Learning from Peers’Stock Prices and Corporate Investment
Laurent Frésard
University of Maryland
R.H. Smith School of Business
College Park, MD 20742
lfresard@rhsmith.umd.edu
Thierry Foucault
HEC, Paris
1 rue de la Liberation
78351 Jouy en Josas, France
foucault@hec.fr
This Version: March 2013
Abstract
We show that the stock market valuation of peers matters for …rms’investment decisions.
In a large sample of …rms, corporate investment is positively related to the market valuation
of peer …rms selling related products. Consistent with a model where managers use the stock
prices of peers as a source of information, this relation is stronger when the level of informed
trading in a …rm’s stock is weak. Also, the link between the investment of a …rm and its own
stock price is weaker when the level of informed trading in its peers’stocks is high, or when
the demand for its products is more correlated with that of its peers’products. Furthermore
the investment of private …rms depends on their peers’ stock prices, but much less so after
they go public. Overall, our results provide new insights on how …nancial markets a¤ect the
real economy.
JEL Classi…cation Numbers: G31, D21, D83
Keywords: Corporate investment, managerial learning, peers, informed trading
We thank Rui Albuquerque, Laurent Bach, Francois Derrien, Maria Guadalupe, Alexandre Jeanneret, Pete
Kyle, Rich Mathews, Sebastien Michenaud, Nagpurnanand Prabhala, Denis Sosyura, Jerome Taillard, David Thesmar, Philip Valta and seminar participants at Aalto University, Babson College, BI Oslo, Copenhagen Business
School, McGill University, the University of Maryland, the University of Warwick and the 2013 Adam Smith
Workshop in Oxford for valuable discussions and suggestions. We also thank Jerry Hoberg and Gordon Phillips
for sharing their TNIC data, and Jay Ritter for sharing its IPO data. All errors are ours.
1
1
Introduction
Firms’ managers, …nancial analysts, bankers, or investment professionals often rely on price
“multiples”(e.g., price-to-book or price-to-earnings ratios) of peer …rms to value new investments.
For instance, survey evidence indicate that corporate executives use peers’valuations for capital
budgeting decisions (e.g., Graham and Harvey (2001)). Hence, one expects …rms’investment to be
in‡uenced by their peers’market valuation (stock price). Whether and why this in‡uence exists
has not received much attention, however. In this article, we address these questions. Speci…cally,
we test the hypothesis that …rms’investment choices are in‡uenced by the market valuation of
their peers (de…ned as …rms exposed to common demand shocks), because this valuation informs
managers about the value of their growth opportunities.
Peers’stock prices aggregate information on future cash-‡ow shocks that are common to a …rm
and its peers. Hence, we propose that managers can learn information from their peers’market
valuation, as they do from their own stock price (see Bond, Edmans, and Goldstein (2012) for
a survey).1 In some situations, peers’valuation might even be the only market signal available
to a manager (consider a private …rm with publicly listed peers), or could be more informative
about a particular investment project than the …rm’s own valuation. Indeed, the latter re‡ects
the …rm’s portfolio of projects and is therefore less informative about the value of a particular
project (e.g. the development of a new product line) than the stock price of …rms focused on this
particular project.
If managers use the valuation of their peers to make investment decisions then …rms’investment and peers’ valuation should co-vary. Evidence thereof is however insu¢ cient to conclude
that managers learn information from peers’stock prices since prices and investment can co-vary
because of unobserved factors (e.g., correlated signals between managers and investors). To address this problem, we exploit the fact that our “learning from peers”hypothesis generates unique
cross-sectional predictions for the relationship between corporate investment and stock prices.
1
Subrahmanyam and Titman (1999) argue that stock prices are particularly useful to managers because they
aggregate investors’dispersed signals about future product demand. This is the case for peers’stock prices since
product demands for related …rms are a¤ected by common shocks (e.g., Menzlis and Ozbas (2010) show that Return
on Assets for individual …rms are positively correlated with Returns on Assets of related …rms).
2
These predictions derive from a simple model in which peers’ valuation complements managers’ knowledge about their investment opportunities.2 In this model, a …rm sells a product
for which demand is uncertain and correlated with the demand for another …rm’s product (its
peer).3 The …rm manager must decide whether to expand production capacity or not. Expansion
is a positive net present value project if future demand for the …rm’s product (about which the
manager is imperfectly informed) is strong enough. As investors trade on private information
about future demand, the …rm’s stock price and its peer’s stock price are informative about this
demand. We compare the relation between the …rm’s investment and each stock price in three
di¤erent scenarios: (i) when the manager uses the information contained in each stock price
(“learning from peers”); (ii) when the manager only relies on his own stock price (“narrow managerial learning”); and (iii) when the manager ignores stock market information (“no managerial
learning”).
The “learning from peers”scenario delivers …ve predictions that do not hold in other scenarios. They all stem from the same mechanism: When a decision-maker learns information from
multiple signals, his decision becomes less sensitive to one speci…c signal when others become
relatively more informative. Accordingly, with learning from peers, the sensitivity of a …rm’s
investment to its peer stock price is inversely related to (i) its own stock price informativeness
and (ii) the quality of “managerial information,” that is, managers’ private information about
their investment opportunities. Symmetrically, the sensitivity of a …rm’s investment to its stock
price decreases with its peer’s stock price informativeness. Furthermore, an increase in the quality of managerial information reduces this sensitivity if its peer’s stock price informativeness is
low enough, and increases it otherwise. Last, the sensitivity of a …rm’s investment to its own
stock price is weakly decreasing in the correlation between the demands for its product and the
product of its peer.
2
Existing models in which …rms learn from stock prices have focused on the case in which …rms learn from their
own stock prices, not the case in which they also learn from their peers (see, for instance, Bresnahan, Milgrom
and Paul (1992), Dow and Gorton (1997), Subrahmanyam and Titman (1999), Goldstein and Guembel (2007),
Foucault and Gehrig (2008), Dow, Güembel and Goldstein (2011); see Bond et al.(2012) for a survey).
3
Peers are not necessarily competing …rms. Firms can be exposed to common demand shocks because they
are vertically related (suppliers/customers …rms) or because their products are complements. For instance, if the
demand for computer hardware is strong, the demand for softwares is likely to be strong as well.
3
As these predictions do not hold if managers ignore stock market information or if they only
rely on their own stock price, they form the backbone of our empirical strategy. We test each
prediction on a large sample of …rms drawn from Compustat over the period 1996 to 2008. We
de…ne the peers of a given …rm using the Text-based Network Industry Classi…cation (TNIC)
developed by Hoberg and Phillips (2011), which is based on …rms’ products description (from
their annual 10Ks). We document that …rms’investment is positively and signi…cantly related to
their peers’valuation (market-to-book ratio) after controlling for their own valuation and other
characteristics.4 The economic magnitude of this correlation is substantial: A one standard
deviation increase in peers’ value is associated with a 5.6% increase in corporate investment,
about 14.5% of the sample average. The sensitivity of …rm’s investment to its peers’stock price
is about half the sensitivity to its own price (a one standard deviation increase in a …rm’s stock
price is associated with a 12.80% increase in its investment in our sample).
The sensitivity of a …rm’s investment to a peer’s stock price disappears once the …rm and its
peer stop operating in the same product space. Furthermore, the investment of a …rm turns out
to be sensitive to its peer’s stock price before the two …rms start operating in the same product
space. This …nding suggests that managers may decide to launch a new product after learning
information about its expected pro…tability from the stock prices of …rms already selling related
products.
Importantly, we …nd empirical support for the cross-sectional implications speci…c to the
“learning from peers” scenario. First, a …rm’s investment is less sensitive to its peers’valuation
when its own stock price is more informative.5 The economic magnitude of this e¤ect is large: A
one standard deviation increase in peers’market valuation is associated with a 1.24% increase in
a …rm’s investment when its own stock price is highly informative versus a 6.80% increase when
its stock price is uninformative. Symmetrically, a …rm’s investment is less sensitive to its own
stock price when its peers’market valuation is more informative.
4
Findings are qualitatively similar if we measure …rms’valuation with price-earnings ratio rather than marketto-book ratio.
5
We use a measure of a …rms’ speci…c return variation as proxy for the level of informed trading in a stock
(as, for instance, in Durnev, Morck and Yeung (2004), Chen, Goldstein and Jiang (2007), and Bakke and Whited
(2010)).
4
We also study the e¤ect of managerial information on investment-to-price sensitivities, using
the trading activity of the main insiders of the …rm (its top …ve executives) as a proxy for the
quality of managerial information. The investment of a …rm is more sensitive to its peers’stock
prices when its managers appear less informed. In addition, the sensitivity of a …rm’s investment
to its own stock price is not signi…cantly related to the quality of managerial information, on
average, as found in Chen, Goldstein, and Jiang (2007). Yet, consistent with our hypothesis, when
peer stock prices are not very informative, an increase in the amount of managerial information
has a negative e¤ect on the sensitivity of a …rm’s investment to its stock price. This e¤ect is
reversed when peer stock prices are highly informative.
Furthermore, and as uniquely predicted by the learning from peers hypothesis, the sensitivity
of a …rm’s investment to its own stock price increases when it is less similar to its peers (they share
fewer common words in the description of their products or their cash ‡ows are less correlated).
Less surprisingly, the sensitivity of a …rm investment to its peers’ market valuation increases
when the …rm is more similar to its peers or when its sales are more correlated with those of its
peers.
Finally, we look at the investment behavior of a sample of …rms before and after they go
public. We show that, prior to their IPO, the investment of these …rms is highly sensitive to their
public peers’ valuation. However, this sensitivity drops signi…cantly once they become public.
This evolution is consistent with our “learning from peers” hypothesis: Once publicly listed, a
…rm can learn information from its own stock price and should therefore put less weight on the
price of its peers as a source of information. In contrast, if managers do not learn information
from their peers, the sensitivity of a …rm investment to its peers’valuation should either remain
unchanged (if managers do not learn from stock prices) or increase after going public (if managers
only learn from their own stock price).
Several empirical studies already show that the information contained in a …rm’s stock price
a¤ects its real decisions.6 Our paper, however, is the …rst to establish that there is a link between
6
See for instance Durnev, Morck and Yeung (2004), Luo (2005), Chen, Goldstein, and Jiang (2007), Fang, Noe
and Tice (2009), Bakke and Whited (2010), Ferreira, Ferreira, and Raposo (2011), Edmans, Goldstein and Jiang
(2012), or Foucault and Frésard (2012).
5
a …rm’s investment, its peers’market valuation, and the informativeness of this valuation. This
link has far reaching implications. In particular, a shock to the stock price of a …rm or to its
informativeness should a¤ect investment decisions of other …rms and vice versa. This externality
generates a new channel through which the stock market could a¤ect aggregate investment and
the allocation of resources in the economy (see Subrahmanyam and Titman (2013)).
According to the learning hypothesis, variations in stock prices have a causal e¤ect on corporate investment because managers actively learn from the stock market. We do not claim to have
identi…ed this e¤ect. Identi…cation is very challenging since stock price variations are endogenous
and can be related to investment through various, non mutually exclusive, channels. As in related
studies (e.g., Chen, Goldstein, and Jiang (2007)), our approach is to show that cross-sectional
variations in the sensitivity of a …rm investment to stock prices are consistent with implications
of the learning hypothesis. Focusing on peer …rms helps to sharpen tests of this hypothesis for
two reasons. First, the learning hypothesis has unique cross-sectional predictions regarding the
e¤ect of stock prices informativeness on the sensitivity of a …rm investment to its own stock price
and its peers’stock prices. Second, the sensitivity of a …rm’s investment to its stock price can be
a¤ected by …nancial constraints because stock overvaluation relaxes these constraints (see Baker,
Stein, and Wurgler (2003), or Graham and Campello (2011)). The …nancial constraint channel,
however, should not a¤ect the sensitivity of a …rm investment to the stock price of its peers. Thus,
by looking at peers, we neutralize one possible confounding factor for the relationship between
stock prices and investment.
Overall, our …ndings indicate that peers’market valuations matter in shaping the investment
behavior of …rms. Hence, they also add to the growing empirical literature emphasizing the
importance of peers in …rms’decision making (e.g., Gilbert and Lieberman (1987) or more recently
Leary and Roberts (2012), and Hoberg and Phillips (2011)). Fracassi (2012) and Dougal, Parsons,
and Titman (2012) provide evidence of “peer e¤ects” in investment decisions, that is, situations
in which a …rm investment is in‡uenced by the investment of related …rms. These papers consider
di¤erent channels for these e¤ects (CEOs’social networks or geographical proximity). Our paper
does not seek to identify such peer e¤ects. It suggests, however, that investment decisions of
6
related …rms might be linked because they learn from each other stock prices.
In the next section, we derive our main testable implications using a simple model of investment in which managers learn information from their stock price, the stock prices of their peers,
and other sources. In Section 3 we describe the data and discuss the methodology. In Section
4 we present the empirical …ndings and in Section 5 we conclude. Proofs are collected in an
appendix at the end of the paper. An Internet Appendix provides additional theoretical and
empirical results.
2
2.1
Hypotheses Development
Model
We consider two …rms A and B that are exposed to common demand shocks for their product.
The timing of the model is described in Figure 1.
[Insert Figure 1 about here]
At date 1, investors trade shares of …rms A and B at prices pA1 and pB1 . At date 2, …rm
A has the option to expand its production capacity. Capacity expansion requires making an
irreversible investment, K, at date 2 before observing demands for …rms A’s and B’s products.
These demands and …rms’cash-‡ows are realized at date 3.
Firms’cash ‡ows. At date 3, demand dj for the product of …rm j can be High (H) or Low
(L) with equal probabilities. Firms’demands share a common factor, that is,
Pr(dA = H jdB = H ) = Pr(dB = H jdA = H ) = ,
where
6= 12 .
