2d + σ2ε(35)

p1,2 = d̄+αρσ2d

ασ2d + σ2ε| {z }0<γd2<βd2,s1

¡s1 − d̄

¢− az

n(1 + ρ)

σ2d¡σ2ε + α (1− ρ)σ2d

¢ασ2d + σ2ε

. (36)

22

Define returns

r1,k = p1,k − p0,k (37)

r2,k = p2,k − p1,k. (38)

Because the informative signal is not incorporated fully at t = 1, equilibrium prices exhibit cross-

momentum.

Cov (r1,1, r2,2) = Cov¡γd1

¡s1 − d̄

¢,¡βd2,s1 − γd2

¢ ¡s1 − d̄

¢+ ηd2,s1

¢(39)

= γd1¡βd2,s1 − γd2

¢ ¡σ2d + σ2ε

¢(40)

23

References

[1] Brennan, Michael, Marasimhan Jegadeesh, and Bhaskaran Swaminathan, 1993, Investment

analysis and the adjustment of stock prices to common information, Review of Financial

Studies 6, 799-824.

[2] Callahan, Tyrone, 2003, Speculative markets with an unknown number of insiders, unpublished

manuscript, University of Texas.

[3] Chan, Louis, Narasimhan Jegadeesh, and Josef Lakonishok, 1996, Momentum strategies, Jour-

nal of Finance 51, 1681-1713.

[4] Daniel, Kent, David Hirshleifer, and Avanidhar Subrahmanyam, 1998, A theory of overconfi-

dence, self-attribution, and security market under- and over-reactions, Journal of Finance 53,

1839-1885.

[5] Fama, Eugene, and Kenneth French, 1997, Industry costs of equity, Journal of Financial

Economics 43, 153-193.

[6] Froot, Kenneth, 1989, Consistent covariance matrix estimation with cross-sectional dependence

and heteroskedasticity in financial data, Journal of Financial and Quantitative Analysis 24,

333-355.

[7] Grossman, Sanford, and Joseph Stiglitz, 1980, On the impossibility of informationally efficient

markets, American Economic Review 70, 393-408.

[8] Holden, Craig, and Avanidhar Subrahmanyam, 2002, News events, information acquisition,

and serial correlation, Journal of Business 75, 1-32.

[9] Hong, Harrison, Terrence Lim, and Jeremy Stein, 2000, Bad news travels slowly: Size, analyst

coverage, and the profitability of momentum strategies, Journal of Finance 55, 265-295.

24

[10] Hong, Harrison, and Jeremy Stein, 1999, A unified theory of underreaction, momentum trad-

ing, and overreaction in asset markets, Journal of Finance 54, 2143-2184.

[11] Hong, Harrison, Walter Torous, and Rossen Valkanov, 2003, Do industries lead stock markets?

unpublished manuscript, Princeton University.

[12] Hou, Kewei, 2002, Industry information diffusion and the lead-lag effect in stock returns,

unpublished manuscript, Ohio State University.

[13] Hou, Kewei, and Tobias Moskowitz, 2002, Market frictions, price delay, and the cross-section

of expected returns, unpublished manuscript, University of Chicago.

[14] Jegadeesh, Narasimhan, and Sheridan Titman, 1993, Returns to buying winners and selling

losers: Implications for stock market efficiency, Journal of Finance 48, 65-91.

[15] Johnson, Timothy, 2002, Rational momentum effects, Journal of Finance 57, 585-608.

[16] Kyle, Albert, 1985, Continuous auctions and insider trading, Econometrica 53, 1315-1335.

[17] Lo, Andrew, and Craig MacKinlay, 1990, When are contrarian profits due to stock market

overreaction? Review of Financial Studies 3, 175-208.

[18] Mech, Timothy, 1993, Portfolio return autocorrelation, Journal of Financial Economics 34,

307-344.

[19] Moskowitz, Tobias, and Mark Grinblatt, 1999, Do industries explain momentum? Journal of

Finance 54, 1249-1290.

[20] Rogers, W., 1993, Regression standard errors in clustered samples, Stata Technical Bulletin

13, 19-23.

[21] Wooldridge, Jeffrey, 2002, Econometric analysis of cross section and panel data, MIT Press.

