Bank of Canada staff working papers provide a forum for staff to publish work-in-progress research independently
from the Bank’s Governing Council. This research may support or challenge prevailing policy orthodoxy. Therefore,
the views expressed in this paper are solely those of the authors and may differ from official Bank of Canada views.
No responsibility for them should be attributed to the Bank.
DOI: https://doi.org/10.34989/swp-2024-1 | ISSN 1701-9397 ©2024 Bank of Canada
Staff Working Paper/Document de travail du personnel—2024-1
Last updated: January 8, 2024
Extreme Weather and Low
-
Income Household Finance:
Evidence from Payday Loans
by Shihan Xie,
1
Victoria Wenxin Xie
2
and Xu Zhang
3
1
University of Illinois Urbana-Champaign
[email protected]du
2
Santa Clara University
wxie@scu.edu
3
Financial Markets Department
Bank of Canada
XZhang@bankofcanada.ca
i
Acknowledgements
We thank Jialan Wang, Julia Fonseca, and Peter Han for creating the Gies Consumer and Small
Business Credit Panel (GCCP), and we thank the Gies College of Business for supporting this
dataset. We acknowledge the support from the Alfred P. Sloan Foundation through the
National Bureau of Economic Research Household Finance small grant program. We thank
Jason Allen, Dan Bernhardt, Richard Carson, Matthew Cole, Tatyana Deryugina, Julia Fonseca,
Jean-Sébastien Fontaine, Toan Phan, Juan Sanchez, Brigitte Roth Tran, and seminar participants
at the University of Birmingham, Southwestern University of Finance and Economics, and Bank
of Canada for helpful comments and suggestions. Minyoung Cho provided excellent research
assistance. The views expressed herein are those of the authors and not necessarily those of
the Bank of Canada.
ii
Abstract
This paper explores the impact of extreme weather exposures on the financial outcomes of
low-income households. Using a novel dataset comprising individual-level payday loan
applications and loan-level information, we find that extreme temperature days—both hot and
cold—lead to surges in demand for payday loans. An increase in the number of days with
extreme heat results in an increase in delinquency and default rates and a reduction of total
credit issued, indicating a contraction in loan supply. These effects are especially noticeable for
online payday loans. Our findings highlight the heightened financial vulnerability of low-
income households to environmental shocks and underscore the need for targeted policies.
Topics: Climate change; Credit and credit aggregates
JEL codes: Q54, G50
Résumé
Cette étude explore l’incidence de l’exposition aux événements météorologiques extrêmes sur
la situation financière des ménages à faible revenu. À partir d’un nouvel ensemble de données
sur les demandes individuelles de prêts sur salaire et sur les prêts, nous constatons que les
journées de chaleur ou de froid extrême entraînent une montée soudaine des demandes de
prêts sur salaire. Une hausse du nombre de journées de chaleur extrême cause une
augmentation du taux de défaillance et de défaut, et une réduction du crédit total octroyé, ce
qui indique une contraction de l’offre de crédit. Ces effets sont particulièrement évidents pour
les prêts sur salaire en ligne. Nos résultats mettent en évidence la vulnérabilité financière accrue
des ménages à faible revenu aux chocs environnementaux et la nécessité d’élaborer des
politiques ciblées.
Sujets : Changements climatiques; Crédit et agrégats du crédit
Codes JEL : Q54, G50
1. Introduction
Extreme temperature events such as heat waves are significantly increasing in both fre-
quency and length in the United States (EPA, 2022). There are many reasons why low-
income households may face additional financial hardship during extreme temperature days
due to unequal exposure (Barreca, Park, and Stainier, 2022; Benz and Burney, 2021), re-
sulting in lost wages, health issues with associated medical costs, inadequate emergency
funds, outdated infrastructure and appliances, higher energy bills, and limited knowledge of
available insurance options. Our paper investigates the financial implications of these chal-
lenges on low- to middle-income households using a novel dataset on payday loans, the most
popular alternative credit product used for short-term liquidity by these income groups.
We investigate the causal effects of extreme temperatures on payday loan market out-
comes. Our individual-level dataset covers both storefront and online payday loan applica-
tions. We explore the impact of extreme weather days on a variety of payday loan market
outcomes, including the number of inquiries, credit approval, default, and delinquency rates.
The results shed light on both how demand for and access to alternative credit products
for low-income households are affected by extreme temperature events. We find that having
more extreme temperature days increases the demand for payday loans but decreases the
total credit issued. Additionally, we show that the negative impact on vulnerable households
can be amplified through potentially disastrous debt cycles, featuring increasing delinquency
and default rates.
Our study complements an emerging literature on the impacts of natural disasters on
household finance (Del Valle, Scharlemann, and Shore, 2022; Gallagher and Hartley, 2017;
Gallagher, Billings, and Ricketts, 2020). Using data from consumer credit panels, student
loans, and credit card transactions, these studies find that spikes in credit card borrowing
after natural disasters are modest and short-lived, and that households ultimately reduce
their total debt due to their use of insurance to repay their debts rather than to rebuild. Our
paper makes two significant departures. First, the literature so far has focused on large nat-
1
ural disasters, which are associated with Federal Emergency Management Agency (FEMA)
declarations and subsequent aid programs. In contrast, we examine extreme temperature
days, which do not lead to any FEMA declaration. Extreme temperature events tend to
occur more frequently than major natural disasters like earthquakes and volcanic eruptions
and can affect vast areas, impacting large populations and extensive agricultural zones. Re-
peated extreme weather events can cause incremental damage. As we show below, they also
lead to sizable and significant deterioration of household financial health. Second, exist-
ing studies on climate change and household finance offer limited evidence on low-income
households. This is partly because previous studies use traditional credit sources, such as
credit cards, and hence miss the impacts on individuals who lack access to credit card lines.