The cash ‡ow of …rm B at date 3;
otherwise. The cash ‡ow of …rm A is
L
A
I
(1)
B,
H
B
is
H
A+
if demand for its product is high and
I
(
L
B
(<
H
B)
K) if demand for its product is high and
K if demand for its product is low, where I is a variable equals to 1 if …rm A invests
7
at date 2 and zero otherwise. Thus, the net present value,
date 2 is:
=
where
8
>
>
>
>
>
<
>
>
>
>
>
:
, of …rm A’s investment decision at
K (> 0) if dA = H;
(2)
K
if dA = L;
is the incremental sales revenues for …rm A if it invests at date 2 and demand for its
product is high at date 3.7
The manager of …rm A. When the manager of …rm A invests at date 2, he has two sources
of information about the demand for its product at date 3. First, he observes stock prices at
date 1. Second, the manager receives a signal sm on demand at date 3. This signal is perfect
with probability
and uninformative with probability (1
managerial information” and to
). We refer to signal sm as “direct
as the quality of direct managerial information. This signal
captures managers’private information about their investment opportunities.
At date 2, for a given investment decision, I, the expected value of …rm A is:
VA (I) = E(
A jsm ; pA1 ; pB1 )
+I
E( jsm ; pA1 ; pB1 ):
(3)
The manager of …rm A chooses I to maximize …rm value and the …rm faces no …nancing constraints. To focus on the interesting case, we assume that in the absence of any information, the
expected NPV of capacity expansion is negative. That is:
A.1 : E( ) = K(
where RH =
K.
RH
2
1)
0,
(4)
Hence, if the manager had no information, he would not invest at date 2.
Moreover, for the moment, we assume that the correlation in demands for both …rms is such that
7
For simplicity, we assume that expanding production capacity is useless if demand is low. The idea is that in
this case the existing capacity of …rm A is su¢ cient to serve all its clients. It is straigtforward to extend the model
to the more general case in which investing at date 2 generates additional sales revenues even when demand is low.
The important point is that expanding capacity has a negative NPV in this case.
8
if the manager of …rm A learns that demand for …rm B is high then he invests. That is:
A.2: E( jdB = H ) = K( RH
or
>
1
RH
>
1
2
1) > 0;
(5)
(the second inequality follows from A.1). We relax assumption A.2 in Section 2.4.
The Stock Market. There are three types of investors in the stock market: (i) a continuum
of risk-neutral speculators, (ii) liquidity traders with an aggregate demand, zj , for …rm j, and (iii)
risk neutral dealers. Liquidity traders’aggregate demand is uniformly distributed over [ 1; 1].
Subrahmanyam and Titman (1999) argue that individuals are well placed (e.g., through their
consumption experience) to obtain information not readily available to managers about demand
for a …rm products. They view this information as being obtained serendipitously, “that is,
by luck and without cost” and treat the number of investors with serendipitous information as
exogenous. Following their approach, we take as given the fraction
j
of speculators receiving a
signal sbj 2 fH; Lg on demand for the product of …rm j 2 fA; Bg.8 For simplicity, we assume
that this signal is perfect. Finally we denote by sbj = ?, the information set of a speculator who
receives no signal about stock j.
After receiving her information on stock j, a speculator decides to buy or to sell one share
of this stock, or to stay put. We denote by xij (b
sj ) 2 f 1; 0; +1g the size of the order submitted
by speculator i when she has received signal sbj . Signal sbj is also informative about the payo¤ of
the other stock since
6= 12 , which opens the possibility that an informed investor trades in both
assets. To simplify the analysis, we assume that a speculator only trades the stock for which she
receives information. This is in fact optimal for speculators if there is a …xed cost of trading per
asset (see Section A.5 in the Internet Appendix).9
Let fj be the order ‡ow (the sum of speculators and liquidity traders’net demand) for stock
8
In the Internet Appendix (Section A.4), we consider the case in which speculators must pay a cost c to receive
a signal. In this case, j is endogenous and variations in j across stocks are then driven by variations in the cost
of information, c, and the volatility of …rm cash ‡ows (uncertainty on future demands).
9
Albuquerque et al. (2008) estimate a model in which investors receive both market-wide private information
and …rm-speci…c information. In our model, when 6= 1, private information is …rm speci…c but it contains a
market-wide component if 6= 12 . The market wide component is a source of correlation in informed investors’
trades across stocks, as found empirically by Albuquerque et al. (2008), even if informed investors are di¤erent
across stocks.
9
j:
fj = zj +
Z
1
xij (b
sj )di
(6)
0
This order ‡ow is absorbed by dealers at a price such that they just break even (as in Kyle
(1985)) given the information contained in the order ‡ow. That is, for a given investment policy
I of …rm A,
pA1 (fA ) = E(VA (I) j fA ),
and
pB1 (fB ) = E(
B
j fB ).
(7)
Using the Law of Iterated Expectations, the stock prices of …rms A and B at date 0 are:
pA0 (I) =E(VA (I)) and pB0 (fB ) =E(
1 is denoted
2.2
pj = pj1
B ).
The change in price of stock j from date 0 to date
pj0 .
Investment decisions and stock prices
In this section, we study how the optimal investment policy of …rm A is related to stock prices in
equilibrium. This step is straightforward in our model but it is key for our empirical implications.
For brevity, we provide the equilibrium in the market for stock B in the Appendix (see Lemma
1). For our purpose, the main feature of this equilibrium is that he stock price of …rm B is
informative about the demand for …rm A’s product if
of informed trading in stock B,
B,
B
> 0 and even more so when the level
increases.
Firm A’s optimal investment decision at date 2, I (sm , pA1 , pB1 ) solves
I (smj ; pA1 ; pB1 ) 2 ArgmaxI2f0;1g VA (I);
(8)
where VA (I) is given in (3). The stock price of …rm A depends on the manager’s investment
decision, which itself depends on the stock price of …rm A. In equilibrium, speculators and
dealers in stock A take this feedback into account. Formally, a stock market equilibrium for …rm
A is a set fxA ( ), pA1 ( ); I ( ))g such that (i) the trading strategy xA ( ) maximizes the expected
pro…t for each speculator, (ii) the investment policy I ( ) maximizes the expected value of …rm
A at date 2 (solves (8)), and (iii) the pricing rule pA1 ( ) solves (7) given that speculators use the
10
trading strategy xA ( ), and the manager of …rm A follows the investment policy I ( ).
We provide the stock market equilibrium for …rm A in Lemma 2 in the Appendix. The main
property of this equilibrium is provided in Figure 2. This …gure shows the optimal investment
policy of the …rm as a function of the change in its price and that of …rm B between dates 0 and
1 when the manager does not receive direct managerial information.10 The price changes can
take three di¤erent values in equilibrium depending on the realization of the order ‡ow in this
stock (see Lemma 1 and 2). Clearly the optimal investment policy of …rm A depends on both the
price of stock A and the price of stock B as long as
B
is not zero. In particular, if the price of
stock A decreases moderately but the price of B increases then the manager of …rm A optimally
invests. Actually, an increase of the stock price of …rm B is a good signal about future demand
for the product sold by …rm A, that compensates the bad signal sent by a moderate drop in price
for stock A. Investment is also optimal, whatever the price of stock B, if the price of A increases.
[Insert Figure 2 about here]
2.3
Empirical implications
We now use the characterization of equilibrium stock prices for …rms A and B and the optimal
investment policy of …rm A to derive predictions that we test in the next section. We focus on
the implications of the model for the covariance between investment and stock prices.11 This is
natural since the in‡uence of stock prices on investment is often measured by regressing investment on (normalized) stock prices. However, investment and stock prices can be correlated even
in the absence of managerial learning from stock prices, for instance because speculators’signals
and managers’signals are positively correlated.
Hence, to isolate predictions that only arise when managers learn from peers’ prices, we
compare the e¤ects of
j,
, and
on the covariance between investment and stock prices in
three di¤erent scenarios: (i) managers ignore stock market information; (ii) managers learn solely
10
When the manager receives direct managerial information, his optimal investment policy is independent of
stock prices: he invests if he learns that demand is high and does not invest if he learns that demand is low.
11
The model has other testable implications. For instance, the model implies that the order ‡ow in stock A
should forecast the return from date 1 to date 3 on stock B and vice versa. Tookes (2008) provide evidence of
cross-stock price impacts of order ‡ow for related …rms.
11
from their own stock price; and (iii) managers learn from their stock price and that of their peers.
Table 5 in Section 4.3 – where we test the cross-sectional implications derived in this section –
provides a synoptic view of our predictions.
2.3.1
No managerial learning
First consider the case in which managers ignore the information contained in stock prices,
either because they are fully informed ( = 1), or because they fail to realize that stock prices
contain information. We focus on the latter case (“inattentive managers scenario”) because it
= 1.12 Let
encompasses the …rst as a special case when
j
L
j
H
j
=
=2.
Corollary 1 (benchmark 1: no managerial learning) In the absence of managerial learning, the
covariance between the price of stock B and the investment of …rm A is equal to:
cov(KI; pB1 ) = [
Hence, when
1
2,
B (2
1)]
B
K
2
0:
(9)
this covariance is positive and it increases with (i) the quality of managerial
information ( ), (ii) the fraction of informed speculators in stock B (
B ),
and (iii) the correlation
in demands for the products of …rms A and B ( ). Moreover the covariance between the price of
stock A and the investment of …rm A is also positive and equal to:
cov(KI; pA1 ) =
A( A
+
2
(
K))
K
2
0:
(10)
Hence, this covariance increases with (i) the quality of managerial information ( ), (ii) the
fraction of informed speculators in stock A (
informed speculators in …rm B(
A ),
and (iii) is independent of the fraction of
B ).
12
In the absence of managerial learning, the stock price of …rm B at date 1 is unchanged. The stock price of
…rm A is di¤erent because this price depends on the investment policy followed by the manager at date 2, which
depends on whether or not the manager of …rm A relies on stock prices as a source of information. In particular,
when he ignores stock prices, the manager of …rm A never invests unless sm = H (Assumption A.1). We account
for this in deriving Corollary 1.
12
Thus, even in the absence of managerial learning, the investment of a …rm is correlated with
its peers’stock price and its own stock price when
j
> 0. The reason is simply that managers’
signals are correlated with speculators’signals. Thus, we refer to this channel as the “correlated
information” channel. When investment and stock prices are related only through this channel,
cov(KI; pj ) increases with
because a higher
makes the correlation between speculators’and
the manager’s signals stronger.
2.3.2
Narrow managerial learning
Another possibility is that the manager only uses his stock price as a source of information, ignoring that of his peers. This scenario (“narrow managerial learning”) is another useful benchmark
to better isolate the predictions that arise only when managers also learn from their peer stock
prices.
Corollary 2 (benchmark 2: narrow managerial learning). When the manager of …rm A only
uses the information contained in the price of his stock, the covariance between the price of stock
B and the investment of …rm A is equal to:
cov(KI; pB1 ) = [( + (1
Hence, it is positive if
1
2.
)
A )(2
1)
B]
B
K
2
0:
(11)
It increases with the fraction of informed speculators in …rm A and
the e¤ ects of the other parameters ( ,
B
and ) on cov(KI; pB1 ) are as when the manager of
…rm A does not learn information from prices (Corollary 1). Moreover the covariance between
the price of stock A and the investment of …rm A is also positive and equal to:
2
6
cov(KI; pA1 ) = 6
4
|
A( A
+ (
{z2
K)) + A (1
} |
Correlated information
)(
A
+(
K
2{z
)((2
(1
Managerial learning from stock A
)
3
7K
7 : (12)
} 2
A ))5
Hence, it decreases with the quality of managerial information ( ) while the e¤ ects of
B
A
and
on cov(KI; pA1 ) are as when the manager of …rm A does not learn information from prices
(Corollary 1).
13
With narrow managerial learning, the manager obtains an informative signal about future
demand (either directly or from the stock market) with probability
just
+ (1
)
A,
rather than
as in the case with no managerial learning. For this reason, his investment decision becomes
more strongly linked to the the future demand for …rm B and therefore its peer stock price when
A
increases (…rst part of Corollary 2). Thus, the model implies that an increase in the level of
informed trading in a …rm’s stock (here
A)
should strengthen the sensitivity of its investment
to peer stock prices, whether managers learn information from stock prices in a narrow sense, as
assumed here, or in a broader sense (see Corollary 3 below).
Now, consider the link (measured by cov(KI; pA1 )) between the stock price of …rm A and its
investment decision. With narrow managerial learning, this link is stronger than in the absence
of managerial learning because a high stock price for A induces the manager to invest when he
does not receive direct managerial information. This e¤ect is captured by the second component
in the expression for cov(KI; pA1 ). This component decreases with
: the manager puts less
weight on the signal conveyed by his stock price when direct managerial information is of higher
quality. As a result, the net e¤ect of an increase in the quality of managerial information on the
covariance between the investment of …rm A and its stock price is negative while the opposite
prediction holds in the absence of managerial learning (Corollary 1).
2.3.3
Learning from peers’stock prices
Now, we turn to the more general case in which both the stock price of …rm A and the stock price
of …rm B in‡uence the investment decision of the manager of …rm A, as formalized in Lemma 2
in the Appendix and shown in Figure 2.
Corollary 3 In equilibrium, the covariance between the investment of …rm A and the stock price
of …rm B is equal to:
0
B
cov(KI ; pB1 ) = @( + (1
|
)
{zA
)(2
1)
Correlated information
B
}
+
14
(1
|
)
B (1
{z
A)
}
Managerial learning from stocks A and B
1
C
A
BK
2
. (13)
Thus, it is positive since
B (
B ).
>
1
2
and it increases in the fraction of informed speculators in stock
Moreover, it decreases in (i) the likelihood that the manager of …rm A receives direct
managerial information, , and (ii) the fraction of informed speculators in stock A (
A)
if
< 1.
Equation (13) splits the co-variation between the investment of …rm A and the stock price of
…rm B in two components. The …rst component is the contribution of the correlated information
channel: this contribution is just equal to the covariance between the investment of …rm A and
the price of its peer with narrow managerial learning. The second component is the incremental
contribution of the learning from peers channel. It is positive because an increase in the stock
price of …rm B conveys a positive signal about future demand for …rm A’s product and can
thereby induce the manager of …rm A to scale up capacity at date 2.