25

Figure 1: Cross-industry momentum beyond the first month

−5 0 5 10 15 20 25 30 35−0.04

−0.02

0

0.02

month after event

cum

ulat

ive

aver

age

exce

ss re

turn

s

High Minus Low Indst. with Upstream Shock High Minus Low Upstream Portfolios

−5 0 5 10 15 20 25 30 35−0.04

−0.02

0

0.02

month after event

cum

ulat

ive

aver

age

exce

ss re

turn

s

High Minus Low Downstream Shock Indst.High Minus Low Downstream Portfolios

Figure 2: Performance of upstream and downstream strategies (1964-2002)C

umul

ativ

e Ex

cess

Ret

urn

1965 1970 1975 1980 1985 1990 1995 2000

0

0.5

1

1.5

2

2.5

3 High - Low US Cross-momentumHigh - Low DS Cross-momentum

Figure 3: Comparison with SMB, HML and MOM strategies (1964-2002)

Cum

ulat

ive

Exc

ess

Ret

urn

1965 1970 1975 1980 1985 1990 1995 2000

0

1

2

3

4

5 High − Low US Cross−momentumUpstream Indstry Own MomentumSMBHMLMOM

Cum

ulat

ive

Exc

ess

Ret

urn

1965 1970 1975 1980 1985 1990 1995 2000

0

1

2

3

4

5 High − Low DS Cross−momentumDownstream Indstry Own MomentumSMBHMLMOM

Industry Industry name Mean Std. dev. Mean Std. dev. Mean Std. dev.

1 Livestock and livestock prod -- -- -- -- -- --2 Other agricultural products -- -- -- -- -- --3 Forestry and fishery produc -- -- -- -- -- --4 Agricultural, forestry, and fis -- -- -- -- -- --5 Iron and ferroalloy ores min -- -- -- -- -- --6 Nonferrous ores mining 0.060 0.325 0.056 0.170 0.030 0.2427 Coal mining 0.105 0.305 0.055 0.177 0.032 0.1388 Crude petroleum and natura 0.033 0.239 0.031 0.203 0.063 0.1589 Stone and clay mining and -- -- -- -- -- --10 Chemical and fertilizer mine -- -- -- -- -- --11 New construction -- -- -- -- -- --12 Maintenance and repair con 0.101 0.374 0.054 0.179 0.039 0.18513 Ordnance and accessories 0.074 0.217 0.064 0.202 0.071 0.22114 Food and kindred products 0.079 0.162 0.064 0.176 0.072 0.21715 Tobacco products -- -- -- -- -- --16 Broad and narrow fabrics, y 0.048 0.237 0.052 0.174 0.072 0.23417 Miscellaneous textile goods 0.087 0.339 0.049 0.184 0.065 0.20918 Apparel 0.070 0.244 0.053 0.204 0.081 0.23019 Miscellaneous fabricated te 0.070 0.429 0.058 0.203 0.065 0.20320 Lumber and wood products 0.092 0.288 0.060 0.176 0.062 0.17921 Wood containers -- -- -- -- -- --22 Household furniture 0.093 0.238 0.069 0.198 0.062 0.25823 Other furniture and fixtures 0.080 0.225 0.058 0.198 0.048 0.21924 Paper and allied products, e 0.048 0.184 0.062 0.175 0.069 0.18725 Paperboard containers and 0.066 0.253 0.050 0.175 0.067 0.16326 Printing and publishing 0.071 0.199 0.056 0.176 0.077 0.21627 Chemicals and selected che 0.046 0.190 0.057 0.160 0.060 0.17628 Plastics and synthetic mate 0.043 0.208 0.052 0.175 0.059 0.19229 Drugs, cleaning and toilet p 0.079 0.172 0.063 0.179 0.082 0.24230 Paints and allied products 0.077 0.223 0.051 0.173 0.064 0.18831 Petroleum refining and relat 0.069 0.174 0.038 0.214 0.062 0.18632 Rubber and miscellaneous 0.055 0.204 0.051 0.176 0.067 0.18633 Leather tanning and finishin -- -- -- -- -- --34 Footwear and other leather 0.052 0.249 0.062 0.189 0.082 0.23435 Glass and glass products 0.067 0.266 0.055 0.168 0.073 0.17436 Stone and clay products 0.040 0.222 0.058 0.169 0.055 0.19237 Primary iron and steel manu 0.005 0.234 0.059 0.170 0.056 0.18838 Primary nonferrous metals m 0.028 0.263 0.054 0.168 0.058 0.18839 Metal containers 0.067 0.227 0.029 0.212 0.077 0.15740 Heating, plumbing, and fabr 0.056 0.214 0.035 0.199 0.062 0.19641 Screw machine products an 0.050 0.243 0.031 0.201 0.062 0.19242 Other fabricated metal prod 0.065 0.202 0.041 0.193 0.068 0.19243 Engines and turbines 0.033 0.296 0.038 0.196 0.059 0.17544 Farm and garden machinery 0.060 0.223 0.042 0.201 0.064 0.23345 Construction and mining ma 0.060 0.233 0.044 0.191 0.089 0.25346 Materials handling machine 0.060 0.287 0.049 0.190 0.080 0.20947 Metalworking machinery an 0.043 0.244 0.048 0.190 0.055 0.19148 Special industry machinery 0.054 0.348 0.051 0.191 0.067 0.188

