Alok 2019

Do Fund Managers Misestimate Climatic
Disaster Risk?
Shashwat Alok
Indian School of Business
Nitin Kumar
Indian School of Business
Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

Russ Wermers
Robert H. Smith School of Business, University of Maryland
We examine whether professional money managers overreact to large climatic disasters.

We find that managers within a major disaster region underweight disaster zone stocks to
a much greater degree than distant managers and that this aversion to disaster zone stocks
is related to a salience bias that decreases over time and distance from the disaster, rather
than to superior information possessed by close managers. This overreaction can be costly
to fund investors for some especially salient disasters like hurricanes and tornadoes: a long-
short strategy that exploits the overreaction generates a significant DGTW-adjusted return
over the following 2 years. (JEL G10, G11, G23, G40, G41)
Received December 7, 2017; editorial decision September 24, 2019 by Editor Harrison
Hong. Authors have furnished an Internet Appendix, which is available on the Oxford
University Press Web site next to the link to the final published paper online.
Introduction
Policy makers and institutional investors across the world are increasingly
concerned about the impact of climate change risks on the broader economy, in
general, and capital markets, in particular.1 At a recent UN Foundation Investor
Summit on Climate Risk, the UN Secretary General, Ban Ki-Moon, remarked
that “. . . investors need to know how the impacts of climate change can
affect specific companies, sectors and financial markets as a whole.”2 Similarly,
We thank Harrison Hong (Editor) and two anonymous referees. We also thank Andrew Karolyi, Jose Scheinkman,
and Sumudu Watugala (discussant) and participants at the RFS Climate Finance Conference 2017 and 2018
for helpful comments. Any remaining errors or omissions are our own. Send correspondence to Shashwat
Alok, Indian School of Business, Hyderabad, Telangana 500111, India; telephone: 91-40-23187188. E-mail:
[email protected].
1 Climate risk can be broadly classified into three categories (Hjort 2016): (a) environmental risks, such as loss
of biodiversity, melting of ice sheets; (b) policy risks, such as when and how new regulations will be imposed,
and their impacts; and (c) economic risks, defined as the uncertainty impact of climate risk on the economy, in
general, and financial markets in particular. Our study concerns the economic risk associated with climate risk.
2 See Ki-moon (2016).
© The Author(s) 2020. Published by Oxford University Press on behalf of The Society for Financial Studies.
All rights reserved. For permissions, please e-mail: [email protected].
doi:10.1093/rfs/hhz143
[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1146 1146–1183

Do Fund Managers Misestimate Climatic Disaster Risk
the Institutional Investors Group on Climate Change (IIGCC), a coalition of

investors managing over E13 trillion in assets, argue in a recent policy paper
that “climate risk needs to be better reflected in the price of risk so that a shift
in capital can be encouraged.”3 However, other than a few recent papers (e.g.,
Giglio et al. 2015; Bansal et al. 2016; Hong et al. 2019; Daniel et al. 2016),
there is limited empirical research on the asset pricing implications of such
climate risk.
Corporations’ risk exposures to climate change are typically associated
with their atmospheric greenhouse gas emissions, but as Hong et al. (2019)
note, “. . . climate change risks need not be so narrowly confined to carbon

exposures. Vulnerability of corporations’ production processes to natural
disasters amplified by climate change can impose significant damage to
corporate profits . . . .” One might ask a natural follow-up question: can investors
correctly estimate the implications of these complicated (and rare) climate-
disaster-related risks for their portfolio holdings? This question assumes
economic importance. Given the large and increasing holdings and trading
volume of all institutional funds, the funds themselves are often the marginal
price setting traders in the market (Wermers 1999; Nofsinger and Sias 1999;
Dasgupta et al. 2011).4 However, if presumably rational money managers
misestimate the disaster impact on firms and, further, trade in such firms based
on this misestimation, it may adversely affect the informational efficiency of
stock prices. Beyond potentially affecting the returns of their fund shareholders,
such stock price distortions can more broadly affect the real economy through
the inefficient allocation of capital (Subrahmanyam and Titman 2001; Wurgler
2000).
How might a mutual fund manager misestimate the impact of a climatic
disaster? We draw on the extant academic finance and psychology literature
and hypothesize that misestimation by a manager could potentially be due to
salience bias. Salience bias is the tendency to overweight probabilities based
on the ease with which events can be recalled (i.e., irrationally overweighting
more readily available information). In the presence of such a bias, subjects
overestimate the risk of salient events based on vividness, proximity, or
emotional impact (Tversky and Kahneman 1973).
Both theory and empirical research guide our study. Bordalo et al. (2012)
model choice under salient risks, and argue that investors may overweight
the probability of a rare downside event when it is salient, and may act
in an overly risk-averse manner with respect to assets exposed to such a
rare event. In our setting, this implies that mutual fund managers exposed
3 http://www.iigcc.org/files/publication-files/IIGCC_2016_Financial_Regulation_paper_v15.pdf
4 Institutional funds hold more than 60% of domestic equity and account for around 70% of trading volume
(Gompers and Metrick 2001; Bennett et al. 2003; Boehmer and Kelley 2009). The more recent report in Investment
Company Institute Fact Book 2018 shows that the total net assets of U.S. investment companies is close to $22.5
trillion, of which $18.7 trillion is held in open-ended mutual funds. The shares held by mutual funds represent
more than 30% of outstanding shares in the U.S. market.
1147

The Review of Financial Studies / v 33 n 3 2020
to (or otherwise made keenly aware of) the consequences of a rare but
devasting climatic event may overestimate the probability of such disasters
in the future, and, consequently, underweight stocks recently exposed to such
climatic events in their portfolios. We call this the “salience hypothesis.”
We also recognize that professional money managers have better information
than most individual investors. This implies that a money manager can
more correctly estimate the impact of climatic disasters and make a rational
portfolio allocation decision. For instance, Coval and Moskowitz (2001)
show that mutual fund managers possess superior information about their
local investments. We call this the “information hypothesis,” and note that

the salience and information hypotheses are overlapping, and not mutually
exclusive.
Further, a clean empirical identification of our research question still poses
other econometric challenges: the lack of clearly defined counterfactual fund
portfolios in the absence of exposure to a climatic disaster. Take, for instance,
the fact that most major climatic disasters in the United States occur near
the coastlines. It also may be true that mutual funds that invest in firms
located in coastal areas are different in some unknown way from other funds.5
Our empirical design addresses this issue (and the salience vs. information
hypotheses) by exploiting the distance of funds from the disaster zone as a
source of exogenous variation in the salience of the disaster strike for fund
managers, where the “disaster zone” is the set of counties that are directly hit
by a climatic disaster as identified in the Spatial Hazard and Loss Database for
the United States (SHELDUS) at Arizona State University.6 Specifically, we
employ a difference-in-differences strategy and compare the portfolio decisions
of mutual funds headquartered within 100 miles of the headquarters of firms in
the disaster zone (treatment group funds) to those located further away (control
group funds).
We begin by examining the change in portfolio weights of stocks of firms
headquartered in a disaster zone following disaster strikes, both by (treatment
group) funds located near a disaster zone and by those (control group) funds
located farther away. To the extent that the salience of disasters declines with
distance from the disaster zone, we expect to observe an overreaction by our
treatment group funds relative to the control group with respect to stocks in the
disaster zone. Consistent with this idea, we find that while, on average, there is a
post-disaster decrease in portfolio weights of disaster zone stocks for all funds,
both close and far, the decrease is significantly greater for close (treatment-
group) funds: a result that is robust to controlling for a host of firm and fund
5 Perhaps, for example, portfolio managers who are “sensation seekers” choose to invest in firms located in a
region with a higher unconditional probability of a climatic disaster. Of course, for this omitted variable to be
problematic in our setting, sensation seeking would need to be correlated with the tendency to exhibit salience
bias.
6 In the prior example, a sensation-seeking fund manager who invests in risky stocks could be located either near
or far from a coastline.
1148

characteristics (e.g., fund size, recent returns, and flows; firm leverage, size,
and profitability) and fixed effects (firm, fund, year-quarter, disaster-quarter,
county, and stock-season fixed effects). Next, we find that this local-fund
underweighting of disaster zone stocks is sharpest up to four quarters after
the disaster, before gradually diminishing over subsequent quarters, a finding
that is consistent with the implications of salience bias (a salient event loses
importance in the mind of an investor as time passes and other salient events
arrive). Importantly, portfolio weights of disaster zone stocks for control group
and treatment groups funds do not have any differential predisaster quarter
trends.

However, we recognize that our above tests do not directly address whether
the portfolio reallocation of disaster zone funds is due to (a) the manager’s
misestimation of the impact of disaster risk on firms due to salience bias
(salience hypothesis) or (b) an informational advantage, as the manager is
herself close to the disaster zone and, hence, may have better information about
its economic impact (Coval and Moskowitz 2001; Kacperczyk and Seru 2007;
Bernile et al. 2015) (information hypothesis). Accordingly, we next conduct a
test that compares the portfolio decisions of close and far funds with respect
to holdings of stocks located in neighboring (“near-disaster zone”) counties of
the disaster zone counties. The underlying assumption is that, to the extent that
fund managers are especially likely to overinterpret the effects of a climatic
disaster at or near its epicenter, we would expect that the salience bias would
be much weaker with respect to firms located in near-disaster zones, relative
to firms within the disaster zone. We do not find that the firms located in the
near-disaster zone are underweighted more heavily by close funds relative to
distant funds. These results further strengthen the causal interpretation of our
findings, and show that underweighting of disaster zone stocks decreases with
time and distance.
In additional tests, we consider a number of alternative interpretations for
our findings and conclude that the evidence for them is not compelling. First,
we investigate whether close funds underweight disaster zone stocks merely
because they have to cater to the preferences of their investor clientele. The idea
is that fund managers may reduce their portfolio investments in disaster zone
stocks, not because they are themselves biased, but because their investors may
want to reduce their exposure to disaster zone stocks. We do not find compelling
evidence of this possibility.
Second, we address the potential concern that the observed underweighting
may be mechanically driven by an overall drop in stock prices of disaster zone
stocks. We test this potential explanation in two ways. First, we explicitly
control for quarterly stock returns and find our results to be robust. In other
words, our results persist even after holding stock prices constant in computing
(pseudo) portfolio weights. Second, we change our LHS variable from weight
to (a) the number of shares held by funds, and (b) the value of shares traded,
and find similar results. Our findings are also robust to managerial experience,
1149

fund size, fund age, socioeconomic characteristics of mutual fund clientele,

and alternative distance cutoffs in defining “close” funds.
We conduct two additional tests to further explore whether our results are
more consistent with the salience hypothesis or the information hypothesis.
These tests are based on the idea that, if the underweighting of disaster
zone stocks is driven by access to superior information regarding the future
performance of such stocks, then we should expect a drop in the performance
of disaster zone stocks in the periods after the disaster strikes. In the first test,
we perform a difference-in-differences test, and compare pre- and post-disaster
profitability of disaster zone firms, relative to firms located in near-disaster

zones. Importantly, because disaster zone funds reduce their weights on firms
located in the disaster zone, while we do not observe such an effect for firms
located in the near-disaster zone, the information hypothesis would suggest
that firms in the near-disaster zone should outperform firms in the disaster zone
around hurricane strikes. To test this, we use two proxies for firms’ performance:
ROA and Sales growth. We find that the post-disaster change in the performance
of firms in the disaster zone relative to those in the near-disaster zone is
statistically indistinguishable from zero. This suggests that the temporary
underweighting of the disaster zone stocks by close funds is consistent with the
salience hypothesis, and not with the information hypothesis.
In the second test, we examine the abnormal return performance of stocks
in the disaster zone. Specifically, we evaluate the subsequent performance of
stocks that are underweighted by funds just after climatic disaster events. If
the underweighting of disaster zone firms by close funds is consistent with
the salience hypothesis, we would expect the most underweighted stocks
to exhibit nonnegative risk-adjusted returns during post-disaster quarters, on
average, while future average underperformance by such stocks would be
more consistent with the information hypothesis. In fact, for the most salient
and devastating disasters—hurricanes and tornadoes—we find that disaster
zone stocks that are most underweighted by disaster zone funds subsequently
outperform stocks that are overweighted by disaster zone funds, indicating a
potential “price pressure” effect from the underweighting by local funds.7 This
statistically and economically significant return reversal in the underweighted
portfolio of disaster zone firms by proximate funds is consistent with the
salience bias hypothesis.
1. Related Literature and Hypotheses