When the manager learns from his peer stock price and his own stock price, the covariance
between the price of stock B and the investment of …rm A decreases with the fraction of informed speculators in …rm A,
A,
and the quality of direct managerial information, . The same
mechanism is at work in both cases. The manager has three sources of information: (i) direct
managerial information, (ii) his stock price, (iii) the stock price of its peer. As
or
A
increase,
the informativeness of the two …rst sources of information becomes relatively higher relative to
the third. Hence, the manager relies relatively less on the third source of information and as a
result its investment becomes less correlated with the stock price of …rm B. This substitution
e¤ect is absent when the manager ignores the information contained in stock prices (no managerial learning) or when he only uses the information contained in his own stock price (narrow
managerial learning). Thus, assessing the direction of the e¤ect of
and
A
on cov(KI ; pB1 )
empirically o¤ers a sharp way to test for the “learning from peers”hypothesis. Indeed, a positive
e¤ect is consistent with other scenarios but not with the scenario in which managers learn from
their peers.
Corollary 4 In equilibrium, the covariance between the investment of …rm A and the stock price
15
of …rm A is given by:
0
B
B
B
B
B
cov(KI ; pA1 ) = B
B
B
B
B
@
|
|
A (1
)(
A
+(
A( A
K))
}
Correlated information
K
2
+ (
{z2
1
+
)((2
B
(1
{z
)(
A (1
Managerial learning from stocks A and B
C
C
C
C
CK
C .
C 2
C
C
C
)
+
))
B
B
A
}
(14)
For all parameter values, the covariance between the investment of …rm A and the stock price
of …rm A is positive. This covariance increases in (i) the fraction of informed speculators in
stock A (
A ),
and (ii) it decreases in the fraction of informed speculators in stock B (
B ),
if
< 1. Moreover, it decreases in the likelihood that the manager of …rm A receives managerial
information,
if
B
< bB and increases in this likelihood if
B
> bB where bB =
2(1 )(1
1+2(1 )(1
A)
A)
.
Hence, the co-variation between the investment of …rm A and its own stock price is equal to
a baseline component, due to the correlated information channel, plus an additional component
due to the learning from peers channel. The covariance between the investment of …rm A and its
stock price increases in the fraction of informed speculators in stock A because this fraction has
a positive e¤ect on both components.
We obtain two additional predictions that are speci…c to the “learning from peers” scenario.
First, the covariance between the investment of …rm A and its stock price should decline when
the level of informed trading in stock B increases. Indeed, such an increase strengthens the
informativeness of the price of stock B, which leads the manager of …rm A to rely more on this
price as a source of information.
Second, the quality of direct managerial information, , has an ambiguous e¤ect on the covariance between the investment of …rm A and its stock price: it is negative if
B
is small and
positive otherwise. Indeed, an increase in the quality of direct managerial information strengthens the correlated information channel (the …rst component in Equation (14)) and weakens the
learning from peers channel (the second component in Equation (14)). When
second negative e¤ect dominates (Corollary 2). As
16
B
B
is small, the
increases, the size of this negative e¤ect
becomes smaller because the manager of …rm A relies relatively less on his stock price. Hence,
when
B
is high enough, the …rst e¤ect dominates and cov(KI ; pA1 ) starts increasing with
(as
with no managerial learning).
Chen, Goldstein, and Jiang (2007) …nd that the sensitivity of a …rm’investment to its stock
price is negatively related with a proxy for managerial information (see their Table 3). However,
this relationship is not statistically signi…cant in their sample. Corollary 4 suggests a possible
explanation. The e¤ect of managerial information on the investment-to-price sensitivity of a
…rm depends on the level of informed trading in its peers. Hence, the unconditional e¤ect (i.e.,
the average e¤ect across …rms with di¤erent
B s’)
may well be indistinguishable from zero.
According to Corollary 4, a stronger test should allow the e¤ect of managerial information on
the investment-to-price sensitivity to di¤er according to
B,
the level of informed trading in peer
…rms. We implement such a test in Section 4.3.2.
2.4
The role of correlation in …rm demands ( )
1=RH (Assumption A.2). We now analyze the case in which
So far we have assumed that
< 1=RH . First, suppose that
1
2
< 1=RH . In this case, demands for products of …rms A and
B remain positively correlated. However, the informativeness of the price of stock B about the
demand for …rm A’s product is too small to in‡uence the manager’s investment decision.13 Hence,
this decision only depends on his own signal and his own stock price, as in the narrow managerial
learning scenario. The only di¤erence is that when
peer stock price is rational. Hence, when
the e¤ects of parameters
Second, suppose that
A,
<1
B,
and
1
2
1
2
< 1=RH , manager’s inattention to its
< 1=RH , the predictions of the model regarding
are given in Corollary 2.
1=RH < 12 . This case is symmetric to the case
> 1=RH . The
demands for the products of both …rms are negatively correlated. Thus, intuitively, a low (resp.,
high) realization of the stock price for …rm B signals a strong (resp. low) demand for …rm A.
Accordingly, the covariance between the investment of …rm A and the stock price of …rm B is
13
Indeed, E( A jrB > 0 )=E( A jdB = H )= K( RH 1) < 0. Thus, even when the price of stock B reveals that
the demand for product B is strong, the manager of …rm A chooses not to invest. Thus, his decision cannot be
in‡uenced by the price of stock B.
17
negative. However, in absolute value, this covariance is identical to that obtained when
> 21 ,
for all possible cases considered in previous sections (see the Internet Appendix, Section A.3).
Thus, the predictions derived in the previous sections hold for jCov(KI; pB1 )j. The expressions
for Cov(KI; pA1 ) are also unchanged in the various cases considered in the previous section.
Last, the case 1
1
2
1=RH <
is symmetric to the case in which
1
2
< 1=RH . Indeed,
in this case, the price of …rm B is not informative enough to in‡uence the investment decision
of …rm A. Thus, the predictions of the model regarding the e¤ects of parameters
A,
B,
and
are given in Corollary 2, except that they hold for jCov(KI; pB1 )j.
The correlated information and the learning components in the covariance between …rm A
investment and the stock price of B increase in . Thus, the covariance between the investment
of …rm A and the stock price of …rm B (cov(KI ; pB1 ) increases in
whether or not the manager
of …rm A learns from the stock price of …rm B (see equations (9), (11), and (13)).
In contrast, the e¤ect of
on the covariance between the investment of …rm A and its stock
price (cov(KI ; pA1 ) depends on whether the manager learns information from its peer or not,
as shown by the next corollary.
Corollary 5 The covariance between the investment of …rm A and its stock price (cov(KI ; pA1 )
weakly declines in
1
2
if and only if the manager of …rm A uses peer stock prices as a source
of information. Otherwise this covariance does not depend on .
When the manager of …rm A uses information contained in the stock price of …rm B, his
investment is relatively less driven by his own stock price than when he does not use the information contained in his peer stock price. Hence, the covariance between a …rm investment and
its stock price should be smaller when the correlation between the demand for the …rm product
and the demand for its peer products becomes higher in absolute value (that is
1
2
is higher),
as claimed in the …rst part of Corollary 5. As this prediction is speci…c to the case in which
managers learn both from their stock price and their peer stock price, it o¤ers another way to
test whether managers learn information from their peer stock prices.
18
2.5
Timing
In the model, the markets for stocks A and B clear simultaneously. As a result, at date 1, the
price of stock A does not re‡ect the information contained in the price of stock B when the
manager makes his investment decision (at time t = 2). Suppose instead that dealers in stock A
adjust their date 1 valuation after observing the price of stock B, at some date . If date
is
posterior to the date at which the investment decision is made, the model is obviously unchanged.
If instead date
occurs between dates 1 and 2 (the investment date), we have:
pA = E(VA (I ) j pA1 ; pB1 );
(15)
where pA is dealers’ valuation for stock A at date . Given our assumptions, the manager of
…rm A makes exactly the same investment decisions whether he conditions his investment policy
on pA only or on fpA ; pB1 g because the price of stock A at date
is a su¢ cient statistic for the
information contained in fpA ; pB1 g about the future demand for …rm A.
This case, however, is quite special for at least two reasons. First, in reality, information
contained in prices di¤uses only gradually across stocks (e.g., Hou (2007), Hong, Torous, and
Valkanov (2007), Cohen and Frazzini (2008), Menzli and Ozbas (2010)), so that there is cross…rms return predictability. In these conditions, managers should improve their forecasts of fundamentals by using the information contained in their peers’stock price, not yet impounded in
their own stock price, as in the model.
Second, and maybe most important, a …rm stock price re‡ects the value of its product portfolio. Thus, it is a noisy signal of the incremental value of a growth opportunity in one speci…c
product within the …rm’s portfolio, especially if the latter contributes little to the aggregate cash
‡ow.14 In this case, stock prices of “pure players”or at least …rms with cash-‡ows more correlated
with the …rm investment project provide more accurate signals than the stock price of the …rm,
even if the latter re‡ects the information available in other stock prices. We provide a simple
illustration of this point in the Internet Appendix (see Section A.6).
14
See Bresnahan, Milgrom, and Paul (1992) for a similar point.
19
3
3.1
Data and Methodology
Sample construction and de…nition of peer …rms
Our empirical analysis is based on a sample of U.S. public …rms. To test the main predictions
of the model (Corollaries 3 and 4), we must pair each …rm in our sample with other …rms (the
empirical counterpart of …rm B in our model) selling similar products. To this end, we use
the new Text-based Network Industry Classi…cation (TNIC) developed by Hoberg and Phillips
(2011).15
This classi…cation is based on textual analysis of product descriptions from …rms’10-K statements …led yearly with the Securities and Exchange Commission (SEC). Speci…cally, Hoberg and
Phillips (2011) calculate …rm-by-…rm similarity measures by parsing the product descriptions
from the …rm 10-Ks and forming word vectors for each …rm to compute a measure of product
similarity for every pair of …rms in each year. The similarity measure is based on the (relative)
number of words that two …rms’product description have in common. It ranges between 0% and
100%. Intuitively, the more …rms use common words in their product description, the more similar are their products. Using this similarity measure, Hoberg and Phillips (2011) construct TNIC
industries using a minimum similarity threshold. They de…ne each …rm i’s industry to include all
…rms j with pairwise similarities relative to i above a pre-speci…ed minimum similarity threshold
chosen to generate industries with the same fraction of industry pairs as 3-digit SIC industries
(equals to 21.32%). We de…ne as “peers” all the …rms that belong to the same TNIC industry
in a given year. The classi…cation covers the 1996 to 2008 period because TNIC industries are
based on the availability of 10-K annual …lings in electronically readable format.
Hoberg and Phillips (2011)’s TNIC industries have three important features. First, unlike
industries based on the Standard Industry Classi…cation (SIC) or the North American Industry
Classi…cation System (NAICS), they change over time. In particular, when a …rm modi…es
its product range, innovates, or enters a new product market, the set of peer …rms changes
accordingly. Second, TNIC industries are based on the products that …rms supply to the market,
15
The data can be found at: http://www.rhsmith.umd.edu/industrydata/industryclass.htm
20
rather than its production processes as, for instance, is the case for NAICS. Thus, …rms within
the same TNIC industry are more likely to be exposed to common demand shock, as in our model.
Third, unlike SIC and NAICS industries, TNIC industries do not require relations between …rms
to be transitive. Indeed, as industry members are de…ned relative to each …rm in the product
space, each …rm has its own distinct set of similar …rms. This provides a richer de…nition of
similarity and product market relatedness.
For the set of …rms with available TNIC industries, we gather …rms’stock price and return
information from the Center for Research in Securities Prices (CRSP), investment and other
accounting data from Compustat. We exclude …rms in the …nancial industries (SIC code 60006999) and utility industries (SIC code 4000-4999). We also exclude …rm-year observations with
negative sales, missing information on total assets, capital expenditure, …xed asset (property,
plant and equipment), and (end of year) prices. We detail the construction of all the variables in
Table 1. To reduce the e¤ect of outliers all ratios are winsorized at 1% in each tail.
[Insert Table 1 about here]
Table 2 presents descriptive statistics on the sample. The sample includes 46,162 …rm-year
observations (7,521 distinct …rms). For a given …rm, the average number of peers is 65 (#peers),
while the median is 28. Summary statistics for the main variables used in the analysis resembles
those reported in related studies. The average (median) rate of investment (capital expenditures
divided by lagged PPE) is 39.2% (24.2%). Following other studies on the sensitivity of investment
to stock price, we use Tobin’s Q as a proxy for a …rm (normalized) stock price. Q is de…ned as
a …rm’s stock price times the number of shares outstanding plus the book value of assets minus
the book value of equity, scaled by book assets. There is a large heterogeneity in Q in our
sample: It ranges from 0.55 to 13.79 with a mean of 2.18 and a median of 1.54. Also, the
average and median values of total assets are $1.4 billion and $169 million, respectively. We also
report summary statistics for peers’characteristics (average across all peers for each …rm-year).
They are very close from their own …rm counterpart, although aggregation lowers their standard
deviation.
21
[Insert Table 2 about here]
3.2
Empirical methodology
To empirically measure the co-variation between a …rm’s investment and (i) the stock prices of its
peers (cov(KI ; pB1 )) and/or (ii) its own stock price (cov(KI ; pA1 )), we adapt a standard linear
investment equation. Speci…cally, we estimate several speci…cations of the following baseline
equation:
Ii;t =
i
+
t
+ Q
i;t 1
+ Qi;t
1
+ 'X
i;t 1
+ Xi;t
1
+ "i;t ;
(16)
where the subscripts i and t represent respectively …rm i (…rm A in the model) and the year,
while the subscript
i represents a (equally-weighted) portfolio of peer …rms based on the TNIC
industries (…rm B in the model). The dependent variable Ii;t is a measure of corporate investment
in year t, which in the baseline speci…cation, is the ratio of capital expenditure in that year scaled
by lagged …xed assets (property, plant, and equipment). The variable Qi;t
stock price of …rm i in year t
1 as de…ned above. The variable, Q
1
is the normalized
i;t 1 ,
is the normalized
stock price of …rm i’s peers, computed as the average Q of all …rms belonging to the same TNIC
industry as …rm i in year t
1, except …rm i.16
In the baseline speci…cation, the vector Xi;t
1
includes control variables known to correlate
with investment decisions. Following previous research, we include the natural logarithm of
assets (ln(T Ai;t
1 ))
to control for the impact of size on corporate investment. Moreover, to
account for the well documented relationship between cash ‡ows and investment, we include cash
‡ows (CFi;t
1)
as an additional control variable. We also include the characteristics of peers
(their average size, ln(T A
i;t 1 ),
and cash ‡ows, CF
i;t 1 )
to further control for product market
characteristics. In addition, we account for time-invariant …rm heterogeneity by including …rm
…xed e¤ects ( i ) and time-speci…c e¤ects by including year …xed e¤ects ( t ). As a result, the
coe¢ cient
measures how, over time, the average …rm’s investment is tied to its peers’ stock
price. We allow the error term ("i;t ) to be correlated within …rms and correct the standard errors
16
Following common practice, we impose a lag of one year between the measurement of investment and the
measurement of Q (e.g., Chen, Goldstein, and Jiang (2007)). In the Internet Appendix (Table IA1), we show that
the …ndings holds when we consider no lag between investment and Q.
