Managerial Attributes and Market Feedback E ectshuangsy/files/FireSale/FireSale.pdf · 1 Introduction Financial markets often experience large non-fundamental drops, the recovery

Managerial Attributes and Market Feedback Effects *

Suman Banerjee�

Shiyang Huang�

Vikram Nanda§

Steven Chong Xiao¶

This version: March 6, 2020

Abstract

We show that managerial learning from stock prices can lead to feedback loop vulnerability:specifically, liquidity induced trading can impose a negative externality on the firm’s investmentdecisions, inducing unconstrained investors to also sell their stock holdings. Interestingly, over-confident managers that ignore stock price information may be less vulnerable to price impact ofliquidity driven trades. Our empirical evidence strongly supports the model’s assumptions andpredictions: First, investment decisions of overconfident CEOs are significantly less responsive tostock price fluctuations. Second, the price impact of liquidity shocks e.g., mutual fund fire salesis substantially smaller for firms with overconfident CEOs.

JEL Classification Code: G23, G32, G34Keyword: Market feedback, CEO overconfidence, Fire sale

*We acknowledge the thoughtful comments of Alex Edmans, Marcin Kacperczyk, Itay Goldstein, Wei Jiang, LiyanYang, Jan Schneemeier, Russ Wermers, Haifeng Guo, Mark Humphery-Jenner, Fengfei Li, Vikas Mehrotra, Saurin Patel,and Eric Tan. We also thank the seminar participants at American Finance Association Annual Meeting 2017, AsianFinance Association Annual Meeting 2017, Annual Conference of the Swiss Society for Financial Market Research2017, European Financial Management Association (EFMA) Greece 2017 Meetings, Northern Finance AssociationAnnual Meeting 2016, FMA Asia 2016, Southern University of Science and Technology, Southern Methodist University,University of Oklahoma, University of New South Wales, Stevens Institute of Technology, University of Texas at Dallas,University of Wyoming and Indian Institute of Management, Calcutta. Shiyang Huang gratefully acknowledges thefinancial support from the Hong Kong General Research Fund (GRF project code: 17503317). Suman Banerjeegratefully acknowledges SER Research Grant from the University of Wyoming.

�School of Business, Stevens Institute of Technology; Tel: 307-761-3315; Email: [email protected]�Faculty of Business and Economics, The University of Hong Kong; Tel: +852-39178564; Email: [email protected]§Naveen Jindal School of Management, University of Texas at Dallas; Tel: 404-769-4368; Email:

[email protected]¶Naveen Jindal School of Management, University of Texas at Dallas; Tel: 972-883-5056; Email:

[email protected]

1

Electronic copy available at: https://ssrn.com/abstract=3550319

1 Introduction

Financial markets often experience large non-fundamental drops, the recovery from which takes

considerable time. An example is the 1994 peso crisis, during which Mexican lending rates rose from

21% to 91% within four months, and then 30 months passed before loan rates returned to their

pre-crisis level.

Individual stocks also frequently experience non-fundamental “fire sales” from mutual funds which

are subject to investor redemptions (see, e.g., Coval and Stafford, 2007; Ellul, Jotikasthira, and

Lundblad, 2011; Pulvino, 1998). Price declines associated with the fund-flow-driven fire sales can

persist for several months. Although liquidity provision appears to be profitable for these “fire sale”

stocks, other mutual funds do not appear to be willing to absorb these non-fundamental price drops.

These economic phenomena are intriguing, but not well understood.1

In this paper, we argue that a plausible explanation for the severity and persistence of these

seemingly non-fundamental shocks could be the response of economic agents who rely on financial

market signals (e.g., the stock prices) when making real investment decisions. The notion is that

the learning behavior of economic agents can create a feedback loop that temporarily distorts firms’

investment decisions and exacerbates liquidity shocks and market fragility. We develop our argument

in the specific context of non-fundamental liquidity shocks in the stock market. As we will show,

rational firm managers’ investment response to stock prices can amplify liquidity shocks in the

financial market.

We propose a stylized model to formalize the aforementioned argument. To highlight the impact

of market feedback on stock prices and to develop testable hypotheses, we assume that managers differ

in the extent to which they learn from their firms’ stock prices about the quality of their investment

opportunities. Whereas a rational manager gleans investment information from the firm’s stock price,

overconfident managers might stick to their beliefs about the quality of investment opportunities and

ignore stock price information. A test of our hypothesis, based on heterogeneity in the way managers

learn from the market, provides compelling empirical support for our arguments.

In our model, a group of mutual funds is endowed with shares of the firm’s stock. Mutual funds

receive a private signal about the quality of a firm’s investment opportunities and decide whether

to retain their shares or to sell them in the stock market. A subset of these mutual funds, however,

1The financial crisis of 2007–2008 may be another example. When the U.S. market experienced a credit crunch,hedge funds hoarded large amounts of cash in their portfolio and were not willing to support the market even if theliquidity provision was profitable. As estimated by the Economist (2008), the hedge fund industry held cash from $156 billion to $184 billion between July and August 2008.

1


face a redemption shock that forces them to liquidate their holdings, regardless of their private

information. Funds not subject to a liquidity shock (i.e., liquidity-unconstrained funds) then decide

whether to sell their shares based on their private information, their alternative opportunities, and

the firm’s stock price.

When rational managers learn from firms’ stock prices, liquidity shocks that some funds face

spill over onto the selling decisions of other (liquidity-unconstrained) funds. Specifically, a liquidity

shock, even when it is publicly observable, affects the quality of stock price information and induces

firm managers to cut investment. Lower firm investment, in turn, will diminish the terminal value of

the firm and make it less attractive for liquidity-unconstrained funds to hold the firm’s shares. As a

result, some funds could sell their holdings even when their private signal about the firm’s investment

opportunities is positive and they are not subject to redemption shocks. This fire sale externality

generated through the feedback channel amplifies the negative price impact of the liquidity shock.

In a word, our model ties market fragility to the response of market participants to stock market

conditions. An interesting counterfactual emerges: if firm managers could ignore market price signals

in these circumstances, the feedback channel would be shut down and the fire sale externality of the

liquidity shock would be muted.

Heterogeneity among firm managers in terms of their sensitivity to the stock price information

provides a way to test the model’s predictions. In particular, we contend that the degree of managerial

overconfidence could plausibly indicate the attention that a manager is likely to pay to market

information. The rationale is that, by definition, overconfident CEOs (henceforth, OC CEOs) are

expected to place greater weight on their prior beliefs and less on stock price information. This

rationale is well supported by the prior empirical literature which documents that overconfident

managers tend to overestimate returns (or their ability to deliver returns) and underestimate the

underlying risk of new investment opportunities: as a result, they tend to invest in novel projects in

situations in which a rational or non-overconfident manager would hesitate to do so.2

We empirically test the model’s prediction in the context of mutual fund fire sales. The literature

documents a substantial price pressure on individual stocks when investor withdrawals from mutual

2Overconfidence can be a desirable trait in managers when, for instance, valuable, but risky investments need tobe made in less certain situations. The downside is that overconfidence can lead to faulty assessments of investmentvalue and risk, often resulting in sub-optimal investment decisions. The double-edged nature of confidence is evident inthe extant literature. Confidence is essential for success in myriad domains, including business (see, e.g., Galasso andSimcoe, 2011; Hirshleifer, Low, and Teoh, 2012; Johnson and Fowler, 2011; Puri and Robinson, 2007; Simsek, Heavy,and Veiga, 2010). Not surprisingly, CEOs tend to be more confident than the lay population (Graham, Harvey, andPuri, 2013). But Malmendier and Tate (2008) find that overconfident CEOs tend to undertake acquisitions that createsignificantly less shareholder wealth. Also, Malmendier and Tate (2005a) show that overconfident CEOs spend moreof their cash flows on capital expenditures, reflecting their greater propensity to invest.

2


funds provoke the concerted selling of individual stocks held in relatively large quantities by the

affected funds (see, e.g., Coval and Stafford, 2007; Ellul, Jotikasthira, and Lundblad, 2011; Lou,

2012). Our model suggests that feedback from stock prices to real investments could drive the

significant price drops associated with mutual fund fire sales. A testable hypothesis is that the price

impact of a mutual fund fire sale depends on the degree to which firm managers rely on stock market

signals in making their investment decisions.

In our empirical analysis, we start by establishing empirical support for our premise that over-

confident managers and other managers significantly differ in their investment response to stock

prices, for example, in their investment-Q sensitivity (Chen, Goldstein, and Jiang, 2007). The em-

pirical evidence strongly supports our premise that overconfident CEOs are less responsive to stock

price fluctuations. Using the options-based measure (Malmendier and Tate, 2005a) and press-based

measure (Hirshleifer, Low, and Teoh, 2012) of CEO overconfidence, we show that firms run by

overconfident CEOs exhibit significantly lower investment-Q sensitivity.

We also examine the market-learning channel in the context of mergers and acquisitions. For

example, Luo (2005) and Kau, Linck, and Rubin (2008) find that managers learn from the market

when making acquisition decisions: they are more likely to cancel a merger and acquisitions deal

when the market negatively reacts to the announcement. Based on our model, we conjecture and

provide evidence that overconfident CEOs are less likely to cancel their announced acquisitions in

response to negative market reactions to the announcements. Further, we test whether rational, or

non-overconfident, managers adjust their investment policies in response to (non-fundamental) price

movements due to mutual fund fire sales. We find that non-overconfident managers do cut their

investments in response to the stock price drops driven by mutual fund fire sales. Overconfident

managers are, however, less sensitive in their investment response to such stock price movements.

Overall, the empirical evidence strongly supports the notion that overconfident CEOs are less re-

sponsive to stock price fluctuations. This supports us to use overconfident CEOs as a counterfactual

to study the impact of market learning on stock price fragility.

Our model suggests that overconfident CEOs can mitigate the adverse effect of mutual fund

fire sales on stock prices, because fire sales have little effect on their CEOs’ investment policies.

We construct a mutual-fund-flow-driven trading measure in light of the literature (e.g., Coval and

Stafford, 2007; Edmans, Goldstein, and Jiang, 2012; Lou, 2012), and our tests confirm the key model

prediction that the price impact of mutual fund fire sales is significantly smaller for firms managed

by CEOs who are less sensitive to market signals. The effect is economically significant: the price

3


impact of mutual fund fire sales is almost completely muted for stocks with overconfident CEOs. The

results are robust in our regression tests when we control for various firm characteristics that are

relevant to firms’ price stability during fire sales, such as stock illiquidity, proxied by the Amihud’s

(2002) measure and analyst coverage.

We conduct several robustness tests to corroborate the results. First, we control for various

firm and CEO characteristics as well as industry fixed effects. The results hold in a multivariate

regression as well as in propensity score-matched settings. Second, to address the endogeneity concern

regarding mutual fund flows, we use the mutual fund trading scandal of the 2003–2004 period as an

alternative setting for mutual fund fire sales (see, e.g., Anton and Polk, 2014; Koch, Ruenzi, and

Starks, 2016). We find that the stock prices of firms with overconfident CEOs are less affected by the

redemption shock to mutual funds implicated in the trading scandal. Third, we find similar results

using alternative empirical measures of mutual fund flow-driven trading (e.g. Lou, 2012; Wardlaw,

2018) as well as an alternative measure of managers’ reliance on market signals based on firms’

historical investment-Q sensitivity. Fourth, we show that the weaker price impact of mutual fund

fire sales for stocks with overconfident CEOs cannot be explained by corporate actions, such as firms’

equity financing, share buyback, or managerial voluntary disclosure.

We also test a key mechanism of our model by examining the trading by liquidity-unconstrained

funds in response to fire sales by funds that suffer redemption shocks. We find that mutual funds not

subject to flow redemption tend to buy fire sale stocks only when firms are managed by overconfident

CEOs. This evidence directly supports the mechanism in our model: market participants have

greater concern about the loss in firm value when firm managers are more responsive to stock price

movements during fire sales. But such a concern is mitigated by overconfident CEOs who tend to

disregard stock price information in their investment decision-making.

Our paper contributes to several areas of the finance literature. First, our finding contributes to

the growing literature on market feedback effects (see, e.g., Bond, Edmans, and Goldstein (2012) for

a survey). On the one hand, the theoretical front has seen considerable development in understanding

the implications of feedback effects.3 On the other hand, documenting empirical evidence on market

learning is challenging, because the act of learning creates a feedback loop between market prices and

firms’ decisions that is difficult to disentangle empirically. Previous studies have tackled endogeneity

issues by testing cross-sectional predictions that are consistent with market learning (see, e.g., Chen,

3See, for example, Bond, Goldstein, and Prescott (2010); Dow, Goldstein, and Guembel (2015); Dow and Gorton(1997); Dow and Rahi (2003); Edmans, Goldstein, and Jiang (2012); Foucault and Gehrig (2008); Goldstein andGuembel (2008); Goldstein, Ozdenoren, and Yuan (2013a); Subrahmanyam and Titman (1999, 2001).

4


Goldstein, and Jiang, 2007; Foucault and Fresard, 2014) or by exploiting exogenous variations in

stock prices due to mutual fund fire sales (see, e.g., Edmans, Goldstein, and Jiang, 2012). A recent

study by Berger (2017) reviews the empirical finding that mutual fund fire sales affect future firm

policies and suggests that this relation might be driven by the selection bias of mutual fund flows.

The study raises further questions about what drives the changes in firm policies and price declines

after severe mutual fund outflows. Our study contributes to this debate by showing that the market

feedback mechanism can potentially explain the joint relation between fund flows, firm policies, and

the associated price drops. Specifically, we examine cases in which fund redemption shocks have

already occurred,4 and exploit heterogeneity in managers’ propensity to respond to stock market

signals due to behavioral biases, such as overconfidence. Supportive of our theoretical arguments,

the empirical evidence shows that mutual fund outflows trigger significant price drops only when

firms are willing to alter investment expenditures in response to stock price movements.

Our proposed mechanism is also related to the theoretical literature that studies how learn-

ing affects macroeconomic outcomes (see, e.g., Amador and Weill (2010); Fajgelbaum, Schaal, and

Taschereau-Dumouchel (2017); Ordonez (2009); Van Nieuwerburgh and Veldkamp (2006); Veldkamp

(2005)). Van Nieuwerburgh and Veldkamp (2006) proposes a model with incomplete information to

study the asymmetry in booms and busts in business cycles. In the model of Van Nieuwerburgh and

Veldkamp (2006), economic activity generates useful information about the state of the economy,

and agents undertake more economic activity in good times than in bad ones. Because economic

activity declines and thus information is abundant, asset prices quickly adjust to negative shocks,

and a sudden crash occurs. When times are bad, scarce information and high uncertainty slows

agents’ reactions, so the market slowly recovers from the bust. Fajgelbaum, Schaal, and Taschereau-

Dumouchel (2017) proposes a model of endogenous uncertainty and business cycles. In the model

of Fajgelbaum, Schaal, and Taschereau-Dumouchel (2017), agents learn from the actions of oth-

ers, thereby generating slow information flows and high uncertainty in times of low activity. Slow

information flows during bad times further discourage investment.

We also contribute to the managerial overconfidence literature. We confirm that CEO overcon-

fidence can lead to excessive investment (see, e.g., Banerjee, Humphery-Jenner, and Nanda, 2015;

Malmendier and Tate, 2005a, 2008). However, our study suggests that overconfident managers may

be better at steering the firm in difficult market situations that break from the normal functioning

of the market, such as on account of large-scale selling by institutional investors and the resultant

4 As such, our empirical finding cannot be driven by selection bias concerning the likelihood of outflow shocks.

5


price pressure. As such, managerial overconfidence could play a beneficial role in stabilizing stock

prices during market disruptions.

This paper is organized as follows. Section 2 describes the model and trading mechanisms. Section

3 describes the data. Section 4 presents our main empirical results and discusses various robustness

tests. Section 6 concludes. Detailed derivations, data definitions, and additional robustness tests

have been relegated to Appendices A1, A2, and A3, respectively.

2 Model

2.1 Model preliminaries

We propose a stylized model in which managers differ in the extent to which they learn information

from stock market prices. Managers can be one of two types: rational managers who make invest-

ment decisions based on information in their firm’s stock price or overconfident managers who have

strong beliefs in their own ability and do not learn from stock prices. As we will show, the invest-

ment decisions and market valuation of firms with rational managers, unlike those of overconfident

managers, tend to be vulnerable to fire sales resulting from redemption shocks to mutual funds.

We consider a two-period economy with time indexed by t = 1 and 2. For convenience, we model

the economy as having one publicly traded firm and one firm manager. The salient events are as

follows. Prior to t = 1, the firm has an asset-in-place and a growth opportunity. While the value of

the asset-in-place is fixed at K (> 0), the growth opportunity delivers a payoff V on date t = 2 such

that:

V = I · ρ− 1

2e · I2. (1)

Here, I is the investment level; e > 0 is a investment cost parameter; and ρ is a productivity

parameter.

Among investors in the firm’s stock is a continuum of risk-neutral mutual funds endowed (in

aggregate) with one share of the firm’s stock. For simplicity, mutual funds are taken to be constrained

in the quantity of shares they can hold due to resource or regulatory constraints, and their holdings

at the start of date 1 are the largest positions they can assume.