In this section, we briefly review related literature, develop testable hypotheses,
and then discuss our empirical methodology.
7 Of course, given the nature of our tests—limited to mutual fund trading inferred through quarterly holdings
data—we cannot conclusively determine that mutual fund underweighting alone drives this return reversal effect.
1150

1.1 Related literature

Our paper is related to the nascent body of work on climate risk, which rely
on rational asset pricing versus deviations from purely rational pricing. Bansal
et al. (2016) argue that market prices accurately reflect long-run climate risks,
as proxied by temperature fluctuations. Giglio et al. (2015) estimate the term
structure of discount rates in real estate markets to help inform the appropriate
discount rates to be used in valuing investments required for mitigating climate
change. Along similar lines, Daniel et al. (2016) show how asset pricing theories
can help guide the optimal tax on carbon emissions. Considering deviations of
asset prices or economic agents from pure rationality, Hong et al. (2019) find that

stock prices of firms in the food sector do not accurately reflect the underlying
climate risks, at least in the context of natural disasters. Dessaint and Matray
(2017) analyze whether corporations located in hurricane areas irrationally
hoard excess liquidity following hurricanes strikes (see also Ramirez and Altay
2011). Studies that examine implications of climate risks for investor behavior
include Hirshleifer (2001), Levitt and List (2007), Goetzmann et al. (2014),
and Kaplanski and Levy (2010). In the context of fund managers, Bernile et al.
(2017) examine risk aversion of U.S.-domiciled fund managers that oversee
international equity funds after they personally experience a disaster that does
not affect their fund’s non-U.S. holdings; such managers become more risk-
averse after their disaster experience, reducing their fund’s volatility.
Our study aims to complement this literature by using a large sample of
data on the actual trading behavior of a sample of arguably market price-
setting investors, mutual fund managers, to determine whether such managers
misestimate the economic risk associated with climatic disasters, and whether
this, as a result, induces irrational biases in their portfolio decisions. In addition,
we aim to show whether any such biased portfolio decisions might induce
localized and short-lived time-varying inefficiencies in stock prices of firms in
the disaster zone.
1.2 Hypotheses
Bordalo et al. (2012) model choice under salient risk, with investors who
overweight the probability of a left-tail event when it is salient and act in an
overly risk-averse manner. In our setting, this implies that mutual fund managers
may overestimate the probability of salient (but rare) disasters that affect their
portfolio firms in the future (e.g., following the occurrence of a major climatic
disaster) and, consequently, underweight such firms in their portfolio. We call
this the salience hypothesis. Coval and Moskowitz (2001), however, find that
fund managers possess significant informational advantages with respect to
firms located in their proximity; these fund managers overweight nearby firms
and earn abnormal returns from their local holdings. In our setting, this implies
that fund managers may possess superior information regarding proximate
firms, and, consequently, may underweight local disaster firms if they expect
such firms to underperform in the near future as a result of a disaster strike. We
1151

Figure 1
Empirical design

call this the information hypothesis. We summarize our hypotheses as follows:
H1: Mutual funds located close to firms in the disaster zone subsequently
underweight such firms, relative to funds located further away from the disaster
zone. H2A (Information Hypothesis): If mutual fund managers underweight
disaster zone stocks because of superior information, then such stocks should
underperform in the near future. H2B (Salience Hypothesis): If mutual fund
managers underweight disaster zone stocks because of salience bias, then such
stocks should not underperform in the near future.
We now discuss our empirical methodology. The main econometric challenge
in evaluating whether portfolio decisions of fund managers are susceptible to
salience bias is obtaining the counterfactual portfolio in the absence of such a
bias. In other words, it is difficult to empirically distinguish biased portfolio
decisions from rational decisions that may be driven by access to superior
information. Our research design aims to circumvent these issues by focusing
on changes in the portfolios of fund managers around salient climatic disasters.
Specifically, we employ a difference-in-differences strategy that compares the
portfolio decisions of funds located close to the disaster zone (treatment group
funds) to those located farther away (control group funds), with respect to their
common holdings of firms headquartered in the disaster zone.
Figure 1 highlights this empirical strategy. We compare the change in weights
on disaster zone stocks (such as firm A) around disaster strikes in the portfolios
of close funds (such as fund 1) relative to more distant funds (such as fund
2). The identifying assumption is that disaster strikes are more salient for
funds located closer to the disaster zone, and consequently, such funds will
underweight stocks of firms headquartered in the disaster zone around the
time of disaster strikes. Our analysis controls for fund and time fixed effects.
Fund fixed effects ensure that our regressions are identified through within-fund
variation that absorb all time-invariant differences across mutual funds. Time
fixed effects further control for aggregate macroeconomic shocks.
To implement our difference-in-differences test, we use the county location
of fund companies and portfolio-holding firm headquarters to calculate the
distance between fund-firm pairs and then classify the funds into two groups.
If the distance between fund and firm headquarters is less than 100 miles,
we classify the fund-firm pairs as CLOSE ms (treatment) or as FAR (control)
1152

otherwise.8 The unit of observation in our analysis is a fund-firm-quarter.

Formally, our baseline specification is as follows:
WEIGHT mst = β0 +β1 CLOSE ms +β2 POST t +β3 (CLOSE ms ×POST t )+
(1)
Xs,t−1 +Xm,t−1 +μm +δt +mst ,
where WEIGHT mst is the weight of stock s in the portfolio of mutual fund m
at the end of quarter t. In these baseline tests, we focus on the two calendar
quarters prior to two quarters after the disaster.9 POST t equals 1 for the disaster
quarter, Q, and the two quarters following the disaster, Q+1 and Q+2, and

0, for the two quarters before the disaster quarter, Q−2 and Q−1. Note that
POST t equals 1 for the disaster quarter, Q, because WEIGHT mst is measured
at the end of the quarter t = Q. For instance, if the disaster strikes on February
15, 2005, POST t = 1 for the disaster quarter Q = 2005 : Q1.
CLOSE ms equals 1 if the headquarters of mutual fund m is located within
100 miles of the headquarters of firm s and otherwise equals 0. Xs,t−1 and
Xm,t−1 are vectors of lagged firm-level and fund-level covariates measured at
the end of the quarter t −1. μm is a vector of fund fixed effects that absorb
all time-invariant differences across funds. This implies that the regressions
are identified through within-fund variation in portfolio decisions around a
hurricane disaster. Finally, δt are year-quarter fixed effects that control for
aggregate macroeconomic shocks. The coefficient of interest is β3 which
measures the extent of underweighting of disaster zone stocks by close mutual
funds relative to distant funds.10
A negative β3 coefficient indicates that funds close to the disaster zone
decrease their portfolio investments in disaster zone stocks more than distant
funds do. The key identifying assumption for consistency of β3 is the presence
of predisaster parallel trends in portfolio weights of disaster zone stocks in the
portfolios of both the treatment (CLOSE) and control (F AR) funds. Panels A
and B of Figure 2 show, for the set of mutual funds described in Section 2, the
average portfolio weights of close versus far funds in disaster zone stocks at
each event quarter-end. Note that close funds hold larger weights of disaster
8 For robustness, we also use alternative cutoff distances of 50, 150, 200, and 250 miles (see Section 3.4).
9 A focus on the two quarters before and the two quarters after the disaster mitigates the likelihood of our results
being driven by idiosyncratic small or large weights in any given quarter. For robustness, we also work with
Q−1 to Q+1, Q−1 to Q+2, Q−1 to Q+3, Q−2 to Q+1, Q−2 to Q+3, and Q−3 to Q+3 and find similar
results (see Section 3.4).
10 The underweighting by CLOSE
ms funds is given by
β2 +β3 = E(WEIGHT |POST = 1,CLOSE = 1)−E(WEIGHT |POST = 0,CLOSE = 1),
and the underweighting by FAR funds is given by
β2 = E(WEIGHT |POST = 1,CLOSE = 0)−E(WEIGHT |POST = 0,CLOSE = 0).
Thus, the relative underweighting by close funds relative to distant funds is given by β3 .
1153


Figure 2
Time-series trend in portfolio weights of disaster zone stocks
Panel A of this figure plots the mean values of portfolio weights of disaster zone stocks for both the close fund-
firm and far fund-firm groups, that is, treated and control groups at different quarters around the disaster events.
Panel B plots the demeaned values (demeaned by county, year, disaster quarter and fund) of portfolio weights.
The close fund-firm group is represented by the solid line, and the far fund-firm group is represented by the
dashed line.
zone stocks, WEIGHT mst , consistent with Coval and Moskowitz (2001). In
addition, both close and far funds decrease their weight on disaster zone stocks
prior to the disaster, consistent with both types of funds reducing their exposure
during a season with higher unconditional disaster risk, such as hurricanes
in late summer. These parallel pretreatment trends provide support for the
use of difference-in-differences (DID) strategy in this context to estimate the
causal effect of the disaster events. Further, we also use a distributed lag model
to formally test for the absence of differential pre-trends in our data. This
procedure and the results, which do not reject equal pre-trends, are discussed
in detail in Section 3.1.2.
We note that our research design draws on the experimental setup of
Dessiant and Matray (2017; DM), with some important differences. First, while
DM focus on hurricanes, our study includes 16 climatic disasters (including
1154

hurricanes). Second, DM compare the cash hoarding behavior of neighboring

area firms with respect to firms located in the rest of the United States in the
aftermath of a hurricane strike, a setup that allows the exclusion of firms in the
disaster area, which may well have real operational cash needs. In our empirical
setup, we compare the portfolio trades of close and far funds with those of firms
headquartered in the disaster area. We include tests that compare this DID (close
vs. far funds, before and shortly after the disaster) with a similar DID that uses
firms headquartered in nearby, but not directly hit, counties. Our treatment and
control groups are defined as nearby and distant mutual funds, whereas DM’s
are nearby and distant firms; in our study, it is crucial to compare two sets of

stocks held by nearby versus distant funds, both of which are within 100 miles
of the close funds, but only one of which is directly affected by the disaster. This
design is needed to identify the potential salience bias of local fund managers
separately from the potential effect of superior private information that they
may possess about the profitability of local firms, both relative to distant fund
managers.
2. Data and Descriptive Statistics

2.1 Climatic disasters
We obtain data on climatic disasters from SHELDUS. Among other things,
SHELDUS provides information on names, dates, property damages, and
county locations of major climatic disasters in the United States. We obtain data
on the sixteen disasters listed in SHELDUS. Table 1 reports summary statistics
for these disasters. Column 2 reports the total property damages (in 2016 $
billions) caused by these disasters during our sample period of 1995–2016.
Table 1
Summary statistics: Climatic disasters
Disaster Total damages Average event damage County × Occurrence

($ billions) ($ billions)
Hurricane/tropical storm 2,409.635 92.678 594
Flooding 1,156.303 13.139 5,821
Tornado 351.165 4.445 482
Wildfire 92.923 3.716 174
Wind 142.890 1.623 3,158
Winter weather 74.197 1.324 4,196
Coastal 23.415 .780 509
Landslide 27.869 .753 142
Severe storm/thunderstorm 45.256 .551 6,954
Hail 18.555 .299 1,569
Tsunami/seiche .505 .168 5
Drought .804 .057 501
Lightning .351 .005 2,306
Fog .007 .001 75
Avalanche .016 .001 143
Heat .008 .001 567
This table reports information on the major climatic disasters in the United States during our sample period
(1995–2016). We obtain data from SHELDUS, maintained by Arizona State University. Reported damages are
in billions of 2016 dollars.
1155

Column 3 reports the average damage (in 2016 $ billions) associated with each
unique disaster event. The top-four disasters, based on the average damage
imposed per occurrence, are hurricanes/tropical storms,11 floods, tornadoes,
and wildfires. These are likely to be more salient disasters, and together account
for over 90% of the total damages caused by all climatic disasters during our
sample period. Finally, Column 4 reports the number of unique combinations
of County × Disaster. Figures 3 and 4 show the geographical distribution of
disaster counties.