22
as in Petersen (2009).
Coe¢ cients
and
in equation (16) measure respectively the co-variation between the in-
vestment of …rm i and the (average) stock price of its peers and the co-variation between the
investment of …rm i and its stock price, holding other variables a¤ecting investment constant.
We use estimates of these coe¢ cients to test the main implications of the model. We proceed in
two steps.
First, we establish that estimates for
and
are both statistically signi…cant in our sample.
This is indeed a necessary condition for the “learning from peers” channel to play a role (see
Corollaries 3 and 4). This condition is not su¢ cient, however. Indeed, the investment of a …rm
should co-vary with its stock price and the stock price of its peers in part because investors’
information is correlated with managers’information about their investment opportunities (this
is captured by the “correlated information”component in the expressions for covariances between
stock prices and investment in our model). Thus, Qi and Q
i
in equation (16) play a dual role:
they are both (i) proxies for unobserved managers’ private information about their investment
opportunities and (ii) determinants of investment if managers learn information from stock prices.
Hence, in a second step, we test whether cross-sectional patterns for
and
are related to
measures of price informativeness or proxies for managerial information as predicted by Corollaries
3 and 4, with a special focus on predictions speci…c to the scenario in which stock prices in‡uence
investment (the “learning from peers” scenario).
4
4.1
Empirical Findings
Corporate investment and stock market prices
Table 3 presents estimates for various speci…cations of the baseline investment equation (16).
The sensitivity of a …rm investment to its peers’stock price ( ) is positive (equal to 0.051). It
is highly signi…cant with a t-statistic of 11.09 and the economic magnitude of the relationship
between the investment of a …rm and the stock price of its peers is substantial as well. To see
this, consider a one standard deviation increase in the stock price of peers (1:109). This shock
23
is associated with a 5:6% increase in investment on average, about 14.5% of the sample average
ratio of capital expenditures on …xed capital (
sd(Q i )=0.051 1.109=5.6%).
As found in many other studies (e.g. Baker, Stein, and Wurgler (2003), Chen, Goldstein, and
Jiang (2007), or Campello and Graham (2011)), our estimation reveals a positive and signi…cant
relation between a …rm’s stock price (Qi;t
1)
and its investment (Ii;t ). On average, the estimated
investment-to-price sensitivity, , is 0.067 with a t-statistic of 23.68. In terms of magnitude, the
e¤ect of peers’price on a …rm investment is about half that of its own stock price. Indeed, a one
standard deviation increase in the …rm own price (1.919) is associated with a 12.8% increase in
investment (
sd(Qi )=0.067 1.919=12.8%).17
[Insert Table 3 about here]
The rest of Table 3 checks the robustness of these …ndings to several changes in the baseline
speci…cation. In column 3, we modify the proxy for peers’ stock price and characteristics (size
and cash ‡ow) by using their median values instead of their average values. In Columns 4 and 5,
peers of a given …rm are de…ned as …rms within the same 3-digit SIC or 4-digit NAICS instead
of …rms within the same TNIC industry. We continue to observe that the investment of a …rm
co-varies positively with the stock price of its peers.
We perform additional robustness tests that we report in the Internet Appendix. We show that
our inference is robust to various forms of …xed e¤ects (e.g. the inclusion of industry and industryyear …xed e¤ects), di¤erent forms of clustering, and di¤erent de…nitions of corporate investment
(Tables IA2-IA4). Also, our conclusions remain identical if we use price-to-earning ratio (PE) or
(idiosyncratic) stock returns instead of the market-to-book ratio as alternative proxies for stock
prices (Table IA5). We also check that the investment of a …rm is not sensitive to the stock
price of unrelated …rms (Figure IA1), that is, randomly selected …rms, which therefore should
not be exposed to common demand shocks (that is,
=
1
2
in the model). The Internet Appendix
presents additional speci…cations (Table IA6) where we include a larger set of control variables
(such as leverage, cash holdings, asset tangibility, sale growth, lagged capital expenditures, and
17
The correlation between Qi and Q i is 0.40 in the sample. Thus the coe¢ cient on Q
that is distinct from the information in Qi .
24
i
captures an information
future (3 years) returns of …rms and their peers). The main results remain una¤ected.
As pointed by Erickson and Whited (2000, 2012), Qi imperfectly measures the information
in stock prices relevant to managers, which biases the estimate of
downward and may lead to
spurious correlation of investment with other variables (e.g., cash-‡ows or, in our case, peers’
valuation). They propose an estimator of
that accounts for this error-in-measurement problem.
The estimation of equation (16) using this technique is reported in the Internet Appendix, for
brevity (see Table IA7).18 As expected, we …nd that estimates of
are signi…cantly larger when
we use the Erickson and Whited’s estimator. However, we still …nd a positive and signi…cant
coe¢ cient for . Moreover, the magnitude of this coe¢ cient is similar to that reported in Table
3.
Overall, the investment of a …rm is positively and signi…cantly correlated with its market
valuation and its peers’market valuation. The …rst empirical …nding is well known. The second
is new to our paper.
4.2
Dynamic peers classi…cation
By using TNIC industries, the set of peers for each …rm is changing over time. So we can gauge
how the investment of a …rm is related to the stock price of past, present, and future peers.
Consider a …rm B which becomes a peer of …rm A at date t. This entry of …rm B in the set
of peers is like a positive shock on
in the model. Hence, we should observe an increase in the
sensitivity of …rm A investment to the stock price of …rm B. Symmetrically, the investment of
…rm A should become less sensitive to the stock price of a peer when this …rm exits its set of
peers. These e¤ects should hold whether or not managers learn information from stock prices
(see Corollary 5) as long as the price of a peer (new or old) contains information correlated with
the cash ‡ows of …rm A’s incremental investment projects.
Now consider the case in which …rm B is not yet a peer of …rm A at date t but will become
one at some point in the future, say date t + 2. This might happen because the manager of …rm
A decides to enter the product space of …rm B at date t + 1, after learning information from …rm
18
We thank Toni Whited for providing the Erickson and Whited estimator for unbalanced panel (stata command
XTEWreg) on her website.
25
B’s stock price (and other sources of information) at date t. In this case, the investment of …rm
A should co-vary with the stock price of …rm B before …rm B becomes a peer of …rm A.
To test these implications, we construct for each …rm-year four distinct sets of peers: (i) new
peers, (ii) still peers, (iii) past peers, and (iv) future peers. For …rm i and year t, we de…ne new
peers as …rms that are in the same TNIC industry as …rm i in year t but that were not in year
t
1. Following the same logic, past peers are …rms that were in the TNIC industry of …rm
i in year t
1 but that are not anymore in year t while still peers are …rms that were in the
TNIC industry of …rm i in year t
1 and continue to be in year t. Finally f uture peers are …rms
that are not in the TNIC industry of …rm i in year t but will be therein in year t + 1 (or t + 2,
depending on speci…cation). Then, we compute the average price (market-to-book ratio) for each
set of peers and use this price instead of the average stock price of all current peers in equation
(16).
[Insert Table 4 about here]
Table 4, present the results. In columns 1, we observe a positive and signi…cant relation
between …rms’ investment and the stock price of their new peers. The estimated coe¢ cient is
0.022 with a t-statistic of 7.39.19 In contrast, column 2 shows that the investment of …rms stop
being sensitive to the stock price of …rms that exit their set of peers (the sensitivity is 0.005 with
a a t-statistic of 1.77). Finally, results regarding still peers largely mirror the baseline results
(presented in Table 2) with a large coe¢ cient on the stock price of still peers (0.036 with a
t-statistic of 7.02).
In columns 4 and 5, we report the sensitivity of investment to the stock price of one year or
two years ahead f uture peers. The sensitivity of …rms’investment to the prices of their one year
ahead (respectively two years ahead) future peers is equal to 0:018 (0:014) and is statistically
signi…cant. These empirical …ndings …t well with a scenario in which …rms’ managers extract
information about the pro…tability of breaking into a new product space from the valuation of
…rms already active in this space.
19
The sample size is smaller than that used in previous tests because we exclude from the analysis …rms that
keep the same peers over the entire sample period.
26
4.3
Cross-sectional tests
We now turn to the cross-sectional implications that are unique to the learning from peers hypothesis. We examine how the co-variation between investment and stock prices varies crosssectionally with (i) measures of informed trading (
A
and
B ),
(ii) measures of managerial in-
formation other than stock prices ( ), and (iii) measures of the correlation in demand shocks
between a …rm and its peers ( ).
[Insert Table 5 about here]
Table 5 summarizes the cross-sectional predictions of the model for
and . We highlight with
an “ ”the predictions that enable us to distinguish the scenario (“learning from peers”) in which
…rms’investment decisions are in‡uenced by their peers’market valuation from the scenarios in
which this in‡uence does not exist (“no managerial learning”, and “narrow managerial learning”).
As discussed in Section 2.4, our cross-sectional predictions for the e¤ects of peer stock prices hold
for the absolute value of . Empirically, we always …nd that
for the average peer, the case in which
4.3.1
Informed trading (
>
1
2
is positive, which suggests that,
is the relevant one empirically.
j)
We …rst examine how investment-to-price sensitivities ( and ) varies with the level of informed
trading in the stock of a …rm (
A)
and the stocks of its peers (
B ).
We use a measure of
price informativeness as proxy for the level of informed trading in a stock. As in many other
papers (e.g., Durnev, Morck, and Yeung (2004), or Chen, Goldstein, and Jiang (2007)), we
measure the informativeness of a …rm stock price by its …rm-speci…c return variation (or price
non-synchronicity), de…ned as
i;t
= ln((1
2 )=R2 ), where R2 is the R2 from the regression
Ri;t
i;t
i;t
in year t of …rm i’s weekly returns on the market portfolio returns and the peers’portfolio returns
(both portfolios being value-weighted). The idea, due to Roll (1988), is that trading on …rmspeci…c information makes stock returns less correlated and thereby increases the fraction of total
volatility due to idiosyncratic returns. The level of informed trading in a stock is higher when
is high.
27
In our sample, the correlation between
i
and
i
(the average …rm-speci…c return variation
of the peers for stock i) is equal to 0:421. To test the model’s predictions, we need to measure the
e¤ect of
i
on the co-variation between investment and stock prices, while holding constant
and vice versa. To do so, we …rst regress
as
+
i.
Similarly, we regress
i
on
our tests, we analyze how a change in
+
variation of, say,
i
is independent of
i
i
on
each year and de…ne the regression residuals
i
i
i
+
i
or
a¤ects the estimates of of
+
i
In
and . As the
by construction, this amounts to varying the level of
informed trading in …rm i’s peers while keeping the level of informed trading in stock i,
Thus, to sum up, we use
+
i .
each year and de…ne the regression residuals as
+
i
as a proxy for
B
and
+
i
as a proxy for
i,
…xed.
A.
[Insert Table 6 about here]
To gauge the e¤ect of informed trading in peers’stocks on the co-variation between a …rm’s
investment and stock prices, we assign observations that are in the bottom and top quintiles of
the distribution of
+
i
into low and high groups, each year. Then, we estimate our baseline
regression (equation (16)) for each group and compare the coe¢ cient estimates of
and
across
groups.20 Panel A of Table 6 presents the results. For brevity, we only display the estimated
coe¢ cients for
of
i
and
i
and , their statistical and economic signi…cance, as well as the average values
(bottom two lines).21 A prediction speci…c to the learning from peer hypothesis is
that the sensitivity of a …rm’s investment to its own stock price should be lower when the level of
informed trading in its peers is high. This is indeed what we …nd in Panel A: the estimate of
lower in the high group than in the low group (0.070 versus 0.059). The di¤erence in
is
between
the two groups is marginally signi…cant (p-value of 0:063). In economic terms, the di¤erence is
signi…cant: a one standard deviation increase in a …rm’s stock price is associated with a 13.12%
increase in investment when the level of informed trading in peers’stocks is low while the same
shock raises investment by 11.12% when the level of informed trading in peers’stocks is high.
20
We partition …rms based on the informativeness variable measured at date t 1, i.e., contemporanously to the
stock price (Q) in equation (16). Moreover, we focus on only the lowest and highest quintile to limit the impact of
measurement errors in the price informativeness proxies (see Greene (2008)).
21
By construction the average i is similar across the low and high groups (2.514 vs. 2.494), but the average
i is not. It amounts to 1.244 in the low group vs. 3.566 in the high group.
28
Moreover, a …rm’s investment turns out to be more sensitive to the stock prices of its peers
when peers’ markets have a higher level of informed trading, as implied by both the learning
channel and the correlated information channel (Corollaries 1 and 2). All else equal (i.e., keeping
i
…xed), these results show that …rms rely less on their own stock price and more on the stock
prices of their peers when peers’stock prices are more informative.
In Panel B, we study how changes in the level of informed trading in a …rm’s own stock (
A)
a¤ects the sensitivity of its investment to its own stock price and the stock price of its peers
(holding
B
constant). The …rst row shows that the co-variation between a …rm’s investment
and the stock price of its peers decreases in the amount of informed trading in its own market,
as uniquely predicted by the scenario in which …rms learn from their peers’stock price. Indeed,
the coe¢ cient on
is large and signi…cant in the low group (0.064 with a t-statistic of 10.31), but
insigni…cant in the high group (0.011 with a t-statistic of 1.621). The di¤erence is economically
large. A one standard increase in peers’stock prices generates a modest 1.24% increase in a …rm’s
investment when its own market has a high level of informed trading. In contrast, the same shock
is associated with a 6.80% increase in a …rm’s investment when the level of informed trading in
its own stock is low. The second row con…rms the results of Chen, Goldstein, and Jiang (2007)
and Bakke and Whited (2010): a …rm’s investment is markedly more sensitive to its own stock
price when its market displays a higher level of informed trading. This observation is consistent
both with the correlated information channel and the learning channel.
