Noisy Stock Prices and Corporate Investment

Abstract

Firms significantly reduce their investment in response to nonfundamental drops in the stock price of their product-market peers. We argue that this results stems from managers’ limited ability to filter out the noise in the stock prices when using them as signals about their investment opportunities. Ensuing losses of capital investment and shareholders’ wealth are economically large and even affect firms not facing severe financing constraints or agency problems. Our findings offer a novel perspective on how stock market inefficiencies can affect the real economy, even in the absence of financing or agency frictions.

Received December 14, 2016; editorial decision July 30, 2018 by Editor Itay Goldstein. Authors have furnished an Internet Appendix, which is available on the Oxford University Press Web site next to the link to the final published paper online.

Stock prices respond to fundamental shocks (news) and nonfundamental shocks (noise). In this paper, we provide evidence that nonfundamental shocks to stock prices affect corporate investment because managers have limited ability to separate information from noise when using stock prices as signals about their prospects. Thus, as suggested by Morck, Shleifer, and Vishny (1990), stock prices are “faulty informants” for corporate managers. Our findings have a novel important implication, namely the noise in stock prices matters for real decisions even when managers focus on maximizing long-term firm value and face no financial constraints. This possibility considerably broadens the set of firms that can be affected by stock market inefficiencies.

To empirically isolate the faulty informant channel, we use a new approach and consider the response of a firm’s investment to the noise in its product market peers’ stock prices instead of its own stock price. In this way, we can analyze the effect of the noise in stock prices, controlling for the effect of firms’ own stock price. Doing so is important because existing theories (described below) predict that managers may alter their investment in response to noise in their own stock price (that is, abnormally high or low investors’ valuation given the actual investment opportunities of the firm) for other reasons than the faulty informant role of prices (e.g., financial or agency frictions). However, these theories do not predict that the noise in the stock prices of other firms should affect a firm’s investment, holding its stock price constant. In contrast, the faulty informant channel does—provided that managers use these prices as signals and have limited ability to filter out the noise therein.¹

The specification and the interpretation of our main test is grounded in a standard investment model in which a manager chooses how much to invest in a growth opportunity. The optimal investment increases in her expectation of the payoff of this opportunity, which depends on her internal information, the firm’s own stock price, and the stock prices of its peers (i.e., firms with correlated fundamentals). We show that the noise in stock prices affect the manager’s expectation, and therefore her investment decision, if and only if she cannot perfectly filter it out from stock prices. If, instead, she can, investment is sensitive to stock prices (because they convey relevant information), but not to the noise in these prices.

These observations have an important implication for testing whether peers’ stock prices truly are faulty signals and not just fundamental signals. Namely, estimating the sensitivity of a firm’s investment to its peers’ stock prices (e.g., Foucault and Frésard 2014) is inadequate to test the faulty informant channel because investment can be sensitive to peers’ stock prices even if these are not faulty signals. Instead, one must estimate the sensitivity of a firm investment to the noise in its peers’ stock prices. Indeed, in theory, this sensitivity differs from zero if and only if stock prices provide faulty signals. Our model suggests to estimate the sensitivity of a firm’s investment to the noise in stock prices by projecting its investment on an observable component of this noise (observable ex post by the econometrician, but not ex ante by the manager), and its orthogonal component.

We implement this approach using a panel of U.S. firms over the period 1996–2011. For each firm-year, we identify product-market peers using the Text-Based Network Industry Classification (TNIC) developed by Hoberg and Maksimovic (2015). We decompose the annual stock price (Tobin’s q) of each firm into a nonfundamental component and its orthogonal component.² We measure the nonfundamental component of a firm’s stock price as the predicted value of a regression of this price on hypothetical sales of the firm’s stock by mutual funds experiencing large investors’ redemptions. These sales are hypothetical in the sense that they are derived assuming that, in response to redemptions, mutual funds rebalance their portfolios to keep the distribution of their holdings constant. Like Edmans, Goldstein, and Jiang (2012), we find that these fire sales are associated with large negative price pressures that eventually revert, consistent with the view that these sales represent nonfundamental demand shocks.

As uniquely predicted by the faulty informant channel, we show that a firm’s investment is sensitive to the noise component of its peers’ stock prices, after controlling for its own stock price. The average firm in our sample cuts its investment in fixed capital by 1.5% (a 4.3% decrease relative to the average level of investment) following a 1-standard-deviation decrease in the nonfundamental component of its peers’ stock prices. We quantify the resultant loss in aggregate investment to be about

29 billion per year. According to the faulty informant hypothesis, this loss generates an opportunity cost for shareholders that corresponds to forgone investments when managers become “excessively” pessimistic about their growth opportunities after observing a drop in their peers’ valuations. However, in truth, this drop contains no fundamental information. Using various approaches, we estimate shareholders’ losses to be in the range of

$\$$

0.9 to

$\$$

3.7 billion per year.

Furthermore, cross-sectional variations in the sensitivity of a firm investment to the noise in its peers’ stock price are consistent with the faulty informant channel. In particular, a firm’s investment is more sensitive to the noise in its peers’ stock prices when managers are (1) less likely to filter out this noise from prices (e.g., when the firm’s ownership by mutual funds does not overlap with that of its peers) and (2) more likely to rely on peers’ stock prices as an additional source of information (e.g., when the firm’s fundamental is more correlated with its peers’ fundamentals, or when managers have less precise internal information).

The correlation between a firm’s investment and the noise in its peers’ stock prices might be spurious, stemming from omitted variables in our baseline specification. We address this concern in several ways. First, we estimate our baseline specification at the division level for multi-division firms. This approach is particularly powerful because it enables us to control (with firm-year fixed effects) for any time-varying unobserved firm-specific characteristics that could be correlated with both the noise in a firm’s peers’ stock price and its overall investment (e.g., the firm’s financing constraints, its CEO’s incentives, or the noise in its own stock price). Confirming our baseline results conglomerates reduce the capital allocated to one division relative to others when that division’s peers experience nonfundamental drops in their stock price.

In addition, we account for possible omitted variables by directly assessing three alternative explanations for our main finding. First, we consider a “financing channel”, that is, the possibility that the noise in peers’ stock prices influences a firm’s investment by affecting its financing conditions, rather than its managers’ beliefs. However, we find that the noise in peers’ stock prices is not significantly related to various proxies for the cost and availability of external financing. Moreover, our baseline results continue to hold when controlling for these proxies in our regressions. Thus, the financing channel cannot explain our results.

We also consider the possibility that nonfundamental shocks to peers’ stock prices affect managers’ personal incentives to invest (the “pressure channel”), without necessarily changing their beliefs about their growth opportunities. For example, a drop in peers’ stock prices might (1) alter the intensity of product-market competition, (2) trigger industry consolidation via acquisitions (Edmans, Goldstein, and Jiang 2012), (3) affect managerial compensation (through relative performance evaluation), or (4) change CEO turnover (e.g., through the threat of takeovers). However, we generally find no effect of nonfundamental shocks to peers’ stock prices on proxies for the pressure channel, and when an effect is present (like in the case for industry consolidation), controlling for these proxies does not change our main result.

Finally, we consider the possibility that managers react to changes in their peers’ investment rather than to changes in their stock prices (the “investment-mimicking channel”). Indeed, a nonfundamental shock to peers’ stock prices could lead a firm’s peers to cut their investment. Strategic considerations could then lead the firm’s manager to invest less if investments decisions are complements, even if the manager’s perceptions of her firm’s fundamentals are not influenced by the noise in her peers’ stock prices. However, a firm’s investment remains sensitive to the noise in its peers’ stock price when restricting attention to firms whose peers do not change investment in response to nonfundamental shocks to their stock price.

Our findings contribute to the literature on the real effects of nonfundamental shocks to stock prices (see Baker and Wurgler 2012 for a survey) by highlighting a new friction through which these shocks can distort firms’ investment. Existing research has proposed two channels through which nonfundamental shocks to a firm’s own stock price can have real effects. First, investors might have incorrect beliefs regarding a firm’s growth opportunities, pushing its market valuation away from fundamentals. This distortion might then lead to inefficient investment decisions (e.g., overinvestment when market valuations are excessively high) if managers maximize their stock price (i.e., follow investors’ distorted beliefs) rather than the actual value of the firm given its true prospects (see, for instance, Stein 1996 for theory and Polk and Sapienza 2009 for evidence). Second, when their stock price deviates from fundamentals, managers of financially constrained firms might opportunistically issue new shares and undertake investments that could not be funded otherwise (see, for instance, Baker, Stein, and Wurgler 2003). Importantly, these channels do not predict that managers should react to the noise in peers’ stock prices (after controlling for their own stock price) since these do not matter for financing (Baker, Stein, and Wurgler 2003) and are not relevant for their shareholders (Stein 1996). In contrast, the faulty informant channel does (see our model). Thus, the noise in stock prices influences real decisions even when managers maximize long-term firm value and face no financing constraints. In this sense, our findings imply that the universe of firms for which stock market inefficiencies matter is much broader than what the literature suggests.

Ozoguz and Rebello (2013), Foucault and Frésard (2014), and Yan (2015) show empirically that firms’ investment is sensitive to peers’ stock prices and find that cross-sectional variations in this sensitivity is consistent with managers learning information from stock prices (as other papers have found by considering the sensitivity of a firm investment to its own stock price; see Bond, Edmans, and Goldstein 2012 for a survey). However, as we show theoretically, this finding does not imply that nonfundamental shocks to peers’ stock price have real effects. Indeed, investment should be sensitive to peers’ stock prices even if managers can perfectly filter out the noise from these prices. In other words, the “informant” role of stock prices identified by existing research does not imply that these prices are sometimes faulty in this role. Thus, testing whether stock prices are faulty informant requires a different approach. Our contribution is to develop such an approach and to motivate it based on theory.