2.2 Fund and firm data
We obtain data on actively managed, open-ended diversified U.S. equity mutual
funds from CRSP Survivor-Bias-free U.S. Mutual Fund database. Our sample
starts from January 1995. We eliminate index funds by using the CRSP-defined
index fund flags and by screening the names of funds for words such as “Index”
or “S&P.” We further remove funds whose names include words such as “ETF.”
The net (after-expense) monthly return comes from CRSP. To avoid multiple
counting of funds that have more than one class, we value-weight fund-class
returns using prior month total net assets to obtain fund level net returns.
Similarly, we also value-weight expense and turnover ratios. Fund size is the
sum of total net assets of all fund classes. Fund age is in years and is computed as
of the month end relative to the fund’s earliest first offer date. We exclude funds
with an erroneous negative age. We also obtain ZIP codes of fund headquarter
locations from CRSP, which are converted to the county level.
We obtain snapshots of quarterly holdings of funds from the Thomson
Reuters mutual fund holdings database. Because our focus is on U.S. equity
mutual funds, we exclude all funds whose Thomson objective code is one of the
following: International, Municipal Bonds, Bond, or Preferred. For funds that
do not report quarterly, we extrapolate the previous quarter shareholdings (e.g.,
adjusting for splits) to the current quarter. This is done for, at most, one quarter
to avoid excessively stale data. Holdings disclosures before a quarter end are
carried forward to the quarter end. From the fund-quarter portfolios identified
through the holdings data, we remove all funds whose total net assets (TNA)
are less than $5 million. We do not necessarily eliminate fund-quarters with
missing TNA because these observations are sometimes for funds that have
large previously disclosed TNA; we use the most recently reported number.
Because our focus is on diversified funds, we eliminate funds holding fewer
than 10 stocks. These funds are unlikely to be diversified. By design, we have to
restrict ourselves to the calendar quarters that correspond to the disasters listed
in Table 1. We combine the CRSP sample with the Thomson Reuters holdings
sample using the MFLINKS data set developed by Wermers (2000) and updated
11 SHELDUS categorizes hurricanes and tropical storms together, as hurricanes are also tropical storms with
sustained wind speeds over 74 miles per hour.
1156

[16:30 30/1/2020 RFS-OP-REVF190150.tex]

Page: 1157
Figure 3
1157
Disaster counties and the distribution of firms

This figure shows the geographical distribution of disaster counties in our sample. Bubble size represents the number of firms within the disaster county. The average logged size (in $millions)
1146–1183
of a firm located in the county is represented by shading. Lighter (darker) shades represent firms in the lowest (highest) quartile of size distribution.
[16:30 30/1/2020 RFS-OP-REVF190150.tex]

1158

Page: 1158
Figure 4
Disaster counties and the distribution of close funds
1146–1183
This figure shows the geographical distribution of disaster counties in our sample. Bubble size represents the number of funds within 100-mile radius of the disaster county (close funds). The
average logged size (in $millions) of a fund located close to the county is represented by shading. Lighter (darker) shades representing funds in the lowest (highest) quartile of size distribution.
Table 2
Summary statistics: Mutual funds
A. Fund-quarters (full sample)
(1) (2) (3) (4)
Variable NOBS Mean SD Median
Fund size ($ millions) 127,303 1,083 2,473 238
Fund age 127,303 13.66 12.72 1.16
Manager experience 96,294 7.54 5.58 6.00
Expense ratio 123,331 .013 .004 .012
Turnover ratio 123,331 .858 .725 .66
B. Treatment versus control funds

Treatment (close) Control (far) Diff p -value
Fund size 979 1,088 −109 .00
Fund age 12.91 13.70 −.79 .00
Manager experience 7.66 7.53 .13 .13
Expense ratio .013 .012 .01 .00
Turnover ratio .81 .86 .05 .00
Panel A reports summary statistics over the full sample of fund-quarters. Panel B reports characteristics of
treatment and control group funds separately, as well as a test of their differences. All variables are winsorized
at the 1st and 99th percentiles.
by Wharton Research Data Service. Our combined sample consists of 3,268

unique funds. Of these, 1,700 funds are within 100 miles of the disaster zone.
We obtain data on firm financials and headquarters’ county locations from
the Compustat Quarterly files.12 We eliminate all firms in the financial services
industry (SIC codes 6000–6999) and those firms for which the county location
of headquarters is missing. Further, we restrict our sample to the period 1995–
2016, because not many funds are in a disaster zone before 1995. This is
expected, because growth in the mutual fund industry increased substantially,
starting in the mid-1990s.
2.3 Descriptive statistics

Table 2 presents summary statistics for our data set. Panel A reports summary
statistics over all fund-quarters (Q−2 to Q+2) in the sample. The average
fund size in the full sample is $1.08 billion, while the median size is about $239
million. The average age of a typical fund is about 13.7 years. The average fund
manager has approximately 7.5 years of experience with her current fund. The
annual expense ratio is 1.30% and turnover is 86%, reflecting that our sample
consists solely of actively managed funds.
In panel B, we compare the sample of funds that are close (treatment group)
to the disaster zone, to those that are far away (control group). We find that
12 Ideally, we would like to know the actual location of the fund and plants of the firm. Unfortunately, we do not
have access to the plant or fund locations. Hence, we follow prior studies that also identify disaster zone firms
and funds based on the headquarters’ location (see, for instance, Chaney et al. 2012; Barrot and Sauvagnat 2016;
Dessaint and Matray 2017). Further, to the extent that the actual plant location may be geographically distant
from the headquarters, we are likely to misclassify nondisaster firms as disasters firms. If anything, this is likely
to bias our estimates against finding any effects (Barrot and Sauvagnat 2016).
1159

the funds in control and treatment groups are economically similar to each
other. They differ slightly in fund size, age, expense ratio, and turnover ratio.
Accordingly, we control for these differences in our regression specifications
in the next section.
3. Results
We begin by discussing the portfolio response of fund managers to climatic
disasters and then examine alternative explanations for our findings. Next, we
distinguish between the information and salience hypotheses. Finally, we briefly

report findings from a number of robustness tests.
3.1 Portfolio response to disasters

3.1.1 Do portfolio managers overreact? We begin our empirical analysis by
examining whether managers overreact to major salient climatic disasters. That
is, we ask whether funds close to the disaster zone underweight disaster zone
stocks to a greater degree than funds that are further away from the disaster
zone. This relative underweighting by close funds can be either due to the
saliency bias or because they possess superior information. Our tests are based
on specification 1 in Section 1.
We examine the portfolio response to the most damage-causing disasters
(hurricanes/tropical storms, tornadoes, and wildfires). Table 3 presents the
results from these tests. Focusing on Column 1, the positive and significant
coefficient on CLOSE ms indicates that, on average, fund managers exhibit a
preference for close stocks, that is, stocks of firms close to the funds receive
greater weights in their portfolios, as indicated in Figure 2. The coefficient
on POST t shows that, on average, even distant funds sell stocks located in
the disaster zone after the disaster strike. This is consistent with the idea that
these stocks may be adversely affected by the disaster or that salience bias may
affect both close and far funds. The coefficient of interest is the interaction
term, CLOSE ms ×POST t , which measures the differential portfolio response
of close and distant funds to the stocks in the disaster zone around the hurricane
strike.
The coefficient on CLOSE ms ×POST t is negative and statistically significant
with a p-value below 1%. Consistent with hypothesis H1, the coefficient on
the interaction term indicates that funds close to the disaster zone reduce their
portfolio weights on disaster zone stocks by approximately 0.09%, as compared
to the reduction in weight by funds that are far from the disaster zone. While
this magnitude may seem somewhat small, for a typical fund in our sample, this
translates into an 1.5% drop in dollar value of each disaster stock.13 In Column
13 The median size of a mutual fund portfolio in our sample is $239 million. The dollar value invested in the median
disaster zone stock by close funds is $2.05 million (unreported in tables). So a reduction in overall portfolio
weight of .09% translates into .09∗$239million
$2.05million
≈ 1.5% reduction in the dollar value of a stock holding.
1160

Table 3
Portfolio response to climatic disasters (hurricanes, tornadoes, and wildfires)
(1) (2) (3) (4) (5) (6)
Closems .405 .066 .086 .072 .093 .091
(.000) (.000) (.000) (.000) (.000) (.000)
Postt −.021 −.041 −.040 −.041 −.042 −.034
(.000) (.000) (.000) (.000) (.000) (.000)
Closems × POSTt −.089 −.045 −.046 −.050 −.051 −.046
(.000) (.000) (.000) (.000) (.000) (.000)
Debt/assetss,t−1 .009 .005 .008
(.321) (.578) (.418)
LBMs,t−1 .055 .044 .038
(.000) (.000) (.001)

Lsizes,t−1 .228 .231 .234
(.000) (.000) (.000)
ROAs,t−1 .881 .769 .612
(.000) (.000) (.000)
Sales growths,t−1 −.001 .003 −.005
(.808) (.591) (.359)
Momentums,t−1 .056 .064 .069
(.000) (.000) (.000)
Expense ratios,t−1 −5.025 −2.054 −2.191
(.093) (.469) (.441)
Turn ratios,t−1 −.030 −.031 −.033
(.010) (.008) (.005)
Lfundsizes,t−1 −.070 −.081 −.081
(.000) (.000) (.000)
Fund returnss,t−1 −.202 −.519 −.670
(.004) (.000) (.000)
Net flows,t−1 −.000 −.000 .001
(.008) (.000) (.000)
Stock returnss,t .333
(.000)
Constant 1.019 .707 −2.688 1.266 −2.121 −2.144
(.000) (.000) (.000) (.000) (.000) (.000)
Year FE No Yes Yes Yes Yes Yes
Disaster quarter FE No Yes Yes Yes Yes Yes
Fund FE No Yes Yes Yes Yes Yes
County FE No Yes Yes Yes Yes Yes
N 1,184,461 1,184,461 1,138,551 1,138,075 1,094,620 1,094,406
Adj R 2 .004 .514 .570 .515 .572 .575
This table reports the coefficient estimates from the following regression model:
WEIGHT mst = β0 +β1 CLOSE ms +β2 POST t +β3 (CLOSE ms ×POST t )+Xs,t−1 +
Xm,t−1 +μm +δt +mst ,
where m refers to a fund, s refers to a firm, and t refers to a quarter. We focus on two quarters before to two
quarters after the disaster. CLOSE ms equals 1 for fund-firm pairs that are less than 100 miles away from each
other and 0 otherwise. POST t equals 1 for the disaster quarter and the two following quarters and 0 otherwise.
The data span the period 1995–2016. Standard errors are robust to heteroscedasticity and clustered at the fund
level. p-values are reported in parentheses.
2, we repeat our tests after controlling for year, disaster-quarter, and fund and
county fixed effects and obtain qualitatively similar results.
In Column 3, we follow Kang and Stulz (1997) and repeat the results
after controlling for the log of firm size (LSI ZE), log of 1 + book-to-
market ratio (LBM), price momentum (MOM), profitability (ROA), sales
growth (SALESGROW T H ), and leverage (DEBT /ASSET S), in addition
1161

to fixed effects. All firm level covariates are measured at the end of the
quarter t −1.14 We find that our results are robust to controlling for these firm
level characteristics. In Column 4, we control for fund characteristics, such as
expense ratio, turnover ratio, fund size, fund returns (average monthly return
over quarter t −1), and net flows during quarter t −1. In Column 5, we control
for both firm and fund characteristics, and find our results to be robust. Finally,
in Column 6, we also control for the contemporaneous stock returns and find
similar results. In other words, our results persist even after holding stock prices
constant.
We exclude floods from the above tests, as prior evidence suggests that