1
These marginalized, overwhelmingly low-income households are more likely to suffer the
most financially from extreme temperature days.
To overcome the data challenges mentioned above, we use two unique payday loan
datasets from Clarity.
2
Our first applicant-level dataset includes 1 million consumers ran-
domly selected from Clarity’s database between 2012 and 2019. It provides comprehensive
information on application date, income, age, ZIP code of residence, among other things.
For the same group of borrowers, we have a second loan-level dataset that contains rich
information on approved loan characteristics, including highest credits, maturation, delin-
quency, and default dates. To construct daily weather outcomes, we use satellite-based data
from ERA5-Land. It contains the daily temperature and precipitation with 0.1
◦
× 0.1
◦
hor-
izontal resolution covering the continental United States from 2012 to 2019. We merge the
ERA5-Land climate data with payday lending market outcomes at the Census ZIP Code
Tabulation Area (hereafter ZCTA) by month level. The resulting measures of ZCTA-level
weather exposure capture variation across both ZCTA and calendar year-month periods.
1
For example, Miller and Soo (2021) show that low-income households are more likely to use high-interest
credit products like payday loans and have limited access to formal credit.
2
Clarity is an agency that specializes in alternative financial services data as part of the major credit
reporting agency, Experian.
2
We adopt a fixed effects temperature-bin approach, a well-established empirical design in
the climate economy literature
3
to causally identify the impact of extreme weather days on
payday loan market outcomes. The main independent variable counts the number of days
with daytime (between 8am and 8pm) mean temperature falling into different temperature
bins.
4
Our empirical design controls for a rich set of fixed effects that may correlate with
extreme weather days, allowing us to identify the marginal effects of extreme heat and cold
days relative to the local average. We first examine the impact on payday loan demand and
credit approval. We then explore how extreme temperature exposure affects the performance
of existing payday loan accounts. Finally, we differentiate between the impact of extreme
temperature days on online versus storefront payday loan markets.
There are three main transmission mechanisms through which extreme weather days can
affect individual demand and repayment for payday loan products. First, extreme weather
days may decrease labor income (Colmer, 2021; Graff Zivin and Neidell, 2014; Heal and
Park, 2016), potentially increasing payday loan demand and default rate. Second, energy
costs, adopting heating or cooling devices, and health expenditure needs rise with extreme
weather shocks, which could also increase payday loan demand (Deschenes, 2022; Zivin
and Shrader, 2016; Park, Pankratz, and Behrer, 2021).
5
We evaluate the income versus
expenditure channels by including borrower income as a control or not. Conversely, the third
mechanism suggests that extreme weather might deter individuals from accessing payday
loans. For example, extreme weather may affect transportation (Roth Tran, 2023), making
it challenging for households to visit a payday loan storefront in person, thereby diminishing
3
For example, Deschˆenes and Greenstone (2007); Graff Zivin and Neidell (2014); Barreca et al. (2016);
Cohen and Dechezleprˆetre (2022).
4
Figure 1 depicts the average annual distribution of daytime mean temperatures across 16 temperature
bins over the 2012–2019 period. We refer to days with daytime mean temperature above 33
◦
C as extreme
heat days. Around 2.5% ZCTA-day observations in our sample are “extreme heat days” by this definition.
Similarly, we refer to days with daytime mean temperature below -9
◦
C as extreme cold days. Around 1.0%
ZCTA-day observations in our sample are “extreme cold days” by this definition.
5
According to Woolf et al. (2023), heat event days are responsible for almost 235,000 emergency de-
partment visits and more than 56,000 hospital admissions for heat-related or heat-adjacent illness, adding
approximately $1 billion in health care costs each summer.
3
overall applications. We provide empirical evidence on this channel by comparing storefront
payday loans with online payday loans.
We find that having more extreme temperature days in a month increases payday loan
demand. Total inquiries for payday loans significantly increase with more extreme heat or
cold days. One extra extreme heat day (with daytime mean temperature above 33
◦
C) is
associated with a 0.009 increase in inquiries, while one extra extreme cold day (with daytime
mean temperature below -9
◦
C) is associated with a 0.011 increase in inquiries. In other
words, one standard deviation increase in the number of extreme heat days leads to about
a 0.4% increase in total inquiries relative to the baseline. This impact is consistent with
several potential channels, such as extreme temperature days lowering household income
and/or increasing household expenditure on health- or energy-related costs.
Next, we examine the impact of extreme weather on credit issuance and the performance
of existing loans in payday loan markets. We find asymmetric effects of extreme heat and
cold days. Extreme heat days significantly reduce total credit issued and accounts opened. In
particular, one extra extreme heat day is associated with a 0.4% drop in credit issued. Having
more extreme heat days in a month also leads to deteriorating performances of existing
payday loans, as we observe significant increases in delinquency and default rates. However,
extreme cold days do not affect the credit issuance or performance of existing payday loan
accounts. Taken together, our results suggest that payday loan lenders reduce credit supply
during extreme heat days out of concern for an increase in default and delinquency rates.
Furthermore, we document the asymmetric effects of extreme heat and cold days on labor
incomes. We show that extreme heat days negatively affect income, whereas extreme cold
days do not. The decline in borrower quality during extreme heat days is consistent with our
observations of reducing credit supply and worsening loan performance. Our main results are
robust to multiple sensitivity checks. For example, we show that using alternative definitions
of temperature bins, including lags of the dependent variables, or using a balanced panel to
address selection does not affect our results.