Following Edmans, Goldstein, and Jiang (2012), other papers have used mutual funds’ fire sales as an instrument to analyze the effect of nonfundamental shocks to stock prices on corporate policies. For instance, Hau and Lai (2013) show that firms respond to severe drop in their stock price due to mutual fund fire sales during the 2007–2009 crisis by cutting investment. Lou and Wang (2014) show that this finding holds in normal times as well. Closer to our paper, Williams and Xiao (2017) report that suppliers cut research and development (R&D) spending following a nonfundamental decline in their customers’ market value (instrumented by large mutual funds’ outflows). However, these observations are consistent with all possible channels through which noise in stock prices might affect investment, including the faulty informant channel.³ Thus, they do not constitute strong evidence in favor or against this channel.⁴

1. A Test of the Faulty Informant Channel

This section presents the framework that guides the specification and interpretation of our empirical tests. In particular, Proposition 1 provides the investment model that we estimate in our tests.

1.1 The investment model

Thus, the optimal investment is equal to the manager’s expectation of the marginal return on her investment. To form this expectation, the manager of firm |$i$| has access to several sources of information (signals) at date 1. Her set of signals about |$\theta _{i}$| is |${\it{\Omega}} _{1}=\{s_{m_{i}},P_{i},P_{-i},s_{u_{i}},s_{u_{-i}}\}$| and, conditional on |$\theta _{i}$|⁠, all these signals are independent. We describe each in turn. First, the manager possesses internal (private) information about the fundamentals of her growth opportunity. We denote this signal by |$s_{m_{i}}= \widetilde{\theta }_{i}+\chi _{i}$| where |$\chi _{i}$| is an error term. We assume that the manager has imperfect internal information (⁠|$\sigma _{\chi _{i}}^{2}>0$|⁠), because, otherwise, she would not rely on other sources of Information, such as stock prices.

Second, the manager can obtain external information from the firm’s own stock price, |$P_{i}$|⁠, and its peers’ stock prices, |$P_{-i}$| (index |$-i$| refers to the product-market peers of firm |$i$|⁠). Peers are firms whose fundamentals are correlated with |$\widetilde{\theta }_{i}$|⁠. We assume that |$P_{i}=$||$\widetilde{\theta }_{i}+u_{i}$|⁠, where |$u_{i}$| is the noise in firm |$i$|’s stock price. Similarly, |$P_{-i}=\widetilde{\theta }_{-i}+u_{-i}$| where |$\widetilde{\theta }_{-i}$| is peers’ fundamentals and |$u_{-i}$| is the noise (from the viewpoint of firm |$i$|’s manager) in peers’ stock price.⁵

We define |$\widetilde{\theta }_{-i}=\rho _{i}\widetilde{\theta }_{i}$|⁠, where |$\rho _{i}$| is a constant, which is equal to either |$1$| or |$-1$|⁠. The sign of |$\rho _{i}$| determines whether the fundamentals of firm |$i$| and its peers are positively (⁠|$\rho _{i}=1$|⁠) or negatively correlated (⁠|$\rho _{i}=-1$|⁠). In reality, this sign can be negative or positive. Indeed, in our sample, a firm and its peers are related horizontally, which means that their products are probably substitutes to some extent. Thus, holding total demand for these products constant, an increase in sales for one firm (e.g., because of a lower product price) tends to lower sales for others. However, variations in total demand for their products (common demand shocks) tend to affect their sales in the same way. The first effect is a source of negative correlation in firms’ sales while the second effect is a source of positive correlation. In our data (see Section 5), the second effect dominates for most firms since the correlation between peers’ sales is positive on average and negative for less than |$10\%$| of firms. In any case, our test of the faulty informant channel requires only the correlation between the fundamentals of a firm and its peers to be different from zero (see also Footnote 7).

Last, the manager receives a signal |$s_{u_{i}}=u_{i}+\eta _{i}$| about |$u_{i}$| and a signal |$s_{u_{-i}}=u_{-i}+\eta _{-i}$| about |$u_{-i}$|⁠, where |$\eta _{i}$| and |$\eta _{-i}$| are the errors in these signals. The variances of these errors, |$\sigma _{\eta _{i}}^{2}$| and |$\sigma _{\eta _{-i}}^{2}$|⁠, determine the extent to which the manager can filter out the noise in her own stock price and her peers’ stock price. For instance, a lower value of |$\sigma _{\eta _{i}}^{2}$| means that the manager is able to interpret more accurately the source of variations in her own stock price.⁶

The faulty informant hypothesis posits that managers use information from stock prices but cannot perfectly filter out noise in stock prices, that is, that |$\sigma _{\eta _{i}}^{2}>0$| and |$\sigma _{\eta _{-i}}^{2}>0$|⁠. Our main purpose is to develop a test of this hypothesis. To this end, we take a “reduced form” approach in the sense that we directly postulate that the signals conveyed by stock prices about a firm’s fundamentals can be decomposed into (1) a component that is informative about the firm’s fundamentals and (2) a component that is not. That is, we directly assume that |$P_{i}=$||$\widetilde{\theta }_{i}+u_{i}$| and |$P_{-i}=\widetilde{\theta }_{-i}+u_{-i}$| instead of deriving these expressions from investors’ equilibrium actions and market clearing conditions. We further motivate this approach and its potential limits in Section II.C.

 

Lemma 1 describes how the manager’s signals at date 1 affect her investment decision. As the manager’s private information about the fundamentals of her growth opportunity, |$\widetilde{\theta }_{i}$|⁠, is imperfect, she can improve her estimate of these fundamentals by using information from stock prices. For instance, suppose that peers’ stock prices are informative (⁠|$\sigma _{u_{-i}}<\infty $|⁠). Then, in this case, |$b_{-i}\neq 0$|⁠. Indeed, a change in peers’ stock prices affect the manager’s beliefs about the fundamentals of her growth opportunity and therefore her investment. The sign of this influence (⁠|$b_{-i}$|⁠) is the same as the sign of the correlation between the fundamentals of firm |$i$| and the fundamentals of its peers, that is, |$\rho _{i}$| (see the proof of Lemma 1, ). Indeed, an increase in peers’ stock price is a positive (resp., negative) signal about firm |$i$|’s fundamentals if peers’ fundamentals covary positively (resp., negatively) with firm |$i$|’s fundamentals. In contrast, the manager’s response to her own stock price is always positive (⁠|$b_{i}\geq 0$|⁠).

Moreover, |$c_{-i}\neq 0$|⁠, if the manager possesses information about the noise in peers’ stock prices (⁠|$\sigma _{\eta _{i}}^{2}<\infty $|⁠). Thus, the manager’s signal about the noise in her firm’s stock price, |$s_{u_{i}}$|⁠, affects her investment decision as well. By itself, this signal is uninformative about her fundamentals, |$\widetilde{\theta }_{i}$|⁠. However, it helps the manager in filtering out the noise contained in peers’ stock price, thereby improving the precision of her estimate of the marginal return on her investment. The sign of coefficient |$c_{-i}$| is always opposite to that of |$b_{-i}$| (see the proof of Lemma 1) because the investor uses her signal about the noise in peers’ stock prices to filter it out from these prices.

2. Testing that Stock Prices Are Faulty Signals

We cannot directly estimate Equation (5) to obtain estimates of the sensitivities of investment to noise (⁠|$\alpha _{i}$| and |$\alpha _{-i}$|⁠) because we do not perfectly observe the nonfundamental and fundamental components of firms’ stock prices. However, we can circumvent this problem insofar as we can measure part of the nonfundamental component of stock prices. To see why, let |$u_{-i}=u_{-i}^{o}+u_{-i}^{no}$|⁠, where |$u_{-i}^{o}$| is the component of the noise in peers’ stock price that can be measured by the econometrician. We assume that |$u_{-i}^{o}$| and |$u_{-i}^{no}$| are independent and normally distributed with means zero and variances |$\lambda _{-i}\sigma _{u_{-i}}^{2}$| and |$(1-\lambda _{-i})\sigma _{u_{-i}}^{2}$|⁠, respectively (⁠|$\lambda _{-i}\in \lbrack 0,1]$|⁠). We decompose the noise in firm |$i$|’s stock price in the same way (⁠|$u_{i}=u_{i}^{o}+u_{i}^{no}$|⁠). Also let |$P_{-i}^{\ast }=\widetilde{\theta }_{i}+u_{-i}^{no}=P_{-i}-$|E|$(P_{-i}\left\vert u_{-i}^{o}\right) $| where the second equality follows from the definition of |$P_{-i}$|⁠. Thus, |$P_{-i}^{\ast }$| is the residual of a regression of |$P_{-i}$| on |$u_{-i}^{o}$|⁠. Similarly, we define |$P_{i}^{\ast }= \widetilde{\theta }_{i}+u_{i}^{no}=P_{i}-$|E|$(P_{i}\left\vert u_{i}^{o}\right) $|⁠. the following implication.

 

Equation (6) is obtained by projecting the optimal investment of firm |$i$| (given in Equation (5) on a set of explanatory variables (⁠|$P_{i}^{\ast }$|⁠, |$u_{i}^{o}$|⁠, |$P_{-i}^{\ast }$|⁠, and |$u_{-i}^{o}$|⁠) that can be measured empirically. The term |$\epsilon _{i}$| is the residual variation in investment that cannot be explained by these variables. In the model, it is orthogonal to the explanatory variables in Equation (6). Thus, in principle, one can obtain unbiased estimates of the true influence of the noise in stock prices on investment, that is, |$\alpha _{i}$| and |$\alpha _{-i}$|⁠, by estimating Equation (6) with ordinary least squares (OLS) regressions. This approach forms the backbone of our empirical tests.