people respond differently to floods as compared to the other disasters. (In
untabulated tests, we find only weak evidence of any portfolio overreaction to
floods.) Our finding of a muted response to floods is consistent with Boustan
et al. (2012), who find that, although people are less likely to move into
areas that are hit by tornadoes, they are more likely to move into areas hit by
flooding. They argue that the government response in rebuilding flood-prone
areas may have resulted in individuals becoming more immune to such events.
The overall effect of floods is much smaller compared to the other disasters,
throughout the remainder of our paper, so we focus on the most salient disasters
(hurricanes/tropical storms, tornadoes, and wildfires), unless stated otherwise.
3.1.2 Temporal dynamics of portfolio response to disasters. We now

examine the temporal dynamics of portfolio changes in response to disasters.
Specifically, we are interested in two questions. First, when does the differential
response of close funds, relative to distant funds, begin, and, second, for how
long does this differential response last? Note that the absence of pre-trends
(differential response before the disasters) in the outcome variable, which we
will confirm shortly, is a necessary condition for the validity of our difference-
in-differences setting. Formally, we run the following regression specification:
WEIGHT mst =β0 +β1 (CLOSE ms ×P RE[−4,−3])+β2 CLOSE ms +
β3 (CLOSE ms ×POST [0,2])+β4 (CLOSE ms ×POST [3,4])+
β5 (CLOSE ms ×POST [5,6])+θ0 P RE[−4,−3]+
θ1 POST [0,2]+θ3 POST [3,4]+θ3 POST [5,6]+Xs,t−1 +Xm,t−1 +
μm +δt +mst ,
(2)
14 LSI ZE is the log of market equity. MOM is the cumulative 12-month return of a stock, excluding the immediate
past month. ROA is the ratio of earnings before interest, depreciation, and taxes to the book value of total assets.
SALESGROW T H is the percentage quarterly change in firm sales, as compared to the same fiscal quarter of
the prior year.
1162

where m refers to a fund, s refers to a firm, and t refers to a quarter. We focus

on four quarters prior to six quarters after a disaster. That is, if the disaster
strikes in quarter t = Q, we focus on quarters Q−4 to Q+6. POST [0,2] is a
dummy variable that equals 1 for quarters Q, Q+1, and Q+2, and 0 otherwise.
POST [3,4] and POST [5,6] are defined analogously. P RE[−4,−3] equals 1
for quarters Q−3 and Q−4, and 0, otherwise. The omitted category in these
tests are quarters Q−2 and Q−1. Therefore, the coefficients on the interaction
terms CLOSE ms ×POST t compares the difference in portfolio weights of
disaster zone stocks between close and far funds in other quarters, relative
to quarters Q−2 and Q−1. The remainder of control variables is the same as

in specification 1.
Table 4 reports the results. First, focusing on Column 1, consistent with the
absence of pre-trends, we find that there is no differential trend between close
and distant funds with respect to disaster zone firms before the disaster strikes.
The coefficient on CLOSE ms ×P RE[−4,−3] is statistically indistinguishable
from zero.
Second, focusing on the effect of the disasters during the period after the
disaster, we find that the underweighting by close funds, relative to distant funds,
in disaster zone stocks decreases progressively through time. For instance,
the coefficient on the interaction term CLOSE ms ×POST [0,2], in model 1,
is −.089. This coefficient remains similar in the next two quarters (-.079),
before decreasing to a statistically insignificant −.048. Similar results obtain
in Column 2, although the magnitude on the coefficients are smaller, when we
also include control variables and fixed effects. Overall, we conclude that, as
time passes, the differential portfolio response of close versus distant funds
relative to the disaster zone stocks diminishes. This result is consistent with the
predictions of the salience hypothesis: as salient events pass into the past, their
impact decreases with time.
3.1.3 Do managers learn? Bernile et al. (2017) show that disasters affect the
risk attitude of fund managers. They find that, compared to fund-managers not
experiencing a disaster, fund managers in the disaster area reduce their portfolio
volatility due to their becoming more (irrationally) averse to risk, in general,
including risks that are unaffected by the disaster; this risk aversion declines
with repeated exposures to disasters. In our context, we wish to know whether
portfolio overreaction varies with the prior disaster experience of managers. For
each fund in our sample, we first compute the number of major disasters that
a fund manager has experienced at any given point in time during our sample
period. We, then, split our sample into quartiles based on the prior disaster
experience of managers in a fund.
Table 5 shows the results of this test. Columns 1 to 4 show results for
quartiles 1 to 4, respectively. We find that, for all managers, the coefficient on the
interaction term, CLOSE ms ×POST t , is negative and statistically significant.
However, as we move from Columns 1 to 4 (i.e., as managers’ prior experience
1163

Table 4
Dynamics of portfolio response to disasters
(1) (2)
Closems × Post[0,2] −.089 −.046
(.000) (.000)
Closems × Post[3,4] −.079 −.042
(.000) (.001)
Closems × Post[5,6] −.048 −.005
(.053) (.736)
Closems .405 .078
(.000) (.000)
Closems × Pre[-4,-3] .005 .012
(.743) (.233)

Post[0,2] −.021 −.031
(.000) (.000)
Post[3,4] .048 −.012
(.000) (.000)
Post[5,6] .066 −.011
(.000) (.004)
Pre[-4,-3 .039 .064
(.000) (.000)
Constant 1.019 −2.187
(.000) (.000)
Year FE No Yes
Disaster quarter FE No Yes
Fund FE No Yes
County FE No Yes
Controls No Yes
N 2,092,240 1,953,941
Adj R 2 .004 .580
WEIGHT mst =β0 +β1 (CLOSE ms ×P RE[−4,−3])+β2 CLOSE ms +β3 CLOSE ms ×POST [0,2]+
β4 (CLOSE ms ×POST [3,4])+β5 (CLOSE ms ×POST [5,6])+θ0 P RE[−4,−3]+
θ1 POST [0,2]+θ3 POST [3,4]+θ3 POST [5,6]+Xs,t−1 +Xm,t−1 +μm +δt +mst ,
where m refers to a fund, s refers to a firm, and t refers to a quarter. We focus on four quarters before to six quarters
after the disaster. That is, if the disaster strikes in quarter t = Q, we focus on quarters Q−4 to Q+6. POST [0,2]
is a dummy variable that equals 1 for quarters Q, Q+1, and Q+2 and 0 otherwise. POST [3,4] and POST [5,6]
are defined analogously. P RE[−4,−3] equals 1 for quarter Q−3, and Q−4 and 0 otherwise. Controls are the
same as those used in model 5 of Table 3. The data span the period 1995–2016. Standard errors are robust to
heteroscedasticity and clustered at the fund level. p-values are reported in parentheses.
to disaster increases), we find that the magnitude of the coefficient decreases.

For the third and fourth quartiles, the coefficient is statistically insignificant.
Our findings suggest that managers learn or otherwise become less affected by
salience and that they exhibit less overreaction as they become more disaster
experienced.
3.2 Alternative explanations

We now examine three alternative explanations for our findings. We elaborate
below.
3.2.1 Are the results mechanically driven by the drop in stock prices? One
potential concern regarding our tests is that our results may be mechanically
1164

Table 5
Do managers learn? Portfolio response to disasters based on the prior disaster experience of fund
manager
Disaster experience quartile First Second Third Fourth
(1) (2) (3) (4)
Closems .130 .101 −.026 .063
(.002) (.008) (.424) (.078)
Postt −.068 −.051 −.040 −.030
(.000) (.000) (.000) (.000)
Closems × POSTt −.052 −.063 −.027 −.022
(.023) (.004) (.148) (.196)
Debt/assetss,t−1 .047 .034 −.031 −.001
(.041) (.115) (.152) (.970)

LBMs,t−1 .018 .028 .040 .036
(.470) (.211) (.038) (.076)
Lsizes,t−1 .266 .241 .216 .203
(.000) (.000) (.000) (.000)
ROAs,t−1 .949 .624 .533 .870
(.000) (.001) (.008) (.000)
Sales growths,t−1 .013 .001 −.010 −.014
(.275) (.926) (.400) (.233)
Momentums,t−1 .071 .077 .075 .084
(.000) (.000) (.000) (.000)
Expense ratios,t−1 −4.360 −3.014 −1.924 −5.164
(.412) (.576) (.797) (.442)
Turn ratios,t−1 −.065 −.017 .005 .011
(.017) (.413) (.840) (.733)
Lfundsizes,t−1 −.069 −.108 −.117 −.111
(.000) (.000) (.000) (.000)
Fund returnss,t−1 −1.025 −.378 −.527 −.383
(.000) (.004) (.000) (.004)
Net flows,t−1 −.000 −.001 −.000 −.001
(.256) (.011) (.376) (.019)
Constant −2.138 −2.030 −1.228 −1.490
(.000) (.000) (.001) (.000)
Year FE Yes Yes Yes Yes
Disaster quarter FE Yes Yes Yes Yes
Fund FE Yes Yes Yes Yes
County FE Yes Yes Yes Yes
N 202,374 180,285 153,878 166,667
Adj R 2 .577 .592 .609 .564
WEIGHT mst = β0 +β1 CLOSE ms +β2 POST t +β3 (CLOSE ms ×POST t )+Xs,t−1 +Xm,t−1 +μm +δt +mst ,
other, 0 otherwise. POST t equals 1 for the disaster quarter and the two following quarters, and 0 otherwise. We
split the sample into four quartiles based on the number of prior disaster events experienced by the managers.
Columns 1–4 report results for the first to fourth quartiles. The data span the period 1995–2016. Standard errors
are robust to heteroscedasticity and clustered at the fund level. p-values are reported in parentheses.
driven by a temporary drop in stock price of disaster zone firms around disasters.
This would automatically lead to a drop in portfolio weights of these stocks even
in the absence of any actual trading by the close funds. This is an interesting
puzzle in itself, as it implies that other investors may be selling the disaster
zone stocks in response to the disasters, driving prices downward. Although
1165

Table 6
Portfolio response to disasters (#shares held and dollar value of trades)
Traded value
Shares ($ millions)
(1) (2) (3) (4)
Closems .031 .022 .091 .059
(.124) (.005) (.125) (.177)
PostT −.011 −.016 −1.519 −1.569
(.000) (.000) (.000) (.000)
Closems x POSTT −.010 −.009 −.283 −.222
(.041) (.014) (.007) (.000)
Debt/assetss,t−1 −.029 −.042
(.000) (.141)

LBMs,t−1 −.006 −.042
(.518) (.153)
Lsizes,t−1 −.107 −.060
(.000) (.000)
ROAs,t−1 −.033 −1.277
(.622) (.000)
Sales growths,t−1 .005 .099
(.167) (.002)
Momentums,t−1 .010 −.077
(.001) (.000)
Expense ratios,t−1 3.773 4.928
(.018) (.163)
Turn ratios,t−1 −.017 .026
(.001) (.034)
Lfundsizes,t−1 .173 −.202
(.000) (.000)
Fund returnss,t−1 −.069 −.336
(.066) (.184)
Net flows,t−1 −.000 .006
(.815) (.000)
Constant .271 .945 .676 2.554
(.000) (.000) (.000) (.000)
Year FE No Yes No Yes
Disaster quarter FE No Yes No Yes
Fund FE No Yes No Yes
County FE No Yes No Yes
N 1,202,956 1,110,947 820,278 756,963
Adj R 2 .000 .602 .029 .050
Ymst = β0 +β1 CLOSE ms +β2 POST t +β3 (CLOSE ms ×POST t )+Xs,t−1 +Xm,t−1 +μm +δt +mst ,
where m refers to a fund, s refers to a firm, and t refers to a quarter. The dependent variable in Columns 1 and
2 is # shares held, which is defined as the ratio of total number of shares of stock s held by fund m in quarter
t to the total number of shares of stock s outstanding. The dependent variable in Columns 3 and 4 is Traded
value, which is the dollar value (in millions) of the shares bought or sold. We focus on two quarters before to two
The data span the period 1995–2016. Standard errors are robust to heteroscedasticity and clustered at the fund
level. p-values are reported in parentheses.
our results in Table 3 already control for stock returns,15 we go a step further
and repeat our tests with two alternative dependent variables in Table 6.
15 We interpret these results with caution. To the extent that stock returns are affected by disasters, we avoid
controlling for them in the baseline regression, because doing so would lead to the “bad control” problem and
confound our coefficient estimates (see Angrist and Pischke 2008, section 3.2.3).
1166

In models 1 and 2, our dependent variable is SH ARESmst scaled by total

shares outstanding of stock s in quarter t, where SH ARESmst is the number
of shares of stock s held by fund m in quarter t. In models 3 and 4, we further
evaluate the robustness of our tests by changing to LHS variable to dollar value
of trades.16 In models 1–4, we find that consistent with the hypothesis H1 and
our results from Table 3, the statistically significant coefficient on the interaction
term shows that the funds close to the disaster zone reduce their holdings of
disaster zone stocks more than that by the funds that are far away from the
disaster zone.