At t = 1, the financial market opens, and investors trade the firm’s stock. At this time, mutual

funds privately receive information about the growth opportunity and make their trading decisions

(a detailed description follows). With probability q (where 0 < q < 1), mutual funds privately learn

that ρ is b < 0, whereas, with complementary probability 1− q, they learn that ρ is g > 0. At t = 2,

6


payoffs are realized, and all agents consume their payoffs. Hence,

ρ ={ g probability 1− q,

b probability q.(2)

Assumption 1. g · (1− q) + b · q = m > 0.

As we will see, the ex ante positive m implies that the firm manager will make a non-zero investment

if no new information is received on date t = 1.

Information structure

Although all mutual funds obtain a private signal about whether ρ is g or b, the firm manager does

not directly observe the signal. We rule out direct communication between mutual funds and the

firm manager. This is reasonable assumption , because any information communicated would not

necessarily be credible. Mutual fund managers could, for instance, provide managers with distorted

information and hence profit by trading in the stock market. This is particulary true in our model

with a large population of mutual funds. Thus, the firm manager only learns about mutual fund

information from the stock price at t = 1.

Fire sale

We assume that each mutual fund can face an exogenous liquidity withdrawal shock at t = 1. A

mutual fund facing such a redemption shock, sells all of its shares. Mutual funds’ liquidity shocks are

independent of each other and independent of project productivity ρ. The probability of a liquidity

withdrawal shock is denoted by f , where 0 ≤ f ≤ 1. Because there is a continuum of mutual funds,

the fraction of funds selling their stock holdings due to liquidity shocks is f . Aggregate mutual

fund holdings are normalized to 1, meaning this is also the quantity of liquidity-driven stock sales

of mutual funds. We assume that f is common knowledge to all agents in the economy, which is

consistent with the reality that mutual funds’ portfolio holdings are disclosed each quarter.

In addition to f , we denote by α the fraction of funds not affected by liquidity shocks but that

endogenously choose to sell off their stock positions. This may occur if, for instance, funds have access

to outside opportunities that are potentially more profitable than holding their stock positions. In

the analysis below, we will examine the effect of f on α in equilibrium and conduct comparative

statics of the expected stock price at t = 1 with respect to f . We note that f = 0 represents an

economy without fire sales and serves as a benchmark for our analysis.

7


Outside opportunities

We also assume that each mutual fund has an outside opportunity at t = 1, that is, an investment

alternative to holding on to the firm’s stock. These outside opportunities are heterogenous across

mutual funds. Specifically, we assume that for any fund i, where (i ∈ [0, 1]), its outside opportunity

ci is drawn from a uniform distribution [0, C]. Hence, in our setup, mutual funds make two types of

selling decisions: selling in response to a liquidity withdrawal shock and selling of mutual funds not

subject to a liquidity shock, but selling to invest in a relatively lucrative outside opportunity. The

fraction of the first type is (f), and the fraction of the second type is α(1− f).

Trading mechanism

The total mass of mutual funds that sell their shares is given by α(1−f)+f . While f is an exogenous

shock, the fraction α is endogenous and will be solved for as an equilibrium outcome. For any mutual

fund i not subject to a liquidity withdrawal shock, the change in holdings can be expressed as below.

Recall that, by assumption, funds are constrained and do not increase their stock holdings:

xi,1 ={ 0 hold position

−1 sell position.(3)

When the stock market opens at t = 1, mutual funds submit market orders and the stock price

is determined by the market-clearing condition. (See, e.g., Goldstein et al., 2013b; Goldstein and

Yang, 2019) Other than mutual funds, non-fundamental demand is for the stock denoted by N . In

the spirit of Goldstein et al. (2013b), the non-fundamental demand is assumed to take the following

form:

N = µ+ µ− λP1. (4)

Here, µ is deterministic, whereas µ is random. We interpret µ as investors’ expectations about

firm value. The variable µ is intended to capture the non-fundamental stochastic demand (e.g.,

investors’ sentiment or liquidity shocks). Hence, we assume that a component of the net demand

for the stock comes from investors that either do not have information or do not learn from the

market price. Several theories (micro-foundations) support the notion that these investors may not

learn information from prices, for example, differences in opinion (Banerjee, 2011; Banerjee and

Kremer, 2010) and behavioral bias (Eyster et al., 2019). For tractability, we assume that the random

variable µ follows a uniform distribution over [−∆,∆]. We note that the uncertainty in µ plays an

8


important role: it prevents the stock price from fully revealing mutual funds’ private information.5

The parameter λ measures market liquidity or price elasticity.

At t = 1, the stock price P1 is endogenously determined based on the following market-clearing

condition: ∫ 1

i=0xi,1 di+N = 0. (5)

Given the anonymous identities of the mutual funds we label, without loss of generality, the mutual

funds that face a liquidity shock as those i ∈ [1−f, 1]. Hence, mutual funds i ∈ [0, 1−f) do not face

a liquidity shock and can decide whether or not to sell their current holdings. The market-clearing

condition yields

P1 =µ

λ+X

λ, (6)

where X =∫ 1−fi=0 xi,1di− f + µ.

Firm manager type and investment strategy

As already noted above, the firm manager can choose a non-negative investment I to maximize firm

value. The firm manager’s optimization problem can be expressed as

maxI

I · ρ− 1

2e · I2. (7)

After the stock market clears on date t = 1, the firm manager can learn information from the

stock price about the productivity parameter ρ and select the investment level accordingly. However,

whether or not firm managers learn from stock prices depends on their type. Rational managers

have an unbiased belief about the distribution of ρ as shown in Equation (2) and may update their

beliefs after observing the market price P1. We denote the investment of rational firm managers

by Ir = max(0, E(ρ|P1)e ). By contrast, overconfident managers have an optimistic view about the

probability of success (i.e. q = 0). As a result, they take the expected productivity parameter to

be g and do not change their belief with market signals (Malmendier and Tate, 2005a). Hence, the

overconfident manager always chooses Ioc = ge , which does not depend on the market-clearing price

P1.

To ensure that firm value is always nonnegative, we derive the sufficient condition under which

the lowest possible firm value is positive. The minimum firm value occurs when ρ is b and the

5Alternative modeling choices, such as introducing uncertain f , could prevent market prices from fully revealingmutual funds’ private information. But the economic intuitions will be qualitatively the same.

9


firm makes its largest investment. The maximum investment level that can be optimal is ge , so a

sufficient condition to ensure a positive firm value is that the firm’s assets-in-place are greater than

the negative project value at this investment level:

Assumption 2. K > −(ge · b−

12g2

e

).

Discussion about model setup

Our setup has several stylized assumptions. The assumptions are made to simplify the analysis and

deliver the economic intuition in a clear way. We briefly discuss some of these assumptions below.

First, we follow Goldstein, Ozdenoren, and Yuan (2013b) and Goldstein and Yang (2019) and

assume that mutual funds submit market orders in a slightly modified Kyle (1985) setting. If mutual

funds submit limit orders, mutual funds’ selling decisions become far more complicated without a

change in the economic intuition.

Second, we assume that every mutual fund at t = 1 is constrained to either sell its shares or hold

on to them. Ruling out the possibility that a mutual fund buys shares at t = 1 provides a simple

way to focus on the fire sale externality. In our model setup, one of our objectives is to show that

liquidity-unconstrained mutual funds tend to liquidate their holdings when there is a fire sale and

the firm manager is rational and sensitive to stock prices. We believe that our economic mechanism

would work if we introduced an additional class of investors that had outside opportunities and could

buy shares but were subject to short-sale constraints. However, introducing another set of investors

complicates the analysis without providing additional economic insights.

Third, our current focus is on how the learning behavior of firm managers amplifies the price

impact of non-fundamental shocks (f). We could introduce additional periods into the current static

model to examine the slow recovery of the fundamental shock. For instance, project productivity

could persist across periods and could gradually become public information. As a result, the market

values of firms with rational and overconfident managers could gradually converge even if the liquidity

shock f temporarily affects the market value of firms with rational managers.

2.2 Equilibrium analysis

We solve for the equilibrium at t = 1. For all variables, we use superscripts oc and r to denote the

equilibrium solutions in economies with overconfident managers and rational managers, respectively.

Meanwhile, we use g and b to denote the equilibrium solutions in an economy when ρ = g or b,

respectively.

10


2.2.1 Equilibrium at t = 1, when the firm manager is overconfident

Overconfident managers do not update their information about investment quality from the firm’s

stock price and always believe productivity to be g. Thus, they choose an investment level of ge .

Next, we pin down mutual funds’ demand decisions and the equilibrium stock price at t = 1. Mutual

funds subject to liquidity shocks (the faction is f) are forced to sell their shares. Other mutual funds

(the fraction is (1 − f)α), that is, those not subject to liquidity withdrawal shocks, may sell their

shares depending on investment productivity and their alternative opportunities.

Scenario 1: ρ is b When ρ = b, firm value is ge · b−

g2

2e +K, where ge is the investment level chosen

by the overconfident manager and g2

2e is the cost of investing. If mutual fund i holds the stock until

t = 2, its net payoff (accounting for opportunity cost ci) is

g

e· b− g2

2e+K − ci. (8)

If mutual fund i sells the stock at t = 1, its expected payoff is E(P1|ρ = b). When the fraction of

mutual funds selling their current positions is α, inserting α into the price function in Equation (6)

and taking expectations yields

E(P1|ρ = b) =µ− f − α(1− f)

λ. (9)

We now characterize α. We observe that when an mutual fund unaffected by liquidity shocks has a

more valuable outside opportunity (i.e., higher c), it will be more willing to sell its shares. Following

Goldstein et al. (2013b), the equilibrium solution requires us to determine an equilibrium cutoff

c∗ ≤ C, such that a unaffected mutual fund i with an outside opportunity ci > c∗ will sell its holdings.

Hence, if c∗ is the equilibrium cutoff, the fraction of mutual funds that do not face liquidity shocks

but still decide to sell their holdings is

α = 1− c∗

C. (10)

Inserting the above expression for α into Equation (9) yields

E(P1|ρ = b) =µ− 1

λ+c∗(1− f)

λC. (11)

Comparing E(P1|ρ = b) and the outside opportunity, mutual fund i’s selling decision can be

summarized as follows. Here, the cutoff is given by c∗,ocb , with the superscript oc and subscript b

11


denoting manager type and productivity, respectively.

xi =

0 if ge · b−

g2

2e +K − ci > µ−1λ +

c∗,ocb (1−f)

λC

−1 if ge · b−

g2

2e +K − ci ≤ µ−1λ +

c∗,ocb (1−f)

λC

In equilibrium, funds with ci = c∗,ocb are indifferent between selling and holding the stock. We can,

therefore, solve for c∗,ocb as

c∗,ocb =ge · b−

g2

2e +K − µ−1λ

1 + 1−fλC

. (12)

Intuitively, when firm value is sufficiently large, mutual funds that are not subject to liquidity shocks

will retain their stock holdings. When firm value is sufficiently low, these mutual funds will sell their

stock holdings. Thus, mutual funds’ selling decisions depend on firm value and outside opportunities.

Three cases can be characterized as follows.

Lemma 1. When the firm manager is overconfident and ρ = b, there exist three possible cases,

depending on c∗,ocb :

Case 1: when c∗,ocb ≥ C, α∗,ocb = 1 and the expected stock price is µ−fλ ;

Case 2: when 0 < c∗,ocb < C, α∗,ocb = 1− c∗,ocbC and the expected stock price is µ−1

λ +c∗,ocb (1−f)

λC ;

Case 3: when c∗,ocb ≤ 0, α∗,ocb = 0 and the expected stock price is µ−1λ .

Here, c∗,ocb =ge·b− g

2

2e+K−µ−1

λ

1+ 1−fλC

.

Scenario 2: ρ is g We now characterize the equilibrium when ρ = g. In this scenario, firm value

is ge · g −

g2

2e + K, where ge is the investment level chosen by the overconfident manager. If mutual

fund i (unaffected by liquidity shocks) retains its shares until t = 2, its net payoff per share (net of

the opportunity cost) is

g2

2e+K − ci.

Following steps similar to those in Scenario 1, we arrive at the following lemma:

Lemma 2. When the firm manager is overconfident and ρ = g, there exist three potential cases,

depending on c∗,ocg :

Case 1: when c∗,ocg ≥ C, α∗,ocg = 1 and the expected stock price is µ−fλ ;

Case 2: when 0 < c∗,ocg < C, α∗,ocg = 1− c∗,ocg

C and the expected stock price is µ−1λ +

c∗,ocg (1−f)λC ;

12


Case 3: when c∗,ocg ≤ 0, α∗,ocg = 0 and the expected stock price is µ−1λ .

Here, c∗,ocg =g2

2e+K−µ−1

λ

1+ 1−fλC

.

2.2.2 Equilibrium at t = 1, when the firm manager is rational

In this section, we pin down the equilibrium in an economy with a rational firm manager. Again,

we focus on the selling decisions of mutual funds not subject to liquidity shocks. The fraction of

unaffected mutual funds that sell their current holdings is denoted by α∗,rρ , where r indicates a

rational manager and ρ the productivity.

In Equation (6), the market-clearing condition is given by

P1 =µ

λ+X

λ,

where X =∫ 1−fi=0 xi di − f + µ. Note that the information revealed by the stock price P1 is equiv-

alent to observing X. Hence, when firm managers observe price P1, their investment becomes

Ir = E(ρ|P1)e = E(ρ|X)

e . To analyze the firm manager’s learning problem E(ρ|X), we note the follow-

ing two facts about the distribution of X conditional on ρ. Recall, that µ is taken to be uniformly

distributed over [−∆,∆]:

Fact 1: When ρ is b, the total order, X, is an element in the range [−α∗,rb (1− f)− f −∆,−α∗,rb (1−

f)− f + ∆]

Fact 2: When ρ is g, the total order, X, is an element in the range [−α∗,rg (1− f)− f −∆,−α∗,rg (1−

f)− f + ∆].

Figure 1 illustrates the distributions of the order flow . To prevent the stock price from fully

−α∗,rb (1− f)− f −∆ When ρ = b −α∗,rb (1− f)− f + ∆

−α∗,rg (1− f)− f −∆ When ρ = g −α∗,rg (1− f)− f + ∆

Figure 1: The distribution of the total order flow X

revealing mutual funds’ private information, we assume some overlap between the two ranges (cor-

responding to ρ = b and ρ = g). In terms of economic intuition, a larger (∆) can be interpreted as

13


greater uncertainty or noise in non-fundamental demand that makes learning about fundamentals

more difficult.

−α∗,rg (1− f)− f −∆ < −α∗,rb (1− f)− f + ∆

A sufficient condition to ensure that we always have some overlap between the two ranges (corre-

sponding to ρ = b and ρ = g) in Figure 1 is

Assumption 3. ∆ ≥ 12 .

We employ a guess-and-verify approach to solve for the equilibrium. µ follows a uniform distri-

bution, so firm managers will update their beliefs about ρ as below:

E(ρ|P1) = E(ρ|X) (13)

=

b when X ∈ [−α∗,rb (1− f)− f −∆,−α∗,rg (1− f)− f −∆]

m when X ∈ [−α∗,rg (1− f)− f −∆,−α∗,rb (1− f)− f + ∆]

g when X ∈ [−α∗,rb (1− f)− f + ∆,−α∗,rg (1− f)− f + ∆].

Thus, optimal investment level Ir is

Ir =

0 when X ∈ [−α∗,rb (1− f)− f −∆,−α∗,rg (1− f)− f −∆]

me when X ∈ [−α∗,rg (1− f)− f −∆,−α∗,rb (1− f)− f + ∆]

ge when X ∈ [−α∗,rb (1− f)− f + ∆,−α∗,rg (1− f)− f + ∆].

(14)

Scenario 1: ρ is b . When ρ = b, firm value is Ir · b− 12

(Ir)2

e +K. If mutual fund i holds the stock

until t = 2, its net payoff (net of the opportunity cost) is

Ir · b− 1

2(Ir)2e+K − ci. (15)

From mutual funds’ perspective: X ∈ [−α∗,rg (1− f)− f −∆,−α∗,rb (1− f)− f + ∆]. As indicated in

Equation (13), when the realization of X is in the overlapping region, that is, X ∈ [−α∗,rg (1− f)−

f −∆,−α∗,rb (1− f)− f + ∆], the firm manager’s assessment of the productivity is E(ρ|X) = m and

investment is Ir = me . This investment occurs with the following probability: 1− (α∗,rb −α

∗,rg )(1−f)

2∆ . On

the other hand, there is no investment when X < −α∗,rg (1− f)− f −∆ and E(ρ|X) = b.

Using Equation (15) and the probability of X being in the overlapping region, the expected value

of the investment is VL = (me · b −12(me )2e) ∗ (1 − (α∗,rb −α

∗,rg )(1−f)

2∆ ). Following Equation 12, we have

c∗,rb =K+VL−µ−1

λ

1+ 1−fλC

. We can characterize the equilibrium as follows:

14


Lemma 3. When the firm manager is rational and ρ is b, there exist three potential cases, depending

on c∗,rb :

Case 1: when c∗,rb ≥ C, α∗,rb = 1 and the expected stock price is µ−fλ ;

Case 2: when 0 < c∗,rb < C, α∗,rb = 1− c∗,rbC and the expected stock price is µ−1

λ +c∗,rb (1−f)

λC ;

Case 3: when c∗,rb ≤ 0, α∗,rb = 0 and the expected stock price is µ−1λ .

Here, c∗,rb =K+VL−µ−1

λ

1+ 1−fλC

and VL = (me · b−12(me )2e) ∗ (1− (α∗,rb −α

∗,rg )(1−f)

2∆ ).