The sign of the sensitivity of a firm investment to its own stock price is always positive (⁠|$\alpha _{i}\geq 0$|⁠). In contrast, the sign of the sensitivity of a firm investment to its peers’ stock price (⁠|$\alpha _{-i}$|⁠) is the same as the sign of the correlation (⁠|$\rho _{i}$|⁠) between the fundamentals of a firm and its peers.⁷ Our key prediction is that this sensitivity should be different from zero (⁠|$\alpha _{-i}\neq 0$|⁠) if managers cannot fully distinguish whether variations in these prices are due to changes in fundamentals or noise (⁠|$\sigma _{\eta _{i}}^{2}>0$| and |$\sigma _{\eta _{-i}}^{2}>0$|⁠). Our tests focus on this prediction.

3. Remarks

3.1 Investment-to-noise sensitivity versus investment-to-price Sensitivity

Growing evidence indicates that managers learn information from their own stock prices and their peers’ stock prices (see Bond, Edmans, and Goldstein 2012 for a survey; Ozoguz and Rebello 2013; Foucault and Frésard 2014). Inferences in this literature are based on estimations of the sensitivity of investment to stock prices and cross-sectional patterns in this sensitivity.

3.2 Alternative channels

In our model, the noise in stock prices influences managers’ beliefs about their growth opportunities and thereby their investment decision. As explained in the introduction, there might be other channels (financing or agency frictions), absent from our model, through which the noise in a firm’s own stock price influences its investment. In our estimation of Equation (6), the effects of these alternative channels on a firm’s investment will be picked by |$u_{i}^{o}$|⁠. Thus, if they operate, the OLS estimate of |$\alpha _{i}$| should overestimate the effect of the noise in a firm’s own stock price on its investment due to the faulty informant channel. In contrast, the alternative channels considered so far in the literature do not not predict that the investment of a firm should be sensitive to the noise in its peers’ stock price, after controlling for the firm’s own stock price. For instance, what matters for a firm’s financing is the actual price at which it can issue new shares, not the price of its peers’ shares. Thus, coefficient |$\alpha _{-i}$| provides a cleaner estimate of the extent to which managers’ beliefs are influenced by the noise in stock prices. For these reasons, our test of the faulty informant channel focuses on this coefficient and whether it is significantly different from zero as uniquely predicted by the faulty informant channel.

3.3 Who observes what?

Variables |$u_{i}^{o}$| and |$u_{-i}^{o}$| in Equation (6) are nonfundamental shocks to stock prices that econometricians can measure ex post (possibly, a long time after they happened). Importantly, however, these shocks are not observed by managers when they happen. Indeed, managers’ beliefs would not be influenced by the effect of these shocks on stock prices if they could observe them perfectly. Hence, one could not use them to assess whether managers’ beliefs are influenced by the noise in stock prices.

For simplicity, we have assumed that managers have a signal about the aggregate nonfundamental shock to their stock price (e.g., |$u_{-i}=u_{-i}^{no}+u_{-i}^{o}$|⁠) rather than two separate signals (e.g., one about |$u_{-i}^{no}$| and one about |$u_{-i}^{o}$|⁠). For this reason, the sensitivity of their investment to, say, |$u_{-i}^{o}$| is the same as the sensitivity of their investment to |$u_{-i}^{no}$| (equal to |$\alpha _{-i}$|⁠) and does not depend on the size (⁠|$\lambda _{-i}$|⁠) of |$u_{-i}^{o}$| relative to |$u_{-i}^{no} $| (see the proof of Proposition 1; for the same reason, |$\alpha _{i}$| does not depend on |$\lambda _{i}$|⁠). This assumption can be relaxed by allowing managers to have different (imperfect) signals about, say, |$u_{-i}^{o}$| and |$u_{-i}^{no}$|⁠. In this case, the sensitivity of a firm investment to |$u_{-i}^{o}$| is different than its sensitivity to |$u_{-i}^{no}$|⁠. Yet, the faulty informant channel still implies that the sensitivity of a firm investment to |$u_{-i}^{o}$| should be different from zero. This is what we test empirically.

3.4 Ex post inefficiencies

In our model, the ex ante expected value of the growth opportunity (i.e., E|$(G(K_{i},\widetilde{\theta }_{i})$|⁠) is higher when the manager uses all available sources of information than when she only uses her internal signal about |$\widetilde{\theta }_{i}$|⁠. Thus, relying on signals from stock prices is ex ante efficient even when stock prices can be faulty signals. However, this can lead to decisions that sometimes are inefficient ex post. For instance, consider a drop in peers’ stock prices due to a negative nonfundamental shock. On average, this drop leads the manager to underinvest relative to what would be optimal. With the benefit of hindsight, this decision appears inefficient, because it results in an opportunity cost for shareholders (a missed positive net present value opportunity). In Section IV.B, we quantify the cost of this inefficiency for the firms in our sample.

3.5 Reduced form for stock prices

For brevity, we have directly assumed that the signals conveyed by stock prices about a firm’s fundamental can be decomposed into an informative and an uninformative component about this fundamental. As stock prices are endogenous, our approach naturally raises the question of whether our central prediction (⁠|$\alpha _{-i}\neq 0$| if |$\sigma _{\eta _{i}}^{2}>0$| and |$\sigma _{\eta _{-i}}^{2}>0$|⁠) would hold in an equilibrium model in which all firms learn from their own and peers’ stock prices.

Recent papers (Foucault and Frésard 2014; Huang and Zeng 2015; Schneemeier 2017) have considered models of this type. In these models (like in models with a single firm, e.g., Subrahmanyam and Titman 1999 or Goldstein, Oznedoren, and Yuang 2013), stock prices depend on investors’ expectations regarding managers’ investment decisions, which in turn depend on stock prices (because stock prices contain information new to managers). Thus, stock prices both reflect investment decisions and influence (“feedback on” ) these decisions. Given these two-way interactions, solving for equilibrium prices and investment decisions is technically complex, even more so in a multiple firms environment. However, in all these models, the signals conveyed by stock prices about a firm’s fundamentals can be written as the sum of a fundamental and a nonfundamental component (as an example, in the Internet Appendix (Section B.1), we provide this decomposition in Schneemeier’s (2017) equilibrium model).⁸ In this paper, we only focus on implications of the faulty informant channel that derive from this fundamental/nonfundamental decomposition of stock prices and the fact that managers are imperfectly informed about each component. Thus, we consider implications that are robust to many possible specifications of equilibrium models in which managers learn from stock prices.

One limitation of this reduced-form approach is that we cannot test more subtle implications of the faulty informant channel, that is, those that depend on equilibrium considerations and the exact specification of the model (e.g., whether or not managers can observe all prices, the exact structure of correlation in firms’ payoffs). For instance, in equilibrium models, the noise-to-signal ratios in stock prices (e.g., |$\sigma _{u_{i}}^{2}/(\sigma _{\theta _{i}}^{2}+\sigma _{u_{i}}^{2})$|⁠) are endogenous and depend on deep parameters (e.g., the mass of informed investors and the precision of their signals). This feature plays no role in deriving our test of the faulty informant hypothesis (see Proposition 1 above) but it could be used to derive additional implications. Another example is offered by Schneemeier (2017). In his model, a manager uses the stock prices of its peers of peers to form its beliefs because these prices help her in filtering the noise in her own stock price (i.e., the price of its peers of peers plays the role of |$s_{u_{-i}}$| in our model). Schneemeier (2017) shows that this feature may or may not lead to situations in which a firm investment depends on the noise of unrelated firms depending on whether managers observe all prices or just a subset of prices. Such predictions are interesting but more model-specific. Testing them is therefore beyond the scope of our paper.

4. Data and Methodology

This section explains how we estimate Equation (6) and test whether stock prices have a faulty informant role (i.e., whether |$\alpha_{-i} \neq 0$|⁠).

4.1 Identifying product-market peers

We identify the product-market peers of each publicly listed firm using the Text-Based Network Industry Classification (TNIC) developed by Hoberg and Phillips (2015). This classification starts in 1996 and relies on textual analysis of the product description sections of firms’ 10-K filings (Item 1 or Item 1A). Every year, Hoberg and Phillips (2016) compute a measure of product similarity for every pair of U.S. public firms based on the number of common words in their product description. This measure ranges between 0% and 100%. Intuitively, the more common words two firms use in describing their products, the more similar they are. Hoberg and Phillips (2015) define each firm |$i$|’s product-market peers to include all firms |$j$| with pairwise similarities relative to |$i$| above a pre-specified minimum similarity threshold—chosen to generate industries with the same fraction of industry pairs as three-digit SIC industries.⁹

Our sample comprises all firms present in TNIC from 1996 to 2011. For each firm in the sample, we define its set of “peers” in a given year as all firms that belong to its TNIC network in that year. As firms’ relations in TNIC are not transitive (firm B can be a peer of firm A without all peers of B being peers of A), each firm has a unique specific set of peers that varies over time. We obtain stock price and return information from the Center for Research in Securities Prices (CRSP). Investment and other accounting data are from Compustat.