Upstream return

Table ISummary Statistics

Downstream returnIndustry return

This table presents summary statistics for 77 industry accounts of the Input-Output Benchmark Survey of the Bureau of EconomicAnalysis. Firms are assigned to industries based on the industry account-SIC code dictionary provided in the Survey. Industryreturns are monthly (July 1963 - December 2002) and value-weighted based on market capitalization of firms. Downstream(upstream) returns are computed using downstream (upstream) industry returns and value-weighted according to industry sales(purchases) reported in the Use Table of the BEA Survey.

Industry Industry name Mean Std. dev. Mean Std. dev. Mean Std. dev.Upstream return

Table ISummary Statistics

Downstream returnIndustry return

This table presents summary statistics for 77 industry accounts of the Input-Output Benchmark Survey of the Bureau of EconomicAnalysis. Firms are assigned to industries based on the industry account-SIC code dictionary provided in the Survey. Industryreturns are monthly (July 1963 - December 2002) and value-weighted based on market capitalization of firms. Downstream(upstream) returns are computed using downstream (upstream) industry returns and value-weighted according to industry sales(purchases) reported in the Use Table of the BEA Survey.

49 General industrial machiner 0.058 0.204 0.045 0.193 0.056 0.18450 Miscellaneous machinery, e 0.049 0.279 0.047 0.194 0.060 0.19351 Computer and office equipm 0.057 0.256 0.072 0.216 0.079 0.22252 Service industry machinery 0.062 0.232 0.050 0.192 0.061 0.20753 Electrical industrial equipme 0.073 0.236 0.053 0.193 0.055 0.19654 Household appliances 0.049 0.236 0.052 0.187 0.093 0.26155 Electric lighting and wiring e 0.045 0.219 0.052 0.189 0.070 0.20356 Audio, video, and communi 0.055 0.296 0.083 0.240 0.057 0.16357 Electronic components and 0.105 0.327 0.056 0.181 0.063 0.20958 Miscellaneous electrical ma 0.041 0.260 0.063 0.198 0.060 0.19059 Motor vehicles and equipme 0.048 0.223 0.056 0.193 0.091 0.26260 Aircraft and parts 0.073 0.234 0.057 0.197 0.065 0.19261 Other transportation equipm 0.084 0.254 0.053 0.192 0.063 0.20162 Scientific and controlling ins 0.058 0.208 0.070 0.210 0.074 0.22863 Ophthalmic and photograph 0.037 0.212 0.070 0.198 0.080 0.20964 Miscellaneous manufacturin 0.039 0.240 0.059 0.191 0.079 0.21765 Transportation and warehou 0.057 0.210 0.069 0.173 0.063 0.17066 Communications, except ra 0.039 0.170 0.072 0.214 0.072 0.19867 Radio and TV broadcasting 0.119 0.237 0.060 0.232 0.086 0.25568 Electric, gas, water, and san 0.032 0.142 0.071 0.207 0.061 0.18269 Wholesale and retail trade -- -- -- -- -- --70 Finance and insurance 0.069 0.191 0.066 0.200 0.051 0.19371 Real estate and rental 0.008 0.240 0.085 0.229 0.073 0.20572 Hotels, personal and repair 0.105 0.314 0.062 0.185 0.070 0.19873 Business and professional s 0.086 0.255 0.057 0.183 0.064 0.18674 Eating and drinking places 0.070 0.222 0.068 0.160 0.064 0.19375 Automotive repair and servi 0.101 0.310 0.058 0.176 0.066 0.19276 Amusements 0.059 0.265 0.064 0.198 0.105 0.21377 Health, educational, and so 0.085 0.309 0.057 0.183 0.073 0.193