3.2.2 Are the results related to a particular fund characteristic? A second
possible explanation is that our results are driven by a particular characteristic
of a mutual fund, and not necessarily by portfolio manager overreaction. That
is, our findings may be related to a particular type of fund, such as more
concentrated funds, small funds, relatively young funds, or funds managed
by less-experienced managers. Small funds are likely to be overinvested
in local stocks, resulting in excessively risky portfolios (Pool et al. 2012).
Thus, they may be particularly sensitive to temporary shocks stemming from
disasters and find it optimal to reduce the exposure to disaster stocks. If only
small funds exhibit overreaction, then this may cast doubt on the economic
impact of overreaction. Similarly, the overreaction may be coming from more
concentrated funds who want to reduce exposure to the disaster-affected stocks.
It is also possible that newer funds or funds managed by less-experienced
managers may disproportionately sell these stocks.
To examine these possibilities, we divide our sample by fund characteristics
(concentration, fund size, fund age, and manager experience) with below- and
above-median values of each fund characteristic and then estimate Equation (1)
in the two subsamples. Table 7 shows the results. We find that our results come
from both concentrated and well-diversified funds. Similarly, the overreaction
is seen in both small and large funds, young and old funds, and funds managed
by both young and experienced managers.
3.2.3 Are fund managers catering to the withdrawal requests of investors?

A third alternative explanation is that fund managers may reduce their portfolio
investments in disaster zone stocks not because they are themselves biased,
but because they may be catering to the preferences of their investors. For
example, retail investors may exhibit salience bias and overreact to the disasters
by liquidating their investments in funds with greater investments in the disaster
zone stocks. Alternatively, the local bias of individual investors (preference
for stocks located in close proximity; see Ivković and Weisbenner 2005) may
16 We obtain the shares traded by fund m in quarter t by comparing the shares held by the fund in quarter t and t +1
(adjusted for, e.g., stock splits).
1167

Table 7
Fund characteristics and portfolio response to disasters
Concentration Number of stocks Fund size Fund age Manager experience
Above Below Below Above Above Below Above Below Above Below
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Closems .122 .025 .103 .039 .101 .072 .067 .094 .063 .097
(.000) (.030) (.000) (.029) (.000) (.003) (.001) (.000) (.004) (.000)
Postt −.064 −.024 −.062 −.023 −.047 −.037 −.039 −.045 −.048 −.041
(.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000)
Closems x Postt −.074 −.014 −.053 −.028 −.038 −.059 −.059 −.045 −.046 −.051
(.000) (.049) (.000) (.001) (.003) (.000) (.000) (.000) (.001) (.000)
Fund controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes

Firm controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Disaster quarter Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Fund FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
County FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 541,539 553,081 539,959 554,661 518,714 575,906 527,164 567,395 401,886 417,369
ADJRSQ .461 .496 .428 .385 .597 .553 .635 .537 .574 .587
In this table, we analyze whether the portfolio response to disasters varies with fund concentration, fund size, fund
age and manager experience. We use two measures for fund concentration/diversification: Herfindahl-Hirschman
index (HHI) based on the weights allocated to each stock in the fund’s portfolio and the number of stocks in the
portfolio. Specifically, Column 1 (2) reports the results for the subsample of funds with HHI concentration above
(below) the median, and Column 3 (4) reports the results for the subsample of funds with the number of stocks
below (above) the median. Column 5 (6) reports the results for the subsample of funds with fund size above
(below) the median, and Column 7 (8) reports the results for the subsample of funds with fund age above (below)
the median. Column 9 (10) reports the results for the subsample of funds with fund manager’s experience above
(below) the median. We run the following regression model for each of the subsamples:
Where m refers to fund, s refers to firm, and t refers to quarter. We focus on the two quarters before to two
other, and 0 otherwise. POST t equals 1 for the disaster quarter and the two following quarters and 0 otherwise.
Controls are the same as in model 5 of Table 3. The data span the period 1995–2016. Standard errors are robust
to heteroscedasticity and clustered at the fund level. p-values are reported in parentheses.
induce a preference for local mutual funds (Bailey et al. 2011), and, to the extent
that such local investors may themselves have over invested in local firms, they
may simply want to reduce their exposure to local firms. Thus, close funds
may experience outflows driven by the behavioral bias of their investors, and,
consequently, managers may reduce their holdings in disaster zone firms to
assuage such investors and curb outflows (a clientele effect). This is interesting
in itself, as it suggests that, even though fund managers may themselves be
rational, flow-driven trading pressures created by behavioral biases of their
investors may still cause them to trade in a biased way.17
To examine the above possibility, we test whether our results are driven
by some specific socioeconomic characteristic of fund investors. To fix ideas,
let us say that the only investor characteristic that influences their investment
17 Coval and Stafford (2007) show that flow-driven mutual fund trading pressure can cause temporary stock
mispricing.
1168

preference is age. For instance, older investors and those closer to retirement
may be particularly risk-averse and prefer lower volatility in their portfolios. If
disasters result in higher volatility in disaster zone stocks, these investors may
rationally prefer to hold funds with less exposure to such stocks. Alternatively,
younger investors have less experience, and, consequently, may be more likely
to exhibit salience bias. Thus, if we find that our results hold only for the
subsample of funds with higher proportion of younger or older investors, it may
cast doubt on our thesis of salience bias in the behavior of mutual fund managers.
Conversely, if our results hold in both subsamples, it would be difficult to argue
in favor of a clientele effect.

Based on this idea, we reestimate our baseline Equation (1) in subsamples
based on a host of socioeconomic characteristics of the local populace.
Specifically, to design our tests, we follow the prior literature which highlights
that socioeconomic characteristics influence investment decisions of retail
investors; for example, Kumar (2009) shows that age, income, education, and
race influence investment preferences of individuals. In particular, he finds
that investors who are poorer, younger, belong to a specific minority (African
American and Hispanic), live in counties with higher unemployment rates
and low-income are more likely to make suboptimal investment decisions.
Perhaps this suboptimal behavior includes such investors overreacting to a
natural disaster.
We proxy for the client base of the funds using the socioeconomic
characteristics of the local populace in the disaster county. For this purpose, we
obtain data on income and poverty estimates from SAIPE (Small Area Income
& Poverty Estimates), data on the civilian unemployment rate from the Bureau
of Labor Statistics (BLS), and data on age and race from the U.S. Census
Bureau.
Table 8 shows the results. In all specifications, we control for the extent of
damage caused by a disaster.18 In Columns 1 and 2, we split our sample based
on whether the unemployment rate in the disaster county is above or below
the full sample median and reestimate our baseline Equation (1) in the two
subsamples. Consistent with our baseline estimates reported in Table 3, we
find that the coefficient on the interaction term, CLOSE ms ×POST t is negative
and statistically significant in both subsamples. Similarly, in Columns 3 and 4;
5 and 6; 7 and 8; and 9 and 10, we split our sample based on the proportion of
poor, proportion of older adults (ages 45 and older), proportion of black, and
proportion of Hispanic populations, respectively. We find that the coefficient
estimates are statistically significant and qualitatively similar in magnitude
across all subsamples. These tests provide suggestive evidence against the
18 We collect data on the damage caused by a disaster in a county from SHELDUS.
1169

Table 8
Socioeconomic clienteles and portfolio response to major disasters
Unemployment rate Prop poor Prop Elderly Prop black Prop Hispanics
Below Above Below Above Below Above Below Above Below Above
(1) (2) (3) (4) (5) (6) (7) (8) (9) (10)
Closems .108 .070 .089 .111 .129 .068 .086 .106 .087 .104
(.000) (.003) (.000) (.000) (.000) (.001) (.000) (.000) (.000) (.000)
Postt −.046 −.035 −.041 −.045 −.035 −.054 −.039 −.045 −.045 −.039
(.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000)
Closems x Postt −.060 −.028 −.055 −.039 −.038 −.052 −.042 −.073 −.052 −.047
(.000) (.051) (.000) (.024) (.018) (.000) (.000) (.000) (.000) (.001)
Property damage −.000 .000 −.000 .000 .000 .000 .000 .000 .000 .000

(per capita)
(.876) (.105) (.207) (.004) (.120) (.075) (.349) (.000) (.002) (.941)
Constant −2.057 −1.719 −2.295 −1.576 −2.105 −1.940 −2.112 −1.787 −1.923 −1.982
(.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000)
Fund controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Firm controls Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Year FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Disaster quarter FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
Fund FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
County FE Yes Yes Yes Yes Yes Yes Yes Yes Yes Yes
N 625,955 431,359 587,829 469,485 598,778 484,010 614,137 468,646 631,959 450,829
Adj R 2 .570 .590 .573 .588 .575 .588 .574 .586 .578 .579
In Columns 1 and 2, we split the sample based on whether the unemployment rate in the disaster county is above
or below the full sample median. Similarly, in Columns 3 and 4; 5 and 6; 7 and 8; and 9 and 10, we split the
sample based on the proportion of poor, proportion of older adults (ages 45 and greater), proportion of black,
and proportion of Hispanic populations, respectively. DAMAGE is the damage caused by a disaster (USD 2016
per capita) in a firm’s county. Controls are the same as those used in model 5 of Table 3. The data span the period
1995–2016. Standard errors are robust to heteroscedasticity and clustered at the fund level. p-values are reported
in parentheses.
clientele effect. Overall, we conclude that the evidence in favor of a clientele

effect is not compelling.19
3.3 Is underweighting by portfolio managers rational or driven by

salience?
Our results, so far, indicate that close mutual fund managers underweight
disaster zone stocks, while distant fund managers do not, and that this behavior
appears to be driven by some commonality among close managers rather than
19 We also conduct one additional test (untabulated). We explicitly examine whether close funds experience greater
outflows relative to distant funds. Again, in these tests, we focus on two quarters prior to two quarters after the
disaster. Each observation represents a unique fund-quarter. The dependent variable in these tests is net flow. Net
[T N Am,t −T N Am,t−1 (1+Rm,t )]∗100
flow for fund m in quarter t is defined as Net Flowm,t = T N Am,t−1 , where T NAm,t
represents the total net assets of fund m at the end of quarter t, and Rm,t is the CRSP net return of fund m during
quarter t. The coefficient on the interaction term CLOSE ms ×POST t does not suggest a differential change in
net flows into close funds, as compared to net flows into distant funds, following a natural disaster.
1170

some differentiating characteristic of their funds or clientele. Accordingly, we

now focus on two competing alternative hypotheses (to the null of no difference
in the reaction of close vs. distant mutual funds to disasters) regarding the
common motivation of close portfolio managers: the information hypothesis
and the salience hypothesis. We first examine the portfolio response to near-
disaster firms and then perform two tests. These tests are based on post-disaster
profitability and return performance of disaster zone firms, conditional on being
underweighted by close versus far mutual funds.
3.3.1 Portfolio response to disaster versus near-disaster firms. We analyze

