Y. Chang et al.

Journal of Banking and Finance 166 (2024) 107233

disaster if the corresponding country has experienced a top-20 disaster that equals 1 for 2005–2008 (i.e., when Hurricane Katrina occurred and
during the previous 6 years. Subsequently, 15 disasters remain in our 3 years after), and 0 for 2002–2004 (i.e., the 3 years before Hurricane
sample period.29 Based on these disaster events, we create an indicator Katrina occurred) for firms in the affected states (treatment states). For
variable, POST, which equals 1 for postdisaster years in affected coun­ firms in unaffected states (control states), POST_AFSTATE is always equal
tries (defined as treatment countries) and 0 for predisaster years. For to 0. We then re-estimate Eq. (2) using REP/PAYOUT as the dependent
countries that did not experience any of the 15 disaster events (defined variable and POST_AFSTATE as the variable of interest. The results are
as control countries), POST is always equal to 0. We then estimate the reported in Table 6 Column (5). The coefficient of POST_AFSTATE is
following OLS regression with an event window of [− 3, +3] years sur­ 0.043 and significant at the 1 % level, suggesting that relative to US
rounding each disaster: firms in unaffected areas, firms in states with severe physical damage
significantly increase their payout flexibility.
PAYOUTi,j,t = α + β⋅POSTj,t + Controlsi,j,t + Fixed Effects + εi,j,t
To offer more granular evidence, in the second additional DiD test,
(2) we measure the distance between a firm’s location and the closest
county that was most severely affected by Hurricane Katrina. We first
where i, j, and t index firm, country, and year, respectively. PAYOUT is
search the historical track of Hurricane Katrina on the National Oceanic
one of the three payout measures, i.e., DIV/S, REP/S, or REP/PAYOUT.
and Atmospheric Administration’s (NOAA) Historical Hurricane Tracks
Controls denotes the set of firm and country characteristics included in
webpage.31 We identify seven counties that were most severely affected
Eq. (1). Fixed Effects denotes both firm and year fixed effects. Once firm
by Hurricane Katrina on land (with cyclone intensity categorized as H3
fixed effects are controlled, POST captures changes in payout policies
and sustained winds of 96–112 knots); 4 counties were in Louisiana (i.e.,
within the treatment countries after the disasters. We also adjust the
Plaquemines Parish, St. Bernard Parish, St. Tammany Parish, and
standard errors for clustering effects at the firm level.
Orleans Parish) and 3 were in Mississippi (i.e., Hancock County, Harri­
Table 6 reports the results. The coefficient of POST is significantly
son County, and Jackson County). We cross-check with a Tropical
negative in Column (1), where the dependent variable is DIV/S, sug­
Cyclone Report on Hurricane Katrina issued by the National Hurricane
gesting that firms reduce their dividends after severe disasters. Columns
Center (NOAA)32 and ensure that the seven most affected counties
(2) and (3) show that POST has positive and significant coefficients
identified are accurate. We then measure the distance between a firm
when the dependent variables are REP/S and REP/PAYOUT, supporting
and the closest affected county based on zip codes.33 We then divide all
that climate-related natural disasters cause firms to use more
firms into NEAR and FAR groups based on the sample median value of
repurchases to increase their payout flexibility. We also examine the
distance and construct two indicator variables: POST_AFCOUNTY
dynamic evolution of payout flexibility surrounding severe disasters to
(NEAR) (POST_AFCOUNTY[FAR]) equals 1 for firms close to (far from) a
rule out the possibility that firms already start using more repurchases
most severely affected county in 2005–2008 (i.e., when Hurricane
before disasters (i.e., preexisting trends). Specifically, we replace POST
Katrina occurred and 3 years after), and 0 for 2002–2004 (i.e., the 3
with four dummy variables: The 3 years leading up to the disaster
years before Hurricane Katrina occurred). We include these two indi­
(POST_− 2_TO_0) and the first year (POST1), the second year (POST2),
cator variables as variables of interest and re-estimate Eq. (2). The re­
and the third year (POST3) after the disaster. Thus, the base period is the
sults in Table 6 Column (6) show that the coefficient of