4
We then explore heterogeneity in payday loan markets over several dimensions. First, we
distinguish between the storefront and online payday loan markets. Correia, Han, and Wang
(2022) show that online loans are associated with higher default rates and, as a result, higher
premiums. We find that the impact of extreme temperature days operates more through
changes in online payday loan markets. In particular, for online payday loan borrowers, we
observe increases in their inquiries, delinquency, and default rates, with decreases in accounts
opened and credit issued with more extreme heat days. In contrast, storefront payday
loan outcomes do not respond significantly to extreme temperature days. One potential
explanation is that the accessibility and ease of application of the online payday loan market
may make it more sensitive to changes in temperature and weather conditions. Finally, we
examine the distributional effects across ZCTAs. Our estimates suggest that counties with
higher population shares of Hispanics, who are predominantly involved in outdoor work, see
larger increases in demand and larger decreases in total credits for payday loans with more
extreme heat days.
We extend our analysis to include alternative subprime credit, including rent-to-own,
installment, and auto loans. Borrowers of these subprime alternative credits take out larger
amounts of credit and bear more negative consequences for their creditworthiness in case of
defaults. We show that extreme heat days increase demand for subprime alternative credits
without negatively impacting credit supply or delinquency rates, suggesting different impacts
of extreme weather exposures across loan types.
Our paper makes three contributions. First, we add to the growing literature on al-
ternative credit markets. Payday loans are a controversial source of liquidity for low- to
middle-income customers (Allcott et al., 2022; Gathergood, Guttman-Kenney, and Hunt,
2019). Payday loan borrowers pay higher interest rates and often struggle to repay the loan
on time, leading to loan renewal and a cycle of accumulating interest and fees, resulting
5
in persistent debt.
6
Mostly based on geographic variation in regulations of and access to
payday loans, a large literature finds mixed results on the effects of payday loans on con-
sumers (Bhutta, Skiba, and Tobacman, 2015; Dobridge, 2018; Di Maggio, Ma, and Williams,
2021; Melzer, 2011, 2018; Miller and Soo, 2021; Morse, 2011; Skiba and Tobacman, 2019).
7
Instead of evaluating the impact of payday loan access, our payday application and payday
transaction data allow us to directly assess the impact of extreme temperatures on payday
loan performance. Our study illustrates that extreme weather exposures, particularly ex-
treme heat days, significantly worsen the performance of existing payday loan accounts. The
impact operates through online payday loan markets.
Second, our paper adds to the recent literature that examines the impact of environmen-
tal shocks and household finance. Previous work examines how natural disasters and the
transition away from fossil fuels affect households in traditional credit markets (Blonz, Tran,
and Troland, 2023; Del Valle, Scharlemann, and Shore, 2022; Gallagher and Hartley, 2017;
Gallagher, Billings, and Ricketts, 2020). We add to this work by focusing on alternative
credit products used by lower-income households. The payday loan applicant- and loan-level
datasets we use allow us to distinguish between the demand for, access to, and performance
of payday loan market products. Although existing work finds that spikes in credit card
borrowing after natural disasters are modest and short-lived, we show that payday loan mar-
kets may operate differently. With more extreme heat days, lower-income households have
a higher demand for payday loan credits while facing tightened credit supply and higher
delinquency rates.
Our paper also contributes to the literature documenting various impacts of extreme
temperature exposures in the United States, including health, employment, time use, school
learning, and firm sales, among others (Deschˆenes and Greenstone, 2011; Wilson, 2019;
6
Burke et al. (2014) find that four out of five payday loans are rolled over or renewed within 14 days.
Their study also shows that most payday loans are issued to borrowers who renew their loans so many times
that they end up paying more fees than they originally borrowed.
7
Fonseca (2023) exploits the time-series variation in the state-level debt collection restrictions and finds
that restricting collections reduces access to mainstream credit and increases payday borrowing.
6
Graff Zivin and Neidell, 2014; Park et al., 2020; Addoum, Ng, and Ortiz-Bobea, 2020).
8
Our analysis shows that extreme temperature exposure, particularly heat stress, also has
significant implications for the financial well-being of lower-income households who rely on
alternative credit products.
The rest of the paper is organized as follows: Section 2 discusses the relevant background
and data. Section 3 describes our research design and presents the main results on payday
loan market outcomes. Section 4 offers several extensions. Section 5 concludes. The appendix
provides further details.
2. Background and data
In this section, we present several unique datasets used in our analysis. Section 2.1 dis-
cusses the background of the U.S. payday loan industry and our payday loan dataset. Section
2.2 presents satellite-based climate data we constructed on temperature and precipitation.
Section 2.3 introduces the background of the Low-Income Home Energy Assistance Program
and its related dataset.
2.1. Payday loan
Payday loans are a short-term source of liquidity used by low- to middle-income indi-
viduals. Typically, the lender advances the borrower $100 to $500 in return for a postdated
check, timed to coincide with the borrower’s next paycheck. Loans usually have two- to
four-week maturities. While payday loans provide flexibility in smoothing consumption over
time, they can also impose a substantial burden. The fees can be as high as $15 to $20 per
$100 principal balance, equivalent to a 400 to 600 annual percentage rate (APR).
There are many reasons why individuals might borrow payday loans despite the outra-
geous interest rates. About 3% of respondents in the Survey of Consumer Finances indicated
that they had taken payday loans. The top reasons for choosing payday loans include emer-
8
See Dell, Jones, and Olken (2014) for a review of the literature.
7
gency needs, convenience, and to pay other bills or loans. A significant fraction of respondents
take payday loans to pay medical bills or utilities.
9
We use lender-reported payday loans from Clarity, an agency that specializes in alter-
native financial services data as part of the major credit reporting agency Experian. Two
samples are involved in our analysis.