We exclude firms in financial (SIC codes 6000-6999) and utility industries (SIC codes 4000-4999). We also exclude firm-year observations with negative sales, sales less than

5 million, or missing information on total assets, capital expenditures, fixed assets (property, plant, and equipment), and (end-of-year) stock prices. Appendix B describes the construction of all variables. We winsorize all ratios at 1% in each tail to reduce the effect of outliers.

4.2 Identifying Nonfundamental shocks to stock prices

We identify nonfundamental variation in stock prices (the empirical analog of |$u_{i}^{o}$| in our model) using stocks’ sales by mutual funds experiencing large outflows (“forced sales). Coval and Stafford (2007) show that these sales represent large negative demand shocks for liquidated stocks that create long-lasting downward price pressure. These sales, which are due to investors’ redemptions, are unlikely to reflect fund managers’ private information about fundamentals, unless fund managers discretely choose the stocks they sell. To mitigate this concern, we follow Edmans, Goldstein, and Jiang (2012) and use hypothetical, rather than actual, sales of mutual funds hit by large outflows as an instrument for nonfundamental variation in stock prices.¹⁰

Specifically, we calculate the hypothetical net selling of a stock (in dollar) by all nonspecialized mutual funds subject to extreme outflows (i.e., greater than 5% of their assets) assuming that in response to these outflows, mutual funds proportionally liquidate their existing holdings and keep the same composition of their portfolios. We define a variable |$MFHS_{i,t}$| equal to this amount of sales for firm |$i$| in year |$t$|⁠, scaled by the total volume of trading, which we label “Mutual Funds Hypothetical Sales.” |$MFHS_{i,t}$| only takes negative values, so that smaller values of |$MFHS_{i,t}$| indicate larger hypothetical sales, and is equal to 0 absent forced sales. By construction, |$MFHS_{i,t}$|⁠, the details of which are exposed in Appendix C, varies with large mutual funds outflows only. It does not depend on choices made by fund managers about which stocks to sell to meet these redemptions, and is unlikely to reflect changes in investors’ views about the firm’s industry because of the exclusion of specialized mutual funds. As such, |$MFHS_{i,t}$| is a plausible proxy for nonfundamental shocks to stock prices.

In support of this claim, Figure 1 displays the relationship between mutual fund hypothetical sales and stock prices in our sample using quarterly data. We define an “event” for stock |$i$| in quarter |$q$| of year |$t$| when |$MFHS_{i,q,t}$| is below the tenth percentile of the sample distribution of |$MFHS_{i,q,t}$|⁠. We then estimate a regression of the quarterly abnormal returns of stocks affected by these events on event-time dummy variables, and plot the cumulated coefficients (i.e., the cumulated average abnormal return, CAAR) around the event. In panel A, we define abnormal returns as stock returns minus the return of the CRSP index. Like Edmans, Goldstein, and Jiang (2012), we find no abnormal decline in stock prices before the event quarter, which mitigates the concern that affected funds own stocks with deteriorating fundamentals. Immediately after the event, stock prices drop by about 10%, then revert in the subsequent quarters and recover after about 2 years. This price reversal supports the use of mutual funds hypothetical sales as a measure of nonfundamental price variation. Indeed, if these shocks were fundamental, the resultant decrease in stock prices would be permanent. In panel B, we define abnormal returns as stock returns in excess of the return of an equally weighted portfolio of product-market peers (based on TNIC). The pattern of stock prices is similar to that observed in panel A, suggesting that |$MFHS_{i,t}$| captures localized (i.e., firm-specific) nonfundamental variation in stock prices.

We perform additional tests that further support using |$MFHS$| as a measure of nonfundamental variation. First, we show that |$MFHS$| is unlikely to capture economy-wide or industry-specific characteristics. Figure 2 displays the average value of |$MFHS$| across firms for each year in our sample and across the Hoberg and Phillips (2016) fixed industry classification. We observe no obvious clustering in any particular time period or industry.¹¹ Second, we find that corporate insiders trade against the price pressure generated by mutual fund fire sales (Figure 3). Specifically, the average quarterly net insider purchases (defined either as insiders’ purchases minus sales, divided by their stock’s turnover, or as the net number of shares purchased) significantly increases in response to downward price pressures triggered by mutual fund sales. This result, which is consistent with that of Ali, Wei, and Zhou (2011) and Khan, Kogan, and Serafeim (2012), who document that, to some extent, managers detect nonfundamental shocks to their own stock price, supports our claim that these shocks are unrelated to firms’ fundamentals. If they were, insiders would not trade against them.

4.3 Decomposing stock prices

We define the estimated residuals from regression (8), |$\overline{Q}_{-i,t}^{\ast }=\widehat{\upsilon }_{-i,t}$|⁠, as a proxy for |$P_{-i}^{\ast }$| in the model.¹³ In all our tests, we refer to |$\overline{MFHS} _{-i,t}$| as the nonfundamental component of peers’ stock price and to |$\overline{Q}_{-i,t}^{\ast }$| as the “fundamental” component of peers’ stock price (even though, like in the theory, |$\overline{Q}_{-i,t}^{\ast }$| is not necessarily completely purged from noise). Proceeding in the same way, we decompose the stock price of each firm |$i$| in each year |$t$| (proxied by |$Q_{i,t}$|⁠, its Tobin’s q in year |$t$|⁠) into a nonfundamental component (⁠|$MFHS_{i,t} $|⁠) and a fundamental component (⁠|$Q_{i,t}^{\ast }$|⁠).

4.4 Econometric specification

Arguably, price pressures induced by mutual fund hypothetical sales might be correlated within industries if funds experiencing extreme outflows have correlated industry allocations.¹⁵ This is not a concern in our setting because we include |$MFHS_{i,t}$| and |$\overline{MFHS}_{-i,t}$| in the regression. Thus, |$\alpha_{0}$| captures the effect of the nonfundamental component of firm |$i$|’s peers’ stock price that is not captured by the nonfundamental component of firm |$i$|’s stock price. Likewise, |$\gamma_{0}$| captures the effect of the information contained in |$\overline{Q}_{-i}^{\ast }$| that is not in |$Q_{i}^{\ast }$|⁠. Table 1 presents the summary statistics for the main employed variables. They are in line with previous research.

The coefficient of interest in Equation (9) is |$\alpha_{0}$|⁠, which is the empirical equivalent of |$\alpha_{-i}$| in the model (i.e., the average value of |$\alpha_{-i}$| across all firms |$i$| in our sample). If managers correctly identify |$\overline{MFHS}_{-i,t}$| as being a nonfundamental shock to their peers’ stock price when they make their investment decision in year |$t$|⁠, then this decision should be unrelated to |$\overline{MFHS}_{-i,t}$| (i.e., the estimated coefficient |$\alpha_{0}$| should equal 0). In contrast, if they cannot do so, their investment should be sensitive to |$\overline{MFHS}_{-i,t}$| (i.e., |$\alpha_{0}$| should be different from zero).

Our test posits that managers do not observe fire sales of their peer stocks by mutual funds, at least when these sales happen (see the discussion in Section II.C.3). We think that this assumption is plausible. Indeed, data on mutual fund holdings are made available only at the end of every quarter. Thus, market participants (including firms’ managers) can, at the earliest, relate stock price movements in a given quarter to mutual funds flows at the end of the quarter only. Moreover, this assumes that managers make the effort of decoding the price movements of their peers by considering mutual fund outflows as a possible cause of these movements. This is unlikely. Indeed, such outflows are only one out of many possible causes for price movements and considering all these causes is costly for managers (it requires time and attention). Finally, we note that the growth of mutual funds is a recent phenomenon and it is likely that, at least in the early part of our sample, corporate managers had insufficient knowledge for considering mutual funds outflows as a possible cause of nonfundamental variations in stock prices.¹⁶ In any case, if corporate managers are fully able to identify the noisy origin of the drops in their peers’ stock price, investment should not respond to such noise; that is, this biases our approach toward finding no effect.

5. The Faulty Informant Role of Stock Prices

5.1 Baseline results

To facilitate interpretation, in estimating Equation (9), we scale all independent variables by their sample standard deviation prior to estimation, such that coefficients represent the estimated change of investment in response to a 1-standard-deviation change in each independent variable. We report the estimates of our baseline specification (9) in the first column of Table 2. Consistent with the faulty informant hypothesis, the coefficient on |$\overline{MFHS}_{-i}$| (⁠|$\hat{\alpha_{0}}$|⁠) is statistically different from zero. The point estimate for |$\alpha_{0}$| is 0.015 with a |$t$|-statistic of 5.50. This means that a 1-standard-deviation decrease in the nonfundamental component of peers’ stock price is associated with a 1.5-percentage-point decrease in firms’ investment (⁠|$4.3\%$| of the average investment level in the sample). The coefficient on the fundamental component of peers’ stock price |$\overline{Q}_{-i}^{\ast }$| (⁠|$\hat{\gamma_{0}}$|⁠) is also significantly positive with a point estimate of |$0.024$|⁠.

The theory predicts that the estimates of |$\alpha_{0}$| and |$\gamma_{0}$| should be positive when the correlation between the fundamentals of firm |$i$| and its peers is positive (see Proposition 1). This is, in fact, the case in our data. For instance, the Correlation between sales of a firm and its peers is 0.6, significant at the 1% level.¹⁷ In addition, Proposition 1 predicts that the coefficient on |$\overline{Q}_{-i}^{\ast }$| should be higher than that on |$\overline{MFHS}_{-i}$| (⁠|$\left\vert \gamma_{0}\right\vert >\left\vert \alpha_{0}\right\vert $|⁠). This is also the case. Indeed, as all explanatory variables are standardized by their standard deviation, we can directly compare the estimate of the coefficient on |$\overline{Q}_{-i}^{\ast }$| to that on |$\overline{MFHS}_{-i}$|⁠. We find that it is about twice bigger, meaning that firm’s investment is roughly two times more sensitive to the fundamental component of its peers’ stock price than to the nonfundamental component.