third year before the disaster. Column (4) reports the dynamic test re­
POST_AFCOUNTY(NEAR) is 0.052 and significant at the 1 % level. In
sults. The coefficient of POST_− 2_TO_0 is almost zero and insignificant,
contrast, the coefficient of POST_AFCOUNTY(FAR) is 0.021 but insig­
suggesting no preexisting trend of increasing payout flexibility before
nificant. This evidence indicates that firms’ response to climate risk by
disasters. In contrast, the coefficients of POST1, POST2, and POST3 are
increasing payout flexibility is stronger if they are close to the disaster
all positive and significant at the 1 % level, indicating that increasing
location. Overall, Table 6 provides ample supporting evidence that
payout flexibility in response to higher climate risk represents a long-
climate risk has a causal effect on corporate payout policies.34,35
term change in firms’ payout policy (at least 3 years after disasters).
Notably, the purpose of the natural disaster tests is not to examine
the effect of media coverage or attention. Instead, this test exploits 4.6. Channel tests
exogenous climate events that impose severe physical risk on firms to
study the causal impact of climate risk. However, also likely is that the This subsection examines the channels that transfer the impact of
physical damage caused by disasters varies across firms in different parts climate risk to firms’ payout flexibility. Our primary argument is that
of large countries. To explore this variation, we focus on the 2005 climate risk affects payout flexibility through firms’ desire to maintain
Hurricane Katrina in the US (one of the 15 most severe disasters in our
sample) and carry out two additional DiD tests. In the first test, we
identify the states directly affected by Hurricane Katina based on the 31
At https://coast.noaa.gov/hurricanes/#map=4/32/-80.
numbers of deaths and injuries.30 As some states affected by Hurricane 32
Retrieved from: https://www.nhc.noaa.gov/data/tcr/AL122005_Katrina.
Katrina recorded zero deaths/injuries, to affirm the physical risk chan­ pdf.
nel, we define affected states as those with at least 1 death or injury; 33
The distance is calculated using the ZIPCITYDISTANCE function in SAS.
these states include Alabama, Florida, Georgia, Kentucky, Louisiana, Firms’ zip codes are sourced from Compustat. The most severely affected
and Mississippi. We construct POST_AFSTATE as an indicator variable counties’ zip codes are obtained from https://www.ciclt.net/sn/clt/capitolimpa
ct/gw_countydet.aspx?state=la&stfips=22&stname=Louisiana&fips=22075.
34
Our paper’s primary climate risk measure focuses on the physical risk of
climate change; therefore, we use climate-related natural disasters as exogenous
29
These disasters include (1) United States 2005 Storm (Katrina), (2) France shocks to physical risk. For completeness, we also conduct a causality test of the
2006 Extreme temperature, (3) Philippines 2006 Storm (Durian) and Landslide, effect of climate regulatory risk on corporate payout policies. Following Seltzer
(4) Germany 2007 Storm, (5) China 2008 Extreme temperature, (6) Australia et al. (2022), we use the Paris Agreement as a shock to regulatory risk. The
2010 Flood, (7) Colombia 2010 Flood, (8) Peru 2010 Extreme temperature, (9) findings are qualitatively similar and available upon request.
35
Russia 2010 Extreme temperature, (10) Brazil 2011 Flood, (11) Indonesia 2013 We also provide further evidence on substituting dividends with
Flood, (12) India 2014 Flood, (13) France 2015 Extreme temperature, (14) repurchases when climate risk is high. Following Grullon and Michaely (2002),
South Africa 2017 Drought, and (15) Argentina 2018 Drought. we calculate the transition probabilities of firms’ payout policy from year t− 1 to
30
The numbers of deaths and injuries data are obtained from the National year t and regress dividend forecast error on the interaction term between
Oceanic and Atmospheric Administration (NOAA) and at the county level. We climate risk and share repurchases, climate risk, share repurchases, and various
aggregate the county-level data into the state level and match them with control variables related to a firm’s payout method. Detailed estimation and
Compustat firm-years based on firms’ headquartered states. results can be found in the Internet Appendix.

12