Scenario 2: ρ is g . Even if ρ = g, because rational firm managers learn information about

productivity ρ from the stock price, they may invest less and have a lower firm value, compared to

economies with overconfident firm managers. To characterize the equilibrium, we need to pin down

the cutoff c∗,rg , at which fund i chooses to sell shares only when ci > c∗,rg .

In the equilibrium with learning and ρ = g, the realization of X can be either in the overlapping

region in which E(ρ|X) = m and investment is Ir = me or in the region in which E(ρ|X) = g and

investment is Ir = ge . As with Scenario 1, we obtain the probability of these two cases and the

associated investment payoff. This allows us to obtain the expected payoff from holding the shares:

F = K

+ VH1︸︷︷︸Firm value with high investment

×(α∗,rb − α

∗,rg )(1− f)

2∆︸︷︷︸Probability of high investment

+ VH2︸︷︷︸Firm value with low investment

× (1−(α∗,rb − α

∗,rg )(1− f)

2∆)︸︷︷︸

Probability of low investment

where VH1 = ge · g −

12(ge )2e = g2

2e , VH2 = me · g −

12(me )2e.

Rational managers cannot always correctly infer the productivity to be ρ = g, so it follows:

Lemma 4. When ρ is g, the firm’s investment in the economy with a rational firm manager is

smaller than or equal to that with an overconfident firm manager. That is, Ir ≤ Ioc.

Comparing the outside opportunity and the expected firm value, mutual funds’ selling decisions can

be summarized as below:

xi ={ 0 if F − ci > µ−1

λ + c∗,r(1−f)λC

−1 if F − ci ≤ µ−1λ + c∗,r(1−f)

λC

(16)

It follows that the cutoff value is such that c∗,rg =F−µ−1

λ

1+ 1−fλC

. We characterize the equilibrium as follows:

15


Lemma 5. When the firm manager is rational and ρ = g, there exist three potential cases, depending

on c∗,rg :

Case 1: when c∗,rg ≥ C, α∗,rg = 1 and the expected stock price is µ−fλ ;

Case 2: when 0 < c∗,rg < C, α∗,rg = 1− c∗,rgC and the expected stock price is µ−1

λ +c∗,rg (1−f)

λC ;

Case 3: when c∗,rg ≤ 0, α∗,rg = 0 and the expected stock price is µ−1λ ;

where F = K + VH2 + (VH1 − VH2) ∗ (α∗,rb −α∗,rg )(1−f)

2∆ .

2.2.3 Comparing economies

In this section, we compare the equilibrium in an economy with a rational manager (non-OC firm)

with an economy with an overconfident manager (OC firm). First, when ρ = b, the OC firm clearly

has a lower E(P1) than does the non-OC firm. In the discussion, we begin with the comparison when

ρ is g, which is more relevant for our purposes. Specifically, we compare α∗,ocg with α∗,rg and c∗,ocg

with c∗,rg .

From the discussion above, the expressions for c∗,ocg and c∗,rg are

c∗,ocg =g2e·g+K−µ−1

λ

1+ 1−fλC

and c∗,rg =F−µ−1

λ

1+ 1−fλC

.

Because F < K+ g2e ·g, we have c∗,rg < c∗,ocg . Focusing on the interior solutions (C > c∗,ocg > c∗,rg > 0),

we find that the comparison between α∗,ocg and α∗,rg can directly yield the following result:

Proposition 1. When ρ is g, mutual funds in the economy with a rational firm manager are more

likely to sell their shares than those with an overconfident firm manager. Meanwhile, the expected

stock price in the economy with a rational firm manager is lower than that with an overconfident firm

manager. That is, α∗,rg > α∗,ocg .

From the discussion above, in the presence of fire sales, managerial overconfidence affects firm value

E(P1) differently when ρ is g than when it is b. In particular, following a fire sale, the firm with

an overconfident manager will have a higher E(P1) compared with the firm with a rational manager

when ρ = g. This suggests that managerial overconfidence can play a beneficial role in stabilizing

stock prices when fire sales occur and economic fundamentals are strong.

Further, we can show that the unconditional expected stock price (i.e., not conditioned on pro-

ductivity ρ) of the OC firm will be larger than that of a non-OC firm under certain conditions:

Corollary 1. When µ−1λ −

g2

2e < K ≤ µ−1λ , the expected stock price (E(P1) of the firm with an

overconfident manager will be higher than that of the firm with a rational manager after a fire sale

shock.

16


We can understand the above result by considering mutual funds’ selling decisions conditional

on ρ. When ρ = g, we know from Proposition 1 that the expected stock price of the firm with

the overconfident manager is higher than that of the firm with a rational manager. When ρ = b

and K ≤ µ−1λ , we can verify from the expressions for α∗,rb and α∗,ocb that selling stock is always

more profitable for mutual funds than holding on to the stock until period 2. This implies that

the expected stock prices of firms with rational and overconfident managers will converge to the

same value when ρ = b and K ≤ µ−1λ . Therefore, the unconditional stock price of the firm with an

overconfident manager will be higher than that of a firm with a rational manager when K ≤ µ−1λ .

2.2.4 Comparative statics: The role of fire sale magnitude (f) in the selling decisions

by liquidity-unconstrained funds.

We examine how the magnitude of the fire sale (f), which is known to all market participants, affects

the selling decisions of mutual funds that have not been affected by redemption shocks. That is,

we conduct comparative statics of α with respect to f . As we will explain, the impact of f on α

manifests in two offsetting ways: liquidity provision and a learning channel.

A larger f puts direct downward price pressure on E(P1). As a result, mutual funds not affected

by redemption shocks find selling their shares at t = 1 less attractive. This is evident, for instance,

in that an increase in f increases the cutoff values c∗ and, hence, lowers the extent of selling (α∗) in

the different scenarios considered. We denote this effect as liquidity provision.

The second effect of f on the selling of shares by mutual funds is through the learning channel.

As we have shown, after a fire sale, the rational manager’s learning from stock prices can make

the firm’s expected stock price (E(P1)) lower than that of a firm with an overconfident manager.

However, when the rational firm manager learns information from P1, the downward price pressure

negatively affects the firm’s investment level and the real firm value. This effect on real firm value

reduces the incentives of mutual funds (not subject to liquidity withdrawal shocks) to hold their

shares until t = 2. This learning effect offsets the negative effect of f on α. We summarize the result

as follows:

Proposition 2. α∗,rg , α∗,ocg have interior values that is, 0 < α∗,rg < 1, 0 < α∗,ocg < 1, when µ−1λ −

g2

2e <

K ≤ µ−1λ , K + m

e g −12(me )2e− µ−1

λ > 0 and K + g2

2e −µ−1λ < C.

In an interior region defined as above, we always have∂α∗,ocg

∂f < 0, and

(1) when 1λC −

12∆C (VH1 − VH2) < 0:

∂α∗,rg∂f > 0.

(2) whereas, when 1λC −

12∆C (VH1 − VH2) > 0:

∂α∗,ocg

∂f <∂α∗,rg∂f < 0.

17


According to Proposition 2, the sensitivity of α∗,rg to f depends on VH1− VH2. When VH1− VH2

is sufficiently large, α∗,rg is increasing with f ; otherwise, it is decreasing with f . As discussed

above, the effect of f on α∗,rg works through two effects: liquidity provision and learning. In fact,

VH1 − VH2 measures the drop of firm value due to the firm’s “mistake” in learning from its stock

price. Specifically, when ρ = g, the firm value would be K +VH1 if the firm manager could perfectly

know the realization of ρ. When the firm manager observes low order flow, even though ρ is g, the

firm value becomes K + VH2. This suggests that VH1 − VH2 reflects the reduction in firm value due

to the firm’s learning behavior. Intuitively, when VH1 − VH2 is sufficiently large, the learning effect

is large and could dominate the effect of liquidity provision. As a result, α∗,rg increases with f .

We can also use Figure 2 to provide intuition. Under conditions specified in Proposition 2, all

mutual funds sell their shares when a bad signal is realized (i.e., αb = 1), because no fund finds

holding the stock profitable relative to its outside opportunity. When f increases, the distribution of

aggregate order flow in the bad scenario (ρ = b) is fixed at [−1−∆,−1+∆], whereas the distribution

of aggregate order flow in the good scenario (ρ = g) shifts to the left. As a result, the overlapping

region of the total order flow becomes larger, making it more difficult for the rational firm manager

to learn. This suggests that even if ρ = g, when f increases, the rational firm manager tends to make

smaller investments, which make it less attractive for mutual funds to retain the stock until t = 2.

−1−∆ When ρ = b −1 + ∆

−α(1− f)− f −∆ When ρ = g −α(1− f)− f + ∆

When f increases, the range of order flow (ρ = g) moves left

Figure 2: The distribution of the total order flow X

This result has important implications. As shown in Proposition 1, E(P1) is higher for the firm

with the overconfident manager than for that with the rational manager when ρ = g. We now show

that, under interior parameter assumptions of Proposition 2, the price gap between overconfident

and rational firms varies (increasing and then decreasing) as f increases. In other words, the fire

sale has differential dampening effects on stock prices for overconfident and rational managers. We

denote the price gap between overconfident and rational firms as GP = E(P ∗,oc1 |ρ = g)−E(P ∗,r1 |ρ =

g) =(α∗,rg −α∗,ocg )(1−f)

λ > 0. The following proposition characterizes the variation in GP as f increases:

18


Proposition 3. In the interior region (Proposition 2): When VH1 − VH2 > 2∆C(1 + 1λC )(1 −√

K+VH2−µ−1λ

K+ g2

2e−µ−1

λ

), as the fraction of mutual funds receiving liquidity withdrawal shocks (f) increases,

the price gap (GP ) between the firm with an overconfident manager and firm with a rational manager

initially increases but eventually decreases. Mathematically, ∂GP∂f > 0 (< 0) for sufficiently small

(large) f .

Proposition 3 states that when VH1 − VH2 is sufficiently large, the price gap is increasing with

f . This is related to Proposition 2, where we show that VH1 − VH2 is related to the importance of

the learning effect. When the learning effect is sufficiently strong, the incentives for mutual funds to

sell shares of the firm with a rational manager are stronger than for the firm with the overconfident

manager. This increases the effect of the liquidity withdrawal shock (f) on the price gap.

Figure 3 illustrates Proposition 3. Two important observations can be made from the figure.

First, conditioning on ρ = g, the expected stock price E(P1) of the firm with an overconfident

manager is higher than that of the firm with a rational manager. For example, when f = 0, the

difference between the stock prices of the two firms is 0.095. When f increases from 0 to 0.3, the

price gap becomes larger. The gap decreases for higher values of f .

Figure 3: This figure provides a numerical example of Proposition 3. The horizontal axis is f and the vertical axisis GP . In this example, C = 3.25, e = 1, δ = 0.5, λ = 1, g = 2.5, q = 0.5, b = −0.5,µ = 2, and K = 1. In thisequilibrium, α∗,rb = α∗,ocb = 1.

3 Data

Our study combines several standard databases. We begin with all U.S. public firms with quarterly

accounting information in Compustat from 1992 to 2016. Data on CEO compensation are from the

Execucomp database; stock price information and mutual fund performance data are from CRSP;

and mutual fund holdings data are from Thomson Reuters. We collect corporate news information

from Factiva. After merging data from the above sources, we have an unbalanced panel with 88,106

19


firm-quarter observations for 2,712 companies that have at least one CEO overconfidence measure

available from 1992 to 2016.

3.1 Overconfidence measures

We use two measures of CEO overconfidence. The first measure is based on CEOs’ vested options

holdings (Malmendier and Tate, 2008), and the second one is a press-based measure (Hirshleifer,

Low, and Teoh, 2012).

The rationale for the first measure is as follows. CEOs’ human capital is extremely undiversified

and shareholders always deliberately tie a large part of CEOs’ wealth to firm performance to incen-

tivize CEOs to exert effort, so a rational CEO would be expected to exercise executive options when

they vest and to diversify by investing in assets not directly connected with the firm’s performance.

In this sense, CEOs’ holdings of vested in-the-money options exhibit a degree of overconfidence in

their ability and in the quality of the firms they manage (see, e.g., Malmendier and Tate, 2005a).6

We use the Execucomp database to construct this measure in the following steps. We first divide the

value of all unexercised vested options (Execucomp item named: opt unex exer est val) by the num-

ber of options (Execucomp item named: opt unex exer num). Next, we scale this value-per-option

by the average strike price, which is the stock price at the end of the fiscal year (Compustat item

named: prcc f) minus value-per-option. Intuitively, this scaled value-per-option measures the extent

to which a CEO retains already vested in-the-money options and is quite similar to the measure

constructed in Malmendier and Tate (2008).7 We follow Malmendier and Tate (2008) and define

a dummy variable that equals one if the ratio exceeds 0.67. For robustness, we also consider a

continuous variable of CEO overconfidence.

For our second measure of CEO overconfidence, we follow Hirshleifer, Low, and Teoh (2012)

and use manually collected data on how the press portrays CEOs in 2000, 2004, and 2006. We

first search for articles referring to CEOs in The New York Times (NYT), Business Week (BW),

Financial Times (FT), The Economist, Forbes Magazine, Fortune Magazine, and The Wall Street

6Malmendier and Tate (2005a, 2008) highlight that holding vested in-the-money options is indeed a behavioral bias,and they find no evidence that such options holdings support any actual private information. This is because thesecompanies tend to underperform the market rather than outperform, which would have been the case with actualprivate information. Further, although CEOs who choose to hold such options are arguably simply well incentivizedto do so, and so should perform better, such an interpretation is inconsistent with the finding both in this paper andin prior work (see, e.g., Malmendier and Tate, 2005a, 2008), that options-based measures of overconfidence are weaklynegatively associated with corporate performance.

7Our measure slightly differs from that in Malmendier and Tate (2008), because the Execucomp database does notprovide the same type of information on options holdings as does the proprietary database of Malmendier and Tate(2008).

20


Journal. For each CEO and sample year, we record the number of articles containing the words

“overconfident” or “overconfidence,” and the number of articles containing the words “optimistic”

or “optimism.” We also record the number of articles containing the words “reliable,” “cautious,”

“conservative,” “practical,” “frugal,” or “steady.” We manually check that these terms are generally

used to describe the CEO in question and separate newspaper articles describing the CEO of interest

as “not confident” or “not optimistic.” We then construct the variable “Net News,” which is equal

to the number of “confident” references minus the number of non-confident references. Finally, we

define a dummy variable that equals one if “Net News” is above the sample median. We use the

press-based measure starting from 2001 and carry over the measure into 2006 for the sample up to

2010. As such, our sample using the options-based (press-based) measure covers firm-year-quarter

observations from 1992 to 2016 (2001 to 2010).8

3.2 Mutual fund flow-induced selling

Following the literature (see, e.g., Coval and Stafford, 2007; Edmans, Goldstein, and Jiang, 2012;

Lou, 2012), we construct a mutual fund flow-induced selling measure to identify non-fundamental

shocks to stock prices due to investor redemption. Based on mutual fund holding data from Thomson

Reuters and fund return data from CRSP, excluding funds that specialize in a particular industry

to address the concern that the measure might reflect industry fundamentals, we construct the flow-

induced trading measure as follows:

FTi,t =S′i,t−1Ft

vi,t, (17)

where S′i,t−1 = (si,1,t−1, ..., si,j,t−1) is a 1× J vector of the number of stock i’s shares owned by each

mutual fund j at the beginning of quarter t, F ′t = (f1,t−1, ..., fj,t−1) is a 1 × J vector of percentage

of fund j’s outflow over quarter t, and vi,t is total share trading volume of stock i in quarter t. The

measure is summed over funds that have outflows greater than -5%. We winsorize fj,t−1 at −90%

to ensure that the measure is not driven by spurious data. In each year-quarter, we define stocks as

being under significant flow-induced selling if FTi,t is in the bottom decile of that quarter. 9

Intuitively, FT measures the projected price pressure due to large outflows from mutual funds

that hold the same stock. The construction of FT assumes mutual funds with outflows proportionally

8Our empirical results are similar if we use the press-based measure only for observations from 2001 to 2006.9 The cross-sectional ranking of FT could alleviate the impact of time trends of FT (see Table A1). This methodology

is also widely used in the literature (see, e.g., Anton and Polk, 2014).

21


liquidate their existing holdings. For example, if a mutual fund holds one million shares of a stock

at the beginning of the quarter and the fund is experiencing a 10% outflow, then FT assumes

that the fund will sell 100,000 shares. This assumption is justified in Lou (2012). Meanwhile, using

predicted trading rather than actual trading could largely mitigates the concern about mutual funds’

endogenous trading decisions, which naturally reflect firms’ fundamentals (Berger, 2017). We further

demonstrate that the differential price impact between those with and without overconfident CEOs

is not driven by endogenous trading decisions by mutual funds (see Section 5.2 ).

3.3 Control variables

In the regression analysis, we include various firm and CEO characteristics. Particularly, we control

for stock liquidity, which is a primary factor for price impact (see, e.g., Banerjee, Humphery-Jenner,

and Nanda, 2018; Chen, Goldstein, and Jiang, 2007; Gopalan, Kadan, and Pevzner, 2012; Mal-

mendier and Tate, 2005a). Firm characteristics include firm size, leverage, profitability, Tobin’s Q,

cash, tangibility, firm age, stock illiquidity (see, e.g., Amihud, 2002), and analyst coverage. CEO

characteristics include the CEO’s age, tenure, cash compensation, and ownership. Appendix A2

provides a detailed description of the variables.