10
The first is an applicant-level dataset, known here-
after as the random Clarity sample, consisting of 1 million consumers randomly drawn from
Clarity’s database from 2012 to 2019. Each individual comes with a unique applicant ID. For
each inquiry, we observe self-reported information obtained during the application process,
including state and ZIP code of residence, income, pay frequency, age, homeowner indicator,
months of residence in current address, and whether this application takes place online or
in a storefront. The second is a loan-level dataset known as the tradeline sample, which
consists of approved loans from 2013 to 2019. For each loan, we observe loan characteristics
such as origination, maturation and delinquency dates, loan size, and amount past due. The
applicant/borrower IDs from the two datasets are linked one-to-one. However, we do not
have a linkage between the application and approval loan at the record level. Though payday
loan lending may require identification, a recent bank account statement, or a recent pay
stub (or verification of other income), the information reported in inquiries is self-reported
by borrowers and may not be verified (Correia, Han, and Wang, 2022).
Out of 1 million unique borrowers who submitted an inquiry for any type of subprime
credit, approximately one-third inquired about payday loans and the rest were for other
products such as installment loans. We focus on payday loans in this study. We follow the
approaches in Correia, Han, and Wang (2022) to clean the data, validate, and construct the
panel. We convert ZIP codes to ZCTA using crosswalks from the U.S. Census. Appendix
Figure B.1 shows the geographical distribution of online and storefront borrowers across the
U.S. by state.
9
See Appendix A.1 for a more detailed discussion.
10
The same dataset has been used in Fonseca (2023); Correia, Han, and Wang (2022).
8
We construct two variables to measure the demand for payday loans at the ZCTA level.
The first variable is total inquiries. It is calculated as the total inquiries made by all individ-
uals who reside in each ZCTA each month. This includes cases where an applicant submits
multiple inquiries per day. We define a second variable called unique inquiries. It is calcu-
lated as the total number of unique individuals making inquiries residing in each ZCTA for
each month. Thus, individuals who make inquiries multiple times in a single day or across
multiple days in a single month will be counted as one unique inquiry.
We use two measures to proxy payday loan performance: delinquency rate and default
rate. We identify a loan as delinquent if it has a non-missing delinquency date. We identify a
loan as in default if it has a non-zero past due amount. The monthly ZCTA-level delinquency
rate and default rate are thus defined as the percent of loans that are delinquent or default,
respectively.
Table 1 provides summary statistics for the full sample in Panel A and the online as well
as the storefront payday loan subsamples in Panel B and Panel C, respectively. Our sample
consists of about 2.4 times total accounts open in the online market relative to the storefront.
In our analysis, we winsorize the top and bottom 1% of credit- and income-related variables
to reduce the effects of extreme outliers. On average, about 1.7 accounts are opened in one
ZCTA each month. The medium loan amount is $475, with $500 from storefront lenders
and $400 from online lenders, respectively. The delinquency rate is on average 8.7% and the
default rate is 7.3%.
Our study mainly focuses on low- and middle-income borrowers. In our full sample,
the median monthly income is $2,500, with the 75th percentile at $3,300. In comparison,
the 2015 American Community Survey reported a U.S. median household average monthly
income of $4,491 and $6,663 for married couples. The storefront payday loan applicant data
show lower incomes, with a median of only $1,494 and a 25th percentile of $921, which is
lower than the $1,000 poverty threshold for a single-person household in 2015.
11
11
https://www.census.gov/library/publications/2016/demo/p60-256.html
9
2.2. Weather Data
Data on temperature and other weather outcomes are from ERA5-Land, which measures
atmospheric variables with enhanced spatial resolution based on the ERA5 climate reanalysis
data provided by the European Centre for Medium-Range Weather Forecasts (ECMWF). We
obtain daytime mean temperature and precipitation with 0.1
◦
× 0.1
◦
horizontal resolution.
To construct monthly weather variables for each ZCTA, we use the ERA5 grid point
closest to the ZCTA centroid. Allowing for non-linear effects, Tempbin
k
m,t
counts the num-
ber of days in a month in which the daytime (between 8am and 8pm local time) average
temperature falls within the specified range k. For our baseline specification, we construct
16 temperature bins: below −9
◦
C, above 33
◦
C, and 14 bins for every 3
◦
C in between.
12
We also include monthly average precipitation data as controls, where daily precipitation is
measured as depths in meters of cumulative precipitation in a day. In the empirical models
presented hereafter, we use the same temperature bins to estimate the relationship between
temperature and the payday loan market and use precipitation as a control variable.
Figure 1 highlights that daytime mean temperatures below -9
◦
C or above 33
◦
C represent
very extreme temperatures in our sample. The bars in Figure 1 represent the average annual
distribution of daytime mean temperatures across 16 temperature bins over the 2012–2019
period. The height of each bar corresponds to the mean number of days of exposure per
year for the average person. Using all ZCTAs of the continental U.S. and all the months,
the average person is exposed to about 8.7 days annually with daytime temperatures below
-9
◦
C, and 4.3 days where daytime temperatures exceed 33
◦
C. One standard deviation in
extreme heat days where daytime temperatures exceed 33
◦
C is around 2 days. Our payday
sample only covers a portion of all ZCTAs, since not every ZCTA reports payday loans
every month. Within the scope of ZCTAs that report payday loans, which comprises 159
12
As robustness checks, we also use other temperature definitions to construct temperature bins, such
as daily maximum temperature or 24-hour mean temperature. Table 2 presents the summary statistics of
daily daytime mean, 24-hour mean, and daily maximum temperatures of ZCTAs with payday loan inquiries
during 2012–2019. The top and bottom 1 percentile of daytime mean temperatures are 34
◦
C and −9
◦
C,
respectively.
10
days annually, the average person in this sample is exposed to about 1.6 days per year
with daytime temperatures falling below -9
◦
C, and 3.2 days where the daytime temperature
exceeds 33
◦
C.