Mean 0.062 0.246 0.056 0.190 0.067 0.202Standard deviation 0.023 0.052 0.011 0.017 0.013 0.026

Number of industries 65 65 65 65 65 65Minimum 0.005 0.142 0.029 0.160 0.030 0.138Maximum 0.119 0.429 0.085 0.240 0.105 0.262

(1) (2) (3) (4) (5)

R upstream,t 0.335 0.333 0.328 0.329 0.327(15.81) (15.62) (15.00) (15.14) (15.12)

R downstream,t 0.208 0.207 0.204 0.205 0.203(13.26) (13.11) (12.91) (13.03) (12.98)

R industry,t-1 0.030 0.026 0.026 0.028(2.11) (1.79) (1.79) (1.93)

R upstream,t-1 0.058 0.063 0.061(2.55) (2.66) (2.77)

R downstream,t-1 0.038 0.045 0.040(2.44) (2.75) (2.46)

R upstream,t-1 * R downstream,t-1 0.335 0.335(0.55) (0.55)

R upstream,t-1 * R industry,t-1 -0.461(1.54)

R downstream,t-1 * R industry,t-1 -0.187(0.65)

R2 0.53 0.54 0.54 0.54 0.54 N obs 31,135 31,135 31,135 31,135 31,135

Industry Return (t)

Table IIPredictive Panel Regressions

This table presents results from regressions of monthly industry returns on contemporaneous and predictive variables.Industry definition is based on industry account definitions of the Benchmark Survey of the Bureau of Economic Analysis.Upstream (downstream) returns consist of upstream (downstream) industry returns and are value-weighted according tobetween-industry commodity flows reported in the Input-Output Matrix. Upstream and downstream industry returns arevalue-weighted based on market capitalization of firms. All regressions include month fixed effects. t-statistics are inparentheses. Underlying standard errors are robust and clustered by month.

(1) (2) (3) (4) (5)

Constant 0.002 0.008 0.002 0.008 0.007(0.58) (2.11) (0.72) (2.05) (1.80)

β market 0.001 0.000 0.001 0.001 0.000(0.39) (0.12) (0.52) (0.21) (0.10)

β HML 0.001 0.002 0.001(0.90) (1.04) (0.35)

β SMB 0.000 0.000 -0.001(0.05) (0.24) (0.48)

R industry,t-1 0.038 0.032 0.029 0.025 0.021(3.34) (2.89) (2.59) (2.28) (1.88)

R upstream,t-1 0.095 0.079 0.074(4.14) (3.55) (3.28)

R downstream,t-1 0.056 0.050 0.051(3.22) (3.00) (3.07)

BM t 0.002(0.95)

LogSize t 0.000(0.05)

Mean R2 0.13 0.17 0.21 0.24 0.28

Industry Return (t)

This table presents results from Fama-MacBeth regressions of monthly industry returns. Specifically, reportedestimates are average estimates from Fama-MacBeth cross-sectional regressions run for each month. Standarderrors assume independence across monthly cross-sectional regressions. Industry definition is based on industryaccount definitions of the Benchmark Survey of the Bureau of Economic Analysis. Upstream (downstream) returnsconsist of upstream (downstream) industry returns value-weighted according to between-industry commodity flowsreported in the Input-Output Matrix. Upstream and downstream industry returns are value-weighted based onmarket capitalization of firms.