3.4 Summary statistics

Table 1 reports summary statistics for the variables. In our sample, 37.3% of the observations

are classified as having an overconfident CEO based on the options-based measure. This is in line

with Malmendier and Tate (2005b), who categorizes 58 of the 113 firms in their sample as having

an overconfident CEO based on the same measure. Our press-based measure defines a CEO as

overconfident if “net news” is above the sample median. These two measures of CEO overconfidence

are positively correlated.

Table A1 in Appendix A3 reports the flow-induced selling for each year in our sample. As shown

in Table A1, FT varies over time and has increased in magnitude in recent years. This is consistent

with the increasing trend in institutional ownership over the past decades.

22


4 Empirical results

4.1 Overconfident CEOs and market learning

We start by justifying the premise that overconfident CEOs are less likely to learn from the stock

market in making corporate policies. We conduct tests in the context of capital and R&D investments

(Chen, Goldstein, and Jiang, 2007) as well as merger and acquisitions(Kau, Linck, and Rubin, 2008;

Luo, 2005).

Our test of firm investment is motivated by Chen, Goldstein, and Jiang (2007). Chen, Goldstein,

and Jiang (2007) document that firms’ investment-Q sensitivity increases with measures of price

informativeness, consistent with the notion that firm managers learn from stock prices when making

investment decisions. Following a similar intuition, we conjecture that overconfident firm managers

underweight the information in stock prices, resulting in lower investment-Q sensitivity. To test this

conjecture, we follow the approach of Chen, Goldstein, and Jiang (2007) and run the following panel

regression:

Investmenti,t+1 =α1 + β1Qi,t + β2Qi,t × Confidencei,t + β3Confidencei,t

+ γ′1CONTROLi,t + φi + ψt + εi,t, (18)

where Investmenti,t+1 is firm i’s total investment in year-quarter t + 1, defined as the sum of

capital expenditure and R&D expenditure divided by the average of current and lagged total assets.

Confidencei,t is the overconfidence measure of firm i’s CEO in year-quarter t. Control variables

include firm and CEO characteristics, including ln(Assets), Leverage, Profitability, Cash, Tangibility,

ln(Firm Age), ln(CEO Age), ln(CEO Tenure), Cash Compensation, and CEO Ownership. ψt denotes

year-quarter fixed effects. We also control for firm fixed effects in the regressions (φi). If overconfident

managers are less responsive to stock prices than are other managers, we would expect β2 to be

significantly negative and β1 to be significantly positive.

Table 2 reports the results. We make two observations. First, consistent with Chen, Goldstein,

and Jiang (2007), investment is positively related to Tobin’s Q, suggesting that managers’ investment

policies are generally in line with the market’s view about investment opportunities. Second, we find

that firms with overconfident CEOs exhibit significantly lower investment-Q sensitivity compared

with other firms. For example, as shown in column 1 with options-based CEO overconfidence, the

estimate of β1 is 0.415, and the estimate of β2 is -0.131. We find similar results when using the

press-based measure of overconfidence (see column 2).

23


To confirm the robustness of our finding, we consider alternative regression specifications in

columns 3 and 4. Specifically, we divide the sample into terciles based on the two measures of

overconfidence. We then interact Tobin’s Q with dummy variables that indicate firms with high,

medium, or low levels of CEO overconfidence. We find that investment-Q sensitivity monotonically

decreases with the level of CEO overconfidence.

Next, we test the market-learning hypothesis in the context of mergers and acquisitions. This

test is motivated by Luo (2005) and Kau, Linck, and Rubin (2008). Luo (2005) and Kau, Linck,

and Rubin (2008) find that managers learn information from the market when making acquisition

decisions: they are more likely to cancel a M&A deal when the market negatively reacts to the

announcement. Similar to the argument for firms’ investment, we conjecture that overconfident

CEOs respond less to announcement-day returns of M&A deals. To test this hypothesis, we collect

all takeover attempts for majority ownership from the Securities Data Company (SDC). Following

Edmans, Goldstein, and Jiang (2012), we exclude acquisitions of partial stakes, minority squeeze-

outs, buybacks, recapitalizations, and exchange offers. In addition, we only keep bids for which

the acquirers had a stake under 50% before the acquisition and end with final ownership over 50%.

After merging acquiring firms’ accounting information with the data on CEO overconfidence, the

final sample consists of 15,196 M&A announcements from 1992 to 2016. We follow Kau, Linck, and

Rubin (2008) and estimate a probit model with a deal withdrawal decision as below:

Pr(Cancel)k = Φ(α1 + β1CARk + β2CARk × Confidencek

+ β3Confidencek + γ′1CONTROLk + φj + ψt + εk), (19)

where Pr(Cancel)k is the probability that M&A attempt k is eventually canceled. CAR is the

market-adjusted cumulative abnormal return from 1 day before to 1, 5, or 10 days after the M&A

announcements. CONTROL refers to deal-level control variables, including Tender Offer Dummy,

Compete Dummy, Litigation Dummy, Lockup of Target Shares Dummy, Target Termination Fee

Dummy, Defense Dummy, Friendly Dummy, Public Target Dummy, Toehold Dummy, and Toehold

Shares; firm-level control variables, including ln(Assets), Leverage, Profitability, Q, Cash, Tangibil-

ity, and ln(Firm Age); and CEO characteristics, including ln(CEO Age), ln(CEO Tenure), Cash

Compensation, and ln(CEO Ownership).10 We also include industry fixed effects (φj) and year fixed

effects (ψt). If non-overconfident managers are more likely than overconfident managers to cancel

an M&A deal in response to a negative stock market reaction, we expect that β1 (β2) should be

10Appendix A3 provides a detailed description of the deal control variables.

24


significantly negative (positive).

As reported in Table 3, the coefficient of CAR is significantly negative, whereas the coefficient

of the interaction term between CAR and Confidence is significantly positive. For example, column

6 shows that a one-percentage-point decrease in the 10-day return after a M&A announcement is

associated with an increase in the likelihood of withdrawal by 0.093 percentage points (or 3.9% from

the mean) for firms with non-overconfident CEOs. However, the announcement return has a near-

zero effect on the likelihood of withdrawal for firms with overconfident CEOs.11 The finding further

justifies our premise that overconfident managers tend to place less weight than non-overconfident

managers on the views (or information) of market participants, as reflected in their firms’ stock price.

Finally, we examine whether managers (overconfident vs. non-overconfident managers) react

differently to non-fundamental price movements from mutual-fund-flow-induced selling. We focus on

firm-year-quarter observations in which the stock is subject to a strong fire sale pressure by mutual

funds (i.e., when FT is in the bottom decile of that year-quarter) and examine the relation between

the change in corporate investment and the change in stock prices around fire sales using the following

regression:

∆Investmenti,t,t+4 =α1 + β1∆Qi,t−1,t+1 + β2∆Qi,t−1,t+1 × Confidencei,t−1 + β3Confidencei,t−1

+ γ′1CONTROLi,t−1 + φj + ψt + εi,t. (20)

As shown in Table 4, the coefficient of ∆Qt−1,t+1 is significantly positive, suggesting that the

stock price declines during mutual fund fire sales are associated with subsequent decreases in corpo-

rate investment. This finding is consistent with our argument that managers adjust their investment

policies in response to non-fundamental shocks to stock prices. Interestingly, the coefficient of the

interaction term, ∆Qt−1,t+1 × Confidencet−1, is significantly negative. This suggests that overcon-

fident managers’ investment policies are less responsive to stock price drops caused by mutual fund

fire sales.

4.2 Overconfident CEOs and the price impact of mutual fund fire sales

Our model shows that firm managers’ learning from stock prices can amplify the price impact of

liquidity shocks on stock prices. When some non-fundamental shocks decrease stock prices, firm

managers potentially cut investment and, consequently, firm value decreases, which in turn induce

11The coefficient for Litigation Dummy perfectly predicts no withdrawal in the sample using the press-based measureof overconfidence and thus is not estimated in columns 4 to 6.

25


selling from liquidity-unconstrained investors. A key implication, therefore, is that the price impact

of a liquidity shock depends on the degree to which a firm’s manager responds to stock prices.

Following the finding that overconfident CEOs are less responsive to stock price movements when

making corporate policies (see Section 4.1 ), in this section, we formally show that overconfidence as

a CEO attribute can mitigate the price impact of fund-flow-induced fire sales on stock prices.

To visually illustrate the effect of CEO overconfidence on the price impact of mutual fund fire

sales, we plot the cumulative abnormal returns (CARs) around mutual fund fire sales on firms with

and without overconfident CEOs. We define a stock in a quarter as experiencing fire sales if FT is in

the bottom decile within the quarter. Then we divide the sample into firms with overconfident and

non-overconfident CEOs. While we consider the options-based measure of CEO overconfidence in

panel A, we consider the media-based measure of CEO overconfidence in panel B. As shown in both

figures, firms with overconfident CEOs have significantly less negative CARs during liquidity shocks.

For example, whereas stocks with non-overconfident CEOs have a close to 1.5% drop in value in the

quarter of mutual fund fire sales, stocks with overconfident CEOs experience a less than 1% drop in

firm value, followed by a quick recovery. These patterns support the model’s prediction that firms

with overconfident CEOs are less vulnerable to non-fundamental price shocks (e.g., mutual fund fire

sales).

To formally test the model’s prediction, we run the following multivariate regression:

CARi,t =α1 + β1Fire Salei,t + β2Fire Salei,t × Confidencei,t−1

+ β3Confidencei,t−1 + γ′1CONTROLi,t−1 + φj + ψt + εi,t, (21)

where CARi,t is the average monthly abnormal return in quarter t adjusted by benchmark portfolios

matched on size (two groups), book-to-market (three groups), and momentum (three groups) (See,

e.g., Fama and French, 1992; Hong, Kubik, and Fishman, 2012). Fire Sale is a dummy variable that

equals one when FT is in the bottom decile in quarter t. CONTROL refers to a set of lagged control

variables, including ln(Assets), Leverage, Profitability, Cash, Tangibility, ln(Firm Age), ln(CEO Age),

ln(CEO Tenure), Cash Compensation, ln(CEO Ownership), ln(Illiquidity), and ln(Analysts). φj and

ψt refer to industry and year-quarter fixed effects, respectively. If firms with overconfident CEOs are

less affected by mutual fund fire sales, we expect that β2 should be significantly positive.

Table 5 reports the results. The estimates in column 2 show that, for the average firm without

an overconfident CEO, the market value declines by 2.1% in the quarter of mutual fund fire sales.

26


More importantly, we find that the coefficient of the interaction term Fire Salet × Confidencet−1

is significantly positive. This suggests that the price impact of mutual fund fire sales is smaller for

firms with overconfident CEOs. In terms of economic magnitude, firms with overconfident CEOs

experience only a 0.6% price drop in the quarter of fire sales (the sum of β1 and β2 in column 2).

These results mirror the patterns in Figure 4.

A potential explanation for our finding in Table 5 is that firms with overconfident and non-

overconfident managers have different characteristics, such as in stock liquidity. To address this

concern, we control for various firm characteristics potentially related to firms’ price declines during

fire sales, such as stock illiquidity, proxied by the Amihud’s (2002) measure and analyst coverage.

In addition, we use the propensity score matching (PSM) approach to control for various observable

firm characteristics. Specifically, among all firm-quarter observations with mutual fund fire sales (i.e.,

those with FT in the bottom decile), we estimate the propensity score for having an overconfident

CEO (the treatment group) based on firm and manager characteristics in Model (21). We then

match each firm with the nearest neighbor from the control group (without overconfident CEOs)

based on the propensity score.

Table 6 reports the PSM results. As shown in Panels A and B, the difference in various character-

istics, especially stock illiquidity and analyst coverage, between firms with and without overconfident

CEOs is insignificant after matching. In this sense, the PSM approach successfully controls for the

heterogeneity in observable firm characteristics. The kernel density distributions of the propensity

scores presented in Figures A1a and A1b in Appendix A3 also confirm the common support con-

dition. Within the matched sample, we find that firms with overconfident CEOs still significantly

have less negative price impacts during mutual fund fire sales compared with other firms. It is worth

noting that since we perform matching within observations in which severe outflows have occurred,

the PSM results reassure us that endogenous fund flow (Berger, 2017) is unlikely to explain our

findings.

Overall, we find strong evidence that stocks with overconfident CEOs are less vulnerable to

mutual fund fire sales. This is consistent with our model’s prediction that the learning behavior of

firm managers could amplify the impact of liquidity shocks in financial markets.

4.3 Natural experiment: Mutual fund late trading scandal of 2003

Our empirical analysis has thus far used flow-induced selling (FT ) to identify mutual fund fire sales.

The identifying assumption is that (large) fund outflows reflect exogenous liquidity shocks to mutual

27


funds, probably because of investors’ demand for liquidity. However, one potential concern is that

fund flows are endogenous to the fundamental value of the stocks in the mutual funds’ portfolios.

To address the endogeneity issue, we use the mutual fund scandal of 2003–2004 as a plausibly

exogenous shock to mutual fund flow (see, e.g., Anton and Polk, 2014; Koch, Ruenzi, and Starks,

2016). Following Zitzewitz (2009) and McCabe (2009), we identify 19 mutual fund families that were

publicly implicated in late trading practices from September 2003 to the end of 2004. These funds

experienced approximately 20% outflow and had to sell stocks in their portfolio in the year following

the revelation of their wrongdoing (Choi and Kahan, 2007). As argued in prior studies (see, e.g.,

Anton and Polk, 2014; Koch, Ruenzi, and Starks, 2016), the stock selling from funds involved in this

scandal was unlikely to have any bearing on firm fundamentals. Using the mutual fund scandal, we

measure a stock’s exposure to implicated fund families based on their ownership as of June 2003 and

examine the impact of implicated funds on stock returns from September 2003 to December 2004.

Table 7 reports the results. We first find that stocks with higher implicated fund ownership in

2003 experienced more negative stock returns in the following year. As shown in column 2, among

firms with non-overconfident CEOs, a 10-percentage-point increase in implicated funds ownership

is related to 2.3-percentage-point lower return per quarter. More importantly, we find that stocks

with overconfident CEOs do not suffer price drops in the aftermath of the trading scandal. These

results provide strong and causal evidence that firms with overconfident CEOs are less susceptible

to mutual fund fire sales.

4.4 Managerial market learning and the externality of fire sales on other in-

vestors

Thus far, we have shown rich evidence that the overconfidence of firm managers could mitigate the

negative price impact of mutual fund fire sales. As our theoretical model shows, the key mechanism

lies in the firm managers’ learning behavior. When firm managers learn information from stock prices,

they respond to mutual fund fire sales in a way that potentially hurts the firm’s fundamental value.

This will in turn decrease the incentives of liquidity-unconstrained investors to trade against the fire

sales (i.e., not holding or even selling the firm’s shares). However, the inattention of overconfident

CEOs to stock market movements can mitigate the negative price impacts of mutual fund fire sales.

To corroborate the evidence for our theory, we directly test this mechanism by examining the trading

behavior of the unaffected mutual funds, that is, those funds not subject to redemption shocks.

28


Specifically, we run the following model:

Other Fund Tradei,t =α1 + β1Fire Salei,t + β2Fire Salei,t × Confidencei,t−1

+ β3Confidencei,t−1 + γ′1CONTROLi,t−1 + φj + ψt + εi,t, (22)

where Other Fund Tradei,t is the net trades by mutual funds without large out flows (-5%) divided by

total trading volume in quarter t. Similar to Model (21), CONTROL denotes a set of lagged control

variables, including ln(Assets), Leverage, Profitability, Cash, Tangibility, ln(Firm Age), ln(CEO Age),

ln(CEO Tenure), Cash Compensation, and ln(CEO Ownership). φj and ψt refer to industry and

year-quarter fixed effects, respectively.

Table 8 reports the results. We find strong evidence that liquidity-unconstrained mutual funds are

not willing to trade against the fire sale among firms with non-overconfident CEOs. The coefficient

of Fire Sale is insignificant, suggesting that investors do not trade against fire sales for firms without

overconfident CEOs. Consistent with our model’s prediction, the coefficient of Fire Sale×Confidence

is significantly positive, suggesting that investors indeed trade against fire sales on firms with over-

confident CEOs.

The finding in Table 8 demonstrates that the learning channel reduces or even reverses the incen-

tive of liquidity-unconstrained mutual funds to trade against fire sales. The diminished incentives to

trade against fire sales among firms with non-overconfident managers could explain the significant

price impact of fire sales and the slow recovery among these firms (see Figure 4).

5 Additional analyses and robustness checks

In this section, we conduct further studies to strengthen our argument. First, we consider and

rule out several alternative explanations for the observed differential price impacts of mutual fund

fire sales between firms with and without overconfident managers. Second, we conduct additional

robustness checks of the measurement error in our key variables (e.g., fire sales). We highlight the

main results in this section and report the detailed description in Appendix A3 for brevity.

29


5.1 Alternative explanations

5.1.1 Proactive actions against non-fundamental price shocks

First, overconfident CEOs could possibly be more concerned about the price impact of mutual fund

fire sales and hence take more aggressive actions against the fire sales. For example, firms could

repurchase shares or voluntarily disclose information to mitigate the impact of non-fundamental

shocks. We test this possibility by estimating the following model:

Actioni,t+1 =α1 + β1Fire Salei,t + β2Fire Salei,t × Confidencei,t−1

+ β3Confidencei,t−1 + γ′1CONTROLi,t−1 + φi + ψt + εi,t. (23)

We consider two types of actions: share repurchase (e.g., Banerjee, Humphery-Jenner, and Nanda,

2018) and voluntary disclosure.