Figures 2a and 2b illustrate the spatial distribution of extreme weather shocks. For
each ZCTA, we average the monthly number of days that had daytime mean temperatures
below -9
◦
C and above 33
◦
C. Extreme heat days occur in southern ZCTAs along the sunbelt
states. The hottest ZCTA in the 95th percentile experienced on average 2 days per month
where the daytime mean temperature was above 33
◦
C. Extreme cold weather concentrates
in the northern ZCTAs. The coldest ZCTA in the 95th percentile experienced on average 3
days per month with daytime mean temperatures below -9
◦
C. Lastly, Appendix Figure B.2
presents the spatial distribution of the daytime mean temperatures across ZCTAs, averaged
over the sample period. From the years 2012 to 2019, the national average of daytime mean
temperature is around 15
◦
C. ZCTAs in the 90th percentile of the temperature distribution
have a daytime mean temperature of around 21.5
◦
C, while ZCTAs in the 95th percentile
have a daytime mean temperature of 23
◦
C.
2.3. LIHEAP program
During extreme weather conditions, energy costs surge, leading to subsequent increases in
utility bills. Low-income households are particularly vulnerable to these price fluctuations,
especially in relation to weather-related energy expenses. In fact, one of the reasons that
households resort to payday loans is to pay bills.
To address these challenges, the Low Income Home Energy Assistance Program (LI-
HEAP)
13
plays a crucial role in providing support to eligible low-income households. LI-
HEAP offers a range of assistance programs, including addressing heating and cooling energy
costs, providing bill payment assistance, offering aid during energy crises, facilitating weath-
erization efforts, and supporting energy-related home repairs.
14
In most cases, approved
13
See https://www.liheap.org/.
14
See Appendix A.2 for more details on the LIHEAP program.
11
households won’t receive payment directly. LIHEAP almost always pays grants directly to
the energy utility.
We assess whether access to LIHEAP can effectively reduce the need for payday loans
among eligible households. Eligible households for this program must meet certain criteria
under federal guidelines. LIHEAP eligibility criteria vary across states and are subject to
annual changes. Specifically, they must have an income less than 150% of the poverty line
or less than 60% of the state’s median income, whatever is greater.
We obtained the state median household income and poverty line used by each state
each year from the Office of Community Services website.
15
Since household size cannot be
observed in the payday loan data, we assume a household size of two. We acknowledge that
this assumption inevitably introduces measurement errors in our estimates of the program’s
effect on reducing payday loan borrowing. In our dataset, approximately 38% of payday
loan borrowers fall within the eligible income range. Panels (a) and (b) of Appendix Figure
B.3 plot the LIHEAP income eligibility cutoff for single-household families across the U.S.
states and the median income of payday loan borrowers in the year 2019.
3. Main empirical results
In this section, we present our main empirical results on how temperature exposures
affect payday lending. We first introduce our identification strategy in Section 3.1. We next
evaluate the demand channel in Section 3.2 and equilibrium outcomes in Section 3.3. We
further investigate the performance of existing accounts in Section 3.4 and examine borrower
characteristics in Section 3.5. Finally, we provide several additional robustness checks on
our results in Section 3.6.
15
See https://www.acf.hhs.gov/ocs/policy-guidance/liheap-information-memoranda.
12
3.1. Methodology
Our primary goal is to identify the effects of temperature exposures on the payday loan
market at the ZCTA level. We adopt a non-linear temperature-bin approach using the
following form:
Outcome
it
= θT
it
+ µ
t
+ η
cy
+ Controls + ε
it
, (1)
where Outcome
it
denotes payday loan-related outcome variables of interest at ZCTA i in
month t; θ is a vector of parameters; T
it
is a vector of climatic variables that we discuss
below; µ
t
is a year-month fixed effect; η
cy
is a county-year fixed effect; and ε
it
is the error
term. Standard errors are clustered at the ZCTA level, which is the geographic unit of
temperature variations.
The climatic variables in T
it
include location and time-specific temperature and precip-
itation measures. We use 16 temperature bins: below -9
◦
C, above 33
◦
C, and 14 bins for
every 3
◦
C in between. We choose -3
◦
C–27
◦
C as the omitted baseline daytime (between 8am
and 8pm) mean temperature bins. We calculate the number of days in a month that the
daytime mean temperature falls within each bin. We also include the monthly average of
daily cumulative precipitation to control for any confounding effects from precipitation. In
what follows, we refer to days with daytime mean temperature below -9
◦
C as extreme cold
days and days above 33
◦
C as extreme heat days.
16
Our adoption of the non-linear temperature-bin specification follows the recent climate
economy literature (Dell, Jones, and Olken, 2014), motivated by facts related to thermal
stress. High temperatures beyond certain thresholds cause worker fatigue and lower task
performance. For example, a meta-analysis of ergonomics literature documents that task
performance losses start to increase in a non-linear manner at high temperature ranges
16
In Section 3.6, we use alternative cutoffs for temperature bins based on 24-hour daytime mean temper-
ature or daily maximum temperature and show that the patterns are similar.
13
(Hancock, Ross, and Szalma, 2007). Other studies also indicate that labor productivity
drops significantly as temperature increases.
17
Energy use and utility bills, which increase
significantly during hot and cold days, are also non-linear functions of temperature.
Our fixed-effect framework allows us to control for confounding shocks that may correlate
with weather variables. To causally identify the effect of temperature exposures on payday
loan market outcomes, we include a rich set of fixed effects to control for confounders and rule
out spurious relationships. First, we consider seasonality, which may correlate within-year
payday loan cycles with temperature fluctuations. Second, a general temperature warming
trend may be correlated with national business cycles during this period. To address both
concerns, we control for year-month fixed effects. Next, payday loan market outcomes may
also be associated with regional business cycles, so we include county-year fixed effects. Our
identifying variations are therefore interpreted as weather shocks, exploiting ZCTA-month
level temperature deviations from the county-year, year-month averages.