On average, a firm’s investment is also significantly and positively related to the nonfundamental component of its own stock price (consistent with empirical findings in Hau and Lai 2013 and Lou and Wang 2014). Specifically, a 1-standard-deviation decrease in |$MFHS_{i}$| is associated with a 1.1-percentage-point drop in investment. In line with previous research, we find a firm’s investment is also highly sensitive to the fundamental component of its own stock price. The coefficient on |$Q_{i}^{\ast }$| is equal to |$0.08$| with a |$t$|-statistic of |$11.78$|⁠.¹⁸

Columns 2 to 5 in Table 2 show that our findings are robust to the methodology used to construct the explanatory variables pertaining to firm |$i$|’s peers (e.g., |$\overline{Q}_{-i}$| and |$\overline{MFHS}_{-i}$|⁠). In Column 2, we obtain these variables as weighted averages of peer-level variables, where the weights are based on the similarity score computed in Hoberg and Phillips (2016) (instead of equal weights like in Column 1). In Column 3, we take the median values of these variables across peers and in Column 4 the equally weighted averages of these variables across the five closest peers of firm |$i$|⁠. Finally, in Column 5, all accounting variables pertaining to firm |$i$|’s peers are simply summed across peers and explanatory variables are built using these sums. Estimates of investment-to-noise sensitivities are similar across these specifications, indicating that our results are not affected by the way we aggregate variables across peers.¹⁹

In sum, and in line with the faulty informant hypothesis, Table 2 shows that firms decrease investment in response to negative nonfundamental shocks to the stock price of their peers. As these shocks are nonfundamental, the observed drop in investment may be transient. To test whether this is the case, we estimate the baseline specification (9) with lagged values of |$\overline{MFHS}_{-i}$| (namely, |$\overline{MFHS}_{-i,t-2}$| and |$\overline{MFHS}_{-i,t-3}$|⁠). To better assess the dynamics of firm |$i$|’s investment in response to nonfundamental variations in peers’ stock price, we also add the contemporaneous and 1-year-ahead values of |$\overline{MFHS}_{-i}$| to this more general specification. Panel A of Figure 4 plots the point estimates (and their confidence interval) for the coefficients on |$\overline{MFHS}_{-i}$| at dates |$t-3$| to |$t+1$|⁠.

In line with the baseline results, investment in year |$t$| responds positively to noise in peers’ stock price measured in year |$t-1$|⁠. However, it responds negatively to the noise in peers’ stock price measured in year |$t-2$|⁠, which indicates that the real effect of noise on investment is transient. That is, following a negative nonfundamental shock in year |$t-2$|⁠, investment decreases in year |$t-1$|⁠, but then increases in year |$t$|⁠, so that the impact of noise is subsequently fully corrected on average. Three years after the shock, a firm’s investment is no longer sensitive to the noise in peers’ stock price (the coefficient for |$t-3$| is insignificant).

In contrast, panel B of Figure 4 shows that the response of a firm’s investment to lagged values of the fundamental component of peers’ stock price is permanent: the realization of this component in a given year has a positive and significant effect on investment in the following year but no significant effect on investment in subsequent years. The patterns in Figure 4 are overall consistent with learning. Indeed, on average, signals received by managers about their previous investment decisions should in fact confirm (resp., invalidate) those sent by fundamental (resp., nonfundamental) shocks to stock prices at the time of their decision.

5.2 Economic magnitude

The main result of the previous section is that nonfundamental variations in peers’ stock price have a direct effect on corporate investment. This finding is consistent with the idea that managers have limited ability to filter out the noise in stock prices when they use the latter as a signal about their growth opportunities. This result suggests a new channel through which the noise in stock prices has a real effect. In this section, we attempt to quantify the economic impact of this effect.

In our experiment, noise in peers’ stock price induce firms to reduce their investment. According to Equation (9), the resultant annual loss in investment as a fraction of PP&E for a firm is |$\alpha_{0} \times \overline{MFHS}_{-i,t-1}$|⁠, that is, in dollar, |$\alpha_{0} \times \overline{MFHS}_{-i,t-1} \times PP\&E_{i,t-1}$|⁠. We can therefore estimate this loss for each firm by using our baseline estimate for |$\alpha_{0}$| (that is, |$0.015$|⁠; see first column of Table 2) together with the actual values of |$\overline{MFHS}_{-i,t-1}$| and |$PP\&E_{i,t-1}$|⁠. Using this approach, we estimate an aggregate loss of investment (the loss per firm summed over all firms and years in our sample) of

440 billion or

$\$$

29 billion per year, which is about 8% of the average capital expenditures per year in Compustat over the same period.

Under the faulty informant hypothesis, this loss in investment represents an opportunity cost for shareholders. Managers do not invest and postpone valuable projects for shareholders because they mistakenly interpret the drop in their peers’ stock price as a bad signal about their fundamentals. Ideally, we could estimate this opportunity cost for a firm by using the net present value (NPV) of all future cash flows per dollar of lost investment for this firm. Unfortunately, this unit NPV cannot be readily observed at the firm level. However, we can establish a range of plausible values for this variable by using two benchmarks. First, we consider the unit NPV of completed cash-financed acquisitions of private firms (by public firms) structured as assets sales, which we estimate from the stock price reaction for the acquirer at the time of the deal announcement. We focus on this type of acquisitions because they are very similar to standard investment projects in fixed assets.²⁰ Using a comprehensive sample of 2,011 such acquisitions from the SDC, we estimate that over our sample period, every dollar invested in these deals yields an average unit NPV of

0.35, obtained as the product of acquirers’ 5-day cumulative abnormal returns and their market capitalization 3 days prior to deal announcement, divided by deal value. Our second benchmark uses the fact that, by definition, a unit NPV is equal to a firm marginal |$q$| minus 1. A firm’s marginal |$q$| is often estimated by its Tobin’s q, which is |$1.96$|⁠, on average, in our sample. Thus, another estimate of the average unit NPV for firms in our sample is

$\$$

0.96.

With an average unit NPV between

0.35 and

$\$$

0.96, the opportunity cost for shareholders of losing

$\$$

440 billion of investment ranges between

$\$$

440 |$\times$|

$\$$

0.35 =

$\$$

154 billion and

$\$$

440 |$\times$|

$\$$

0.96 =

$\$$

422 billion. If this aggregate investment of

$\$$

440 billion is only postponed by 1 year (as suggested by Figure 4) then the opportunity cost for shareholders is the time cost of postponing all cash flows associated with this aggregate investment by 1 year. This time cost is equal to the net present value of this

$\$$

440 billion investment, minus the net present value of the same investment further discounted by 1 year. For example, with a discount rate of 10% and a unit NPV of

$\$$

0.96, the opportunity cost would be

$\$$

422 –

$\$$

422/1.10 =

$\$$

38 billion.

Table 3 reports various estimates of the opportunity cost of transient cuts in investment due to faulty stock prices for various combinations of unit NPVs (between

0.35 and

$\$$

0.96) and discount rates (between 10% and 15%, which corresponds to the average rate used by managers for capital budgeting as reported in the survey of Jagannathan et al. 2016). Our baseline estimate of |$\alpha_{0}$| in Equation (9) implies a total opportunity cost for shareholders ranging between

$\$$

14 billion (unit NPV of

$\$$

0.35 and 10% discount rate) and

$\$$

55.1 billion (unit NPV of

$\$$

0.96 and 15% discount rate), which is equivalent to

$\$$

0.9 and

$\$$

3.7 billion per year. This cost represents between one and four basis points of the overall annual market capitalization over our sample period.

Overall these calculations suggest that the real effects of the faulty informant channel are sizable. Of course, they must be interpreted carefully since they rely on various assumptions, in particular for the unit net present value of investment for the firms in our sample. However, some of these assumptions are conservative in at least three ways. First, we just focus on the effect of noise in peers’ stock price due to mutual funds fire sales (the effect that we can estimate). Other nonfundamental shocks to these prices should also result in inefficient investment decisions according to the faulty informant hypothesis. Second, we do not compute the loss in investment and shareholder value for a given firm due to nonfundamental shocks to its own stock price. This conservatively assumes that the effects of these shocks are due only to other channels identified in the literature (the financial constraints and agency channels described in the introduction). Last, we assume that cuts in firms’ investment due to nonfundamental shocks are transient for all firms and that firms postpone investment by only one year.

5.3 Heterogeneous exposure to the faulty informant channel

Our baseline specification assumes that the effect of noise in stock prices is the same for all firms. However, the faulty informant hypothesis implies that this effect should vary with firms’ characteristics and be stronger for some firms than for others. In particular, using the closed-form solution of |$\alpha_{-i}$| (Equation (A11) in Appendix A), we predict that firms’ investment should be more sensitive to the noise in peers’ stock price when (1) managers are less able to filter out the noise in their peers’ stock price (i.e., when their signal about the noise in peers’ stock price is noisier; |$\sigma _{\eta _{-i}}^{2}$| is higher), (2) peers’ stock prices are more informative, for instance because they are less noisy (i.e., |$\sigma _{u_{-i}}^{2}$| is low) or firm |$i$|’s fundamentals are more correlated with that of its peers (⁠|$\rho_i$| is large in absolute value), and (3) managers have less precise internal information about their fundamentals (i.e., |$\sigma _{\chi _{i}}^{2}$| is high).²¹ Intuitively, in the two last cases, managers rely more on peers’ stock price as a source of information either because these signals are more informative or because the manager has less-precise information from other signals. We check whether these cross-sectional implications hold in our sample.