the portfolio response of close and far funds with respect to firms located
in near-miss counties that are not directly affected by the disaster. We are
interested in analyzing whether close funds also underweight neighboring
zone (in neighborhood of the disaster zone) firms. Specifically, each disaster
zone county is matched with its five closest neighboring counties that are not
directly affected by the disasters. To perform the matching, we first compute
the geographical distance between the centers of disaster counties as reported
in SHELDUS and all nonaffected adjacent counties. The distance is computed
using a Haversine formula from the latitude and longitude coordinates of the
centers of counties. We match each disaster county to its five nearest neighbors
with replacement.20 For ease of reference, we call the set of such counties as
near-disaster zone.
The salience hypothesis suggests that fund managers close to disaster zone
firms rely on the availability heuristic and, thus, overestimate the probability
that a disaster will occur again in the future to disaster zone firms. However, if
fund managers especially overinterpret the effects of a climatic disaster at or
near its epicenter, we would expect that the salience bias would be much weaker
with respect to firms located in near-disaster zones, relative to firms within the
disaster zone. The vividness of a climatic disaster to fund managers who are
local—with respect to disaster zone firms versus near-disaster zone firms—is
plausibly affected by local media (both public media and social media) that
provides detailed coverage and commentary on the effects of the disaster on
localities, such as townships, factories, neighborhoods, and shopping centers.
That is, local funds are likely affected much more by such granular news than
are distant funds, which likely view less granular national news coverage.
To check this, we examine the portfolio response for four groups of fund-firm
pairs: (1) Close × Disaster, (2) Far × Disaster, (3) Close × Near-Disaster, and
(4) Far × Near-Disaster. Close × Disaster represents fund-firm pairs, where
the firm is located in a disaster zone county and the fund holding its stock has
headquarters located within 100 miles of the firm headquarters. Recall that, in
20 A disaster often affects multiple counties in the disaster zone. Therefore, one neighboring county can be matched
to more than one county in the disaster zone, thus, serving as a control for more than one treatment (disaster)
county.
1171

Table 9
Portfolio response to climatic disasters (disaster vs. near-disaster zone firms)
A. Pre- versus post- B. Difference-in-difference-in-differences
Disaster Near-disaster
Close Far Close Far
(1) (2) (3) (4) (5) (6) (7) (8)
Postt −.097 −.042 −.051 −.046 −.047 −.047 −.047 −.047

(.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000)
Closems .008 .008 .008 .008
(.391) (.391) (.391) (.391)
Disasterst −.007 −.007 −.007 −.007

(.008) (.008) (.008) (.008)
Closems × Disasterst .064 .064 .064 .064
(.000) (.000) (.000) (.000)
Postt × Disasterst .006 .006 .006 .006
(.000) (.000) (.000) (.000)
Closems × Postms .003 .003 .003 .003
(.515) (.515) (.515) (.515)
Closems × Postms × −.050 −.050 −.050 −.050
DISASTERst
(.000) (.000) (.000) (.000)
Constant −4.005 −2.083 −2.724 −2.375 1.091 1.091 1.091 1.091
(.000) (.000) (.000) (.000) (.000) (.000) (.000) (.000)
Fund controls Yes Yes Yes Yes No Yes No Yes
Firm controls Yes Yes Yes Yes No No Yes Yes
Year FE Yes Yes Yes Yes Yes Yes Yes Yes
Disaster quarter FE Yes Yes Yes Yes Yes Yes Yes Yes
County FE Yes Yes Yes Yes Yes Yes Yes Yes
Fund FE Yes Yes Yes Yes Yes Yes Yes Yes
N 48,577 1,046,043 129,268 3,330,218 4,913,749 4,913,749 4,913,749 4,913,749

Adj R 2 .626 .573 .598 .565 .499 .499 .499 .499
This table reports the coefficient estimates from the following regression models: A. Pre- versus post-(single
difference)
WEIGHT mst = β0 +β1 POST t +Xs,t−1 +Xm,t−1 +μm +δt +mst
B. Difference-in-difference-in-differences
WEIGHT mst = β0 +β1 CLOSE ms +β2 POST t +β3 DISASTERst +β4 CLOSE ms ×POST t +
β5 CLOSE ms ×DISASTERst +POST t ×DISASTERst +
CLOSE ms ×POST t ×DISASTERst +Xs,t−1 +Xm,t−1 +μm +δt +mst ,
where m refers to a fund, s refers to a firm and t refers to a quarter. We focus on two quarters before to two quarters
after the disaster. CLOSE ms equals 1 for fund-firm pairs that are less than 100 miles away from each other and 0
otherwise. POST t equals 1 for the disaster quarter and the two following quarters and 0 otherwise. DISASTERst
equals 1 for firms headquartered in the disaster zone and 0 for firms headquartered in the neighborhood zone.
Columns 1 and 2 (3 and 4) reports the results for the subsample of firms in disaster (near-disaster) counties.
Columns 1 and 3 (2 and 4) report the results for close (far) funds. The data span the period 1995–2016. Standard
errors are robust to heteroscedasticity and clustered at the fund level. p-values are reported in parentheses.
our baseline tests in Table 3, we compare the reaction of funds located close to
disaster zone firms relative to far funds with regards to disaster zone stocks. In
panel A of Table 9, we perform a single difference test and examine the change
in portfolio weights around the disaster event for the four paired groups. We find
that the extent of underweighting is, by far, greatest for the Close × Disaster
1172

group, as shown by the coefficient on the dummy variable POST t , in each

column. Even far funds underweight disaster zone stocks following an event,
but the extent of underweighting by such funds is roughly half of the magnitude
of that for close funds. Further, the extent of underweighting of near-disaster
zone firms by both close and far funds is comparable in magnitude.
To formally test whether the portfolio response of close and far funds
is different for disaster versus near-disaster zone firms, we also perform
a difference-in-difference-in-differences test: different funds (close vs. far),
different counties (disaster vs. near-disaster), different time periods (before
vs. after the climactic disaster). Panel B of Table 9 summarizes the results

from these tests. The dependent variable, WEIGHT mst , is the weight in the
portfolio of mutual fund m on stock s located in the disaster or near-disaster
zone at the end of quarter t. Again, CLOSE ms equals 1 if the headquarters
of mutual fund m is located within 100 miles of the headquarters of the
firm s, else CLOSE ms equals . DISASTERst equals 1 for firms headquartered
in the disaster zone and 0 for firms headquartered in the near-disaster zone
(fund and firm control variables are defined similar to those in panel A; their
coefficients are not reported in panel B). The key findings of these tests is
that the coefficient on the interaction term, CLOSE ms ×POST t , is statistically
indistinguishable from zero, suggesting that the portfolio response of close and
far funds is similar for near-disaster zone firms. Importantly, the coefficient
on the interaction term, CLOSE ms ×POST t ×DISASTERst , is negative and
statistically significant, implying that the underweighting of firms by close
funds is significantly greater for disaster zone firms. In summary, we conclude
from these tests that saliency impact of climatic disasters on mutual fund
managers decreases with distance.
3.3.2 Impact on profitability. Our results, documented thus far, show that
close mutual funds reduce their weights on firms located in the disaster zone, but
not on firms located in the near-disaster zone. If the underweighting of disaster
zone stocks is driven by access to superior information regarding the future
performance of such stocks, then we would expect a drop in the performance
of disaster zone stocks, relative to near-disaster zone stocks, in the periods
after the disaster strikes. That is, the information hypothesis would suggest that
firms underweighted by a close fund in a near-disaster zone should outperform
underweighted firms in the disaster zone around hurricane strikes.
We employ a multivariate difference-in-differences strategy to evaluate
whether climatic disasters have an adverse incremental impact on the
performance of disaster zone firms relative to near-disaster zone firms. Table 10
reports estimates from the following regression specification:
PERFORMANCE st = β0 +β1 POST t +β2 DISASTERs

(3)
+β3 (POST t ×DISASTERs )+Xs,t−1 +μs +δt +st ,
1173

Table 10
Impact of disasters on firms’ performance
ROAs,t Sales growths,t
(1) (2) (3) (4)
Postt .002 −.002 2.123 2.860
(.459) (.190) (.303) (.319)
Disasterst −.010 .001 .352 2.036
(.359) (.089) (.250) (.315)
Postt × Disasters −.003 −.001 −2.312 −2.838
(.230) (.244) (.295) (.316)
Debt/assetss,t−1 −.029 .088 .764 −2.575
(.349) (.284) (.418) (.329)
LBMs,t−1 −.043 .002 .585 .692

(.309) (.861) (.629) (.363)
Lsizes,t−1 −.000 .014 .867 −2.067
(.824) (.202) (.326) (.355)
Momentums,t−1 −.011 −.004 −.354 .147
(.420) (.594) (.400) (.661)
ROAs,t−1 1.362 .592 −39.609 −103.607
(.002) (.000) (.182) (.340)
Sales growths,t−1 −.029 −.012 24.276 27.994
(.048) (.095) (.279) (.312)
Constant .046 −.223 −12.546 39.578
(.420) (.245) (.346) (.309)
Year FE No Yes No Yes
Disaster quarter FE No Yes No Yes
Firm FE No Yes No Yes
N 129,425 129,425 129,181 129,181
Adj R 2 .010 .564 .000 .023
This table reports the coefficient estimates for the following regression model:
Yst = β0 +β1 POST t +β2 DISASTERst +β3 (POST t ×DISASTERst )+Xs,t−1 +μs +δt +st ,
where s refers to a firm and t refers to a quarter. The dependent variable Y is ROA in Columns 1 and 2 and
Sales growth in Columns 3 and 4. POST t equals 1 for the disaster quarter and the two following quarters and
0 otherwise. DISASTERst is a dummy variable that identifies stocks that are in the disaster zone. It equals 1 for
all stocks that are in the disaster zone and 0 for the stocks in the near-disaster zone. The data span the period
1995–2016. Standard errors are robust to heteroscedasticity and clustered at the fund level. p-values are reported
in parentheses.
where s refers to a stock, and t refers to a quarter. The dependent variable

PERFORMANCE is ROA in Columns 1 and 2 and Sales Growth in Columns
3 and 4. POST t equals 1 for the disaster quarter t = Q and the two following
quarters Q+1 and Q+2 and 0 for quarters Q−1 and Q−2. DI SAST ERs is
a dummy variable that identifies firms in disaster zone counties - taking the
value 1 for all firms in the disaster zone and 0 for all firms in the near-disaster
zone for all quarters from Q−2 to Q+2. The coefficient of interest is β3 , which
measures the post-disaster difference in performance of firms in the disaster
zone, as compared to firms in near-disaster zone.
Focusing on the coefficient on the interaction term POST t ×DI SAST ERs
in Column 1, we observe a negative coefficient, indicating a larger impact
of disasters on “directly hit” counties, but it is statistically insignificant. In
Column 2, we repeat the test, after controlling for fixed effects, and obtain
1174

similar results. In Columns 3 and 4, we repeat these tests using sales growth as
a measure of firms’ performance and obtain qualitatively similar results. These
results are consistent with the idea that both firms that are headquartered in the
disaster zone area and those in nearby areas may experience a similar level of
profitability decrease, because of the types of widespread disruptions that occur
in and around the epicenter of a disaster (power outages, road closures, public
service downtime). Further, it is also possible, that while there may be an adverse
impact of the disasters on local sales, a greater fraction of the sales of these firms
may be coming from other regions not affected by the disaster, thereby limiting
the overall effect of the disasters on firm profitability. Thus, our findings indicate