Table IIIPredictive Fama-MacBeth Regressions

Low (1) (2) (3) (4) High (5) High - Low

EW mean return (annualized) 0.029 0.051 0.060 0.074 0.097 0.068Standard deviation 0.183 0.192 0.191 0.187 0.186 0.080Sharpe ratio 0.157 0.268 0.316 0.396 0.521 0.852

VW mean return (annualized) 0.016 0.055 0.059 0.055 0.081 0.065Standard deviation 0.165 0.180 0.179 0.181 0.173 0.113Sharpe ratio 0.099 0.303 0.330 0.306 0.471 0.573

Low (1) (2) (3) (4) High (5) High - Low



Low (1) (2) (3) (4) High (5) High - Low

Mean ($ million) 7,722 5,439 6,199 6,271 8,351 630Standard deviation 3,509 2,507 2,892 2,818 3,725 4,079

Low (1) (2) (3) (4) High (5) High - Low


Market capitalization of industries in upstream strategy

Market capitalization of industries in downstream strategy

This table presents profitability of several cross-industry momentum strategies. Panel A (Panel B) reports mean and standarddeviation of one-month industry returns ranked into five bins based on previous month upstream (downstream) portfolio returns.Both equal-weighted and value-weighted one-month industry returns are reported. Panel C reports mean and standard deviationof industry market capitalization in each bin.

Monthly ranking of industries based on previous month downstream portfolio return

Panel C: Market capitalization of industries in upstream and downstream strategies

Panel B: One-month momentum strategy based on previous month downstream porfolio return

Table IVCross-Industry Momentum Profitability

Monthly ranking of industries based on previous month upstream portfolio return

Panel A: One-month momentum strategy based on previous month upstream porfolio return

Industry Industry Name Low High Low High

1 Livestock and livestock products -- -- -- --2 Other agricultural products -- -- -- --3 Forestry and fishery products -- -- -- --4 Agricultural, forestry, and fishery services -- -- -- --5 Iron and ferroalloy ores mining -- -- -- --6 Nonferrous ores mining 0.010 0.012 0.033 0.0247 Coal mining 0.011 0.012 0.029 0.0248 Crude petroleum and natural gas 0.027 0.023 0.028 0.0289 Stone and clay mining and quarrying -- -- -- --10 Chemical and fertilizer mineral mining -- -- -- --11 New construction -- -- -- --12 Maintenance and repair construction 0.010 0.007 0.019 0.01413 Ordnance and accessories 0.017 0.018 0.017 0.01514 Food and kindred products 0.010 0.013 0.024 0.02415 Tobacco products -- -- -- --16 Broad and narrow fabrics, yarn and thread mills 0.025 0.023 0.023 0.02417 Miscellaneous textile goods and floor coverings 0.025 0.023 0.019 0.01818 Apparel 0.029 0.025 0.018 0.02219 Miscellaneous fabricated textile products 0.026 0.025 0.016 0.01520 Lumber and wood products, except containers 0.003 0.005 0.014 0.01321 Wood containers -- -- -- --22 Household furniture 0.019 0.020 0.027 0.02623 Other furniture and fixtures 0.012 0.013 0.030 0.02824 Paper and allied products, except containers 0.012 0.013 0.015 0.01625 Paperboard containers and boxes 0.024 0.024 0.013 0.01626 Printing and publishing 0.015 0.017 0.019 0.02027 Chemicals and selected chemical products 0.013 0.014 0.011 0.01228 Plastics and synthetic materials 0.016 0.016 0.018 0.01629 Drugs, cleaning and toilet preparations 0.006 0.008 0.025 0.02630 Paints and allied products 0.018 0.020 0.010 0.01131 Petroleum refining and related products 0.031 0.029 0.012 0.01332 Rubber and miscellaneous plastics products 0.022 0.019 0.001 0.00133 Leather tanning and finishing -- -- -- --34 Footwear and other leather products 0.015 0.017 0.022 0.02235 Glass and glass products 0.011 0.010 0.006 0.00936 Stone and clay products 0.010 0.011 0.008 0.00537 Primary iron and steel manufacturing 0.011 0.011 0.011 0.01138 Primary nonferrous metals manufacturing 0.023 0.019 0.008 0.00839 Metal containers 0.032 0.022 0.023 0.02840 Heating, plumbing, and fabricated structural metal products 0.023 0.017 0.013 0.01341 Screw machine products and stampings 0.024 0.017 0.016 0.01642 Other fabricated metal products 0.013 0.010 0.005 0.00643 Engines and turbines 0.020 0.015 0.019 0.01744 Farm and garden machinery 0.017 0.013 0.023 0.02145 Construction and mining machinery 0.013 0.010 0.031 0.03046 Materials handling machinery and equipment 0.006 0.007 0.018 0.01947 Metalworking machinery and equipment 0.008 0.005 0.009 0.00648 Special industry machinery and equipment 0.008 0.007 0.007 0.01049 General industrial machinery and equipment 0.010 0.009 0.010 0.00650 Miscellaneous machinery, except electrical 0.008 0.004 0.010 0.00851 Computer and office equipment 0.021 0.024 0.017 0.01952 Service industry machinery 0.008 0.008 0.025 0.02553 Electrical industrial equipment and apparatus 0.004 0.005 0.012 0.00954 Household appliances 0.003 0.004 0.026 0.027