3.2. Do temperature exposures drive demand for payday loans?
We first examine whether temperature exposures affect borrowers’ demand for payday
loans. We adopt two measures of payday loan demand—total inquiries and unique inquiries—
as outcome variables in the regression specification outlined in Equation (1). The left panel
of Figure 3 as well as column (2) of Table 3 provide the estimated impact of a day in six
extreme daytime temperature bins on the number of total inquiries, relative to a day with
daytime mean temperature falling in the -3
◦
C to 27
◦
C bins. We find strong evidence that
total inquiries significantly increase with more extreme heat or cold days. In particular, one
extra day with daytime mean temperature above 33
◦
C per month is associated with a 0.009
increase in total inquiries, while one extra day below -9
◦
C is associated with a 0.011 increase
in total inquiries in an average ZCTA. In other words, one standard deviation increase in
17
These include, but are not limited to, evidence from assembly lines, laboratories, meta-analyses, and
self-reported surveys: Adhvaryu, Kala, and Nyshadham (2020), Graff Zivin and Neidell (2014), Heal and
Park (2020), Niemel¨a et al. (2002).
14
the number of extreme heat or cold days per month leads to about a 0.4% increase in total
inquiries relative to the baseline.
18
In general, extreme temperatures lead to higher energy consumption demand and expose
households to higher health risks. Such increases in energy costs and health expenditures
could directly drive up demand for credit, especially for lower-income households. Extreme
weather may also reduce borrowers’ incomes, contributing to increases in demand for payday
loans. We test the income channel in Section 3.5.
Our results on total inquiries could partially be attributed to a rise in borrowers who
make multiple inquiries during extreme temperatures, which might be driven by impulsive
decisions, the practice of loan stacking, desperation, or loan denials. To separate from the
effects of such behavior, we analyze the impact of temperature exposures on unique inquiries,
which measure the number of applicants instead of total applications. The estimation results
are shown in the right panel of Figure 3 and in column (3) of Table 3. Echoing the findings
on total inquiries, we find similar effects. Specifically, the figure shows that one extra day
below -9
◦
C is associated with a 0.002 increase in unique inquiries, and one extra day above
33
◦
C is associated with a 0.003 increase in unique inquiries. Taken together, these results
indicate that extreme temperature exposures lead to increasing demand for payday loans.
3.3. Do temperature exposures affect amount of credit issued?
Next, we analyze the impact of extreme temperatures on equilibrium outcomes in payday
loan markets. Figure 4 presents results using the total number of accounts open (left panel)
and (log) total credit issued (right panel) as outcome variables in specification Equation (1).
We find a strong negative relationship between extreme heat days and the number of accounts
open. In particular, one extra day above 3
◦
C is associated with a 0.006 decrease in the total
number of accounts open in an average ZCTA. In other words, a one standard deviation
18
In our sample, one standard deviation increase in the number of extreme heat days with daytime temper-
ature above 33
◦
C is around 2 days, and the average baseline total inquiries is around 4.66 days. One standard
deviation increase in the number of extreme heat days per month therefore leads to a 0.009 × 2/4.66 = 0.4%
increase in total inquiries relative to the baseline.
15
increase in the number of extreme heat days per month leads to about a 0.7% increase
in accounts open relative to the baseline. Similarly, we find a strong negative relationship
between extreme heat days and total credit: one extra day above 33
◦
C is associated with a
0.4% drop in credit issued.
The total number of accounts opened and total amounts of credit issued are jointly
determined by credit supply and credit demand. Although payday loan inquiries increase
significantly on extreme heat days, equilibrium outcomes suggest a decrease in credit supply
during such periods. This contraction in credit supply may be because of two reasons. First,
payday lenders tend to screen borrowers from very-low-income backgrounds. We specifically
test this income channel in Section 3.5 to see if borrowers’ incomes decline during extreme
heat days. This may lead to increases in default or delinquency rates that induce payday
lenders to reduce credit supply. The following section presents evidence in support of this.
In contrast, we found different effects during extreme cold days. The total number of
accounts opened increases by 0.005, but it is still smaller than the increase in total inquiries.
There is no significant effect of extreme cold days on the total amount of credit granted.
These findings offer suggestive evidence that credit supply is responding to some extent to
increased demand during such periods. In contrast to extreme heat days, payday lenders may
be less concerned about default or delinquency rates during extreme cold days, as borrowers’
incomes might be less affected during extreme cold days.
3.4. Do temperature exposures affect the performance of existing accounts?
We next examine delinquency and default rates associated with extreme weather. Figure
5 reveals that a single extra day with a temperature above 33
◦
C is linked to a 0.09 percentage
point increase in delinquency rate and a 0.1 percentage point increase in default rate. In
other words, a one standard deviation increase in the number of extreme heat days per
month leads to about a 3% increase in default rates relative to the baseline. Our analysis
indicates that the effects of temperature exposure on the default and delinquency rates are
asymmetric between extreme cold and hot days. Specifically, default and delinquency rates
16
rise during hot days but not during cold days. These findings support our hypothesis that
payday lenders reduce credit supply during extreme heat days out of concern for increases in
default or delinquency but not during extreme cold days. Overall, our findings suggest that
credit supply responds differently to extreme temperature days due to asymmetric impacts
on applicant incomes. We investigate this next.
3.5. Do temperature exposures affect applicants’ income?
Borrower income plays an important role in both the supply and demand of payday loan
lending during extreme temperature days. If extreme weather negatively affects borrower
income, we would expect an increase in demand for payday loans. Meanwhile, payday loan
lenders may screen borrowers from very-low-income backgrounds, resulting in a contraction
in payday loan supply. We regress log of monthly ZCTA average income on the weather
exposure measures and other control variables as specified in Equation (1). Figure 6 indicates
a negative and statistically significant impact of extreme heat temperature shocks on income.