that local mutual fund managers overreact to salient information regarding
disaster counties to a greater degree than to information regarding neighboring
counties and not to value-relevant informational differences between directly
hit and neighboring firms.
We note that our result of finding no effect for near-disaster zone firms is
consistent with Dessaint and Matray (2017), who also do not find any impact
of hurricanes on the neighborhood zone firms. For disaster zone firms, we
do find (a point estimate) that there is a differential drop in sales growth of
about 2.3%–2.8% which lies in the confidence interval of estimates reported in
DM.21 However, in our sample these effects are not statistically distinguishable
from zero. There are important differences in empirical design between our
papers that might explain why we find different results. First, our tests include
a wider set of disasters, such as tornadoes and wildfire, which have very focused
geographical effects, unlike hurricanes, which bring widespread destruction to
a geographical area. Second, and more importantly, we examine differences in
profitability in stocks that are held by local mutual funds in disaster counties
versus neighboring counties, whereas DM examine all stocks in neighboring
counties, and disaster counties, and compare their performance with firms in
rest of the United States (other zone firms). DM find that relative to other zone
firms, disaster zone firms are negatively affected by disaster, whereas there is no
impact on neighboring county firms. These differences in the firm sample are
important because mutual funds hold stocks that are different from the universe
of all stocks.
Specifically, mutual funds, on average, hold larger capitalization stocks with
stronger fundamentals, such that they may not be affected as much by a localized
natural disaster. Indeed, in untabulated tests, we are able to replicate the results
of DM using the full sample of disaster zone stocks both held and not held
by local funds and measure their performance relative to firms located in the
rest of the United States. These results are especially strong if we restrict the
treatment sample to disaster zone firms not in our sample (that is firms not
held by local funds) and repeat these tests. Consistent with such firms having
21 Barrot and Sauvagnat (2016) also find a weak evidence of a drop in sales growth of disaster zone firms by about
2.4%–2.6% two quarters after the event.
1175

smaller market capitalization, the median market cap of these nonheld firms is
$46 million. Finally, we restrict the treatment sample to the disaster zone firms
in our sample and find that although there is an economically meaningful drop
in sales growth, the coefficient estimate is not statistically distinguishable from
zero. Consistent with firms held by funds being larger, firms in our sample have
an average market cap of $354 million.
Overall, while the evidence from analyzing the post-disaster profitability
of disaster and near-disaster zone firms in our sample is consistent with the
salience hypothesis, our estimates regarding the impact of disasters on sales
growth in affected firms lie in the confidence interval of prior studies, such as

DM, who find that there is a drop in sales growth of disaster zone firms. Hence,
we interpret these results with caution and examine the impact on stock returns
in Section 3.3.3 to strengthen the evidence in favor of the salience hypothesis.
3.3.3 Impact on stock returns. We now perform a second type of test to

add evidence on the channel that drives the underweighting of disaster zone
stocks by funds close to the disaster zone. The idea is to evaluate the subsequent
return performance of disaster zone stocks. If the underweighting of disaster
zone firms by close funds is consistent with the information hypothesis, then
we would expect underweighted firms to underperform during the post-event
quarters. On the other hand, no underperformance is more consistent with the
salience hypothesis.
We first identify firms located in the disaster zone for each disaster. We then
identify the funds that hold the disaster zone stocks and are within 100 miles
of each of these firm’s headquarters. As before, we call such fund-firm pairs as
CLOSE ms . Then, for each stock i held by close funds, we calculate the average
change in portfolio weight across all close funds that hold the stock around the
disaster-quarter (event-quarter, Q0) as follows:

N
(WQ0,k(i) −(WQ−1,k(i) +WQ−2,k(i) )/2)
k=1
WQ0,i = , (4)
N
where WQ0,k(i) is the weight on stock i in the kth fund’s portfolio at the end of
Q0, while WQ−1,k(i) and WQ−2,k(i) are the pre-event weights on stock i in the
kth fund’s portfolio at the end of quarters Q-1 and Q-2. N is the total number
of close funds that hold stock i.
We then sort stocks into equal-weighted tercile portfolios by WQ0,i , and
track the performance of tercile portfolios for 3 years: 1 year before the event-
quarter (year-1) and for 2 years after the event-quarter (year+1 and year+2). We
calculate risk-adjusted returns, using DGTW benchmarks (Daniel et al. 1997;
Wermers 2004).22 Finally, for each event-quarter, we compute average yearly
22 Benchmark returns are buy-and-hold DGTW returns.
1176

DGTW-adjusted returns for years: year-1, year+1, and year+2, and quarterly
DGTW-adjusted returns for the event-quarter by averaging each equal-weighted
portfolio DGTW-adjusted return over all events. Our events are those climatic
disasters that cause the most damage, that is, hurricanes and tornadoes.23 Such
climatic disasters are arguably the most salient in our database, as hurricanes and
tornadoes are preceded by days of media coverage warning of their impending
arrival; in addition, these climatic disasters are, geographically, widespread in
their direct impact, and have large “footprint” in terms of average damage per
event (see Table 1). Table 11 shows the results. Panel A shows DGTW-adjusted
returns for close funds, and panel B reports results for far funds. Panel C reports

the difference (close-far). Tercile portfolio 1 is the most underweighted portfolio
(lowest WQ0,i ), whereas tercile portfolio 3 is the most overweighted portfolio
(highest WQ0,i ). 1-3 represents the zero-investment long-short portfolio that
is long tercile 1 and short tercile 3. We report Newey-West-corrected p-values
with a lag length of 3 in parentheses.
Panels A and B show that the year-1 return for the 1-3 portfolio are similar
(around −18.0%), and, thus, there is a difference in the performance of
underweighted minus overweighted portfolios of close and far funds prior to
the event-quarter. Panel C further confirms this. The difference is only 0.45%
per year and it is statistically insignificant. As expected, the event quarter zero
returns are large and negative for both close and far funds. This is in line with the
earlier studies, such as Wermers (1999), where trading quarter returns increase
in the direction of the trade (terciles 1 to 3) due to momentum investing by
mutual funds. It is the post-disaster returns that we are most interested in. Here,
we find economically large return reversals in year+1, and year+2. For instance,
for year+1, the 1-3 portfolio return is 11.38% per year with a p-value of .06.
Similarly, year+2, the return is 4.68% per year with a p-value of .03. Thus, in the
2 years following the climatic disaster, the underweighted portfolio by funds
in proximity of the disaster zone outperforms the overweighted portfolio by
about 16%. This evidence is consistent with portfolio managers misestimating
the economic risk associated with disaster zone stocks.
Panel B repeats our analysis for distant funds. Here we do not find significant
evidence of reversals. The year+1 and year+2 returns for the 1-3 portfolio are
6.08% (p-value = .13), and −3.50% (p-value = .21), respectively. Panel C
shows the difference in returns. We find a greater return reversal associated
with the 1-3 portfolio traded by close funds during the post-event years. The
incremental returns for the 1-3 portfolio traded by close funds over far funds
are 5.30% (p-value = .02) for year+1 and 8.19% (p-value = .02) for year+2.
We note, here, that the above evidence that disaster zone stocks
underweighted by disaster-area funds are most prone to reversals does not firmly
establish that these funds are responsible for a temporary “price-pressure”
23 We exclude floods for the reasons mentioned in Section 3.1.1.
1177

Table 11
Underweighting and stock returns
A. Close
Tercile Year-1 Event qtr Year+1 Year+2
1 1.141 −7.890 11.122 2.051
(.695) (.003) (.214) (.273)
2 −.782 −1.946 .148 2.393
(.811) (.257) (.955) (.377)
3 19.339 9.025 −.267 −2.633
(.010) (.001) (.945) (.390)
1-3 −18.198 −16.914 11.389 4.684
(.028) (.000) (.062) (.031)

B. Far
1 1.355 −11.738 7.712 −1.124
(.155) (.001) (.364) (.501)
2 −1.064 −2.053 −1.764 3.843
(.001) (.088) (.568) (.106)
3 2.011 1.453 1.625 2.384
(.009) (.003) (.733) (.457)
1-3 −18.656 −22.191 6.088 −3.508
(.018) (.001) (.130) (.217)
C. Close - far
1 −.214 3.849 3.410 3.175
(.938) (.036) (.003) (.051)
2 9.282 .107 1.912 −1.450
(.031) (.899) (.016) (.240)
3 −.672 −1.428 −1.892 −5.017
(.682) (.079) (.323) (.047)
1-3 .458 5.277 5.302 8.193
(.714) (.032) (.023) (.028)
This table reports quarterly equal-weighted portfolio DGTW-adjusted returns. In panel A (panel B), at the end
of each disaster quarter (event quarter), we sort firms into tercile portfolios on the basis of the average change
N

in stock weight ( (weightQ0,i − (weightQ−1,i +weightQ−2,i )/2)) across N funds that are close to (far
k=1
from) the disaster zone and hold stock i . We then compute equal-weighted DGTW-adjusted returns for each
portfolio for (a) 1 year before the event quarter (year-1), (b) event quarter, and (c) 2 years after the event quarter
(year+1 and year+2). Finally, we compute time-series average of yearly returns (year-1, year+1, and year+2) and
quarterly return (event quarter). Portfolio 1 is the most underweighted portfolio, whereas portfolio three is the
most overweighted portfolio. “1-3” represents a zero-investment long-short portfolio that is long tercile 1 and
short tercile 3. Panel C shows the difference in returns between panels A and B. Newey-West-corrected p-values
are reported in parentheses.
effect. This is somewhat outside the scope of our analysis, as it would involve a
market-wide assessment of the valuation of stocks subject to temporary disaster
risks. For instance, other market participants, outside the mutual fund universe,
could possibly push down prices. In that sense, our evidence in this section
can be interpreted, most precisely, as supporting the idea that mutual fund
managers who experience salience bias and oversell stocks within the disaster
zone likely impose a performance “cost” on their investors in the form of selling
1178

such stocks beyond the value-relevant information of disasters because of the

additional stimulus of salience.
3.4 Robustness
We perform a number of robustness tests. For brevity, we report a concise
summary of these tests.24
3.4.1 Extent of damage. We examine our baseline results by conditioning

disasters on the extent of damages caused by them. We split the sample of

major disasters (hurricanes, tornadoes, and wildfires) based on whether the
total damages in the disaster county is above or below the full sample median.
We also split the sample based on the per capita damages, in the disaster
county. We find stronger results for more damaging disasters. The coefficient on
the interaction term, CLOSE ms ×POST t , is larger for more damage-intensive
disasters. This is consistent with the idea that more damaging disasters are
likely to be more salient. The managers close to such disasters overreact more
by underweighting disaster zone firms more than that by distant managers. We
obtain similar results if we use per capita damages to classify the disasters.
3.4.2 Floods. We earlier noted, in Section 3.1.1, that we find weak evidence of
managerial overreaction to floods. However, when we examine the subsample
of floods that cause more damage (those with above-median total damages and
per capita damages), we find significant evidence that close funds underweight
stocks located in flood-hit counties more than that by distant funds. In another
test, we include floods along with hurricanes, tornadoes, and wildfires, and then
rerun our tests. We find robust results.
3.4.3 Other disasters. We separately examine managerial overreaction to

less prevalent disasters in our database, that is, disasters other than hurricanes,
tornadoes, wildfires, and floods. To the extent that these climatic disasters
account for a small fraction of the overall damages caused by all disasters, even
per capita, such events are less likely to be salient events. Consistent with this
idea, we don’t find a significant incremental reaction by close fund managers
relative to distant managers, for the most part. Of the 12 such disasters, only
droughts and winter weather exhibit significant coefficients on the interaction
term, CLOSE ms ×POST t .
3.4.4 Alternative definition of close funds. In our baseline tests, we define

a fund-firm pair to be close if the distance between them is less than 100 miles.
24 The Internet Appendix provides detailed tables for these tests.
1179