To examine whether overconfident CEOs respond to mutual fund fire sales by repurchasing stocks,

we use the natural logarithm of firms’ stock repurchase over the future four quarters as the dependent

variable for Model (23). If overconfident CEOs are more likely to repurchase stocks in response to

mutual fund fire sales, then β2 should be significantly positive. However, the estimates in columns 1

and 2 of Table A2 show that this is not the case.

Next, to examine whether overconfident CEOs voluntarily disclose information to facilitate price

recovery, we directly test whether overconfident CEOs are more likely to issue managerial earnings

forecasts in response to mutual fund fire sales in Model (23). We use the log number of earnings

forecasts in quarter t + 1 as the dependent variable. The estimates in columns 3 and 4 of Table

A2 show that overconfident CEOs are less likely to issue earnings forecasts in response to mutual

fund fire sales. In other words, the muted price impact of mutual fund fire sales among firms with

overconfident CEOs is unlikely explained by the disclosure policies of overconfidence CEOs.

5.1.2 Financing through the primary market

A second potential explanation is that fire sales reduce firm value by decreasing firms’ abilities in

raising equity capital through the primary market. Overconfident CEOs may be less affected if, for

instance, they always perceive stock prices to be undervalued and thus rely less on equity capital in

the first place. To rule out this possibility, we first examine equity issuance after mutual fund fire

sales.

In Table A3, we estimate Model (23) with firms’ Net Equity Issuance as the dependent variable.

30


The estimates show that whereas the average firm experiences a decrease in equity financing after

mutual fund fire sales, those with overconfident CEOs do not behave differently from the average

firms in terms of equity financing. This suggests that the primary market financing channel is unlikely

to explain the observed differential price impact of mutual fund fire sales.

5.1.3 Corporate governance

A third alternative explanation is that fire sales reduce blockholders’ ownership, which in turn

impairs firm value as it weakens the governance role of blockholders (Edmans, 2009). Firms with

overconfident CEOs may be less affected if they are less responsive to governance in the first place.

We formally conduct a test to address this concern by controlling for the G index of Gompers, Ishii,

and Metrick (2003) as a proxy for corporate governance in Model (21). This index reflects the level

of takeover defenses, and, thus, a higher index value generally reflects weaker corporate governance.

The limited availability of the data on the G index, results in a smaller sample. Nevertheless, the

results in Table A4 show that the difference in the price impact of mutual fund fire sales between

firms with and without overconfident CEOs is significant and is even larger after controlling for G

index. Hence, corporate governance is unlikely to explain the weaker price impact of mutual fund

fire sales on firms with overconfident CEOs.

5.2 Additional robustness checks

In this subsection, we conduct several additional robustness tests of the measurement error in our

key variables.

Table A5 shows that our baseline result is robust to alternative definitions of fire sales. Specifically,

we measure fire sales using the FIT measure by Lou (2012) and the modified fire sales measure by

Wardlaw (2018).12 Using these alternative definitions of fire sales, the cross-sectional difference

between firms with and without overconfident CEOs remains significant.

Although Wardlaw (2018) has raised concerns about the empirical measurement of mutual fund

fire sales in the literature, his arguments cannot explain the cross-sectional variation in price impact

of fire sales that we document, because our results hold within a set of firms that has similar observed

firm characteristics, including liquidity (see PSM results). More importantly, we show that our results

12Both of these measures project mutual funds’ trades based on fund flows and lagged stock holding, and benchmarkthe projected trading volume against the number of shares outstanding. The key difference between the two measuresis that, whereas Lou (2012) aggregate flow-induced trading across all funds, Wardlaw (2018) only use funds withsignificant outflows as in Edmans, Goldstein, and Jiang (2012). Lou (2012) also accounts for asymmetry in the extentto which funds may buy (sell) in response to inflows (outflows).

31


also hold in the context of the mutual fund trading scandal, which is a plausible exogenous shock to

mutual fund flows. Hence, our finding is robust to alternative measures of mutual fund fire sales.

We also provide further analysis in light of the discussion by Berger (2017) of the selection bias in

mutual fund fire sales. Specifically, there are two layers of potential selection biases here. First, large

fund outflows might be endogenous to the performance of the underlying stocks. Second, when funds

experience large fund outflows, they may sell their underlying stocks selectively rather than sell their

holdings proportionally. We focus on stocks that have already experienced large fund outflows, so

the first layer of selection bias is less of a concern. However, the second layer of selection bias could

explain our findings if liquidity constrained mutual funds choose not to sell stocks with overconfident

CEOs for unobservable reasons. To address this concern, we examine mutual funds’ selling behavior

when faced with large fund outflows.

To conduct the test, we first calculate the mutual funds’ selling as follows:

MFSalei,t =m∑j=1

min(Tradei,j,t, 0)

vi,t, (24)

where Tradei,j,t is the number of shares of stock i traded by fund j in quarter t, conditional on the

fund having a greater than 5% outflow. Here, vi,t is the total share trading volume of stock i in

quarter t. We only focus on sales by these funds to capture flow-induced selling. Using this measure,

we test whether mutual funds selectively sell stocks using the following model:

MFSalei,t =α1 + β1FTi,t + β2FTi,t × Confidencei,t−1

+ β3Confidencei,t−1 + γ′1CONTROLi,t−1 + φi + εi,t. (25)

Table A6 reports the results. The coefficient on FT is significantly positive, suggesting that

mutual funds experiencing large fund outflows indeed sell a significant amount of their existing

holdings. Importantly, the estimated coefficient of the interaction term FT ×Confidence is positive,

suggesting that mutual funds with large outflows are more likely to sell stocks with overconfident

CEOs. Hence, mutual funds’ discretion in selling should bias us against finding a muted price drop

for firms with overconfident CEOs.

Next, we show that our findings are robust to alternative measures of abnormal returns. We

perform the baseline analysis using cumulative abnormal returns adjusted by benchmark portfolios

matched on size, the book-to-market ratio, and momentum. As shown in Table A7, our results are

32


unchanged when we adjust stock returns by value-weighted or equally weighted market returns.

Finally, we address the potential concern that our overconfident CEO measures capture unob-

servable firm characteristics (e.g., managers’ private positive information), rather than managers’

propensity to learn from the market. We do this by looking beyond overconfident CEOs and using

a more general measure of market learning: investment-Q sensitivity. All else equal, firm managers

who learn information from stock prices when making investment decisions should exhibit greater

investment-Q sensitivity (Chen, Goldstein, and Jiang, 2007). We, therefore, attempt to measure

investment-Q sensitivity at the firm level and use it as a proxy for market learning. For each firm i,

we run the following time-series regression over the sample period:

Investmenti,t+1 =αi,1 + βi,1Qi,t + βi,2Sizei,t + βi,3Profitabilityi,t + ψi,q + εi,t. (26)

ψi,q refers to fiscal and calendar quarter fixed effects, which are included to control for seasonality in

investment. The coefficient βi,1 clearly measures investment-Q sensitivity. After that, we define a firm

as having significant investment-Q sensitivity if the estimated βi,1 is in the top tercile and is significant

at the 5% level. Using the investment-Q sensitivity to measure firm managers’ learning behavior,

we expect that firms that have significant investment-Q sensitivity should experience greater price

drops during mutual fund fire sales. As reported in Table A8, stocks with significant investment-Q

sensitivity indeed experience greater price drops during mutual fund fire sales. This is consistent

with our main result, where we use CEO overconfidence to measure managers’ propensity to learn

information from stock prices.

We note that the estimated investment-Q sensitivity is a coarse proxy of market learning for

several reasons. First, many factors other than market learning can affect investment-Q sensitivity.

For example, financially constrained and equity-dependent firms may also exhibit greater investment-

Q sensitivity (Baker, Stein, and Wurgler, 2003). Second, given the limited observations on the

investment variable, we have to estimate firm-level investment-Q sensitivity using the entire time

series. Thus, the measure is time invariant and subject to measurement error or forward-looking

bias. For these reasons, we believe that CEO overconfidence, which is an exogenous behavioral bias,

is a more intuitive measure of managers’ learning behavior.

33


6 Conclusion

We study how market learning amplifies non-fundamental shocks in the financial markets. We

argue that the learning channel, whereby managers learn information from the stock prices, can

amplify the impact of liquidity shocks by exerting externalities on other liquidity-unconstrained

market participants. To support this argument, we develop a model of fire sales and test its empirical

implications. In the model, we show that firm managers’ behavioral traits (overconfident or rational)

play an important role in the price impact of liquidity shocks. Rational managers may mistakenly

cut real investment in response to fire sales from mutual funds with liquidity shocks and hence

cause a large drop in the firm’s fundamental value. This diminished firm value deters liquidity

provision and even induces selling from other liquidity-unconstrained mutual funds. In contrast,

overconfident managers tend to be unresponsive to the stock market and to maintain their investment

policies. Their inattention to stock market information encourages other unconstrained mutual funds

to provide liquidity in the presence of mutual funds fire sales.

We further provide strong empirical evidence to support the model’s assumptions and predictions.

First, we justify the premise that overconfident CEOs are less responsive to stock price fluctuations

than are others. Supportive of our model’s predictions, we find that stocks with overconfident CEOs

experience smaller price impacts from fire sales induced by large fund outflows or by trading scandal

shocks. These results demonstrate that the learning channel can amplify the price impact of liquidity

shocks and that managerial overconfidence, associated with less attention to stock prices, can play a

beneficial role in stabilizing stock market prices during market turmoil.

34


References

Amador, M., Weill, P.-O., 2010. Learning from prices: Public communication and welfare. Journal of Political Economy118 (5), 866–907.

Amihud, Y., Jan. 2002. Illiquidity and stock returns: Cross-section and time-series effects. Journal of Financial Markets5 (1), 31–56.

Anton, M., Polk, C., 2014. Connected stocks. The Journal of Finance 69 (3), 1099–1127.

Baker, M., Stein, J. C., Wurgler, J., Aug. 2003. When Does the Market Matter? Stock Prices and the Investment ofEquity-Dependent Firms. The Quarterly Journal of Economics 118 (3), 969–1005.

Banerjee, S., 2011. Learning from prices and the dispersion in beliefs. The Review of Financial Studies 24 (9), 3025–3068.

Banerjee, S., Humphery-Jenner, M., Nanda, V., 2015. Restraining Overconfident CEOs through Improved Governance:Evidence from the Sarbanes-Oxley Act. Review of Financial Studies 28 (10), 2812–2858.

Banerjee, S., Humphery-Jenner, M., Nanda, V., Feb. 2018. Does CEO Bias Escalate Repurchase Activity? Journal ofBanking & Finance.

Banerjee, S., Kremer, I., 2010. Disagreement and learning: Dynamic patterns of trade. The Journal of Finance 65 (4),1269–1302.

Berger, E., 2017. Does Stock Mispricing Drive Firm Policies? Mutual Fund Fire Sales and Selection Bias.

Bond, P., Edmans, A., Goldstein, I., 2012. The Real Effects of Financial Markets. Annual Reviews of FinancialEconomics 4, 339–360.

Bond, P., Goldstein, I., Prescott, E., 2010. Market-based corrective actions. Review of Financial Studies 23 (2), 781–820.

Chen, Q., Goldstein, I., Jiang, W., 2007. Price Informativeness and Investment Sensitivity to Stock Price. Review ofFinancial Studies 20, 620–650.

Choi, S., Kahan, M., 2007. The market penalty for mutual fund scandals. BUL Rev. 87, 1021.

Coval, J., Stafford, E., 2007. Asset fire sales (and purchases) in equity markets. Journal of Financial Economics 86,479–512.

Dow, J., Goldstein, I., Guembel, A., 2015. Incentives for information production in markets where prices affect realinvestment. Working Paper.

Dow, J., Gorton, G., 1997. Stock market efficiency and economic efficiency: Is there a connection? Journal of Finance52 (3), 1087–1129.

Dow, J., Rahi, R., 2003. Informed trading, investment, and economic welfare. Journal of Business 76 (3), 439–54.

Edmans, A., Goldstein, I., Jiang, W., 2012. The Real Effects of Financial Markets: The Impact of Prices on Takeovers.Journal of Finance 67 (3), 933–971.

Ellul, A., Jotikasthira, C., Lundblad, C., 2011. Regulatory pressure and fire sales in the corporate bond market. Journalof Financial Economics 101 (3), 596–620.

Eyster, E., Rabin, M., Vayanos, D., 2019. Financial markets where traders neglect the informational content of prices.The Journal of Finance 74 (1), 371–399.

Fajgelbaum, P. D., Schaal, E., Taschereau-Dumouchel, M., 2017. Uncertainty traps. The Quarterly Journal of Eco-nomics 132 (4), 1641–1692.

Fama, E. F., French, K. R., 1992. The cross-section of expected stock returns. the Journal of Finance 47 (2), 427–465.

Foucault, T., Fresard, L., 2014. Learning from Peers’ Stock Prices and Corporate Investment. Journal of FinancialEconomics 111, 554–577.

35


Foucault, T., Gehrig, T., 2008. Stock price informativeness, cross-listings, and investment decisions. Journal of FinancialEconomics 88 (1), 146–68.

Galasso, A., Simcoe, T. S., 2011. CEO overconfidence and innovation. Management Science 57 (8), 1469–1484.

Goldstein, I., Guembel, A., 2008. Manipulation and the allocational role of prices. Review of Economic Studies 75 (1),133–64.

Goldstein, I., Ozdenoren, E., Yuan, K., 2013a. Learning and complementarities in speculative attacks. Review ofFinancial Studies 78 (1), 263–92.

Goldstein, I., Ozdenoren, E., Yuan, K., 2013b. Trading frenzies and their impact on real investment. Journal of FinancialEconomics 109 (2), 566–582.

Goldstein, I., Yang, L., 2019. Good disclosure, bad disclosure. Journal of Financial Economics 131 (1), 118–138.

Gopalan, R., Kadan, O., Pevzner, M., 2012. Asset liquidity and stock liquidity. Journal of Financial and QuantitativeAnalysis 47 (2), 333–364.

Graham, J. R., Harvey, C. R., Puri, M., 2013. Managerial attitudes and corporate actions. Journal of FinancialEconomics 109 (1), 103–121.

Hirshleifer, D., Low, A., Teoh, S. H., 2012. Are Overconfident CEOs Better Innovators? Journal of Finance 67 (4),1457–1498.

Hong, H., Kubik, J. D., Fishman, T., 2012. Do arbitrageurs amplify economic shocks? Journal of Financial Economics103 (3), 454–470.

Johnson, D., Fowler, J., 2011. The evolution of overconfidence. The Nature 477, 317–320.

Kau, J., Linck, J., Rubin, P., 2008. Do managers listen to the market? Journal of Corporate Finance 14 (1), 347–362.

Koch, A., Ruenzi, S., Starks, L., 2016. Commonality in liquidity: A demand-side explanation. The Review of FinancialStudies 29 (8), 1943–1974.

Kyle, A. S., 1985. Continuous auctions and insider trading. Econometrica 53, 1315–1336.

Lou, D., 2012. A flow-based explanation for return predictability. The Review of Financial Studies 25 (12), 3457–3489.

Luo, Y., 2005. Do insiders learn from outsiders? Evidence from mergers and acquisitions. The Journal of Finance60 (4), 1951–1982.

Malmendier, U., Tate, G., 2005a. CEO Overconfidence and Corporate Investment. Journal of Financial Economics60 (6), 2661–2700.

Malmendier, U., Tate, G., 2008. Who Makes Acquisitions? CEO Overconfidence and the Market’s Reaction. Journalof Financial Economics 89, 20–43.

Malmendier, U., Tate, S., 2005b. Does Overconfidence Affect Corporate Investment? CEO Overconfidence MeasuresRevisited. European Financial Management, 649–59.

McCabe, P. E., 2009. The economics of the mutual fund trading scandal.

Ordonez, G. L., 2009. Larger crises, slower recoveries: The asymmetric effects of financial frictions.

Pulvino, T., 1998. Do Fire-Sales Exist? An Empirical Study of Commercial Aircraft Transactions. Journal of Finance53 (3), 939–78.

Puri, M., Robinson, D. T., 2007. Optimism and economic choice. Journal of Financial Economics 86 (1), 71–99.

Simsek, Z., Heavy, C., Veiga, J. F., 2010. The impact of CEO core self-evaluation on the firm’s entrepreneurialorientation. Strategic Management Journal 31 (1), 110–119.

Subrahmanyam, A., Titman, S., 1999. The going public decision and the development of financial markets. Journal ofFinance 54 (3), 1045–82.

36


Subrahmanyam, A., Titman, S., 2001. Feedback from stock prices to cash flows. Journal of Finance 56 (6), 2389–2413.

Van Nieuwerburgh, S., Veldkamp, L., 2006. Learning asymmetries in real business cycles. Journal of monetary Eco-nomics 53 (4), 753–772.

Veldkamp, L. L., 2005. Slow boom, sudden crash. Journal of Economic theory 124 (2), 230–257.

Wardlaw, M., Oct. 2018. Measuring Mutual Fund Flow Pressure As Shock to Stock Returns. SSRN Scholarly PaperID 3248750, Social Science Research Network, Rochester, NY.

Zitzewitz, E. W., 2009. Prosecutorial discretion in mutual fund settlement negotiations, 2003-7. The BE Journal ofEconomic Analysis & Policy 9 (1).