In particular, one extra day with daytime mean temperature above 33
◦
C is associated with
a 0.14% decrease in monthly income.
In contrast, we find no significant evidence that extreme cold temperature days affect
monthly income. Our finding that cold and hot temperatures have asymmetric effects on
income is in line with existing findings.
19
For example, Graff Zivin and Neidell (2014) find
that time allocated to labor is reduced with more extreme heat days but is non-responsive
at the low end of the temperature distribution.
Despite the importance of the income channel, this is not the only channel through
which extreme weather impacts the payday loan market. In Table B.1, we include income
as a control variable to isolate the direct effects of temperature exposures from the indirect
effects through income. Our baseline results hold after controlling for income, suggesting
19
In many regions, construction is more heavily scheduled during the summer months than winter months.
This is because of factors like favorable weather conditions, extended daylight hours, and optimal setting
behaviors of materials such as concrete and asphalt.
17
other channels, such as increasing energy or health expenditures, also matter.
3.6. Robustness
We perform several additional robustness checks in this section. First, in our main analy-
sis, we define temperature bins based on daytime (between 8am and 8pm) mean temperature:
below -9
◦
C, above 33
◦
C, and 14 3
◦
C bins in between. We omit the bins with daytime mean
temperature between -3
◦
C and 27
◦
C as the baseline. As robustness checks, we use alter-
native methods to define temperature bins. Table B.2 shows the results where we define
temperature bins using daily maximum temperature: below -6
◦
C, above 36
◦
C, and every
3
◦
C bin in between. We see that an additional day with daily maximum temperature above
36
◦
C results in increased demand, decreased credits, and a rise in default and delinquency.
Table B.3 presents results where we define temperature bins using the 24-hour daily mean
temperature: below -12
◦
C, above 30
◦
C, and every 3
◦
C bin in between. Again, the main
results are similar. Having an additional day in a month with 24-hour daily mean temper-
ature above 30
◦
C results in increased demand, decreased credits, and increased default and
delinquency rates. Table B.4 presents results where we calculate the number of days that
the daytime mean temperatures are above the 95th percentile and above 33
◦
C, the daytime
mean temperatures are below the 5th percentile and below -6
◦
C, and the number of days
with daytime mean temperature in between. In summary, these robustness checks show that
our main results are not sensitive to the specific definitions of “extreme temperature.”
Second, to allow for serial correlations in the dependent variables, we add the lags of
the dependent variable as an additional control variable in Table B.5 (using daytime mean
temperature), Table B.6 (using daily maximum temperature) and Table B.7 (using 24-hour
daily mean temperature). The inclusion of the lagged dependent variable as a regressor
significantly reduces our sample size. The estimates on delinquency and default rates are
sometimes imprecisely measured, although always positive. Our other main findings are not
affected.
Third, our primary payday loan dataset consists of an unbalanced panel, which could
18
potentially induce biases in the results since ZCTAs are dropped out of our analysis when
they do not have payday loans reported in a particular month. To address this, we first
performed a robustness check with a fully balanced panel. This involved filling in zeros for
ZCTA-level observations that occasionally do not have any payday loan observations for the
entire study period. As reported in Columns (1)–(4) of Table B.8, our results remained
consistent, demonstrating that our findings are not a consequence of possible biases in the
unbalanced panel data. We have also conducted a robustness check with a partially balanced
panel. Here, we fill in zeros for ZCTAs that occasionally do not have any payday loan
observations between the first and last period observed in the sample. This test allowed
us to maximize our sample size while controlling for potential biases related to unbalanced
data. Columns (5)–(8) of Table B.8 show that our key results hold up.
4. Further discussions
We extend our discussion of extreme weather exposure and credit demand over several
dimensions. Section 4.1 discusses heterogeneities between online and storefront markets.
Section 4.2 conducts heterogeneity analysis to examine the distributional effects of extreme
temperature shocks based on education. Section 4.3 compares our findings on the payday
loan market with other types of subprime credit. Section 4.4 evaluates the impact of the
LIHEAP on payday loan markets.
4.1. Storefront versus online payday loans
To this point, we have not differentiated between the storefront and online payday loan
markets. However, as these two markets have unique characteristics, we might anticipate
heterogeneous effects of temperature exposures. For instance, storefront loan borrowers must
apply in person, which can be challenging during days with extreme weather. To explore
the heterogeneities between the two markets, we repeat our earlier analysis separately for
storefront (Figure 7 and Table B.9) and online (Figure 8 and Table B.10) payday loan
markets.
19
For online payday loans, we observe increases in unique inquiries, delinquency, and default
rates, with decreases in accounts opened and credit issued under more extreme heat days.
Online payday loan inquiries also increase when there are more extreme cold days. Compared
to online borrowing, storefront loan borrowers’ unique inquiries and accounts open are less
responsive to extreme temperatures. We also observe a less significant response in storefront
payday loan default and delinquency rates compared to online loans. It is interesting to
note that higher levels of precipitation decrease inquiries for storefront loans but increase
inquiries for online loans. One potential explanation is that extreme weather discourages or
makes it more difficult for individuals to make in-person trips.
Taken together, our results suggest that the effects of temperature exposures on the
payday loan market operate predominantly through changes in online payday loan markets.
The distinctive characteristics of the online payday loan market, such as its accessibility
and ease of application, may make it more sensitive to changes in temperature and weather
conditions.
4.2. Heterogeneous effects
In this section, we conduct heterogeneity analysis to examine potential distributional
effects. To do so, we link ZCTA-level payday loan application data with neighborhood
demographic compositions. In particular, we use the share of the Hispanic population, given
their dominant employment in the construction sector, and outdoor work in general, with
greater exposure to temperatures. Among all occupation groups, construction workers are
most likely to be affected by extreme heat due to the outdoor nature of their work. As a
result, they are more likely to suffer from income loss and health-related issues due to direct
exposure, and they often lack sufficient protective measures against extreme temperatures.