The general picture that emerges from Table 6 supports the hypothesis that managerial
learning explains part of the positive correlation between corporate investment and stock market
prices. In the Internet Appendix (Table IA8), we show the robustness of the …ndings in Table
6 using two other proxies for the level of informed trading in a stock: one based on Llorente,
Michaely, Saar, and Wang (2002) and another one based on the P IN measure developed by
Easley, Kiefer, and O’Hara (1996).
29
4.3.2
Managerial information ( )
According to the learning from peers hypothesis, an increase in the quality of managerial information ( ) should decrease the sensitivity of a …rm’s investment to its peers’valuation. Moreover,
the e¤ect of a higher
on the sensitivity of a …rm’s investment to its own stock price should
depend on the level of informed trading in its peers’stocks. If this level is high, the e¤ect should
be positive while if this level is low, the e¤ect should be negative. None of these predictions are
obtained if the learning from peers hypothesis does not hold (see Table 5).
We test these two predictions using the trading activity of corporate insiders as a proxy
for managerial information, following Chen, Goldstein, and Jiang (2007). Intuitively, managers
should be more likely to trade their own stock if they possess more private information ( higher).
We measure the intensity of managers’insider trading activity (Insideri;t ) in a given year as the
total number of insider transactions for that year divided by the total year’s transactions. We
obtain the trades of corporate insiders from the Thomson Financial Insider Trading database and
the total number of trades (turnover) from CRSP.22 We follow previous studies (e.g. Beneish
and Vargus (2002), or Peress (2010)) and limit insider trades to open market stock transactions
initiated by the top …ve executives (CEO, CFO, COO, President, and Chairman of the Board).
[Insert Table 7 about here]
Table 7 reports the results from the estimation of equation (16) where observations in the
bottom and top quintiles of the distribution of Insideri are assigned into low and high groups
respectively.23 In the …rst row, we observe that the investment of a …rm is signi…cantly more
sensitive to its peers’stock price when managers have little private information. The estimated
coe¢ cient for
is 0.050 in the low group, and 0.038 in the high group. The di¤erence between
these estimates is signi…cant (p-value of 0.036). This …nding is consistent with the learning from
peers hypothesis and inconsistent with the absence of managerial learning from peers (see Table
22
The Thomson Financial Insider Filing database compiles all insider activity reported the the SEC. Corporate
insiders include those that have "access to non-public, material, insider information" and required to …le SEC form
3, 4, and 5 when they trade in their …rms stock.
23
The last row of Table 7 indicates that the average fraction of insider trading is 0% in the low group, while it
amounts to 9.85% in the high group. The sample average is 1.85%.
30
5).
The estimates for
are not signi…cantly di¤erent between the low and the high group (0.066
vs. 0.062). Chen, Goldstein, and Jiang (2009) also …nd that the sensitivity of a …rm’s investment
to its own stock price is not signi…cantly related to the extent of insider trading. The learning from
peers hypothesis however predicts that the sign of the relation between the quality of managerial
information and the co-variation between a …rm’s investment and its stock price should depend
on the level of informed trading in its peers’ stocks (Corollary 4). To test whether the data
uphold this prediction and to better understand the role of managerial information in explaining
investment-to-price sensitivities, we double-sort …rm-year observations based on (i) the level of
informed trading in peers (measured by
+
i;
see Section 4.3.1) and (ii) the quality of managerial
information for each …rm. As in Section 4.3.1, in each year, we keep the …rms that are in the
lowest and highest quintiles of the distribution of
+
i
and, in each of these two groups, we further
separate …rms in two groups based on the median value of Insideri (in each initial group). We
then estimate
and
for each of the four subgroups separately.
[Insert Table 8 about here]
Panel A of Table 8 displays the results. For low level of informativeness in peers’stock prices
(low
+
i ),
…rms’investment is less sensitive to their own price when managers enjoy more private
information. The coe¢ cient on Qi ,
, is 0.073 when the value of Insider is below the median
(column 1), and 0.062 when the value of Insider is above the median (column 2). In contrast,
when the level of informativeness in peers’ stock prices is high (high
estimates for
+
i ),
the ranking of the
between the two groups is reversed: A …rms’investment is more sensitive to its
own stock price when managers enjoy more private information. In either case, the di¤erence
between groups in estimates for coe¢ cient
is not signi…cant at conventional levels, maybe
because of lack of power due to smaller sub-samples. Yet, remarkably, the direction of the
e¤ects is consistent with the predictions of the learning from peers hypothesis. Furthermore, the
estimates for the coe¢ cients on Q
i
( ) are as expected: Corporate investment co-varies more
with peers’valuation when managers have less private information (columns 1 and 3), and even
31
more so when peers’valuation is highly informative.
4.3.3
Correlated demand for products ( )
If …rms learn information from their peers’valuation, the sensitivity of a …rm investment to its
own stock price should decrease when the correlation in product demands between the …rm and
its peers increases in absolute value (Corollary 5). We now test this prediction.
We use two di¤erent variables to proxy for the correlation between the demands for …rms’
products. First, we use the correlation of cash-‡ows (CF correlation) between a …rm and its
peers. We compute this correlation for each …rm-year using quarterly cash ‡ows over the previous
three years. Second, we use a direct measure of product similarity based on …rms’ product
description. As explained in section 3.1, the TNIC industries developed by Hoberg and Phillips
(2011) are constructed based on similarity scores (ranging from 0% to 100%) for each pair of
…rms that are above a minimum similarity threshold (21.32%). We use these scores to compute
the average similarity score (denoted similarity) of a …rm with its peers in a given year.24 This
is another proxy for .
[Insert Table 9 about here]
For each proxy, we group …rms in quintiles and estimate equation (16) separately for the
lowest and the highest quintiles of …rms. Results are reported in Table 9. The investment of
…rms in the lowest quintile is less sensitive to their peers’valuation than the investment of …rms
in the highest quintile. For instance, the estimate for
is 0.052 (with a t-statistic of 5.72) when
the correlation between a …rm’s cash ‡ow and that of its peers is relatively high and 0.030 (with
a t-statistic of 3.89) when this correlation is relatively low. This ranking of the estimate for
between the two extreme quintiles is consistent with the learning hypothesis but it could also
arise even if …rms do not learn from their peers (see Table 5).
In contrast, the ranking of the estimates for
between the high and the low quintiles is
consistent with the learning hypothesis only. For each proxy for , the investment of …rms in the
24
We are especially grateful to Jerry Hoberg and Gordon Phillips for sharing the data required for this test with
us.
32
low quintile is signi…cantly more sensitive to their stock price. For instance, the estimate for
is 0.063 (with a t-statistic of 16.791) when the average similarity between a …rm and its peers is
low and 0.047 (with a t-statistic of 12.40) when this average similarity is high. This di¤erence in
sensitivity between …rms with low and high values of
cannot be explained if …rms do not rely
on peers’stock prices as a source of information (see Table 5).
4.4
Going public and learning from peers’stock prices
Finally, we provide an additional piece of evidence in favor of our learning from peers hypothesis
by looking at the investment behavior of a subset of …rms before and after they go public. By
de…nition private …rms cannot learn from their own stock price. Hence, they are likely to rely
on their peers’stock price as a source of information. More importantly for our purpose, when
these …rms go public, they obtain the possibility of learning information from their own stock
price. Thus, going public is similar to an increase in
A
in our model (a move from a situation
where a …rm stock price is completely uninformative to a situation in which it is informative). If
the investment decision of a …rm is in‡uenced by its peers’valuation because managers extract
information from these prices, we should therefore observe a drop in the sensitivity of a …rm
investment to its peers’valuation after a …rm IPO. In contrast, we should observe no change in
this sensitivity after an IPO if managers do not learn information from stock prices or an increase
in this sensitivity if they learn information only from their own stock price (see the predictions
regarding
A
in Table 5).
Hence, we compare the investment behavior of …rms before and after their IPO using a sample
of 3,166 U.S. IPOs (provided to us by Jay Ritter) for the 1996 to 2008 period. In this sample,
we identify 1,342 …rms for which we can obtain data on investment and peers’stock prices before
and after these …rms go public (we consider a maximum of …ve years post-IPO).25 We examine
whether the investment of these …rms is sensitive to the stock price of their public peers before
they go public and how this sensitivity changes after …rms become public. To this end, we regress
25
The investment data are from Compustat. As observed in Teoh, Welch and Wong (1998), Compustat typically
contains information for …rms prior to their public o¤ering. We follow their approach and concentrate on …rms for
which we can obtain data for at least one year before the IPO.
33
the investment of IPO …rms (Ii ) on the stock price of their peers (Q i ), a dummy variable P osti
that equals one for the years that follow the IPO, the interaction between Q
i
and P osti , and
the usual control variables (including …rms’own post-IPO stock price).
[Insert Table 10 about here]
Table 10 presents the results of this ‘event-time’analysis. Consistent with the idea that the
stock market matters for private …rms, we …nd that their investment is sensitive to their public
peers’stock price. The coe¢ cient on Q
More importantly, the coe¢ cient on Q
i
i
is positive (0.026) and signi…cant (t-statistic of 4.54).
P osti is signi…cantly negative (-0.016 with a t-statistic
of 2.93). Thus, …rms’investment becomes less sensitive to their peers’valuation once they are
publicly listed. The same result obtains when we add sales growth to further control for …rms’
investment opportunities before they are publicly listed. As explained previously, this evolution of
the sensitivity of investment to peers’valuation is consistent with the scenario in which managers
learn information from their peers’stock prices but not with other scenarios in our model.
5
Conclusion
In this article, we document that a …rm’s investment is positively related to its peers’valuation.
Our hypothesis is that this relationship arises in part because peers’valuation conveys information
to managers about their growth opportunities. We formalize this idea in a simple model and we
derive several cross-sectional predictions about the relationship between investment and stock
prices that hold only if …rms extract information from their peers’ valuation. For instance,
the sensitivity of a …rm investment to its peers’ valuation should decrease with the quality of
managers’private information if the latter learn from their peers’valuation but should increase
with this quality if they don’t. Our tests do not reject the cross-sectional predictions speci…c to
the learning from peers hypothesis.
If …rms learn information about future demand from their peers’ valuation then product
market choices by …rms exert an externality on other …rms. For instance, if a …rm increases the
di¤erentiation between its products and those of its peers, the ability of peers to learn from its
34
stock price is weaker. Analyzing the e¤ects of these learning externalities on product market
choices and …rm valuations is an interesting venue for future research. In particular, they could
be one channel through which the stock market a¤ects product market choices and the degree of
competition within an industry.26
From a di¤erent perspective, learning externalities could also a¤ect the behavior of privately
held …rms. The precision and nature of the information they can gather from the stock prices
of publicly-listed peers could impact their willingness to go public, the type of securities issued,
their listing location, and the type of investors they want to attract.
More generally, it would
be interesting to better understand the characteristics (…rm, product, or stock markets) that
determine the extent to which …rms can learn from each other stock prices. We plan to investigate
these questions in future research.
26
Some papers suggest that the degree of competition within an industry a¤ects stock market returns and price
e¢ ciency (see Peress (2010) and Tookes (2008)). If …rms learn from their peers’valuation, there might also be an
e¤ect of the stock market on …rms’product choices and therefore the degree of competition in an industry.
35
Appendix
Appendix A: Stock market equilibrium for …rms A and B.
In this appendix, we solve for equilibrium prices at date 1 and the manager’s optimal investment decision at date 2. The results in this section are used for the discussion in Section 2.2 and
the empirical predictions, in particular those in Section 2.3.3. We provide the equilibrium in the
markets for stocks A and B in Lemma 1 and 2 below and then we provide a proof of these two
lemma.
Lemma 1 : The stock market equilibrium for …rm B is as follows:
1. A speculator: (i) buys when she knows that demand for …rm B’s product is high (xiB (H) =
+1) and (ii) sells when she knows that demand for …rm B’s product is low (xiB (L) =
1).
2. The stock price of …rm B at date 1 is an increasing step function of investors’net demand
for this stock, fB . Speci…cally, the stock price of …rm B, pB1 (fB ), is
E(
B)
when
1+
fB
B
1
B;
L
B
and
If investors’net demand is relatively strong (fB
when fB <
1
B ),
1+
H
B
when fB > 1
B;
B:
dealers infer that speculators have
received a good signal about the demand for …rm B’s product and the price of stock B is therefore
H
B.
If instead, investors’ net demand for stock B is relatively low (fB
1+
infer that speculators have received a bad signal and the price of stock B is
L
B.
B ),
dealers
Intermediate
realizations for investors’ net demand are not informative and therefore the price of stock B is
just equal to the unconditional expected cash-‡ow of this …rm.
Remember that
pB = pB E(
B)
is the change in the price of stock B from date t = 0 to
date t = 1. Lemma 1 implies:
Pr( pB
Pr( pB
0 jdA = H )
+ (1
=
0 jdA = H )
(1
)(1
B ) + (1
36
B)
)
=
1
(1
1
)
B
B
.
(17)
When
1
2
>
(i.e., …rm A and B have positively correlated demand), this likelihood ratio is strictly
greater than 1 and increases in
B.
That is, as
B
increases, a high stock price for B at date
t = 1 is more likely to occur when the demand for the product of …rm A is high than a low stock
price (and vice versa when the demand for …rm A is low). Hence, the stock price of …rm B is
informative about the demand for …rm A’s product and its informativeness increases with
B,
as
claimed in the text.
Let denote pH
A =
H
A
K); pM
A =E(
+(
1
A) + 2
( + (1
)
B) (
K); and pL
A =
L
A.
Note
M
L
that pH
A > pA > pA .