37


Table 1: Summary Statistics

This table shows the summary statistics of all the variables. We include the sample average, median, 25th,

75th percentiles, and standard deviations of all the variables of interest as well as the control variables.

N Mean P25 Median P75 Std. Dev.

FT 88,063 -0.616 -0.776 -0.283 -0.021 0.946Confidence Options 79,826 0.373 0.000 0.000 1.000 0.484Confidence Press 32,061 0.486 0.000 0.000 1.000 0.500Ln(Assets) 88,106 7.321 6.109 7.192 8.387 1.681Leverage 88,106 0.228 0.053 0.212 0.344 0.192Profitability 88,106 0.036 0.021 0.035 0.051 0.032Q 88,106 1.998 1.157 1.554 2.292 1.366Cash 88,106 0.154 0.024 0.080 0.220 0.179Tangibility 88,106 0.287 0.096 0.213 0.433 0.238Ln(Firm Age) 88,106 2.737 2.181 2.856 3.466 1.018Ln(CEO Age) 88,106 4.006 3.932 4.007 4.094 0.129Ln(CEO Tenure) 88,106 4.021 3.401 4.174 4.804 1.090Cash Compensation 88,106 0.391 0.172 0.313 0.554 0.273Ln(CEO Ownership) 88,106 0.628 0.101 0.297 0.825 0.784Ln(Illiquidity) 86,472 0.691 -0.842 0.583 2.079 2.188Ln(Analysts) 88,106 1.818 1.386 1.946 2.485 0.876Ln(Repurchase) 77,713 2.323 0.000 1.649 4.304 2.448Ln(Guidance) 83,550 0.709 0.000 0.693 1.386 0.743Investment (percent) 88,106 2.383 0.771 1.662 3.111 2.434Withdrawal of Acquisition 15,244 0.024 0.000 0.000 0.000 0.152

38


Table 2: CEO Overconfidence and Investment-Q Sensitivity

This table reports estimates from regressions (Model (18)) where the dependent variable is Investment in quar-

ter t + 1. In columns 1 and 2, interactions between Q and binary variables Confidence Options or Confidence

Press are included in the regressions. In columns 3 and 4, the sample is divided into terciles based on the

continuous measures of two overconfidence proxies. We then include the interactions between Q and dummy

variables indicating firms with high/medium/low CEO overconfidence. Control variables include Ln(Assets),

Leverage, Profitability, Cash, Tangibility, Ln(Firm Age), Ln(CEO Age), Ln(CEO Tenure), Cash Compensation, and

Ln(CEO Ownership). Firm and year-quarter fixed effects are included. t-statistics using firm-clustered standard

errors are shown in brackets. *, ** and *** indicate significance better than 10%, 5%, and 1% respectively.

Dependent Variable: Investment i,t+1

Overconfidence Measure: Options Press Options Press

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

Q × Confidence -0.131*** -0.121***(-4.88) (-2.76)

Confidence 0.348*** 0.232***(6.47) (2.78)

Q 0.415*** 0.385***(12.95) (8.51)

Q × High Confidence 0.281*** 0.259***(12.06) (6.11)

Q × Medium Confidence 0.338*** 0.309***(11.03) (6.17)

Q × Low Confidence 0.491*** 0.391***(12.17) (8.38)

High Confidence 0.200*** 0.106(3.98) (1.03)

Low Confidence -0.315*** -0.181*(-6.05) (-1.84)

Ln(Assets) -0.507*** -0.528*** -0.505*** -0.532***(-10.22) (-6.78) (-10.19) (-6.79)

Leverage -1.429*** -1.562*** -1.427*** -1.563***(-8.24) (-5.96) (-8.29) (-5.92)

Profitability 0.702 2.092** 0.560 2.076**(0.76) (1.99) (0.61) (1.97)

Cash -0.671*** -0.206 -0.671*** -0.194(-3.67) (-0.95) (-3.66) (-0.90)

Tangibility 2.684*** 1.629*** 2.704*** 1.638***(8.77) (3.99) (8.77) (4.01)

Ln(Firm Age) 0.007 -0.055 0.005 -0.048(0.22) (-0.82) (0.17) (-0.72)

Ln(CEO Age) -0.225 -0.226 -0.231 -0.229(-1.15) (-0.94) (-1.19) (-0.95)

Ln(CEO Tenure) 0.048*** 0.039** 0.047*** 0.039**(3.13) (2.05) (3.08) (2.04)

Cash Compensation -0.071 -0.007 -0.065 -0.004(-1.27) (-0.11) (-1.16) (-0.06)

Ln(CEO Ownership) 0.061* 0.044 0.060* 0.044(1.71) (1.07) (1.67) (1.06)

Firm FE Yes Yes Yes YesYear-Quarter FE Yes Yes Yes Yes

Adj. R2 0.660 0.688 0.661 0.688Observations 79,776 32,030 79,776 32,030

39


Table 3: Market Reaction to M&A Announcements and Withdrawal of Acquisitions

This table examines how overconfident and non-overconfident CEOs respond differently to M&A announcement-day

returns in withdrawing the acquisitions. We report estimates from probit regressions (Model (19)) where the dependent

variable is a dummy variable, Withdrawal, which equals one if the acquirer cancels the acquisition. CAR is the market-

adjusted cumulative abnormal return over various time windows around takeover announcements. The control variables

include: Tender Offer, Compete, Litigation, Lockup of Target Shares, Target Termination Fee, Defense, Friendly, Public

Target, Toehold Indicator, and Toehold Shares. Control variables include Ln(Assets), Leverage, Profitability, Q, Cash,

Tangibility, Ln(Firm Age), Ln(CEO Age), Ln(CEO Tenure), Cash Compensation, Ln(CEO Ownership). Industry and

year fixed effects are included. t-statistics using acquirer-clustered standard errors are in shown brackets. *, ** and ***

indicate significance better than 10%, 5%, and 1% respectively. Marginal effects of CAR on the likelihood of cancelling

an M&A deal for the subsamples of overconfident and non-overconfident CEOs are reported for each specification.

Dependent Variable: Withdrawal of Acquisitionsk

Overconfidence Measure: Options Press

CAR Time Window: [-1,1] [-1,5] [-1,10] [-1,1] [-1,5] [-1,10](1) (2) (3) (4) (5) (6)

CAR -1.273 -1.374** -1.879*** -2.549** -2.093** -2.956***(-1.61) (-2.36) (-3.71) (-2.01) (-2.22) (-3.57)

CAR × Confidence 1.325 1.654** 1.568** 4.701*** 2.983** 2.995***(1.25) (2.11) (2.43) (3.13) (2.52) (3.03)

Confidence -0.000 0.005 0.018 0.263** 0.275** 0.304***(-0.00) (0.08) (0.28) (2.40) (2.48) (2.67)

Tender Offer -0.709*** -0.717*** -0.711*** -0.476* -0.476* -0.477*(-3.73) (-3.81) (-3.77) (-1.68) (-1.72) (-1.75)

Compete 1.394*** 1.395*** 1.408*** 1.143*** 1.149*** 1.182***(10.57) (10.56) (10.67) (4.44) (4.47) (4.56)

Litigation 0.453* 0.458* 0.468* 0.000 0.000 0.000(1.89) (1.91) (1.92) (.) (.) (.)

Lockup -1.117*** -1.118*** -1.112*** -0.830 -0.857 -0.888(-3.26) (-3.26) (-3.24) (-1.29) (-1.36) (-1.38)

Target Termination Fee -0.471*** -0.478*** -0.481*** -0.403* -0.395* -0.379(-4.24) (-4.32) (-4.34) (-1.75) (-1.71) (-1.62)

Defense 1.025*** 1.027*** 1.026*** 1.482*** 1.430*** 1.447***(3.45) (3.48) (3.48) (3.65) (3.58) (3.62)

Friendly -1.125*** -1.123*** -1.124*** -1.197*** -1.201*** -1.223***(-11.03) (-11.04) (-11.00) (-6.70) (-6.77) (-6.79)

Public Target 1.072*** 1.078*** 1.065*** 1.187*** 1.171*** 1.123***(11.71) (11.78) (11.60) (5.72) (5.65) (5.38)

I(Toehold>0) 0.550 0.549 0.541 1.910*** 1.883*** 1.901***(1.43) (1.43) (1.40) (3.70) (3.76) (3.81)

Toehold Shares -0.034* -0.034* -0.034* -0.094*** -0.091*** -0.091***(-1.94) (-1.93) (-1.91) (-2.65) (-2.67) (-2.67)

Control Variables Yes Yes Yes Yes Yes YesIndustry Fixed Effects Yes Yes Yes Yes Yes YesYear Fixed Effects Yes Yes Yes Yes Yes Yes

Pseudo R2 0.333 0.334 0.337 0.359 0.357 0.362Observations 14,882 14,882 14,882 5,850 5,850 5,850

Marginal Effect:Non-overconfident CEOs -0.049 -0.053 -0.072 -0.081 -0.066 -0.093Overconfident CEOs 0.002 0.011 -0.012 0.068 0.028 0.001

40


Table 4: CEO Overconfidence and Investment-Q Sensitivity around Mutual Fund Fire Sales

This table reports the estimates of Model (20) using a sample of stock-year-quarter observations that are

under large mutual fund flow-induced selling (with FT in the bottom decile within each quarter). The de-

pendent variable is the change in Investment from quarter t to quarter t + 4. The independent variables of

interest are the interaction term between the change in Q over the quarter and measures of CEO overconfi-

dence. Control variables include Ln(Assets), Leverage, Profitability, Cash, Tangibility, Ln(Firm Age), Ln(CEO

Age), Ln(CEO Tenure), Cash Compensation, Ln(CEO Ownership), and the lagged dependent variable. Indus-

try and time fixed effects (year, fiscal quarter, and calendar quarter) are included. t-statistics using robust

standard errors are in brackets. *, ** and *** indicate significance better than 10%, 5%, and 1% respectively.

Dependent Variable: ∆Investmenti,t,t+4


(1) (2)

∆Qt−1,t+1 0.003*** 0.004***(4.77) (3.10)

∆Qt−1,t+1 × Confidencet−1 -0.002*** -0.002(-2.63) (-1.63)

Confidencet−1 -0.001** 0.000(-2.27) (0.85)

Ln(Assets)t−1 0.000*** 0.000*(2.60) (1.82)

Leveraget−1 0.001 0.001(0.87) (0.79)

Profitabilityt−1 0.006 -0.015(0.87) (-1.42)

Casht−1 0.003** 0.003(2.33) (1.62)

Tangibilityt−1 -0.004*** -0.003*(-3.05) (-1.95)

Ln(Firm Age)t−1 -0.000 -0.000(-1.50) (-0.89)

Ln(CEO Age)t−1 0.000 -0.001(0.13) (-0.57)

Ln(CEO Tenure)t−1 -0.000 -0.000(-0.62) (-0.80)

Cash Compensationt−1 0.001 0.001(0.65) (0.65)

Ln(CEO Ownership)t−1 0.000 0.000(0.00) (1.24)

Industry FE Yes YesYear-Quarter FE Yes Yes

Adjusted R2 0.014 0.015Observations 6,755 2,651

41


Table 5: CEO Overconfidence and CAR under Mutual Fund Fire Sales

This table reports estimates from regressions (Model (21)) where the dependent variable is the quarterly cu-

mulative abnormal return adjusted by benchmark portfolio matched based on size, book-to-market, and

momentum. The independent variables of interest are the interactions between Fire Sale and measures

of CEO overconfidence. Control variables include Ln(Assets), Leverage, Profitability, Q, Cash, Tangibility,

Ln(Firm Age), Ln(CEO Age), Ln(CEO Tenure), Cash Compensation, Ln(CEO Ownership), Ln(Illiquidity),

and Ln(Analysts). Industry and year-quarter fixed effects are included. t-statistics using firm-clustered stan-

dard errors are in brackets. *, ** and *** indicate significance better than 10%, 5%, and 1% respectively.

Dependent Variable: Quarterly CARi,t

Confidence Measures: Option Press(1) (2)

Fire Sale -0.022*** -0.021***(-7.99) (-3.89)

Fire Sale × Confidence 0.022*** 0.015**(5.56) (2.26)

Confidence -0.002 0.011***(-0.87) (4.43)

Ln(Assets) 0.014*** 0.025***(7.64) (8.01)

Leverage -0.016*** 0.000(-2.64) (0.03)

Profitability 0.732*** 0.990***(14.19) (10.01)

Q -0.006*** -0.008***(-4.72) (-2.98)

Cash 0.076*** 0.084***(10.12) (8.24)

Tangibility -0.008 -0.008(-1.56) (-0.90)

Ln(Firm Age) -0.001 -0.002(-1.22) (-1.43)

Ln(CEO Age) -0.015** -0.011(-2.23) (-1.09)

Ln(CEO Tenure) 0.002* -0.001(1.84) (-0.71)

Cash Compensation -0.009** -0.002(-2.41) (-0.43)

Ln(CEO Ownership) -0.001 0.000(-0.40) (0.24)

Ln(Illiquidity) 0.014*** 0.025***(8.95) (8.42)

Ln(Analysts) 0.000 0.004*(0.15) (1.82)



42


Table 6: CEO Overconfidence and CAR under Mutual Fund Fire Sales: Propensity Score Matching

This table conducts the analysis using propensity score matching and reports the differences in quarterly cumulative

abnormal return (CAR) in the quarter of fire sales between firms with and without overconfident CEOs in the matched

sample. Among all the firm-quarter observations with large mutual fund flow-induced selling (FT in the bottom

decile), we estimate the propensity score for being a treated firm (with overconfident CEOs) based on firm and

manager characteristics used in Model (21). We then match each treated firm with a nearest neighbor from the control

group (without overconfident CEOs) based on propensity score. Panels A and B present the results for option-based

and press-based measures of CEO overconfidence, respectively. The p-value of the differences are reported in brackets.

Panel A: Propensity Score Matching (Confidence Option)

Sample: Unmatched Matched

Group: Treated Control Difference Treated Control Difference(1) (2) (3) (4) (5) (6)

Ln(Assets) 6.808 6.888 -0.080 6.886 6.897 -0.010(0.02) (0.80)

Leverage 0.197 0.214 -0.017 0.211 0.208 0.003(0.00) (0.56)

Profitability 0.042 0.030 0.012 0.038 0.038 0.000(0.00) (0.67)

Q 2.198 1.432 0.766 1.829 1.783 0.046(0.00) (0.10)

Cash 0.148 0.120 0.028 0.133 0.134 -0.002(0.00) (0.74)

Tangibility 0.250 0.297 -0.048 0.267 0.263 0.004(0.00) (0.54)

Ln(Firm Age) 2.576 2.910 -0.334 2.688 2.662 0.026(0.00) (0.31)

Ln(CEO Age) 4.001 4.013 -0.012 4.009 4.006 0.004(0.00) (0.38)

Ln(CEO Tenure) 4.437 4.280 0.158 4.400 4.421 -0.021(0.00) (0.40)

Cash Compensation 0.439 0.466 -0.027 0.445 0.443 0.002(0.00) (0.81)

Ln(CEO Ownership) 0.764 0.628 0.136 0.737 0.765 -0.028(0.00) (0.29)

Ln(Illiquidity) 1.185 1.926 -0.741 1.392 1.344 0.048(0.00) (0.40)

Ln(Analysts) 1.582 1.463 0.119 1.536 1.539 -0.003(0.00) (0.91)

Observations 2,594 5,130 7,724 2,025 2,025 4,050

Difference in CAR in the Matched Sample

Variable Treated Controls Difference(1) (2) (3)

Quarterly CAR 0.012 -0.001 0.012**(0.01)

43


Panel B: Propensity Score Matching (Confidence Press)

Sample: Unmatched Matched

Group: Treated Control Control−Treated Treated Control Control−Treated(1) (2) (3) (4) (5) (6)

Ln(Assets) 7.066 6.792 0.274 7.060 7.066 -0.006(0.00) (0.93)

Leverage 0.193 0.205 -0.012 0.208 0.207 0.001(0.06) (0.87)

Profitability 0.039 0.027 0.012 0.033 0.033 -0.000(0.00) (0.88)

Q 1.973 1.392 0.581 1.613 1.587 0.026(0.00) (0.46)

Cash 0.134 0.141 -0.006 0.128 0.126 0.002(0.30) (0.75)

Tangibility 0.264 0.238 0.027 0.259 0.263 -0.004(0.00) (0.72)

Ln(Firm Age) 2.802 2.914 -0.111 2.874 2.863 0.011(0.00) (0.77)

Ln(CEO Age) 3.994 4.013 -0.019 4.004 4.006 -0.001(0.00) (0.82)

Ln(CEO Tenure) 4.321 4.268 0.054 4.305 4.319 -0.014(0.08) (0.73)

Cash Compensation 0.439 0.455 -0.017 0.447 0.440 0.007(0.09) (0.56)

Ln(CEO Ownership) 0.641 0.590 0.051 0.616 0.612 0.004(0.06) (0.91)

Ln(Illiquidity) 0.864 1.919 -1.055 1.234 1.146 0.088(0.00) (0.31)

Ln(Analysts) 1.649 1.395 0.254 1.537 1.576 -0.039(0.00) (0.30)

Observations 1,298 1,518 2,816 885 885 1,770

Difference in CAR in the Matched Sample

Variable Treated Controls Difference(1) (2) (3)

Quarterly CAR 0.017 -0.001 0.018**(0.01)

44


Table 7: Fire Sales by Implicated Mutual Funds in the Later-Trading Scandal.