We use two variables as proxies for managers’ information about the noise in peers’ stock price. First, we conjecture that managers should more easily identify non-fundamental shocks to their peers’ stock price due to mutual funds’ forced sales when their ownership by mutual funds overlaps more with that of their peers. We define the annual overlap of mutual fund ownership as the pairwise cosine similarity between firms’ ownership structure, and compute the average ownership overlap between each firm and its peers in each year. Second, we posit that managers can better identify the noise in their peers’ stock price when financial analysts indicate that these stocks are mispriced (e.g., Sulaeman and Wei 2014). We measure analysts’ estimate of a firm’s mispricing as the average difference between analysts’ target price for that firm and its current stock price, and average this difference across firms’ peers in each year. We then estimate specifications in which we add interaction terms between each of these proxies and all independent variables of our baseline specification (9), and only report in Table 4 the coefficients on |$\overline{MFHS}_{-i,t-1}$| and its interactions. The first two columns reveal that the coefficient on the interaction between |$\overline{MFHS}_{-i,t-1}$| and each proxy are negative and statistically significant, supporting the prediction that firms’ investment is more sensitive to the noise in peers’ stock price when managers are less likely to filter out this noise.

The rest of Table 4 also confirms that firms’ investment respond significantly more to the noise in peers’ stock price when managers are more likely to use these prices as a source of information about their fundamentals; that is, when (1) their fundamentals are more correlated with that of their peers or (2) their peers’ stock price is more informative. In Columns 3 and 4, we focus on the correlation of firms’ fundamentals with that of their peers, measured in a given year as the average pairwise correlation of sales or assets computed using quarterly data over the last 3 years. In Columns 5 and 6, we consider two proxies for the informativeness of peers’ stock price, measured using the ability of current stock prices to forecast future earnings like in Bai, Philippon, and Savov (2016) and the average earnings forecast error of financial analysts. For all measures, the positive coefficients on the interaction term confirm that firms’ investment is significantly more sensitive to the noise in peers’ stock prices when their fundamentals are more correlated with their peers’ fundamentals and when peers’ stock prices are more informative.

In the last two columns, we show that the sensitivity of firms’ investment to the noise in their peers’ stock price is smaller when the precision of managers’ internal information about their firm’s fundamentals is higher. We first use the profitability of insiders’ trades as a proxy for this precision, positing that managers are more likely to generate trading profits if their internal signals are more precise. We measure the profitability of insiders’ trades as the average 1-month market-adjusted return of holding the same position as insiders for each insider’s transactions. Second, we rely on firms’ competitive position within their industry as another proxy for the quality of managerial information, positing that industry leaders possess more precise internal information. As predicted, we observe negative and significant coefficients on the interaction between |$\overline{MFHS}_{-i,t-1}$| and these two proxies.

Overall, the results in Table 4 confirm that the intensity of the faulty informant role of prices is heterogeneous and stronger when managers rely more on signals conveyed by peers’ stock prices and when it is more difficult for them to identify the noise in these signals.

6. Alternative Channels

Our test of the faulty informant hypothesis assumes that nonfundamental shocks to peers’ stock price are not related to other variables that could directly determine a firm’s investment. One concern is that this assumption is not valid and that, for this reason, our findings just reflect time-varying omitted variables in our baseline specification (9), correlated with both investment and noise in peers’ stock price. We address this concern in this section. We first develop a test that directly controls for the potential effect of omitted variables. Next, we discuss several plausible alternative explanations for our findings regarding the sensitivity of investment to noise in peers’ stock price and show that they are not supported by the data.

6.1 Capital allocation within firms

One way to control for time-varying firm-level variables potentially correlated with both a firm’s investment and the noise in its peers’ stock price is to include firm|$\times $|year fixed effects in our baseline regression. This approach is possible using data on investment at the division level in multi-division firms (conglomerates). In this setting, we can test whether noise in peers’ stock price across divisions of a given firm affects capital allocation within this firm in a given year, holding possible firm-level omitted variables constant.²²

We obtain segment level information on annual capital expenditures, total assets, as well as a four-digit SIC code for each segment from Compustat. Because the TNIC classification of Hoberg and Phillips (2016) does not allow identifying product-market peers at the division level, we define peers using three different standard industry classifications (Fama-French 49 industries, two-digit SIC code, and three-digit NAICS code), which allows us to extend our sample back to 1982 (as opposed to 1996 in the baseline test).²³ Within each firm-year (and each industry classification), we then aggregate capital expenditures and total assets by industry. We refer to the resultant firm-industry-year observations as “divisions”. The “peers” of a given division comprise either all the firms operating in the same industry of that division or all the single-division firms in that industry, depending on the specification. This sample includes 4,291 distinct conglomerate firms, operating 11,765 divisions (based on Fama-French industries).

Table 5 presents the results. In Columns 1 to 3, we report estimates of Equation (11) for each of the three industry definitions when the peers of a given division are all firms operating in the same industry (whether single-industry firms or not) while in Columns 4 to 6, we define the peers of a given division as single industry firms operating in firm i’s industry. The coefficient on |$\overline{MFHS}_{-i,d}$| is positive, statistically significant, and almost identical across all specifications. Thus, as predicted by the faulty informant hypothesis, capital allocation across divisions is sensitive to both the fundamental, and the nonfundamental component of the stock price of the peers of each division.

Figure 5 displays the dynamic of these sensitivities (with the same methodology as that used to obtain Figure 4). We observe that fundamental shocks to peers’ stock price have a permanent effect on investment. Moreover, non-fundamental shocks to peers’ stock price also have a permanent effect on the investment in a particular division, suggesting that division managers do not recover the loss of capital allocated to their division due to a misleading interpretation of the noise in their peers’ stock price, even if corrective signals arise in subsequent years. Possible reasons are that division managers lack discretion in making investment decisions or that the capital available in a given year is consumed by the other divisions and is no longer available later on.

In contrast, the effect of the noise in stock prices on the level of investment for all firms in our sample is transient on average (see panel A in Figure 4). However, Figures 4 and 5 are not directly comparable. Indeed, Figure 4 shows the dynamics of the total level of investment for a given firm following a nonfundamental shock to its stock price. In contrast, Figure 5 shows the dynamics of the allocation of investment across divisions within conglomerates following such a shock. Per se, it says nothing about the effect of the noise in stock prices on the total level of investment for conglomerates. In fact, we checked that this effect and the dynamics of investment following nonfundamental shocks to stock prices are not different for conglomerates and single segment firms in our sample.

Overall, this section shows that our main results survive in a setting where these results are, by design, unlikely to stem from omitted variables. We recognize however that this setting is limited to conglomerates (which represent a third of firms in the sample). We thus perform additional analyses to further rule out four possible alternative explanations for our findings.

6.2 Financing channel

Our focus on peers guarantees that the effect of noise on investment cannot reflect a direct financing channel, whereby firms exploit nonfundamental variation in their own stock price by timing their share issuances. Nevertheless, the noise in peers’ stock price could still affect a firm cost of financing (and thereby its investment) for two reasons. First, capital providers might also rely on peers’ stock price to learn information about a firm’s fundamentals and set financing conditions accordingly. Second, a decrease in peers’ stock price might reduce peers’ ability to buy existing assets in the industry (precisely because of the financing channel), which may lower the collateral value of industry-specific assets, and limit firms’ borrowing capacity (e.g., Shleifer and Vishny 1992).

To assess whether our results could reflect such a financing channel, we test whether firms’ cost of financing and actual financing decisions are sensitive to nonfundamental variation in their peers’ stock price. We do so by replacing investment in our baseline specification (9) with various measures of financing costs and policies. We rely on credit default swap (CDS) spreads and spreads on new private debt issues as indicators of the cost of financing. To mitigate the concern that these proxies, which are not available for all firms, may not be representative of the entire population of firms, we also use the measures of financing constraints developed by Hoberg and Maksimovic (2015), which are based on textual analysis of firms’ 10-Ks. In particular, we use their scores of the intensity of debt-market and equity-market constraints, where a higher score indicates more binding constraints. These scores are available for all firms since 1997, including firms that are not seeking access to financing, and as such, are free of any selection bias. However, they might reflect subjective opinions of managers. We thus complement this analysis of the cost of financing with an analysis of actual financing decisions, namely, the amount of new security issuances (debt plus equity) and total payout (share repurchases plus dividends) over total assets.

Table 6 reports the results. In the first four columns, the coefficient on |$\overline{MFHS}_{-i,t-1}$| is either insignificant or significant (for the spread on new private debt issues) but positive, which means that firms’ cost of financing and access to external finance do not deteriorate when peers’ stock price is under negative price pressure. Columns 5 and 6 further show that firms’ financing decisions do not respond to noise in peers’ stock price. Column (7) finally shows that controlling for firms’ financing decisions in our baseline specification barely affects the coefficient on |$\overline{MFHS}_{-i,t-1}$| (0.014 compared to 0.015). In sum, Table 6 shows that our main results cannot be explained by a potential correlation between firms’ financing conditions and the noise in their peers’ stock price.

Another prediction of the financing channel is that the sensitivity of investment to the noise in peers’ stock price should vary with the intensity of financing constraints. Using various proxies for the presence of such constraints, we show in the Internet Appendix (Section C.5) that this is not the case. For instance, the investment of large and mature firms with a credit rating, which are typically the least constrained ones, is as sensitive to noise in peers’ stock price as the other firms. This finding demonstrates that noise in secondary market prices can affect investment even in the absence of financing frictions. This result is fully consistent with the faulty informant channel (which applies to both constrained and unconstrained firms), but not with the financing channel.