Lemma 2 : There is a stock market equilibrium for …rm A in which:
1. A speculator: (i) buys when she knows that demand for …rm A’s product is high (xiA (H) =
+1) and (ii) sells when she knows that demand for …rm A’s product is low (xiA (L) =
1).
2. The stock price of …rm A at date 1, pA1 (fA ), is an increasing step function of investors’
net demand for this stock, fA . Speci…cally, this price is pH
A when fA > 1
1+
A
fA
1
A;
and pL
A when fA <
1+
A;
pM
A when
A.
3. The manager of …rm A optimally invests at date 2 (I = 1) if and only if: (i) his managerial
information indicates a high demand at date 3 (sm = H), or (ii) its stock price at date 1
is pH
A or (iii) the stock price of …rm B is
H
B.
Investors’ net demand for stock A, fA , a¤ects its stock price because this demand is informative about the future demand for …rm A’s product, as for stock B. Moreover, as the level
of informed trading (
A)
in stock A increases, investors’ demand for this stock becomes more
informative. Hence, the informativeness of the stock price of …rm A for its manager increases
with
A.
As the last part of the proposition shows, the investment decision of the manager of
…rm A depends on its own stock price and the stock price …rm B.27
27
The last part of the proposition implies that if the manager of …rm A does not receive managerial information
then he does not invest, except for high realizations of the price for stock A or stock B. Thus, it speci…es the
manager’s optimal decision for all possible prices for stocks A and B, even those that cannot be observed in
equilibrium (prices that are out of the equilibrium path). For these prices, the manager’s beliefs regarding future
demand cannot be computed by Bayes law. We choose these beliefs so that they are identical to those that the
37
Using equation (7) and Lemma 2, we obtain:
pA0 (I ) = E(E(VA (I ) j fA )) = E(pA1 (fA )) = E(
A)
+
1
(
2
K) ( + (1
)(
A
+
B )).
(18)
L
Thus, the value of …rm A at date 0 (before trading) is higher than pM
A and pA and less than
pH
A . We conclude from the last part of Lemma 2 that when the manager has no information, he
invests if and only if (a) the change in price of stock A,
pA , is positive or (b) the change in
price of stock B is positive and the change in price of stock A is moderately negative ( pA =
pM
A
pA0 (I )). These observations yield Figure 2.
Proof of Lemma 1. We show that the strategies described in the lemma form an equilibrium.
Let
(xiB ; sbB ) be the expected pro…t of a speculator who trades xiB 2 f 1; 0; +1g shares of …rm
B when his signal is sbB and let
B
=
H
B
E(
B ) = E(
B)
L
B
=
H
B
L
B
2
.
First, consider an informed speculator who observes sbB = H. If he buys the asset, his expected
pro…t is:
(+1; H) = E(
= Pr(
1+
B
< fB < 1
B
B
pB1 (fB ) jb
sB = H; xiB = 1 )
jb
sB = H; xiB = 1 )(
B)
+ Pr(fB <
1+
B
jb
sB = H; xiB = 1 )(2
When sbB = H, the informed speculator expects other informed speculators to buy the asset and
uninformed speculators to stay put . Hence, using equation (6), he expects:
fB = zB +
B
when sbB = H:
(19)
manager would have if he could directly observe investors’ demands, fA and fB , in each stock (see the proof of
the proposition). In this way the equilibrium described in Lemma 2 is identical to the equilibrium that would be
obtained if the manager could observe investors’demand.
38
B ):
We deduce that
Pr(
1+
B
< fB < 1
B
jb
sB = H; xiB = 1 ) = Pr(
1 < zB < 1
2
B )=(1
B ).
and
Pr(fB <
1+
B
jb
s = H; xiB = 1 ) = Pr(zB <
1) = 0.
Thus,
(+1; H) = (1
B) B
> 0:
If instead, the speculator sells, his expected pro…t is:
( 1; H) = E(
pB1 (fB ) jb
sB = H; xiB =
B
1) =
(1
B) B
< 0,
where we have used the fact that each speculator’s demand has no impact on aggregate speculators’demand since each speculator is in…nitesimal. Thus, the speculator optimally buys the asset
when he knows that demand for …rm B is strong at date 3. A symmetric reasoning shows that
the the speculator optimally sells the asset when he knows that this demand is low at date 3.
Now consider dealers’prices at date t = 1. Remember that pB1 (fB ) =E( B jfB ). If sbB = L,
R
sB )di =
bB = L.
then 0 B xiB (b
B . Thus, equation (6) implies that fB > 1
B is impossible if s
R
sB )di = B . Thus, equation (6) implies that fB > 1
In contrast, if sbB = H, then 0 B xiB (b
B
happens when zB > 1
2
B.
As a result, when dealers observe fB > 1
B,
they deduce that
sbB = H. This implies:
pB1 (fB ) = E(
B
jfB ) = E(
B
jb
sB = H ) =
H
B
for fB > 1
B:
A symmetric reasoning implies:
pB1 (fB ) = E(
B
jfB ) = E(
B
jb
sB = L ) =
39
L
B
for fB <
1+
B:
Finally, for
1+
B
fB
1
B,
we have:
j 1+
B
fB
1
B)
Pr(dB = H j 1 +
B
pB1 (fB ) = E(
B
=
B +(2Pr(dB
=Hj 1+
B
< fB < 1
B)
1)
B.
Now:
=
B
)
Pr( 1 + B < fB < 1
1 + B < fB < 1
B jdB = H ) + Pr(
1
1
B
= .
2
B +1
B
Pr(
=
< fB < 1
1
jdB = H )
1 + B < fB < 1
B
B
jdB = L )
Hence,
pB1 (fB ) = E(
B)
for
1+
B
fB
1
B:
Proof of Lemma 2.
Speculators’ optimal trading strategy. It can be shown that speculators’order placement
strategy is optimal following the same steps as in the proof of Lemma 1. Hence, we skip this step
for brevity.
The manager’s optimal investment policy. Now consider the investment policy for the
manager of …rm A given the equilibrium price function pA1 ( ). If the manager receives managerial
information, he just follows his signal since this signal is perfect. Hence, in this case, he invest
if sm = H and does not invest if sm = L. If he receives no managerial information (that is,
sm = ?), the manager relies on stock prices. If he observes that pA1 = pH
A , the manager deduces
that fA > 1
A.
In this case, investors’ net demand in stock A reveals that demand for
product A is strong, i.e., dA = H (the reasoning is the same as that followed for …rm B when
fB > (1
B );
see the proof of Lemma 1). Thus the manager optimally invests. If instead the
manager observes that pA1 = pL
A then the manager deduces that fA <
1+
A
and he infers
that the demand for the product sold by …rm A will be low at date 3. In this case, the manager
40
optimally abstains from investing at date t = 2.
Finally if he observes pA1 = pM
A then the manager infers that
1+
A
fA
case, investors’demand in stock A is uninformative (that is, Pr(dA = H j 1 +
1
2 ).
A
1
A.
In this
< fA < 1
A )=
Hence, the manager of …rm A relies on the price of stock B as a source of information. If
pB1 =
H
B,
he deduces that fB > 1
B
and that the demand for product B will be high. We
deduce that
E(
pB =
H
B,
pA = pM
A ) = E( jdB = H ) > 0;
where the last inequality follows from Assumption A.2. If pB1 <
fB
1
B,
H
B,
(20)
the manager deduces that
in which case either investors’demand in stock B is uninformative or it signals that
the demand for product B will be low. Hence, when pB1 <
…rm A assigns a probability less than or equal to
1
2
H
B
and pA1 = pM
A , the manager of
to demand for product A being high at date
3. Using Assumption A.1, we deduce that the manager of …rm A does not invest in this case.
We have considered the manager’s optimal decision for prices that can be observed on the
M
L
equilibrium path, that is, pH
j , pj , pj for j 2 fA; Bg. Other prices for stocks A and B are out-of
the equilibrium path and, for these prices, the manager’s belief about the demand for its product
cannot be computed by Bayes rule. Consider the following speci…cation of the manager’s belief
about investors’demand for prices out of the equilibrium path: (i) if he observes pM
A
the manager of …rm A believes investors’ demand for stock A to be
1+
A
pA1 < pH
A,
fA
1
A,
(ii) if he observes pA1 < pM
A , the manager of …rm A believes investors’demand for stock A to be
less than
1+
A,
and (iii) if he observes pB1 <
H
B,
the manager believes that fB
1
B.
It
is straightforward, using the same reasoning as in the previous paragraph that this speci…cation
of the manager’s out-of-equilibrium beliefs leads to the optimal investment policy prescribed by
Lemma 2 for out-of-equilibrium prices.
Collecting these observations, we deduce that the manager’s optimal investment policy at
date 2 is as given in the last part of Lemma 2.
Equilibrium prices. Now consider the stock price of …rm A. We must check that the following
41
equilibrium condition is satis…ed:
pA1 (fA ) = E(VA (I ) jfA );
where I (sm ; pA ; pB ) = 1 if sm = H, or pA = pH
A , or pB1 =
H
B
(21)
and I (sm ; pA ; pB ) = 0, otherwise
(last part of Lemma 2). We check that the price function given in the second part of Lemma 2
satis…es Condition (21).
Suppose …rst that fA
(1
A ).
In this case, investors’net demand in stock A reveals that
demand for product A is strong, i.e., dA = H (the reasoning is the same as that followed for …rm
B when fB > (1
B );
see the proof of Lemma 1). Moreover, the stock price of …rm A is pH
A
according to the conjectured equilibrium. Hence, I = 1 and we deduce that:
H
A
E(VA (I ) jfA ) =
+(
which is equal to pH
A , so that for fA > (1
Now suppose that fA
1+
A.
K) for fA
A ),
(1
A );
Condition (21) is satis…ed.
In this case, investors’net demand in stock A reveals that
demand for product A is low, i.e., dA = L. Moreover, the stock price of …rm A is pL
A according
to the conjectured equilibrium. Hence, I = 0 and we deduce that:
E(VA (I ) jfA ) =
which is equal to pL
A . Hence, for fA
Last, consider the case in which
1+
1+
A
L
A
A,
for fA
1+
A;
Condition (21) is satis…ed.
fA
1
A.
In this case the investors’demand for
stock A is uninformative about the demand for product A, that is Pr(dA = H j 1 +
1
2
A
< fA < 1
(the reasoning is as for …rm B). Moreover, the stock price of …rm A is pM
A according to the
conjectured equilibrium. Hence, the manager of …rm A will invest (I = 1) if and only if (i) he
receives a signal that demand for product A is strong or (ii) he observes a high price for stock B.
42
A )=
We deduce that:
E(I (
K) j 1 +
A
fA
1
A)
= Pr(sm = H j 1 +
+Pr(sm = ? \ fB > 1
A
< fA < 1
j 1+
B
A
A
)(
< fA < 1
K)
A
)(
K).
Now,
Pr(sm = H j 1 +
A
< fA < 1
A
)= ;
2
and
Pr(sm = ? \ fB > 1
= (1
=
B
j 1+
)Pr(fB > 1
(1
B
A
< fA < 1
A)
jdB = H )Pr(dB = H j 1 +
A
< fA < 1
A)
)
2
B
We deduce from the expression for VA (I ) (equation (3)) that,
E(VA (I ) j 1 +
A
fA
which is equal to pM
A , so that for
1
1+
A)
A
= E(
fA
1
A)
+
A,
2
+
(1
)
2
B
(
K);
Condition (21) is satis…ed.
Appendix B: Empirical Implications
Proof of Corollary 1. When the manager of …rm A does not learn information from prices,
his investment policy, I b ( ), is
I b (sm ) =
8
>
< 1 if sm = H
>
: 0 if sm 6= H
:
(22)
As this investment policy di¤ers from that obtained when the manager accounts for the information in stock prices (I b ( ) 6= I ( )), the equilibrium price of stock A in the no managerial learning
43
scenario is di¤erent from that given in Lemma 2. In the Internet appendix (Section A.1), we
show that this price is given by:
pbA1 (fA ) =
8
>
>
>
>
>
>
>
>
>
>
>
>
<
E(
>
>
>
>
>
>
>
>
>
>
>
>
:
H
A
+ (
K)
A)
+ (
K)=2 when
when fA
L
A
1+
A
1
A;
< fA < 1
when fA
1+
A;
(23)
A:
By de…nition, the covariance between the price of stock A and the investment of …rm A in
the benchmark case is:
K
h
cov(I b ; pbA1 ) = K E(I b pbA )
Using equations (22) and (23), we deduce that cov(K
i
E(I b )E(pbA ) :
I b ; pbA1 ) =
A( A
+ 2(
K)) K2 , as
announced in the corollary.
The cash ‡ow for …rm B is the same whether or not the manager of …rm A relies on stock
prices as a source of information. Hence, even if when the manager of …rm A does not use stock
prices as a source of information, the stock price of …rm B is given by Lemma 1. Using this
observation, we easily derive the covariance between the price of stock B and the investment of
…rm A; I b , when there is no managerial learning.
Proof of Corollary 2. For brevity, we derive the expressions for cov(K
I; pj1 ) when the
manager only learns from his stock price in the internet appendix (see Section A.2).
Proof of Corollary 3. The covariance between the price of stock B and the investment of …rm
A in equilibrium is:
K
cov(I ; pB1 ) = K [E(I pB1 )
E(I )E(pB1 )] :
The last part of Lemma 2 implies that I (sm ; pA1 ; pB1 ) = 1 if sm = H, or pA = pH
A , or pB =
44
H
B
and I (sm ; pA1 ; pB1 ) = 0, otherwise. Using Lemma 1, we deduce that:
E(I pB1 ) =
2
(
Moreover, E(pB1 ) = E(
B
B)
+
B (2
1)) +
(1
)
2
(
B A
+
B B (1
2
A
+
A
)):
and
E(I ) =
2
(1
+
)
2
(
A
+
B (1
A )):
(24)
Thus, after some algebra, we obtain
cov(KI ; pB1 ) = (( + (1
)
A )(2
1)
B
+ (1
)
B (1
A ))
BK
2
:
We deduce that:
@cov(I ; pB1 )
@ B
@cov(I ; pB1 )
@ A
@cov(I ; pB1 )
@
= ((2
1)( + (1
=
(1
)
B (1
=
(1
A ) B (1
)
A)
+ (1
) < 0; if
)
B
)(1
A ))
B =2
> 0;
< 1,
< 0.