Table VCross-Industry Momentum Diagnostics: Industry Inclusion Probabilities

Upstream strategy Downstream strategy

This table presents the inclusion probability of industries in upstream and downstream momentum strategies.

Industry Industry Name Low High Low High

Table VCross-Industry Momentum Diagnostics: Industry Inclusion Probabilities

Upstream strategy Downstream strategy

This table presents the inclusion probability of industries in upstream and downstream momentum strategies.

55 Electric lighting and wiring equipment 0.005 0.005 0.007 0.00956 Audio, video, and communication equipment 0.025 0.031 0.020 0.01757 Electronic components and accessories 0.004 0.004 0.017 0.01658 Miscellaneous electrical machinery and supplies 0.011 0.012 0.015 0.01559 Motor vehicles and equipment 0.008 0.012 0.027 0.02860 Aircraft and parts 0.010 0.010 0.022 0.02061 Other transportation equipment 0.008 0.009 0.019 0.01762 Scientific and controlling instruments 0.019 0.021 0.022 0.02463 Ophthalmic and photographic equipment 0.015 0.020 0.011 0.01464 Miscellaneous manufacturing 0.003 0.003 0.012 0.01565 Transportation and warehousing 0.014 0.019 0.002 0.00466 Communications, except radio and TV 0.022 0.024 0.003 0.00367 Radio and TV broadcasting 0.028 0.025 0.024 0.02568 Electric, gas, water, and sanitary services 0.025 0.028 0.001 0.00269 Wholesale and retail trade -- -- -- --70 Finance and insurance 0.018 0.020 0.011 0.01071 Real estate and rental 0.025 0.030 0.014 0.01372 Hotels, personal and repair services (except auto) 0.012 0.013 0.003 0.00373 Business and professional services 0.014 0.016 0.002 0.00374 Eating and drinking places 0.020 0.024 0.002 0.00475 Automotive repair and services 0.011 0.013 0.001 0.00176 Amusements 0.018 0.019 0.020 0.02977 Health, educational, and social services and nonprofits 0.015 0.015 0.005 0.008

Mean 0.015 0.015 0.015 0.015Standard deviation 0.008 0.007 0.008 0.008

Number of industries 65 65 65 65Minimum 0.003 0.003 0.001 0.001Maximum 0.032 0.031 0.033 0.030

Alpha Rmarket - Rf SMB HML MOM R2

(1) Upstream high - low strategy 0.005 -0.007 -0.001 0.052 0.050 0.012(3.45) (0.17) (0.03) (0.82) (1.62)

(2) Upstream high - low strategy 0.005 0.024 0.017 -0.011 0.046(3.20) (0.54) (0.37) (0.20) (1.63)

* NBER recession -0.209 -0.110 0.145 0.022 0.071(2.38) (0.74) (1.16) (0.28)

(3) Downstream high - low strategy 0.005 -0.016 -0.029 -0.003 0.010 0.003(5.09) -(0.53) -(0.90) -(0.05) (0.36)

(4) Downstream high - low strategy 0.005 0.001 -0.017 -0.071 -0.009(4.80) (0.03) (0.52) (1.73) (0.26)

* NBER recession -0.106 -0.054 0.197 0.032 0.036(0.81) (0.49) (1.83) (0.50)

Table VICross-Industry Momentum Diagnostics: Return Exposures

This table presents exposures inherent in cross-industry momentum strategies. Results from regressing monthly upstream and downstreamstrategy returns on excess market return, Fama-French SMB and HML factors, and Carhart MOM factor are reported. Even-numberedspecifications include interaction terms with NBER recessions.