6.3 Pressure channel

Another possibility is that our findings reflect a link between the noise in peers’ stock price and managers’ personal incentives to invest. We call such a link the “pressure channel”. It could have three origins. First, the noise in peers’ stock price might alter the intensity of product-market competition. For instance, a nonfundamental drop in peers’ stock price could trigger industry consolidation through acquisitions (e.g., Edmans, Goldstein, and Jiang 2012 show that a nonfundamental drop in a firm stock price increases its likelihood of being taken over). In this case, after observing a drop in its peers’ stock price, a manager might decide to “wait and see,” postponing new investments in expectation of significant industry changes. Second, drops in peers’ stock price could reduce managers willingness to invest if their compensation is tied to peers’ stock price via relative performance evaluation contracts. Third, a nonfundamental drop in peers’ stock price may affect managers’ career concerns if it increases the threat of a takeover leading to their replacement. In response, managers might undertake actions, such as postponing profitable investment, to temporarily boost their earnings and thereby their short-term stock price (see Stein 1989, 1996).²⁵

Estimates reported in Table 7 do not support the pressure channel. The first two columns indicate that noise in peers’ stock price (⁠|$\overline{MFHS}_{-i,t}$|⁠) does not significantly affect competition intensity, as measured by the (sales based) Herfindahl-Hirschman index or the (log) number of peers.²⁶ Moreover, Column 3 shows that our baseline estimate of the effect of |$\overline{MFHS}_{-i,t}$| does not change if we control for peers’ acquisition activity by adding the fraction of peers making acquisitions and the fraction of targeted peers in our baseline specification.²⁷ Next, using the methodology of Aggarwal and Samwick (1999) to identify the likely use of relative performance evaluation (RPE), we find that the sensitivity of firms’ investment to the noise in peers’ stock price is similar for firms that tie managerial compensation to their peers’ stock price (i.e., RPE=1 in Column 4) and for firms that do not (i.e., RPE=0 in Column 5). Finally, Columns 6 and 7 show that |$\overline{MFHS}_{-i}$| does not affect the likelihood of a firm being taken over or the incidence of a CEO turnover, which casts doubt on the possibility that noise in peers’ stock price be a serious source of career concerns for firm managers.²⁸ Overall, results displayed in Table 7 are largely inconsistent with the idea that our results originate from the interplay between noise in peers’ stock price and managers’ personal incentives.

6.4 Investment-mimicking channel

Once concern is that the sensitivity of a firm’s investment to the noise in peers’ stock price could be explained by firms’ mimicking behaviors (or peer effects) instead of the faulty role of stock prices. Indeed, previous studies report that firms reduce investment in response to nonfundamental drops in their own stock price, for reasons unrelated to the faulty informant channel. Thus, a decrease in peers’ stock price could lead the peers of firm |$i$| to reduce their investment. In turn, this reduction might lead firm |$i$| to also cut its investment, either because its manager uses peers’ investment as a source of information (e.g., Lieberman and Asaba 2006) or because firms’ investment decisions are strategic complements (e.g., Fudenberg and Tirole 1984; Bulow, Geanokoplos, and Klemperer 1985).²⁹ If these mechanisms operate, they might explain, at least partially, why a firm investment is sensitive to the noise in its peers’ stock price and confound the ability of our test to measure the faulty informant channel.

However, in all our specifications so far, we included the lagged investment of a firm’s peer as a control variable. This control is likely to capture the direct influence of peers’ investment on the investment of a firm since the investment of a given firm is known perfectly by other firms when financial accounts are released, that is, with at least a 1-year delay. Moreover, this control excludes the possibility that the effect of |$\overline{MFHS}_{-i,t-1}$| on firm |$i$|’s investment stems from an effect of this variable on lagged investment of a firm’s peer. Nevertheless, in this section, we consider additional analyses (reported in Table 8) to examine whether our findings regarding the effect of the noise in peers’ stock price could just capture peer effects in investment decisions.

First, we test whether investment is sensitive to the noise in peers’ stock price even for sub-samples of firms whose peers do not significantly change their investment (the annual growth of their investment is less than one-tenth of its standard deviation) after nonfundamental shocks to their stock price (Column 1 of Table 8) or even increase it (Column 2 of Table 8). If |$\overline{MFHS}_{-i,t-1}$| was significantly related to investment because of strategic complementarities or substitutability in firms’ investment (or other types of mimicking behavior), the coefficient on this variable should be zero when estimated with these subsamples. Instead, we find that it is significantly positive and similar to that obtained for the entire sample.

Second, in Column 3 of Table 8, we estimate our baseline specification (9) by adding peers’ contemporaneous average investment as a control variable. One caveat is that estimates of this specification are difficult to interpret. Indeed, as previously explained, various theories (including the faulty informant hypothesis) predicts that the average investment of peers should respond to the noise in peers’ stock price (⁠|$\overline{MFHS}_{-i,t-1}$|⁠). If it does, controlling for peers’ contemporaneous investment mechanically biases downward the coefficient on |$\overline{MFHS}_{-i,t-1}$| on firm |$i$|’s investment, even if peers’ contemporaneous investment has no effect on firm |$i$|’s investment. Using Angrist and Pischke’s (2009) terminology, we define peers’ contemporaneous investment as a “bad control.” Their contemporaneous investment biases the estimate of the effect of interest (here the true effect of the noise in peers’ stock price on firm |$i$|’s investment, regardless of the exact reason for this effect). In any case, with this specification, the coefficient on |$\overline{MFHS}_{-i,t-1}$| remains positive and statistically significant at the 1% level. It is smaller than in our main specification (0.007 vs. 0.015), possibly because, peers’ contemporaneous investment picks part of the effect of |$\overline{MFHS}_{-i,t-1}$| on firm |$i$|’s investment, even though it does not directly affect this investment.³⁰

6.5 Correlated noise in prices

A last concern is that the observed component of the noise in peers’ stock price (⁠|$\overline{MFHS}_{-i}$|⁠) could be correlated with the unobserved component of the noise in firm |$i$|’s stock price (the empirical analog of |$u_{i}^{no}$| in our model).³¹ In this case, |$\alpha_0$|—our estimate of |$\alpha_{-i}$|—will pick the effect of the noise in firm |$i$|’s stock price on firm |$i$|’s investment. One way to control for this is to use firm|$\times $|year fixed effects as we do in Section 6.6.1 for conglomerates. Using firm |$\times$| year fixed effects allows to control for any time-varying characteristic of firm |$i$|⁠, including the unobserved noise in its stock price (this is the reason we do not control for the components of this price in specification (11)). Therefore, the tests that we perform in Section 6.6.1 already fully address the above concern. Of course, a limitation is that these tests apply to conglomerates only.

To assess whether the coefficient on |$\overline{MFHS}_{-i}$| in our tests for the entire sample of firms captures the influence of the unobserved noise component of firms’ own stock price on their investment decisions, we estimate our baseline specification (9) with and without |$\overline{MFHS}_{-i}$|⁠. If our results reflect the influence of the unobserved noise component of firms’ own stock price, the coefficient on |$Q_{i}^{\ast }$| should substantially vary across these two specifications (see Oster 2017).³² Columns and (2) of Table 9 indicate that this is not the case. Excluding |$\overline{MFHS}_{-i}$| moves the magnitude of the point estimate on |$Q_{i}^{\ast }$| from 0.081 to 0.083 only. In Columns 3 and 4, we control for firms’ own stock price |$Q_{i}$| before its decomposition into |$Q_{i}^{\ast }$| and |$MFHS_{i}$|⁠. Again, removing |$\overline{MFHS}_{-i}$| barely alters the magnitude of the coefficient on |$Q_{i}$|⁠, suggesting that our results are unlikely to be driven by a correlation between observed and unobserved components of noise in stock prices.

7. Conclusion

In this paper, we provide evidence that a firm’s investment is sensitive to nonfundamental shocks in stock prices. We argue that this finding is consistent with the faulty informant hypothesis: the notion that managers have limited ability to filter out the noise in stock prices when using them as signals about their growth opportunities. We also show that our findings regarding the sensitivity of firms’ investment to the noise in their peers’ stock prices cannot be explained by other mechanisms relying on financing and agency frictions or peers’ effects in investment decisions.

Overall, our results shed a new light on the links between real and stock market efficiency. First, they imply that stock market efficiency matters even for real decisions of firms facing no financing or agency frictions. Second, they suggest a different type of intervention to mitigate possible harmful real effects of stock market inefficiencies. In the literature, stock market inefficiencies lead to inefficient investment decisions because of agency frictions (Stein 1996). Accordingly, investment distortions due to stock market inefficiencies can be mitigated by improving governance systems (see Jensen 2005). In contrast, our findings suggest that nonfundamental shocks to stock prices can generate investment inefficiencies because they influence managers’ beliefs about their growth opportunities. In this case, preventing managers from using stock prices as signals would make things worse on average (see the discussion in Section II.C.4). Instead, to mitigate inefficiencies due to faulty stock prices, one should improve managers’ ability to filter out the noise in stock prices or incentivize them to do so. How to do so would be an interesting question for future research.

Another interesting question is whether nonfundamental shocks to a firm’s stock price could also affect the investment of fundamentally unrelated firms (its peers of peers).³³Schneemeier (2017) shows that such contagion can happen if firms’ managers have both limited ability to filter out the noise in prices and limited attention to stock prices. Testing whether this mechanism is at play in the data would be another interesting venue for future research on the ramifications of the faulty informant channel.