The table reports the result using the Later-Trading Scandal as a plausible exogenous shock to mutual fund flows. The

sample consists of stock-quarter observations from September 2003 to December 2004 (the period when majority of

involved mutual funds were implicated). The dependent variable is the quarterly cumulative abnormal return adjusted

by benchmark portfolio matched based on size, book-to-market, and momentum. The independent variables of interest

are the interactions between Scandal Fund Ownership, the ownership as of June 2003 by mutual funds implicated in

2003 and 2004, and measures of CEO overconfidence. Control variables include Ln(Assets), Leverage, Profitability,

Q, Cash, Tangibility, Ln(Firm Age), Ln(CEO Age), Ln(CEO Tenure), Cash Compensation, Ln(CEO Ownership),

Ln(Illiquidity), and Ln(Analysts). Industry and year-quarter fixed effects are included. t-statistics using firm-clustered


Dependent Variable: CARi,t

(1) (2)

Scandal Fund Ownership -0.166** -0.232***(-2.40) (-2.72)

Scandal Fund Ownership × Confidence 0.229** 0.223**(2.10) (2.04)

Confidence -0.015 0.021**(-1.44) (2.12)

Ln(Assets) 0.021*** 0.018***(4.55) (3.65)

Leverage -0.012 0.004(-0.64) (0.22)

Profitability 0.739*** 0.644***(4.00) (3.22)

Q -0.003 -0.005(-0.67) (-1.08)

Cash 0.058*** 0.054**(2.70) (2.21)

Tangibility 0.004 0.016(0.25) (0.89)

Ln(Firm Age) -0.005 0.000(-1.62) (0.15)

Ln(CEO Age) 0.006 -0.016(0.29) (-0.74)

Ln(CEO Tenure) -0.006* -0.008**(-1.78) (-2.16)

Cash Compensation -0.010 -0.002(-0.95) (-0.20)

Ln(CEO Ownership) 0.004 0.004(0.93) (1.05)

Ln(Illiquidity) 0.017*** 0.014***(4.02) (3.25)

Ln(Analysts) 0.003 -0.000(0.82) (-0.12)



45


Table 8: Mutual Fund Outflows and Trading by Liquidity Unconstrained Funds

This table reports estimates from regressions (Model (22)) where the dependent variable is Other Fund Tradei,t,

where Other Fund Tradei,t is the net trades by mutual funds without large out flows (-5%) divided by total trading

volume in quarter t. and the independent variables of interest are the interaction term between Fire Sale in quarter t

and measures of CEO overconfidence. Lagged firm control variables include Ln(Assets), Leverage, Profitability, Cash,

Tangibility, Ln(Firm Age), Ln(CEO Age), Ln(CEO Tenure), Cash Compensation, CEO Ownership, Ln(Illiquidity) and

Ln(Analysts). Firm time fixed effects (year, fiscal quarter, and calendar quarter) are included. t-statistics using firm-

clustered standard errors are in brackets. *, ** and *** indicate significance better than 10%, 5%, and 1% respectively.

Dependent Variable: Other Fund Tradei,t


Fire Sale 0.122 -0.027(1.31) (-0.17)

Fire Sale × Confidence 0.454*** 0.594***(2.78) (2.98)

Confidence 0.022 -0.177*(0.41) (-1.86)

Ln(Assets) 0.212** -0.073(2.45) (-0.53)

Leverage 0.078 -0.086(0.35) (-0.28)

Profitability -0.976 -1.073(-1.08) (-0.88)

Q -0.016 -0.285***(-0.51) (-5.84)

Cash -0.787*** -0.716*(-3.01) (-1.92)

Tangibility -0.631 -0.210(-1.52) (-0.31)

Ln(Firm Age) -0.582*** -0.598***(-5.65) (-3.36)

Ln(CEO Age) 1.144*** 1.273**(3.33) (2.34)

Ln(CEO Tenure) -0.116*** -0.084(-2.64) (-1.40)

Cash Compensation 0.183* -0.026(1.74) (-0.21)

Ln(CEO Ownership) -0.124** -0.039(-2.04) (-0.46)

Ln(Amihud) 0.169*** -0.014(4.35) (-0.25)

Ln(Analysts) 0.091** 0.134**(2.20) (2.40)

Firm FE Yes YesYear-Quarter FE Yes Yes


46


Figure 4: This figure presents the cumulative abnormal return (CAR) over equal-weight market return around mutualfund flow-driven price pressure.

(a) Option-based Overconfidence

(b) Press-based Overconfidence

47


Internet Appendix for“Managerial Attributes and Market

Feedback Effects”

A1 Appendix: Detailed proofs for the theoretical results

A1.1 Proofs of Lemma 1 and 2

For c∗,ocb , there are three potential cases:

Case 1 (c∗,ocb ≥ C): This corresponds to the case where mutual funds’ outside opportunities are sufficiently low. Thus,all funds hold their shares.Case 2 (c∗,ocb ≤ 0): This corresponds to the case where mutual funds’ outside opportunities are sufficiently high. Thusall mutual funds sell their shares.Case 3 (0 < c∗,ocb < C): This corresponds to the case where mutual funds have intermediate values of outside oppor-

tunities. There exists A unique interior solution of α∗,ocb , where α∗,ocb = 1 − c∗,ocbC

. Inserting α∗,ocb into the expression

of P1 yields the expected stock price as µ−1λ

+c∗,ocb

(1−f)

λC. This completes the proof of Lemma 1.

The proof of Lemma 2 is similar to the proof of Lemma 1, with c∗,ocb replaced by c∗,ocg . �

A1.2 Proofs of Lemma 3, 4 and 5

The proofs of Lemma 3 and 5 follow the same steps as Lemma 1-2, where we replace c∗,oc with c∗,r. Let M = α(1−f)+f .

The proof of Lemma 4 is based on the fact thatM(α

∗,rb

)−M(α∗,rg )

2∆< 1. �

A1.3 Proof of Proposition 1

The proof of Proposition 1 is based on Lemma 4. From Lemma 4, we have that Ioc ≥ Ir and F < K + g2

2e. From

expressions for c∗,ocg and c∗,rg , we can directly verify Proposition 1 (for interior solutions). The complete comparisonbetween economies with overconfident and rational manager firms is as follows:Case 1: When c∗,ocg > c∗,rg ≥ C, α∗,rg = α∗,ocg = 0 and the expected stock price is µ−f

λ.

Case 2: When c∗,ocg ≥ C > c∗,rg > 0, α∗,ocg = 0, α∗,rg = 1 − c∗,rgC

> 0, the expected stock price in the economy

with overconfident firm managers is µ−fλ

and the expected stock price in the economy with rational firm managers isµ−1λ

+c∗,rg (1−f)

λC.

Case 3: When C > c∗,ocg > c∗,rg > 0, 0 < α∗,ocg = 1− c∗,ocgC

< α∗,rg = 1− c∗,rgC

, the expected stock price in the economy

with overconfident firm managers is µ−1λ

+c∗,ocg (1−f)

λCand the expected stock price in the economy with rational firm

managers is µ−1λ

+c∗,rg (1−f)

λC.

Case 4: When C > c∗,ocg > 0 ≥ c∗,rg , 0 < α∗,ocg = 1 − c∗,ocgC

< α∗,rg = 1, the expected stock price in the economy

with overconfident firm managers is µ−1λ

+c∗,ocg (1−f)

λCand the expected stock price in the economy with rational firm

managers is µ−1λ

.Case 5: When 0 > c∗,ocg > c∗,rg , α∗,ocg = α∗,rg = 1, the expected stock price in the economy with overconfident firmmanagers is µ−1

λand the expected stock price in the economy with rational firm managers is µ−1

λ. �

A1.4 Proof of Corollary 1

Let P1,b and P1,g be the price at t = 1 for ρ = b, g, respectively. Here, E(P1) = q · E(P1,b) + (1 − q) · E(P1,g). Wewill provide sufficient conditions to ensure E(P r1 ) < E(P oc1 ), where r denotes the rational manager and oc denotes theoverconfident manager.

Decomposing E(P1), we have two components: E(P1,b) and E(P1,g). We discuss different components separately.

For E(P1,g), from Proposition 1, we know that E(P r1,g) ≤ E(P oc1,g), where the inequality holds when c∗,ocg > 0 (see Case

1-4 in proof of Proposition 1). From Lemma 2, we know c∗,ocg =g2

2e+K−µ−1

λ

1+ 1−fλC

. This implies that g2

2e+K − µ−1

λ> 0 can


guarantee E(P oc1,g) > E(P r1,g).

For E(P1,b), when all mutual funds sell theirs shares, the market-clearing condition yields the prices are the same at

E(P r1,b) = E(P oc1,b) = µ−1λ

. From the proofs of Lemma 1 and 3, we know that when both c∗,ocb (=ge·b− g

2

2e+K−µ−1

λ

1+ 1−fλC

) and

c∗,rb (=K+VL−

µ−1λ

1+ 1−fλC

) are negative, all funds sell their shares in two economies. Given c∗,ocb < c∗,rb , the sufficient condition

for all funds selling shares is c∗,rb < 0. That is K + VL − µ−1λ

< 0. Meanwhile, given VL(= (me· b − 1

2(me

)2e) ∗ (1 −M(α

∗,rb

)−M(α∗,rg )

2∆)) is negative (since b < 0), the sufficient condition to ensure c∗,rb < 0 is K − µ−1

λ< 0.

Taking the above discussion together, µ−1λ− g2

2e< K < µ−1

λcan ensure that the expected P1 in the economy with

rational firm manages is lower than that with overconfident managers. �


We examine∂α∗,OCg

∂fand

∂α∗,rg

∂fin this proof. We focus on the scenario where all mutual funds sell their shares at

t = 1 when receiving bad signals, and there are interior solutions of c∗,rg and c∗,OCg when mutual funds receive goodsignals. Two conditions ensure such an equilibrium. First, K < µ−1

λensures that all mutual funds sell their shares

when receiving bad signals). Second, C > c∗,ocg > c∗,rg > 0 ensures interior solutions of c∗,rg and c∗,OCg . The conditionC > c∗,ocg > c∗,rg > 0 is equivalent to:

c∗,rg =K + VH2 + (VH1 − VH2)

M(α∗,rb

)−M(α∗,rg )

2∆− µ−1

λ

1 + 1−fλC

> 0,

c∗,OCg =K + g2

2e− µ−1

λ

1 + 1−fλC

< C,

which can be satisfied when the following conditions hold:

K +m

eg − 1

2(m

e)2e− µ− 1

λ> 0

K +g2

2e− µ− 1

λ< C

In a brief summary, when K < µ−1λ

, K + meg − 1

2(me

)2e − µ−1λ

> 0 and K + g2

2e− µ−1

λ< C, we have the equilibrium

where all mutual funds sell their shares at t = 1 when receiving bad signals, and there are interior solutions of c∗,rg and

c∗,OCg when mutual funds receive good signals. Focusing on this equilibrium, we examine∂c∗,OCg

∂fand

∂c∗,rg∂f

.

Regarding∂c∗,OCg

∂f, Lemma (2) yields:

∂α∗,OCg

∂f= − 1

C

∂c∗,OCg

∂f= − 1

C

K + g2

2e− µ−1

λ

(1 + 1−fλC

)2

1

λC< 0. (A-1)

We examine∂c∗,rg∂f

using several steps. First, we solve c∗,rg . From Lemma (5), we have

c∗,rg (1 +1− fλC

) = F − µ− 1

λ= K + VH2 + (VH1 − VH2)

M(α∗,rb )−M(α∗,rg )

2∆− µ− 1

λ. (A-2)

Inserting the expressions of M(α∗,rb ) and M(α∗,rg ) yields:

c∗,rg (1 +1− fλC

) = K + VH2 + (VH1 − VH2)(1− f)

2∆− µ− 1

λ− (VH1 − VH2)

(1− c∗,rgC

)(1− f)

2∆.

Finally, we solve c∗,rg as c∗,rg =K+VH2−

µ−1λ

1+( 1λC− 1

2∆C(VH1−VH2))(1−f)

. Based on this expression of c∗,rg , we can get: When

1λC− 1

2∆C(VH1 − VH2) < 0,

∂c∗,rg∂f

< 0 and∂α∗,rg

∂f= − 1

C

∂c∗,OCg

∂f> 0.

When 1λC− 1

2∆C(VH1 − VH2) > 0,

∂c∗,rg∂f

> 0 and∂α∗,rg

∂f= − 1

C

∂c∗,OCg

∂f< 0.

2


We further compare | ∂α∗,rg

∂f| and | ∂α

∗,OCg

∂f|. Specifically, we find that | ∂α

∗,rg

∂f| = 1

C

K+VH2−µ−1λ

[1+( 1λC− 1

2∆C(VH1−VH2))(1−f)]2

[ 1λC−

12∆C

(VH1 − VH2)] < 1C

K+ g2

2e−µ−1

λ

(1+ 1−fλC

)21λC

= | ∂α∗,OCg

∂f|. �


Given the definition of GP , GP =(α∗,rg −α

∗,OCg )(1−f)

λ. Thus, we have ∂GP

∂f= 1

λ(∂α∗,rg

∂f− ∂α∗,OC

g

∂f)(1− f)− (α∗,r

g −α∗,OCg )

λ.

A simple calculation yields:

∂GP

∂f=

1

λ

1

C[

K + VH2 − µ−1λ

[1 + ( 1λC− 1

2∆C(VH1 − VH2))(1− f)]2

−K + g2

2e− µ−1

λ

(1 + 1−fλC

)2].

Intuitively, when f approximates 1, ∂GP∂f

approximates − (α∗,rg −α

∗,OCg )

λ, which is negative.

We turn to characterize the sufficient condition to obtain ∂GP∂f

> 0. We focus on the case where f is close to zero.

Intuitively, whenK+VH2−

µ−1λ

[1+( 1λC− 1

2∆C(VH1−VH2))]2

>K+ g2

2e−µ−1

λ

(1+ 1λC

)2, ∂GP

∂fis positive when f = 0. Using the intermediate value

theorem, whenK+VH2−

µ−1λ

[1+( 1λC− 1

2∆C(VH1−VH2))]2

>K+ g2

2e−µ−1

λ

(1+ 1λC

)2, there exists f∗ such as ∂GP

∂fis positive when f < f∗. This

completes the proof of Proposition 3.�

3


A2 Appendix: Data Definition

1. Confidence Options is the ‘Holder67’ measure (Malmendier and Tate, 2005), an indicator for CEOs that hasan option with five years remaining duration that is at least 67% in the money. The option-based measure ofoverconfidence is measured as the value-per-vested-unexercised option scaled by the average strike price of thoseoptions. Holder67 is then an indicator that equals one from the first year in which the above ratio exceeds 0.67if the ratio exceeds 0.67 on at least two occasions.

2. Confidence Press is a binary variable that equals one if the number of articles refering to the CEO as overconfidentminus the number of articles refering to the CEO as nonconfident is above the sample median.

3. Investment is the sum of capital expenditure and R&D expenditure, divided by average of current and laggedtotal assets.

4. Ln(Assets) is the natural logarithm of total assets.

5. Leverage is the sum of long term debt and debt in current liabilities divided by total assets.

6. Profitability is the operating income before depreciation divided by average of current and lagged total assets.

7. Q is the market value of equity plus total liability minus deferred taxes and investment tax credit, divided bytotal assets.

8. Cash is cash and equivalent divided by total assets.

9. Tangibility is the net total value of property, plant and equipment, divided by total assets.

10. Ln(Firm Age) is the natural logarithm of firm’s age.

11. Ln(CEO Age) is the natural logarithm of CEO’s age.

12. Ln(CEO Tenure) is the natural logarithm of CEO’s tenure in terms of month.

13. Cash Compensation is the sum of CEO salary and bonus, divided by total compensation.

14. Ln(CEO Ownership) is the natural logarithm of one plus CEO’s percentage of ownership excluding options.

15. Ln(Illiquidity) is defined as ln(1 + AvgILLIQ × 109), where AvgILLIQ is an yearly average of illiquidity

measured as the absolute return divided by dollar trading volume:AvgILLIQi,t = 1Daysi,t

∑Daysi,td=1

|Ri,t,d|DolV oli,t,d

.

Daysi,t is the number of valid observation days for stock i in fiscal year t, and Ri,t,d and DolV oli,t,d are thereturn and dollar trading volume of stock i on day d in the fiscal year t.

16. Ln(Number of Analysts) is the natural logarithm of one plus maximum number of analysts following the stockfor the year. It is coded as 0 if there is not coverage from I/B/E/S.

17. Ln(Repurchase) is the natural logarithm of the amount firm spends repurchasing stock over the next fourquarters.

18. Ln(Earnings Guidance) is the log number of earnings forecasts provided by the management in the quarter.

19. MFSale is the total number of shares sold by mutual funds with larger than -5% outflow, divided by the totaltrading volume of the stock.

20. Net Equity Issuance is the difference between sale of common/preferred stock and the purchase of common/preferredstock in the next four quarters scaled by lagged total assets.

21. Other Fund Trade is the net trades by mutual funds without large out flows (-5%) divided by total tradingvolume.

22. Significant Investment-Q Sensitivity : the following time-series regression is estimated for each firm: Investmentt+1 =α+ β1Q+ β2Ln(Assets) + β3Profitability +QuarterFE + ε. Significant Investment-Q Sensitivity equals 1 ifβ1 is in the top tercile and significant at 5% level. We require each firms to have at least 20 years of data to beincluded in the estimation.