This proves Corollary 3.
Proof of Corollary 4. The covariance between the price of stock A and the investment of …rm
A is given by:
K
cov(I ; pA1 ) = K [E(I pA1 )
E(I )E(pA1 )] :
(25)
Calculations yield:
E(I pA1 ) =
2
H
A pA
+ (1
M
A )pA
+(
1
2
)
H
A pA
+ (1
M
A ) B pA
.
Equation (24) gives the expression for E(I ). By de…nition, pA0 = E(pA ). Hence, the expression
for E(pA ) is given by equation (18). Using these observations, we obtain equation (14) from
equation (25).
45
Di¤erentiating equation (14), it is straightforward that
1
@cov(I ; pA1 )
=
@ A
2
A
+(
K) ((1
)(1 +
A
+
B)
@cov(I ;pA1 )
@ B
2
+1
A
< 0 and
)2
+ (1
B B)
> 0,
and
K
@cov(I ; pA1 )
=
@
Clearly,
B
@cov(I ;pA1 )
@
A
increases in
B.
((1
2
)(1
Moreover, for
B
A )(1
= 1,
@cov(I ;pA1 )
@
= 0, it is negative. Thus, there exists a value bB such that
@cov(I ;pA1 )
@
< 0 if
B
< bB and
@cov(I ;pA1 )
@
corollary. Calculations yield bB =
= 0 for
2(1 )(1
1+2(1 )(1
A)
A)
B
.
B)
B
).
2
is positive while for
@cov(I ;pA1 )
@
> 0 if
B
> bB ,
= bB , as claimed in the last part of the
Proof of Corollary 5.
As explained in the text, when 1
1
RH
<
1
RH ,
the equilibrium is identical to the case in
which the manager ignores the information contained in the price of stock B. Thus, in this case,
cov(KI ; pA1 ) is given by equation (12) in Corollary 2. In contrast, when
<1
1
RH
or
>
1
RH ,
investment is in‡uenced by the stock price of B and, as explained in the text, cov(KI ; pA1 ) is
given by equation (14) in Corollary 4. Calculations show that the expression for cov(KI ; pA1 )
in (14) is smaller than in (12). The …rst part of Corollary 5 follows. When the manager of
…rm A ignores the information contained in the stock price of …rm B, cov(KI ; pA1 ) is given by
equations (12) or (10). In either case, it does not depend on , which yields the last part of the
corollary.
46
References
[1] Albuquerque, Rui, Eva De Francisco, and Luis B. Marques, 2008, Marketwide Private Information in Stocks: Forecasting Currency Returns, Journal of Finance 63, 2297-2343.
[2] Baker, Malcolm, Jeremy Stein, and Je¤rey Wurgler, 2003, When Does the Market Matter?
Stock Prices and the Investment of Equity-Dependent Firms, Quarterly Journal of Economics
118, 969-1005.
[3] Bakke, Tor, and Toni M. Whited, 2010, Which Firms Follow the Market? An Analysis of
Corporate Investment Decisions, Review of Financial Studies 23, 1941-1980.
[4] Beneish, Messod D., and Mark E. Vargus, Insider Trading, Earnings Quality, and Accrual
Mispricing, The Accounting Review 77, 755-791.
[5] Bond, Philip, Alex Edmans, and Itay Goldstein, 2012, The Real E¤ects of Financial Markets,
Annual Review of Financial Economics 4, 339-360.
[6] Bresnahan Timothy, Paul Milgrom, and Jonathan Paul, 1992, The Real Output of the Stock
Exchange, in Output Measurement of the Services Sectors, NBER.
[7] Campello, Murillo and John Graham, 2011, Do Stock Prices In‡uence Corporate Decisions?
Evidence from the Technology Bubble, Journal of Financial Economics (forthcoming).
[8] Chen, Qi, Itay Goldstein, and Wei Jiang, 2007, Price Informativeness and Investment Sensitivity to Stock Price, Review of Financial Studies 20, 115-147.
[9] Cohen, Lauren, and Andrea Frazzini, 2008, Economic Links and Predictable Returns, Journal of Finance 63, 1977-2011.
[10] Dougal, Casey, Christopher A. Parsons, and Sheridan Titman, 2012, Urban Vibrancy and
Corporate Growth, Working Paper, University of San Diego.
[11] Dow, James, and Gary Gorton, 1997, Stock Market E¢ ciency and Economic E¢ ciency: Is
There a Connection?, Journal of Finance 52, 1087-1129.
47
[12] Dow James, Itay Goldstein, and Alexander Gümbel, 2011, Incentives for Information Production in Markets where Prices A¤ect Real Investment, Working paper, Wharton School.
[13] Durnev, Art, Randall Morck, and Bernard Yeung, 2004, Value-Enhancing Capital Budgeting
and Firm-Speci…c Return Variation, Journal of Finance 59, 65-105.
[14] Easley, David, Nicholas Kiefer, Maureen O’Hara, and J. Paperman, 1996, Liquidity, Information, and Less-Frequently Traded Stocks, Journal of Finance 51, 1405–1436.
[15] Edmans Alex, Itay Goldstein, and Wei Jiang, 2012, The Real E¤ects of Financial Markets:
the Impact of prices on Takevovers, Journal of Finance 67, 933-971.
[16] Erickson, Timothy, and Toni M. Whited, 2000, Measurement Errors and the Relationship
between Investment and q, Journal of Political Economy 108, 1027-1057.
[17] Erickson, Timothy, and Toni M. Whited, 2012, Treating Measurement Errors in Tobin’s q,
Review of Financial Studies 25, 1286-1329.
[18] Fang Vivian, Thomas Noe, and Sheri Tice, 2009, Stock Market Liquidity and Firm Value,
Journal of Financial Economics 94, 150-169.
[19] Ferreira Daniel, Miguel Ferreira and Clara Raposo, 2011, Board Structure and Price Informativeness, Journal of Financial Economics 99, 523-545.
[20] Foucault, Thierry, and Laurent Frésard, 2012, Cross Listing, Investment Sensitivity to Stock
Price and the Learning Hypothesis, Review of Financial Studies 25, 3305-3350.
[21] Foucault, Thierry, and Thomas Gehrig, 2008, Stock Price Informativeness, Cross-Listings,
and Investment Decisions, Journal of Financial Economics 88, 146-168.
[22] Fracassi, Cesare, 2012, Corporate Finance Policies and Social Networks, Available at SSRN:
http://ssrn.com/abstract=1121503 or http://dx.doi.org/10.2139/ssrn.1121503
[23] Gilbert, Richard L., and Marvin Lieberman, 1987, Investment and Coordination in
Oligopolistic Industries, Rand Journal of Economics 18, 17-33.
48
[24] Goldstein, Itay, and Alexander Guembel, 2008, Manipulation and the Allocational Role of
Prices, Review of Economic Studies 75, 133-164.
[25] Graham John and Campbell Harvey, 2001, The theory and practice of corporate …nance:
Evidence from the …eld, Journal of Financial Economics 61, 187-243.
[26] Greene, William H., 2008, Econometric Analysis, 6th Edition, New York: Prentice Hall.
[27] Hoberg, Gerard, and Gordon Phillips, 2011, Text-Based Network Industries and Endogenous
Product Di¤erentiation, Working Paper, University of Maryland.
[28] Hong, Harrison, Walter Torous, and Rossen Valkanov, 2007, Do industries lead stock markets? Journal of Financial Economics 83, 367-396.
[29] Hou, Kewei, 2007, Industry Information Di¤usion and the Lead-lag E¤ect in Stock Returns,
Review of Financial Studies 20, 1113-1138.
[30] Kyle, Albert, 1985, Continuous Auctions and Insider Trading, Econometrica 53, 1315–1335.
[31] Leary, Mark T., and Michael R. Roberts, 2012, Do Peer Firms A¤ect Corporate Financial
Policy? Working Paper.
[32] Llorente, Guillermo, Roni Michaeli, Gideon Saar, and Jiang Wang, 2002, Dynamic VolumeReturn Relation of Individual Stocks, Review of Financial Studies 15, 1005-1047.
[33] Luo, Yuanzhi, 2005, Do Insiders Learn from Outsiders? Evidence From Mergers and Acquisitions, Journal of Finance 60, 1951-1982.
[34] Morck, Randall, Andrei Shleifer, and Robert W. Vishny, 1990, The Stock Market and Investment: Is the Market a Sideshow?, Brookings Papers on Economic Activity, 157-215.
[35] Menzly, Lior and Oguzhan Ozbas, 2010, Market Segmentation and Cross-predictability of
Returns, Journal of Finance 65, 1555-1580.
[36] Petersen, Mitchell, 2009, Estimating Standard Errors in Finance Panel Datasets: Comparing
Approaches, Review of Financial Studies 22, 435-480.
49
[37] Peress, Joel, 2010, Product Market Competition, Insider Trading, and Stock Market E¢ ciency, Journal of Finance 65, 1-44.
[38] Roll, Richard, 1988, R2 , Journal of Finance 43, 541-566.
[39] Subrahmanyam, Avanidhar, and Sheridan Titman, 1999, The Going-Public Decision and
the Development and Financial Markets, Journal of Finance 54, 1045-1082.
[40] Subrahmanyam, Avanidhar, and Sheridan Titman, 2013, Financial Market Shocks and the
Macroeconomy, Working Paper.
[41] Teoh, Siew Hong, Ivo Welch, and T.J. Wong, 1998, Earnings Management and the Long-Run
Market Performance of Initial Public O¤erings, Journal of Finance 53, 1935-1974.
[42] Tookes, Heather, 2008, Information, Trading, and Product Market Interactions: Crosssectional Implications of Informed Trading, Journal of Finance 63, 379-413.
50
Figure 1: Timing of the model Date 1 Date 2
1. Investors trade shares of firms A and B 2. Stock prices, pA1 and pB1, are realized Date 3 1. The manager of firm A
observes stock prices, pA1 and pB1
and his private signal smA
2. The managers of firm A
decides to invest or not
Demands and firms'
cash‐flows are realized Figure 2: Equilibrium Figure 2 shows for each realization of the order flow in stock A (Y‐axis) and B (X‐axis), the sign of the changes in prices realized between dates 0 and 1 in each stock and the corresponding investment decision for firm A . Table 1: Variable definition Variable Capex PPE Q TA Size CF (cash flow) Sale Ψ Insider CF Correlation Similarity Fluidity Definition Capital expenditures (capx) Property, Plant and Equipement (ppent)
[Book value of assets (at) – book value of equity (ceq) + market value of equity (csho*prcc_f)] / book value of assets (at) Book value of total assets (at)
Logarithm of the book value of assets (at)
Income before extraordinary items (ib) plus depreciation (dp) divided by total assets Total sales (sale) Firm specific return variation computed as ψi,t = ln[(1‐R2i,t)/ R2i,t], where R2i,t represents the R2 from a regression of firm i weekly returns on the returns on the (value‐weighted) market portfolio and a (value‐weighted) portfolio of peers, where peers are defined using the TNIC industries. Total number of insider transactions in a given year divided by the total year’s transaction. We only consider open market stock transactions initiated by the top five executives (CEO,CFO, COO, President, and Chairman of the board) Correlation between firm i’s cash flow and the average cash flow of its peers (using quarterly observation over the past three years) Average similarity score across the peers of firm i. Similarity scores (ranging from 0% to 100%) are from the detailed text‐
based analysis of Hoberg and Phillips (2010) Measure of the stability of a firm’s product mix compared to that of its peers (from Hoberg, Phillips, and Prabhala (2012) Source Compustat Compustat Compustat Compustat Compustat Compustat Compustat CRSP Thomson Financial Insider Trading Datavase and CRSP Compustat Hoberg ‐Phillips data library Hoberg ‐Phillips data library Table 2: Descriptive statistics This table reports the summary statistics of the main variables used in the analysis. We present the mean, median, 25th and 75th percentiles, standard deviation and the number of non‐missing observations. All variables are defined in the Appendix. In the upper panel, we present the statistics for firm‐level observations. In the lower panel, we present statistics for peers’ average (i.e. the average of peers for each firm‐year observation). Peers are defined using the TNIC industries developed by Hoberg and Phillips (2010). The sample period is from 1996 to 2008. Mean 25th Median 75th St. Dev. Obs. 0.468 1.919 0.312 4,577.1 46,162 46,162 46,074 46,162 Own firm 0.392 2.182 ‐0.029 1,418.5 Capital expenditures (Ii) Qi Cash Flow (CFi) Total assets (mio.) 0.130 1.116 ‐0.032 44.1 0.242 1.546 0.067 169.3 0.457 2.425 0.118 733.1 Average of peers # Peers Capital expenditures (I‐i) Q‐i Cash flow (CF‐i) Total assets (mio) 64.89 0.403 2.294 ‐0.031 1,477.1 9 0.234 1.503 ‐0.085 374.2 28 0.362 1.936 0.032 845.6 89 0.518 2.814 0.084 1,739.3 83.05 0.231 1.109 0.169 2,105.4 46,162 46,035 46,162 46,155 46,162 Table 3: Investment‐to‐price sensitivities: Baseline Results This table presents the results from estimations of equation (16). The dependent variable is investment, defined as capital expenditures divided by lagged property, plant and equipment (PPE). Qi is the market‐to‐book ratio of firm i, defined as the market value of equity plus the book value of debt minus the book value of assets, scaled by book assets. Q‐i is the average market‐to‐book of the peers of firm i, where peers as defined using the TNIC industries developed by Hoberg and Phillips (2010). The subscript –i refers to (the average across) firm i’s peers. The other explanatory variables are defined in Table 1. Columns (1) and (2) present the baseline estimations. In column (3) we use the median peers’ characteristics instead of the average. In column (4) the peers of firm i are all the firms that belong to firm i’s 3‐digit SIC industry. In column (5) the peers of firm i are all the firms that belong to firm i’s 4‐