(1) (2) (3) (4) (5) (6) (7) (8)

1 [R own,t-1 ≥ p 80%,t-1 ] 0.0019 0.0016 0.0017 0.0035 -0.0003 0.0018(1.79) (1.55) (1.58) (2.63) (0.16) (1.71)

1 [R own,t-1 ≤ p 20%,t-1 ] -0.0035 -0.0031 -0.0035 -0.0033 -0.0029 -0.0034(3.42) (3.04) (3.28) (2.92) (1.72) (3.28)

1 [R upstream,t-1 ≥ p 80%,t-1 ] 0.0029 0.0026 0.0027 0.0009 0.0042(3.29) (2.96) (2.95) (0.88) (3.02)

1 [R upstream,t-1 ≤ p 20%,t-1 ] -0.0028 -0.0024 -0.0019 -0.0044 -0.0004(3.45) (3.03) (2.28) (4.32) (0.34)

1 [R downstream,t-1 ≥ p 80%,t-1 ] 0.0017 0.0013 0.0016 0.0006 0.0022(2.11) (1.70) (1.97) (0.54) (1.84)

1 [R downstream,t-1 ≤ p 20%,t-1 ] -0.0030 -0.0025 -0.0026 -0.0028 -0.0023(3.72) (3.25) (3.28) (2.64) (2.02)

1 [R upstream,t-1 ≥ p 80%,t-1 ] * 0.0035 1 [R downstream,t-1 ≥ p 80%,t-1 ] (2.14)

1 [R upstream,t-1 ≤ p 20%,t-1 ] * -0.0043 1 [R downstream,t-1 ≤ p 20%,t-1 ] (3.01)

R2 0.53 0.53 0.53 0.53 0.52 0.61 0.45 0.53 N obs 31,135 31,135 31,135 31,135 28,600 15,535 15,600 31,135

Table VIICross-Industry Momentum Profitability - Panel Regression Approach

This table presents panel regressions of monthly industry returns on indicator variables based on previous month own, upstream and downstream industry returns. Industry definition isbased on industry account definitions of the Benchmark Survey of the Bureau of Economic Analysis. Upstream (downstream) returns consist of upstream (downstream) industry returnsand are value-weighted according to between-industry commodity flows reported in the Input-Output Matrix. Upstream and downstream industry returns are value-weighted based onmarket capitalization of firms. Column 5 excludes the month of January. Columns 6 and 7 split the sample into two halves, 1963-1982 and 1983-2002. All regressions include month fixedeffects. t-statistics are in parentheses. Underlying standard errors are robust and clustered by month.

Industry Return (t)

Low (1) (2) (3) (4) High (5) High - Low

EW mean return (annualized) 0.052 0.069 0.061 0.071 0.049 -0.003Standard deviation 0.182 0.186 0.184 0.183 0.183 0.076Sharpe ratio 0.285 0.372 0.334 0.386 0.269 -0.036

VW mean return (annualized) 0.053 0.059 0.047 0.046 0.047 -0.006Standard deviation 0.174 0.174 0.173 0.172 0.179 0.125Sharpe ratio 0.306 0.337 0.270 0.270 0.264 -0.048

Low (1) (2) (3) (4) High (5) High - Low


VW mean return (annualized) 0.050 0.045 0.044 0.063 0.047 -0.003Standard deviation 0.174 0.179 0.186 0.172 0.162 0.115Sharpe ratio 0.290 0.251 0.235 0.365 0.291 -0.027

Low (1) (2) (3) (4) High (5) High - Low

Mean ($ million) 5,280 7,863 9,540 8,241 5,143 -137Standard deviation 2,123 3,275 4,409 3,543 2,129 2,125

Low (1) (2) (3) (4) High (5) High - Low


This table presents profitability of several cross-industry momentum strategies based on stocks with market capitalization above50th percentile of NYSE firms. Panel A (Panel B) reports mean and standard deviation of one-month industry returns ranked intofive bins based on previous month upstream (downstream) portfolio returns. Both equal-weighted and value-weighted one-monthindustry returns are reported. Panel C reports mean and standard deviation of industry market capitalization in each bin.