Appendix A. Proofs

In this appendix, we denote by |$\psi _{i}=\sigma _{\theta _{i}}^{2}\left( \sigma _{\theta _{i}}^{2}+\sigma _{\chi _{i}}^{2}\right) ^{-1}$|⁠, the signal-to-noise ratio for the manager’s own private information, |$s_{mi}$|⁠, about the fundamentals of the growth opportunity, |$\widetilde{\theta }_{i}$|⁠. Similarly, |$\kappa _{i}=\sigma _{\theta _{i}}^{2}\left( \sigma _{u_{i}}^{2}+\sigma _{\theta _{i}}^{2}\right) ^{-1}$| and |$\kappa _{-i}=\sigma _{\theta _{i}}^{2}\left( \sigma _{u_{-i}}^{2}+\sigma _{\theta _{i}}^{2}\right) ^{-1}$| denote the signal-to-noise ratios for the signals conveyed by stock prices. These ratios measure, respectively, the informativeness of firm |$i$|’s own stock price (⁠|$\kappa _{i}$|⁠) and the informativeness of peers’ stock prices (⁠|$\kappa _{-i}$|⁠) about the fundamentals of the growth opportunity. As |$\sigma _{u_{i}}^{2}>0$| and |$\sigma _{u_{-i}}^{2}>0$|⁠, we have |$\kappa _{i}<1$| and |$\kappa _{-i}<1$| (i.e., stock prices never perfectly reveal firm |$i$|’s fundamentals). Finally, |$\phi _{i}=\sigma _{u_{i}}^{2}\left( \sigma _{\eta _{i}}^{2}+\sigma _{u_{i}}^{2}\right) ^{-1}$| and |$\phi _{-i}=\sigma _{u_{-i}}^{2}\left( \sigma _{\eta _{-i}}^{2}+\sigma _{u_{-i}}^{2}\right) ^{-1}$| denote the signal-to-noise ratios for the manager’s signals about the noise in her own stock price and peers’ stock prices, respectively.

Proof of Lemma 1.

The following remark is useful for the proof of Lemma 1.

Remark 1: Let (⁠|$X,\widetilde{\theta}_{i}$|⁠) be a random vector with a multivariate normal distribution. In this case E|$(\widetilde{\theta}_{i}\left\vert X\right) =$| E|$(\widetilde{\theta}_{i})+$| Cov|$(\widetilde{\theta}_{i},X)^{^{t}}{\it{\Omega}} ^{-1}(X-$| E|$(X))$|⁠, where |${\it{\Omega}} ^{-1}$| is the inverse of the variance-covariance matrix of |$X$| and Cov|$(\widetilde{\theta}_{i},X)^{^{t}}$| is the transpose of the (column) vector giving the covariance between |$\widetilde{\theta}_{i}$| and each component of |$X$|⁠.

Special cases. Four special cases are of interest (cases 3 and 4 are discussed in the text).

Case 1: The manager’s private information about |$\theta_i$| is perfect. In this case, |$\sigma _{\chi_{i}}=0$| and therefore |$\psi_i=1$|⁠. It follows that |$b_i^*=c_i^*=0$| and therefore |$b_i=c_i = 0$|⁠. Moreover, |$\sigma _{\chi _{i}^{\ast }}=0$| and therefore, using Equations (A9) and (A10), we have |$b_{-i}=c_{-i}=0$|⁠.

Case 2: The manager’s private information about |$\theta_i$| is imperfect and peers’ stock prices are uninformative. In this case |$\sigma _{\chi_{i}} >0$| and |$\kappa_{-i}=0$|⁠. Thus, using Equations (A9) and (A10), we have |$b_{-i}=c_{-i}=0$|⁠. Moreover, if the firm’s own stock price is informative then |$b_i > 0$| because |$\psi_i < 1$| and |$\kappa_i >0$|⁠. If in addition, the manager of firm |$i$| is informed about the noise in her own stock price then |$\phi_i >0$| and therefore |$c_i < 0$|⁠.

Case 3: The manager’s private information about |$\theta _{i}$| is imperfect and peers’ stock prices are informative. This case is a more general version of Case 2. As the manager is imperfectly informed we have |$\psi _{i}<1$| and |$\psi _{i}^{\ast }<1$|⁠. As peers’ stock prices are informative, we also have |$\kappa _{-i}>0$| and therefore |$b_{-i}\neq 0$| (see Equation (A9)). Moreover, the sign of |$b_{-i}$| is the same as the covariance between firm |$i$|’s fundamentals and its peers’ fundamentals, that is, |$\rho _{i}$|⁠. If, in addition, firm |$i$|’s manager is informed about the noise in her peers’ stock price then |$\phi _{-i}>0$| and therefore |$c_{-i}\neq 0$| (see Equation (A10)). Last, |$c_{-i}=-\phi _{-i}b_{-i}$|⁠. Thus, |$b_{-i}$| and |$c_{-i}$| have opposite signs

Case 4: The manager’s private information about |$\theta _{i}$| is imperfect but the manager has a perfect signal on the noise in its peers’ stock price, that is, |$\sigma _{\eta _{-i}}=0$|⁠. In this case, we deduce from Equations (A9) and (A10) that |$b_{-i}=1$| and |$c_{-i}=-1$|⁠. Moreover, |$a_{i}=b_{i}=c_{i}=0$|⁠. Thus, using Equation (A8), we deduce that |$K_{i}^{\ast }=\theta _{i}$|⁠. It follows from Remark 1 that E|$(K_{i}^{\ast }\mid P_{-i},P_{i})=$|E|$(\theta _{i}\mid P_{-i},P_{i})=\frac{ \tau _{u_{i}}}{\tau _{u_{i}}+\tau _{u_{-i}}+\tau _{\theta _{i}}}P_{i}+\frac{ \rho_i^{-1}\tau _{u_{-i}}}{\tau _{u_{i}}+\tau _{u_{-i}}+\tau _{\theta _{i}}}P_{-i}$| where |$\tau _{u_{i}}$| (resp., |$\tau _{u_{-i}}$|⁠) is the inverse of the variance of |$u_{i}$| (resp., |$u_{-i}$|⁠).

Thus, |$\alpha _{-i}\neq 0$| is strictly different from zero if and only if (1) the manager’s private signal is not perfect (⁠|$\psi _{i}^{\ast }<1$|⁠), (2) peers’ stock prices are informative (⁠|$\kappa _{-i}>0$|⁠), and (3) the manager cannot perfectly filter out the noise in her peers’ stock prices (⁠|$\phi _{-i}<1$|⁠). Moreover, the sign of |$\alpha _{-i}$| is the same as the sign of |$\rho_{i} $|⁠.

Symmetrically, we have |$\alpha _{i}=b_{i}+c_{i}=(b_{i}^{\ast }+c_{i}^{\ast })(1-\rho _{i}b_{-i})=(1-\phi _{i})b_{i}^{\ast }(1-\rho _{i}b_{-i})$| where the first equality is the definition of |$\alpha _{i}$|⁠, the second follows from Equations (A5) and (A6), and the last from the expressions for |$b_{i}^{\ast }$| and |$c_{i}^{\ast }$|⁠. As |$\phi _{i}$|⁠, |$b_{i}^{\ast }$|⁠, |$b_{-i}$| and |$\rho_i$| belong to |$[0,1]$|⁠, we deduce that |$\alpha _{i}\geq 0$|⁠. Moreover, it is strictly positive if and only if (1) the manager’s private signal is not perfect (⁠|$\psi _{i}^{\ast }<1$| so that |$b_{-i}<1$|⁠), (2) firm |$i$|’s stock price is informative (⁠|$\kappa _{i}>0$| so that |$b_{i}^{\ast }>0$|⁠), and (3) the manager cannot perfectly filter out the noise in her firm’s stock price (⁠|$\phi _{i}<1$|⁠).

Proof of Proposition 1

Appendix B

Appendix C. Constructing Mutual Fund Hypothetical Sales (⁠|$MFHS$|⁠)

This appendix explains how, for each stock |$i$|⁠, we construct |$MFHS_{i,t}$|⁠, a measure of hypothetical sales in stock |$i$| in year |$t$| due to large outflows in mutual funds owning the stock. Our approach follows the three-step approach proposed by Edmans, Goldstein, and Jiang (2012).

For helpful comments, we thank Alice Bonaime, Don Bowen, Markus Brunnermeier, Evan Dudley, Alex Edmans, Michael Faulkender, Harrison Hong, Larry Glosten, Xavier Giroud, Itay Goldstein, Alexander Gorbenko, Robin Greenwood, Denis Gromb, Wei Jiang, Pete Kyle, Max Maksimovic, Rich Mathews, Gregor Matvos, Ryan Peters, Clemens Otto, Gordon Phillips, Andrei Shleifer, Adi Sunderam, Sheridan Titman, Fabio Verona, Jeffrey Wurgler, Motohiro Yogo, and Wei Xiong and seminar participants at various institutions and conferences. We thank Jerry Hoberg and Gordon Phillips for sharing their TNIC data, and Jerry Hoberg and Max Maksimovic for sharing their text-based financing constraints data. This work was supported by the Investissements d’Avenir Labex [ANR-11-IDEX-0003/Labex Ecodec/ANR-11-LABX-0047 to T. F.] and the Julis-Rabinowitz Center for Public Policy and Finance [to A. M.]. All errors are the authors alone. Supplementary data can be found on The Review of Financial Studies Web site.

Footnotes

References