digit NAICS industry. The sample period is from 1996 to 2008. The standard errors used to compute the t‐statistics (in squared brackets) are adjusted for heteroskedasticity and within‐firm clustering. All specifications include firm and year fixed effects. Symbols ** and * indicate statistical significance at the 1% and 5% levels, respectively. Investment (capex over lagged PPE) Baseline (1) Median (2) SIC‐3 (3) NAICS‐4 (4) Q‐i CF‐i Size‐i 0.051** [11.09] 0.145** [4.31] ‐0.004 [0.97] 0.054** [9.75] 0.133** [3.96] ‐0.005 [1.26] 0.044** [7.09] 0.211** [4.68] ‐0.012 [144] 0.050** [7.64] 0.183** [3.802] ‐0.007 [0.73] Qi CFi Sizei 0.067** [23.68] 0.356** [18.66] ‐0.063** [8.51] 0.068** [24.07] 0.356** [18.66] ‐0.061** [8.25] 0.070** [24.45] 0.356** [18.42] ‐0.060** [7.98] 0.069** [23.98] 0.355** [18.22] ‐0.060** [8.08] Firm FE Year FE Yes Yes Yes Yes Yes Yes Yes Yes Obs. Adj. R2 43,861 0.36 43,861 0.36 43,867 0.36 43,867 0.36 Table 4: Investment‐to‐price sensitivities: Dynamic peers’ classification This table presents the results from estimations of equation (16) where we change the definition of peers to include dynamic relationships. The dependent variable is investment, defined as capital expenditures divided by PPE. Qi is the market‐to‐book ratio of firm i. Q‐i is the average market‐to‐book of the peers of firm i, where peers as defined using the TNIC industries developed by Hoberg and Phillips (2010). The subscript –i refers to (the average across) firm i’s peers. The other explanatory variables are defined in Table 1. In column (1), peers are defined as firms in the same TNIC industry in year t that we not in year t‐1 (new peers). In column (2), peers are defined as firms that were in the same TNIC industry in year t‐1 but not in year t (past peers). In column (3), peers are defined as firms in the same TNIC industry in year t that we in year t‐1 (old peers). In columns (4) and (5) peers are defined as firms that are not in the same TNIC industry in year t but will be in year t+1 and t+2 respectively (future peers). The sample period is from 1996 to 2008. The standard errors used to compute the t‐statistics (in squared brackets) are adjusted for heteroskedasticity and within‐firm clustering. All specifications include firm and year fixed effects. Symbols ** and * indicate statistical significance at the 1% and 5% levels, respectively. Investment (capex over lagged PPE) New (1) Past (2) Still (3) Future (+1) (4) Future (+2) (5) Q‐i CF‐i Size‐i 0.022** [7.39] 0.156** [4.33] ‐0.007 [1.44] 0.005 [1.77] 0.140** [3.87] ‐0.006 [1.43] 0.036** [7.02] 0.153** [4.33] ‐0.005 [1.25] 0.018** [5.69] 0.142** [3.38] ‐0.007 [1.28] 0.014** [3.93] 0.156** [3.37] ‐0.010 [1.83] Qi CFi Sizei 0.071** [23.03] 0.347** [17.26] ‐0.048** [4.33] 0.062** [19.33] 0.320** [15.56] ‐0.043** [5.12] 0.058** [18.28] 0.314** [15.49] ‐0.044** [5.53] 0.072** [22.13] 0.361** [16.66] ‐0.040** [4.35] 0.074** [21.21] 0.359** [15.13] ‐0.035** [3.37] Firm FE Year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes Obs. Adj. R2 37,579 0.35 34,013 0.31 35,533 0.32 33,713 0.35 28,432 0.369 Table 5: Investment‐to‐price sensitivities: Cross‐sectional predictions This table summarizes the cross‐sectional predictions of the model as explained in Section 2.3. The model generates predictions for the co‐variation between a firm’s investment and its own stock price (β) and the co‐
variation between a firm’s investment and the price of its peers (η) as a function of four parameters: the fraction of informed trading in a firm’s stock πown (πA in the model), the fraction of informed trading in the peer’s stock πpeer
(πB in the model), the private information of managers (γ), and the correlation between the demands for firms’ products. We present the predictions for three scenarios: (i) managers ignore stock market information: (ii) managers only rely on the information contained in their own stock price, and (iii) managers rely on their stock price and the stock price of their peers. We highlight with an “*” the predictions that are unique the scenario in which managers learn from the stock prices of their peers (iii). Sensitivity of Investment to: Own stock price (β) Peer stock price (η) πown
πpeer
γ
|ρ-1/2|
πown
πpeer
γ
|ρ-1/2|
(i) No managerial learning + 0 + 0 0 + + + (ii) Narrow managerial learning + 0 ‐ 0 + + + + (iii) Learning from peers + ‐* +/‐* ‐* ‐* + ‐* + Effect of Table 6: Investment‐to‐price sensitivities: Stock price Informativeness (πA and πB) This table presents the results from estimations of equation (16) across different partitions based on a price informativeness proxy. The dependent variable is investment, defined as capital expenditures divided by PPE. Qi is the market‐to‐book ratio of firm i. Q‐i is the average market‐to‐book of the peers of firm i, where peers as defined using the TNIC industries developed by Hoberg and Phillips (2010). The subscript –i refers to (the average across) firm i’s peers. The other explanatory variables (not reported) are defined in Table 1. The low (high) partition corresponds to the bottom (top) quintile of the partitioning variable (in year t‐1). The partitioning variable is peers’ average (firm) firm‐specific return variation Ψ‐i+ (Ψi+). The partitioning variables are de‐correlated (superscript+) as explained in section 4.3.2. The sample period is from 1996 to 2008. The standard errors used to compute the t‐
statistics (in squared brackets) are adjusted for heteroskedasticity and within‐firm clustering. We report the economic effects (the coefficient estimate multiplied by the standard deviation of the explanatory variable in the relevant partition) in italic. We report p‐values corresponding to (unilateral) tests for difference in coefficients across low and high partitions. All specifications include firm and year fixed effects. Symbols ** and * indicate statistical significance at the 1% and 5% levels, respectively. Panel A : Investment‐to‐price sensitivities as a function of Ψ‐i+ (Low) (High) (p‐value) 0.029** [3.93] 2.98% 0.070** [15.74] 13.12% 7,994 2.514 1.244 0.051** [4.29] 6.06% 0.059** [14.95] 11.12% 7,992 2.494 3.566 0.018** 0.063 Q‐i (η) Qi (β) Obs. Mean (Ψi) Mean (Ψ‐i) Panel B : Investment‐to‐price sensitivities as a function of Ψi+ (Low) (High) (p‐value) 0.064** [10.31] 6.80% 0.047** [16.57] 9.87% 0.011 [1.21] 1.24% 0.069** [12.32] 11.73% 0.000** 0.000** 8,011 ‐0.041 2.351 7,982 5.190 2.403 Q‐i (η) Qi (β) Obs. Mean (Ψi) Mean (Ψ‐i) Table 7: Investment‐to‐price sensitivities: Managerial Information (γ) This table presents the results from estimations of equation (16) across different sample partitions based on a proxy for managerial information. The dependent variable is investment, defined as capital expenditures divided by PPE. Qi is the market‐to‐book ratio of firm i. Q‐i is the average market‐to‐book of the peers of firm i, where peers as defined using the TNIC industries developed by Hoberg and Phillips (2010). The subscript –i refers to (the average across) firm i’s peers. The other explanatory variables (not reported) are defined in Table 1. The low (high) partition corresponds to the bottom (top) quintile of insider trading (Insider) in year t‐1, defined as the number of insider transactions in year t divided by the total year t’s transactions. We limit insider trades to open market stock transactions initiated by the top five executives (CEO, CFO, COO, President, and Chairman of the board). The sample period is from 1996 to 2008. The standard errors used to compute the t‐statistics (in squared brackets) are adjusted for heteroskedasticity and within‐firm clustering. We report the economic effects (the coefficient estimate multiplied by the standard deviation of the explanatory variable in the relevant partition) in italic. We report p‐values corresponding to (unilateral) tests for difference in coefficients across low and high partitions. All specifications include firm and year fixed effects. Symbols ** and * indicate statistical significance at the 1% and 5% levels, respectively. Investment‐to‐price sensitivities as a function of Insider (γ) Q‐i (η) (Low) (High) (p‐value) 0.050** 0.038** 0.036** [6.57] [4.89] 6.15% 4.06% Qi (β) 0.066** 0.062** 0.462 [16.73] [13.69] 13.92% 12.02% Obs. 9,648 8,759 Mean (Insider ) 0.000 0.097 Table 8: Investment‐to‐price sensitivities: Peers’ stock price informativeness (πB) and Managerial Information (γ) This table presents the results from estimations of equation (16) across different sample partitions based on a proxy for managerial information and three proxies for price informativeness of peers. The dependent variable is investment, defined as capital expenditures divided by PPE. Qi is the market‐to‐book ratio of firm i. Q‐i is the average market‐to‐book of the peers of firm i, where peers as defined using the TNIC industries developed by Hoberg and Phillips (2010). The subscript –i refers to (the average across) firm i’s peers. The other explanatory variables (not reported) are defined in the Table 1. The sample is first partitioned into low and high groups based on the bottom and top quintiles of the (de‐correlated) peers’ average price informativeness in year t‐1 (Ψ‐i+, LRMV‐
+
+
i , and APIN‐i as defined in section 4.3.1). Then, we further partition the (low and high) groups in two sub‐groups based on whether the value of Insider is below or above the median. Insider corresponds to the number of insider transactions divided by the total number of transactions. The sample period is from 1996 to 2008. The standard errors used to compute the t‐statistics (in squared brackets) are adjusted for heteroskedasticity and within‐firm clustering. We report the economic effects (the coefficient estimate multiplied by the standard deviation of the explanatory variable in the relevant partition) in italic. We report p‐values corresponding to (unilateral) tests for difference in coefficients across low and high partitions. All specifications include firm and year fixed effects. Symbols ** and * indicate statistical significance at the 1% and 5% levels, respectively. Investment‐to‐price sensitivities as a function of Ψ‐i+ and Insider (γ) Panel A: Low Ψ‐i+ Q‐i (η) High Ψ‐i+ Low Insider High Insider p‐value Low Insider High Insider p‐value 0.039** 0.015 0.353 0.092** 0.010 0.000** [3.22] [2.36] [5.43] [0.40] 4.13% 1.48% 11.13% 1.16% Qi (β) 0.073** 0.062** 0.253 0.051** 0.064** 0.317 [10.62] [7.85] [8.29] 11.38% 10.10% 11.27% 14.78% Obs. 3,943 3,947 3,953 3,960 Mean (Ψi) 2.271 2.752 2.615 2.499 Mean (Ψ‐i) 1.220 1.267 2.858 2.840 Mean (Insideri) 0.000 0.051 0.000 0.037 [9.67] Table 9: Investment‐to‐price sensitivities: Correlation between demands () This table presents the results from estimations of equation (16) across different sample partitions based on proxies for the correlation between the demands for firms’ products. The dependent variable is investment, defined as capital expenditures divided by PPE. Qi is the market‐to‐book ratio of firm i. Q‐i is the average market‐to‐
book of the peers of firm i, where peers as defined using the TNIC industries developed by Hoberg and Phillips (2010). The subscript –i refers to (the average across) firm i’s peers. The other explanatory variables (not reported) are defined in Table 1. The low (high) partition corresponds to the bottom (top) quintile of each partitioning variable (in year t‐1). We use two proxies. CF correlation is the correlation between a firm’s cash flows and the average cash flows of its peers (using quarterly information over the past three years). Similarity is the similarity score (based on firms’ description of their products) provided by Hoberg and Phillips (2010). The sample period is from 1996 to 2008. The standard errors used to compute the t‐statistics (in squared brackets) are adjusted for heteroskedasticity and within‐firm clustering. We report the economic effects (the coefficient estimate multiplied by the standard deviation of the explanatory variable in the relevant partition) in italic. We report p‐values corresponding to (unilateral) tests for difference in coefficients across low and high partitions. All specifications include firm and year fixed effects. Symbols ** and * indicate statistical significance at the 1% and 5% levels, respectively. Q‐i (η) Qi (β) Obs. CFCorr.
Similarity
Investment‐to‐price sensitivities as a function product relatedness ()
CFCorrelation
(Low) (High) (p‐value) 0.030** [3.89] 3.33% 0.058** [14.71] 11.48% 7,546 ‐0.451 0.052** [5.72] 5.35% 0.050** [10.95] 8.40% 7,543 0.597 ‐ ‐ 0.065** 0.201 ProductSimilarity
(Low) (High) (p‐value) 0.031** [5.74] 3.56% 0.063** [16.91] 11.40% 8,767 0.050** [4.32] 6.10% 0.047** [12.40] 10.24% 8,720 ‐ ‐ 0.011 0.057 0.132 0.000** Table 10: Going public firms In this table, we use a subsample of 1,342 firms that go public at some point during our sample period (1996‐
2008). We compare the sensitivity of investment to peers’ stock prices for these firms before and after they go public by regressing investment on peers’ stock price, a dummy variable equal to one after a firm goes public and the interaction between this dummy variable and peers’ stock price. Investment is defined as capital expenditures divided by PPE. Peers’ stock price (Q‐i) is the average market‐to‐book of firm i’s peers, where peers as defined using the TNIC industries developed by Hoberg and Phillips (2010). Post is a dummy variable that equals one after firm i IPO, and zero otherwise. The other control variables (including the firm own stock price post IPO, measured by market‐to‐book) are defined in Table 1. In Column 2, we also control for firms sales growth before and after their IPO. The standard errors used to compute the t‐statistics (in squared brackets) are adjusted for heteroskedasticity and within‐firm clustering. All specifications include firm and year fixed effects. Symbols ** and * indicate statistical significance at the 1% and 5% levels, respectively. Investment (capex over lagged PPE) Q‐i IPO Firms (1) (2) 0.026**
0.020**
[4.54]
[3.44]
Post 0.014
0.012
[0.78]
[0.67]
‐0.016**
‐0.011*
[2.93]
[2.00]
Q‐I x Post Sales Growthi
0.039**
[5.29]
Control variables
Yes
Yes
Firm FE Yes
Yes
Year FE Yes
Yes
Obs. 5,878
5,728
Adj. R2 0.65
0.67