To ensure that our results are not driven by the choice of cutoff, we repeat
our baseline specification with alternative distance cutoffs of 50, 150, 200, and
250 miles to classify close fund-firm pairs. We find that the coefficient on the
interaction term is always statistically and economically significant in these
alternative specifications with the magnitude of the estimate decreasing with
distance.
3.4.5 Placebo test with alternative definition of close funds. We examine

results with a placebo definition of close funds. That is, we define close (treated)

funds as those between 900 to 1,000 miles away and all other funds as control
funds. We expect weak or no results if we use such a definition of close
funds. Indeed, we find that across all these alternative definitions of close
funds, the coefficient on the interaction term, CLOSE ms ×POST t , is statistically
insignificant.
3.4.6 Alternative pre- and post-disaster quarters. To ensure that our results
are not driven by a particular window around climatic disasters, we perform
tests with alternative windows around the disaster quarter. We work with the
following windows around the disaster quarter: [-1Q, +1Q], [-1Q, +2Q], [-1Q,
+3Q], [-2Q, +1Q], [-2Q, +3Q], and [-3Q, +3Q]. We find similar results across all
event windows. The coefficient on the interaction terms, CLOSE ms ×POST t ,
is always significant and similar to our baseline results in Table 3.
3.4.7 Alternative clustering of standard errors. We repeat our managerial

overreaction test with alternative clustering of standard errors. We cluster
standard errors by firm, and by disaster county. We find robust results in both
tests.
3.4.8 Controlling for other fixed effects. We repeat our managerial

overreaction test after adding fixed effects for stock, year-quarter, stock-fund,
and stock-season25 and find our results to be robust.
3.4.9 Are the results due to home bias? One potential concern with our tests
is that our results may be simply a manifestation of home bias. Specifically, we
know from prior studies that, on average, local funds hold a greater share of
local firms in their portfolios (Coval and Moskowitz 1999). This implies that,
even if both local and distant funds sold out 100% of their holdings in disaster
zone stocks, one might still see a greater negative coefficient on CLOSE ms ×
POST t for close funds. To address this concern, we repeat our baseline tests
25 For the stock season, we use four fixed effects per firm, one for each quarter of the year to control for seasonality.
1180

with the natural log of the total dollar value of the shares of a disaster zone
stock held by a fund as the dependent variable. The log transformation allows
us to capture the percentage change in total capital invested in a disaster zone
stock around the disaster events. Thus, a fund that sells half of its one million
shares of a stock will have the same dependent variable value as a fund that sells
half of its 10,000 shares. Our results are robust to this alternative specification:
close funds sell a higher proportion of their predisaster holdings of disaster
zone stocks.

3.4.10 Fund volatility around disasters. Bernile et al. (2017) document a
decrease in risk-taking by portfolio managers following disaster experience.
We examine these results in our context using fund volatility as a measure of
risk-taking. We compute fund volatility as the standard deviation of imputed
daily returns from the ex ante disclosed portfolio. The volatility is obtained
from the daily returns in a quarter. Consistent with the evidence reported in
Bernile et al. (2017), we find that the post-disaster volatility of close mutual
funds decreases after the disaster.
4. Conclusion
We examine whether professional money managers misestimate climatic
disaster risk. The risk associated with climatic disasters can enter a manager’s
portfolio if a disaster affects portfolio firms. Using a difference-in-differences
strategy, we first show that relative to the distant funds, funds closer to the
disaster zone reduce their portfolio holdings of firms located in the disaster
area. We do not observe such a differential underweighting by close funds
relative to the distant funds with respect to the firms located in the neighboring
counties.
Consistent with the fund managers overestimating the adverse impact of
disasters on stocks located in the disaster zone, we find that the bias in their
trading response is transitory and vanishes with time and distance. Moreover,
the response of close funds is not driven by any information advantage they
may possess over the distant funds as we do not find any difference in the
post-disaster profitability across firms in the disaster area and those in the
neighboring counties. The greater underweighting of the disaster zone firms
by close funds relative to distant funds is not driven by flow-driven trading
pressure created by biased investors or due to the drop in stock price of the
disaster zone firms.
Finally, we find that climatic disaster risk misestimation is costly to the fund
investors as it adversely affects portfolio returns. Specifically, a portfolio that
goes long on the disaster zone stocks that experience the sharpest reduction in
weights in portfolios of close funds and goes short on the stocks that experience
the least reduction generates statistically and economically significant positive
risk-adjusted returns after the disaster.
1181

References
Angrist, J. D., and J.-S. Pischke. 2008. Mostly harmless econometrics: An empiricist’s companion. Princeton,
NJ: Princeton University Press.
Bailey, W., A. Kumar, and D. Ng. 2011. Behavioral biases of mutual fund investors. Journal of Financial
Economics 102:1–27.
Bansal, R., D. Kiku, and M. Ochoa. 2016. Price of long-run temperature shifts in capital markets. Working Paper.
Barrot, J.-N., and J. Sauvagnat. 2016. Input specificity and the propagation of idiosyncratic shocks in production
networks. Quarterly Journal of Economics 131:1543–92.
Bennett, J., R. Sias, and L. Starks. 2003. Preferences for stock characteristics as revealed by mutual fund portfolio.

Review of Financial Studies 16:1203–38.
Bernile, G., V. Bhagwat, A. Kecskes, and P.-A. Nguyen. 2017. Are the risk attitudes of professional investors
affected by personal catastrophic experiences? Working Paper.
Bernile, G., A. Kumar, and J. Sulaeman. 2015. Home away from home: Geography of information and local
investors. Review of Financial Studies 28:2009–49.
Boehmer, E., and E. K. Kelley. 2009. Institutional investors and the informational efficient of prices. Review of
Financial Studies 22:3563–94.
Bordalo, P., N. Gennaioli, and A. Shleifer. 2012. Salience Theory of Choice Under Risk. Quarterly Journal of
Boustan, L. P., M. E. Kahn, and P. W. Rhode. 2012. Moving to higher ground: Migration response to natural
disasters in the early twentieth century. American Economic Review 102:238–44.
Chaney, T., D. Sraer, and D. Thesmar. 2012. The collateral channel: How real estate shocks affect corporate
investment. American Economic Review 102:2381–409.
Coval, J., and E. Stafford. 2007. Asset fire sales (and purchases) in equity markets. Journal of Financial Economics
86:479–512.
Coval, J. D., and T. J. Moskowitz. 1999. Home bias at home: Local equity preference in domestic portfolios.
Journal of Finance 54:2045–73.
———. 2001. The geography of investment: Informed trading and asset prices. Journal of Political Economy
109:811–41.
Daniel, K., M. Grinblatt, S. Titman, and R. Wermers. 1997. Measuring mutual fund performance with
characteristic-based benchmarks. Journal of Finance 52:1035–58.
Daniel, K. D., R. B. Litterman, and G. Wagner. 2016. Applying asset pricing theory to calibrate the price of
climate risk. Working Paper.
Dasgupta, A., A. Prat, and M. Verado. 2011. Institutional trade persistence and long-term equity returns. Journal
of Finance 66:635–53.
Dessaint, O., and A. Matray. 2017. Do managers overreact to salient risks? Evidence from hurricane strikes.
Journal of Financial Economics 126:97–121.
Giglio, S., M. Maggiori, M., Stroebel, J., and Weber, A. 2015. Climate change and long-run discount rates:
Evidence from real estate. Working Paper.
Goetzmann, W. N., D. Kim, A. Kumar, and Q. Wang. 2014. Weather-induced mood, institutional investors, and
stock returns. Review of Financial Studies 28:73–111.
Gompers, P., and A. Metrick. 2001. Institutional investors and equity prices. Quarterly Journal of Economics
116:229–60.
Hirshleifer, D. 2001. Investor psychology and asset pricing. Journal of Finance 56:493–94.
1182

Hjort, I. 2016. Potential climate risks in financial markets: A literature overview. University of Oslo.
Hong, H., F. W. Li, and J. Xu. 2019. Climate risks and market efficiency. Journal of Econometrics 208:265–81.
Ivković, Z., and S. Weisbenner. 2005. Local does as local is: Information content of the geography of individual
investors’ common stock investments. Journal of Finance 60:267–306.
Kacperczyk, M., and A. Seru. 2007. Fund manager use of public information: New evidence on managerial skills.
Kang, J., and R. Stulz. 1997. Why is there a home bias? An analysis of foreign portfolio equity ownership in
Japan. Journal of Financial Economics 46:3–28.
Kaplanski, G., and H. Levy. 2010. Sentiment and stock prices: The case of aviation disasters. Journal of Financial

Ki-moon, B. 2016. Remarks at investor summit on climate risk.
Kumar, A. 2009. Who gambles in the stock market? Journal of Finance 64:1889–933.
Levitt, S., and J. List. 2007. What do laboratory experiments measuring social preferences reveal about the real
world? Journal of Economic Perspectives 21:153–74.
Nofsinger, J. R., and R. W. Sias. 1999. Herding and feedback trading by institutional and individual investors.
Pool, V. K., N. Stoffman, and S. E. Yonker. 2012. No place like home: Familiarity in mutual fund manager
portfolio choice. Review of Financial Studies 25:2563–99.
Ramirez, A., and N. Altay. 2011. Risk and the multinational corporation revisited: The case of natural disasters
and corporate cash holdings. Working Paper.
Subrahmanyam, A., and S. Titman. 2001. Feedback from stock prices to cash flows. Journal of Finance 56:2389–
413.
Tversky, A., and D. Kahneman. 1973. Availability: A heuristic for judging frequency and probability. Cognitive
Psychology 5:207–32.
Wermers, R. 1999. Mutual fund herding and the impact on stock prices. Journal of Finance 54:581–622.
———. 2000. Mutual fund performance: An empirical decomposition into stock-picking talent, style,
transactions costs, and expenses. Journal of Finance 55:1655–703.
———. 2004. Is money really ‘smart’? New evidence on the relation between mutual fund flows, manager
behavior, and performance persistence. Working Paper.
Wurgler, J. 2000. Financial markets and the allocation of capital. Journal of Financial Economics 58:187–214.
1183

Alok 2019

Uploaded by

Copyright:

Available Formats

Alok 2019

Uploaded by

Document Information

Original Title

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Alok 2019

Uploaded by

Copyright:

Available Formats

Do Fund Managers Misestimate Climatic

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

We examine whether professional money managers overreact to large climatic disasters.

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1146 1146–1183

the Institutional Investors Group on Climate Change (IIGCC), a coalition of

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1147 1146–1183

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1148 1146–1183

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1149 1146–1183

fund size, fund age, socioeconomic characteristics of mutual fund clientele,

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

1. Related Literature and Hypotheses

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1150 1146–1183

1.1 Related literature

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1151 1146–1183

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1152 1146–1183

otherwise.8 The unit of observation in our analysis is a fund-firm-quarter.

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

β2 +β3 = E(WEIGHT |POST = 1,CLOSE = 1)−E(WEIGHT |POST = 0,CLOSE = 1),

and the underweighting by FAR funds is given by

β2 = E(WEIGHT |POST = 1,CLOSE = 0)−E(WEIGHT |POST = 0,CLOSE = 0).

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1153 1146–1183

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1154 1146–1183

hurricanes). Second, DM compare the cash hoarding behavior of neighboring

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

2. Data and Descriptive Statistics

Disaster Total damages Average event damage County × Occurrence

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1155 1146–1183

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1156 1146–1183

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

Disaster counties and the distribution of firms

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

The Review of Financial Studies / v 33 n 3 2020

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

by Wharton Research Data Service. Our combined sample consists of 3,268

2.3 Descriptive statistics

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1159 1146–1183

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

3.1 Portfolio response to disasters

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1160 1146–1183

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

Xm,t−1 +μm +δt +mst ,

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1161 1146–1183

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

3.1.2 Temporal dynamics of portfolio response to disasters. We now

WEIGHT mst =β0 +β1 (CLOSE ms ×P RE[−4,−3])+β2 CLOSE ms +

β3 (CLOSE ms ×POST [0,2])+β4 (CLOSE ms ×POST [3,4])+

β5 (CLOSE ms ×POST [5,6])+θ0 P RE[−4,−3]+

θ1 POST [0,2]+θ3 POST [3,4]+θ3 POST [5,6]+Xs,t−1 +Xm,t−1 +

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1162 1146–1183

where m refers to a fund, s refers to a firm, and t refers to a quarter. We focus

Downloaded from https://academic.oup.com/rfs/article/33/3/1146/5735311 by EPF Lausanne user on 04 February 2022

[16:30 30/1/2020 RFS-OP-REVF190150.tex] Page: 1163 1146–1183