Table VIIICross-Industry Momentum Profitability - Large Stocks Only








Low (1) (2) (3) (4) High (5) High - Low



Low (1) (2) (3) (4) High (5) High - Low



Low (1) (2) (3) (4) High (5) High - Low


Low (1) (2) (3) (4) High (5) High - Low

Mean ($ million) 5,999 7,835 7,534 7,616 5,899 -100Standard deviation 2,752 3,344 3,446 3,332 2,684 2,647

This table presents profitability of several cross-industry momentum strategies based on stocks with market capitalization above20th percentile of NYSE firms. Panel A (Panel B) reports mean and standard deviation of one-month industry returns ranked intofive bins based on previous month upstream (downstream) portfolio returns. Both equal-weighted and value-weighted one-monthindustry returns are reported. Panel C reports mean and standard deviation of industry market capitalization in each bin.

Table IXCross-Industry Momentum Profitability - Excluding Micro Stocks








(1) (2) (3)

R industry,t-1 0.033 0.033 0.031(2.25) (2.23) (2.08)

R upstream-(0-33%),t-1 0.069 0.062(2.48) (2.23)

R upstream-(33-66%),t-1 0.034 0.034(2.71) (2.68)

R upstream-(66-100%),t-1 0.011 0.009(1.15) (0.89)

R downstream-(0-33%),t-1 0.049 0.045(2.82) (2.66)

R downstream-(33-66%),t-1 0.011 0.010(1.09) (0.92)

R downstream-(66-100%),t-1 0.017 0.015(1.77) (1.68)

R2 0.53 0.53 0.53 N obs 31,135 31,135 31,135

This table presents results from regressions of monthly industry returns oncontemporaneous and predictive variables. Industry definition is based on industryaccount definitions of the Benchmark Survey of the Bureau of Economic Analysis.Upstream (downstream) industries are first sorted according to their sales(purchases) to (from) the reference industry in ascending order and then allocated toone of three portfolios (0-33%, 33-66% and 66-100%). The 0-33% portfolio iscomprised of upstream (downstream) industries whose sales (purchases) are thesmallest and add up to 33% of the total purchases (sales) of the reference industry.The 33-66% portfolio is comprised of the next set of upstream (downstream)industries whose sales (purchases) constitute another 33% of the total. The 66-100%portfolio is comprised of the remaining industries. Upstream and downstream industryreturns are value-weighted based on market capitalization of firms. All regressionsinclude month fixed effects. t-statistics are in parentheses. Underlying standard errorsare robust and clustered by month.

Table XLarge versus Small Supply Chain Relationships

Industry Return (t)

(1) (2) (3)

ROA market 0.227 0.299 0.148(5.62) (4.92) (3.57)

ROA upstream 0.457 0.361(11.92) (11.07)

ROA downstream 0.404 0.291(8.46) (8.23)

R2 0.52 0.51 0.53 N obs 2,225 2,225 2,225

Table A-IIndustry Return on Assets

Industry Return on Assets

This table presents results from panel regressions of annual industry ROA (return onassets) on contemporaneous market, downstream and upstream ROAs. Industrydefinition is based on industry account definitions of the Benchmark Survey of theBureau of Economic Analysis. Upstream (downstream) ROA consists of upstream(downstream) industry ROAs and are value-weighted according to between-industrycommodity flows reported in the Input-Output Matrix. Industry ROA is defined as thesum of cash flow divided by the sum of assets of firms in the industry. Assets aremeasured with COMPUSTAT item 6. Cash flow is defined as the sum of earningsbefore extraordinary items (item 18) and depreciation and amortization (item 14)following Kaplan and Zingales (1997). All specifications include industry fixed effects.t-statistics are reported in parentheses. Underlying standard errors are robust andclustered by year.

cross-industry momentum - semantic scholar · 2015-07-29 · cross-industry momentum ∗ lior...

Documents