23. Tender Offer : a binary variable that equals one if it is a tender offer, zero otherwise.

24. Compete: a binary variable that equals one if a third party made a competing offer while this bid is pending,zero otherwise.

25. Litigation: a binary variable that equals one if a litigation is filed as a result of the transation, zero otherwise.

4


26. Lockup of Target Shares: a binary variable that equals one if the bidder is granted an option to purchase sharesin the target, zero otherwise.

27. Target Termination Fee: a binary variable that equals one if the bid includes a bidder payable termination fee,zero otherwise.

28. Defense: a binary variable that equals one if the target employs a defensive tactic against a takeover attempt,zero otherwise.

29. Friendly : a binary variable that equals one if the target board recommends the offer, zero otherwise.

30. Public Target : a binary variable that equals one if the target is a public firm, zero otherwise.

31. Toehold Indicator : a variable that equals one if the bidder has a prior toehold in the target, zero otherwise.

32. Toehold Shares: is the percentage of toehold if the bidder has a prior toehold in the target, zero otherwise.

5


A3 Appendix: Additional Empirical Results

Table A1: Mutual Fund Fire Sales by Year

This table shows the percentage of observations with mutual fund fire sales, defined as FT in the bottom decile.

Year FT (in percentage)

1992 -0.1581993 -0.2011994 -0.2761995 -0.2801996 -0.2851997 -0.1961998 -0.4681999 -0.9832000 -0.5062001 -0.1652002 -0.4082003 -0.3202004 -0.3802005 -0.6522006 -0.5602007 -0.5832008 -0.5952009 -0.7292010 -0.6382011 -0.6242012 -1.0052013 -1.1432014 -1.0122015 -1.2262016 -0.849

6


Table A2: Stock Repurchase and Earnings Forecast under Mutual Fund Fire Sales

In columns 1 and 2, the dependent variable is Ln(Repurchase) over the next four quarters. In columns 3 and 4,

the dependent variable is the natural logarithm of one plus the number of earnings forecasts in quarter t + 1. The

independent variables of interest are the interactions between Pressure and measures of CEO overconfidence. Control

variables include Ln(Assets), Leverage, Profitability, Cash, Tangibility, Ln(Firm Age), Ln(CEO Age), Ln(CEO Tenure),

Cash Compensation and Ln(CEO Ownership). Firm and year-quarter fixed effects are included. t-statistics using firm-

clustered standard errors are in brackets. *, ** and *** indicate significance better than 10%, 5%, and 1% respectively.

Dependent Variable: Ln(Repurchase)i,t+1 Ln(Earnings Guidance)i,t+1

Overconfidence Measure: Options Press Options Press

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

Fire Sale 0.079** 0.055 0.036*** 0.052***(2.50) (1.09) (4.21) (3.64)

Fire Sale × Confidence -0.033 0.016 -0.021 -0.046**(-0.63) (0.21) (-1.52) (-2.32)

Confidence 0.093** 0.101 0.014 0.018(2.52) (1.47) (1.62) (1.16)

Ln(Assets) 0.775*** 0.721*** 0.095*** 0.130***(15.74) (7.49) (8.45) (6.36)

Leverage -2.708*** -3.421*** -0.033 -0.109*(-15.73) (-13.66) (-0.83) (-1.70)

Profitability 8.294*** 8.625*** 1.106*** 1.519***(13.58) (9.37) (7.56) (6.39)

Cash 1.288*** 1.370*** -0.231*** -0.163**(6.61) (4.40) (-4.45) (-2.29)

Tangibility -0.104 -0.776 -0.374*** -0.205*(-0.32) (-1.35) (-5.01) (-1.69)

Ln(Firm Age) 0.158*** 0.530*** 0.060*** 0.025(3.15) (4.42) (4.78) (0.81)

Ln(CEO Age) -0.342 -0.753* -0.185*** -0.319***(-1.34) (-1.82) (-3.11) (-3.26)

Ln(CEO Tenure) 0.037* 0.054* 0.017*** 0.017**(1.77) (1.77) (3.35) (2.27)

Cash Compensation -0.074 -0.012 -0.064*** -0.027(-1.17) (-0.13) (-3.78) (-1.21)

Ln(CEO Ownership) -0.049 -0.079 0.004 0.039**(-1.14) (-1.09) (0.32) (2.34)

Firm FE Yes Yes Yes YesYear-Quarter FE Yes Yes Yes Yes

Adjusted R2 0.616 0.663 0.603 0.592Observations 85,803 33,845 95,466 38,550

7


Table A3: Equity Financing under Mutual Fund Fire Sales

This table reports estimates from regressions where the dependent variable is Net Equity Issuance over

the next four quarters. The independent variables of interest are the interactions between Fire Sale

and measures of CEO overconfidence. Control variables include Ln(Assets), Leverage, Profitability, Cash,

Tangibility, Ln(Firm Age), Ln(CEO Age), Ln(CEO Tenure), Cash Compensation and Ln(CEO Owner-

ship). Firm and year-quarter fixed effects are included. t-statistics using firm-clustered standard er-

rors are in brackets. *, ** and *** indicate significance better than 10%, 5%, and 1% respectively.

Dependent Variable: Net Equity Issuancei,t+1


(1) (2)

Fire Sale -0.002** -0.002(-2.12) (-1.00)

Fire Sale × Confidence -0.002 -0.001(-0.81) (-0.37)

Confidence 0.009*** -0.001(6.14) (-0.53)

Ln(Assets) -0.033*** -0.034***(-12.21) (-7.62)

Leverage 0.084*** 0.123***(8.39) (10.00)

Profitability -0.307*** -0.175***(-6.07) (-2.65)

Cash -0.074*** -0.078***(-5.80) (-5.02)

Tangibility -0.008 -0.016(-0.58) (-0.75)

Ln(Firm Age) -0.013*** -0.018***(-5.79) (-5.25)

Ln(CEO Age) -0.017* -0.019(-1.81) (-1.00)

Ln(CEO Tenure) -0.000 0.000(-0.05) (0.21)

Cash Compensation -0.004 -0.002(-1.04) (-0.59)

Ln(CEO Ownership) 0.003 0.004(1.57) (1.27)



8


Table A4: CEO Overconfidence and CAR under Mutual Fund Fire Sales: Controlling for Governance

This table examines the robustness of the muted price impact of mutual fund fire sales on firms with overconfident

CEOs after controlling for the level of corporate governance as proxied by G-Index (Gompers, Ishii, and Metrick,

2003). The dependent variable is the quarterly cumulative abnormal return adjusted by benchmark portfolio matched

based on size, book-to-market, and momentum. The independent variables of interest are the interactions between

Fire Sale and measures of CEO overconfidence. Other control variables include Ln(Assets), Leverage, Profitability,






Fire Sale -0.022*** -0.017**(-5.64) (-2.49)

Fire Sale × Confidence 0.025*** 0.016*(4.10) (1.94)

Confidence -0.002 0.015***(-0.81) (5.11)

G Index -0.000 0.000(-1.00) (0.45)

Ln(Assets) 0.006** 0.016***(2.52) (4.29)

Leverage -0.022** 0.003(-2.35) (0.21)

Profitability 0.590*** 0.840***(6.59) (5.17)

Q -0.010*** -0.011**(-4.64) (-2.53)

Cash 0.073*** 0.074***(6.74) (5.90)

Tangibility -0.009 -0.015(-1.22) (-1.29)

Ln(Firm Age) -0.003* -0.001(-1.69) (-0.69)

Ln(CEO Age) -0.008 -0.010(-0.77) (-0.79)

Ln(CEO Tenure) 0.002 -0.002(0.99) (-0.78)

Cash Compensation -0.007 0.000(-1.42) (0.07)

Ln(CEO Ownership) -0.000 -0.000(-0.19) (-0.21)

Ln(Illiquidity) 0.007*** 0.016***(3.15) (4.25)

Ln(Analysts) -0.002 -0.000(-0.97) (-0.12)


Observations 28836 14346Adjusted R2 0.021 0.030

9


Table A5: Alternative Mutual Fund Fire Sale Measures

This table reports estimates from regressions where the dependent variable is the quarterly cumulative abnormal

return adjusted by benchmark portfolio matched based on size, book-to-market, and momentum. In columns 1 and

2, Fire Sale is a dummy variable that equals one if the flow-induced trading measure following Lou (2012) is in

the bottom decile. In columns 3 and 4, Fire Sale is a dummy variable that equals one if the flow-induced trading

measure following Wardlaw (2018) is in the bottom 5%. Control variables include Ln(Assets), Leverage, Profitability,





Fire Sale Measures: Lou (2012) Wardlaw (2018)

Confidence Measures: Option Press Option Press(1) (2) (3) (4)

Fire Sale -0.013*** -0.003 -0.010** -0.007(-4.40) (-0.65) (-2.36) (-0.86)

Fire Sale × Confidence 0.009** 0.012* 0.014** 0.018*(1.98) (1.77) (2.29) (1.72)

Confidence -0.000 0.010*** -0.000 0.011***(-0.26) (4.39) (-0.13) (4.68)

Ln(Assets) 0.013*** 0.025*** 0.013*** 0.025***(7.36) (7.95) (7.42) (7.94)

Leverage -0.014** 0.001 -0.014** 0.002(-2.34) (0.14) (-2.34) (0.16)

Profitability 0.727*** 0.984*** 0.725*** 0.983***(14.10) (9.91) (14.06) (9.92)

Q -0.006*** -0.008*** -0.006*** -0.008***(-4.88) (-3.04) (-4.83) (-3.01)

Cash 0.077*** 0.085*** 0.077*** 0.085***(10.27) (8.33) (10.29) (8.33)

Tangibility -0.008 -0.007 -0.008 -0.007(-1.59) (-0.79) (-1.54) (-0.81)

Ln(Firm Age) -0.001 -0.002 -0.001 -0.002(-1.39) (-1.37) (-1.27) (-1.37)

Ln(CEO Age) -0.016** -0.011 -0.016** -0.012(-2.27) (-1.08) (-2.25) (-1.10)

Ln(CEO Tenure) 0.002* -0.002 0.002* -0.001(1.77) (-0.87) (1.71) (-0.84)

Cash Compensation -0.009** -0.002 -0.009** -0.002(-2.48) (-0.42) (-2.45) (-0.43)

Ln(CEO Ownership) -0.000 0.001 -0.000 0.001(-0.37) (0.38) (-0.31) (0.37)

Ln(Illiquidity) 0.014*** 0.025*** 0.014*** 0.025***(8.61) (8.34) (8.66) (8.34)

Ln(Analysts) 0.000 0.004* 0.000 0.004*(0.26) (1.83) (0.20) (1.84)

Industry FE Yes Yes Yes YesYear-Quarter FE Yes Yes Yes Yes


10


Table A6: Mutual Fund Outflows and Actual Sales

This table reports estimates from regressions where the dependent variable is MFSale in quarter t and the independent

variables of interest are the interaction term between FT (the continuous variable) in quarter t and measures of

CEO overconfidence. Control variables include Ln(Assets), Leverage, Profitability, Cash, Tangibility, Ln(Firm Age),

Ln(CEO Age), Ln(CEO Tenure), Cash Compensation, CEO Ownership, Ln(Illiquidity) and Ln(Analysts). Firm

and time fixed effects (year, fiscal quarter, and calendar quarter) are included. t-statistics using firm-clustered


Dependent Variable: MFSalei,t


FT 0.442*** 0.489***(44.71) (19.81)

FT × Confidence 0.054*** 0.051(3.46) (1.63)

Confidence 0.033*** 0.013(3.87) (0.79)

Ln(Assets) 0.002 -0.026(0.17) (-1.48)

Leverage 0.053* 0.076(1.83) (1.51)

Profitability -0.568*** -0.059(-5.32) (-0.37)

Q -0.006 -0.028***(-1.41) (-3.75)

Cash -0.006 -0.055(-0.19) (-1.07)

Tangibility 0.025 0.039(0.53) (0.46)

Ln(Firm Age) 0.007 0.031(0.79) (1.34)

Ln(CEO Age) -0.025 -0.105(-0.59) (-1.45)

Ln(CEO Tenure) -0.005 0.006(-0.84) (0.71)

Cash Compensation -0.003 0.002(-0.18) (0.08)

Ln(CEO Ownership) 0.014* 0.016(1.80) (1.12)

Ln(Illiquidity) 0.005 -0.020***(1.04) (-2.99)

Ln(Analysts) 0.003 -0.015*(0.56) (-1.80)



11


Table A7: Alternative Benchmarks for Cumulative Abnormal Returns

In columns 1 to 4 (5 to 8), the dependent variable is the quarterly cumulative abnormal return adjusted by

value-weighted (equal-weighted) market returns. The independent variables of interest are the interactions between

Fire Sale and measures of CEO overconfidence. Control variables include Ln(Assets), Leverage, Profitability, Q,

Cash, Tangibility, Ln(Firm Age), Ln(CEO Age), Ln(CEO Tenure), Cash Compensation, Ln(CEO Ownership),




Benchmark: Value-weighted Market Return Equal-weighted Market Return

Confidence Measures: Option Press Option Press(1) (2) (3) (4) )

Fire Sale -0.026*** -0.031*** -0.026*** -0.031***(-9.22) (-5.49) (-9.27) (-5.48)

Fire Sale × Confidence 0.019*** 0.019*** 0.020*** 0.019***(4.68) (2.68) (4.73) (2.71)

Confidence -0.000 0.011*** -0.000 0.011***(-0.10) (4.53) (-0.24) (4.50)

Ln(Assets) 0.018*** 0.032*** 0.018*** 0.031***(9.95) (9.80) (9.90) (9.71)

Leverage -0.012** 0.004 -0.012** 0.004(-1.98) (0.40) (-1.96) (0.42)

Profitability 0.792*** 1.045*** 0.795*** 1.044***(15.02) (10.35) (15.06) (10.34)

Q -0.005*** -0.006** -0.005*** -0.006**(-4.00) (-2.17) (-3.97) (-2.20)

Cash 0.080*** 0.085*** 0.080*** 0.085***(10.37) (7.62) (10.35) (7.59)

Tangibility -0.005 0.001 -0.005 0.001(-0.84) (0.09) (-0.84) (0.12)

Ln(Firm Age) -0.001 -0.002 -0.001 -0.002(-1.28) (-1.43) (-1.29) (-1.48)

Ln(CEO Age) -0.013* -0.009 -0.013* -0.009(-1.79) (-0.84) (-1.80) (-0.84)

Ln(CEO Tenure) 0.002* -0.001 0.002* -0.001(1.68) (-0.33) (1.75) (-0.36)

Cash Compensation -0.008** -0.004 -0.008** -0.004(-2.04) (-0.78) (-1.98) (-0.75)

Ln(CEO Ownership) -0.002 -0.001 -0.002 -0.001(-1.17) (-0.50) (-1.19) (-0.48)

Ln(Illiquidity) 0.020*** 0.034*** 0.020*** 0.034***(12.49) (11.32) (12.43) (11.22)

Ln(Analysts) 0.000 0.004** 0.000 0.004**(0.11) (2.19) (0.08) (2.13)

Industry FE Yes Yes Yes YesYear-Quarter FE Yes Yes Yes Yes


12


Table A8: Investment-Q Sensitivity and CAR under Mutual Fund Fire Sales

This table reports estimates from regressions where the dependent variable is the average monthly market-

adjusted abnormal return. The independent variables of interest are the interactions between Fire Sale

and Significant Investment-Q Sensitivity, an dummy variable indicating firms with significant estimated

investment-Q sensitivity. Control variables include Ln(Assets), Leverage, Profitability, Q, Cash, Tangibility,

Ln(Firm Age), Ln(CEO Age), Ln(CEO Tenure), Cash Compensation, CEO Ownership, Ln(Illiquidity), and

Ln(Analysts). Industry and year-quarter fixed effects are included. t-statistics using firm-clustered stan-

dard errors are in brackets. *, ** and *** indicate significance better than 10%, 5%, and 1% respectively.


(1) (2)

Fire Sale -0.008* -0.011**(-1.74) (-2.26)

Fire Sale × Significant Inv-Q Sensitivity -0.021** -0.021**(-2.18) (-2.17)

Significant Inv-Q Sensitivity -0.003 -0.004(-0.93) (-1.04)

Ln(Assets) -0.002** 0.009***(-2.05) (3.61)

Leverage -0.006 -0.016(-0.61) (-1.51)

Profitability 0.605*** 0.668***(9.52) (10.33)

Q -0.015*** -0.010***(-8.95) (-5.15)

Cash 0.058*** 0.065***(4.04) (4.47)

Tangibility 0.007 0.013(0.82) (1.39)

Ln(Firm Age) -0.003 -0.003(-1.00) (-0.95)

Ln(CEO Age) -0.028** -0.029**(-2.16) (-2.20)

Ln(CEO Tenure) 0.004** 0.004**(1.98) (2.37)

Cash Compensation -0.017*** -0.022***(-2.86) (-3.62)

Ln(CEO Ownership) -0.001 -0.002(-0.35) (-0.83)

Ln(Illiquidity) 0.010***(5.16)

Ln(Analysts) 0.002(0.78)



13


Figure A1: This figure presents the kernal density estimate for the predicted likelihood of CEO overconfidence.

(a) Option-based Overconfidence

(b) Press-based Overconfidence

14


Documents

Managerial Attributes and Market Feedback E ectshuangsy/files/FireSale/FireSale.pdf · 1 Introduction Financial markets often experience large non-fundamental drops, the recovery