In the following analysis, we include the interaction of temperature bins and county-level
demographic measures as additional variables to test whether disadvantaged neighborhoods
20
are disproportionally affected by extreme weather. We consider the following specification:
Outcome
it
= θT
it
+ γT
it
× Hispanic
c
+ µ
t
+ η
cy
+ Controls + ε
it
, (2)
where Hispanic
c
is a dummy variable that takes the value 1 if county c in which ZCTA i
belongs falls within the top 5th percentile in the share of Hispanics among all counties in
the year 2012, and is 0 otherwise. We use the same set of fixed effects and control variables
as in our baseline specification. Our primary interest is in the coefficient γ, which measures
the differential impacts based on county characteristics.
Table B.11 reports the estimates of γ’s for seven dependent variables. The results indicate
that counties with a larger share of the Hispanic population see a more significant increase
in demand for payday loans with an increased number of extreme heat days, along with
decreases in total credit. Extreme weather imposes a more considerable impact on households
in more disadvantaged neighborhoods, and as a consequence it drives up their demand for
payday loans more. The results also suggest a contraction in payday loan supply leading to
a decrease in total credits.
4.3. Other subprime alternative credits
While our main focus is on the payday loan industry, our dataset contains information
on other types of subprime credit products from Clarity. This section briefly describes
findings for alternative subprime credit, including rent-to-own, installment, auto, and auto
title loans. Compared to payday loans, these alternative subprime credit products are more
broadly available in the U.S., and borrowers usually take out larger amounts of credit through
each application. Interest rates on these loans could be very high if the borrower has bad
credit records. Unlike payday loans, defaults on these loans negatively affect borrowers’
creditworthiness.
Table B.12 reports the impact of temperature shocks on borrower income, the total
number of inquiries, accounts opened, and default rates. Similar to our findings on payday
21
loans, borrower income decreases while demand for credit increases during extreme heat
days. However, in contrast to payday loan lending, the number of accounts opened increases
and there is no impact on total credits approved, because the increases in extreme heat days
are not associated with greater delinquency. Figure 9 illustrates these results. This evidence
suggests that lenders are more willing to extend credit like installment loans when demand
increases due to extreme temperatures. To summarize, our results show that the impact of
weather exposures differs significantly according to the type of loan products.
4.4. Low-Income Home Energy Assistance Program
One major policy initiative aimed at helping households with energy needs during ex-
treme weather conditions is the Low-Income Home Energy Assistance Program (LIHEAP).
As discussed previously, one potential factor contributing to the rising demand for payday
loans during extreme temperatures is the increase in heating or cooling costs. Given the
negative impacts of extreme weather days on household payday loan market outcomes, we
now investigate whether LIHEAP helps improve payday loan market performance.
Our research design exploits the eligibility thresholds of LIHEAP, which vary across
states and are proportional to the state-level median income, as detailed in Section 2.3. Our
specification resembles a fuzzy regression discontinuity design, in that we compare payday
loan market outcomes for households slightly above versus below the eligibility thresholds
within a narrow bandwidth.
20
We use the following specification to offer evidence of the
impact of LIHEAP on payday loan borrowing:
Outcome
it
= θT
it
+ βTreat
it
+ µ
t
+ η
cy
+ Controls + ε
it
, (3)
where Treat
it
is a dummy variable that equals 1 if the payday loan applicant/borrower i
is eligible for the LIHEAP program in month t and equals 0 otherwise. In addition to
20
However, it is worth noting that we do not have information on the household-level LIHEAP enrollment
status, which prevents us from executing a full RDD design.
22
precipitation, year-month fixed effects, and county-year fixed effects, we also control renter
fixed effects and age group fixed effects. Standard errors are clustered at the ZCTA level.
We restrict our sample to payday loan borrowers whose incomes are within a narrow range
of the LIHEAP income eligibility cutoff. The eligibility cutoff varies based on household
size, with larger households having lower cutoffs per person. However, household size is not
available for payday loan borrowers in our data. We assume a household size of two to define
Treat
it
, and acknowledge that this assumption could introduce measurement errors in our
analysis.
In our main analysis, we adopt a bandwidth of $1,000 around the eligibility income thresh-
old in each state. We end up with a sample of 32,555 inquiry records and about 2,200 records
of loans. As reported in Table 4, we found no significant evidence that eligibility for LIHEAP
changes most payday loan market outcomes, such as credits, default, or delinquency rates.
We have two measures of inquiries at the individual level, total inquiries and days inquired,
which measure the total number of inquiries an individual made and the number of days an
individual made inquiries within each month, respectively. The estimates on inquiries are
significant and negative, suggesting that access to LIHEAP could lower household demand
for payday loans by offering an additional source of income subsidy when extreme weather
occurs.
There are many reasons why we may not see a significant impact of LIHEAP on other
payday loan market outcomes that are related to the program’s design and accessibility across
different states. For instance, in states like Arizona and California, LIHEAP is available
only once a year or 12-month period but accessible throughout the year; in Wisconsin, it is
a once-a-year benefit available only between October and May; in Ohio, there is a 12-week
application processing time. Households may choose to apply or receive the LIHEAP benefits
during a period without extreme weather conditions or fail to use LIHEAP to smooth their
credit demand over the course of the year. Targeting the accessibility and availability of
LIHEAP during extreme weather days may help alleviate the negative impacts on payday
23
loan markets. As mentioned previously, the lack of information on the number of household
members could also bias our results toward zero in the case of classical measurement errors.
Evidence in this section is merely suggestive and should be taken with caution.
5. Conclusion
Payday loans are a controversial short-term source of liquidity for low- to middle-income
customers. In this paper, we study how extreme temperatures affect household credit mar-
kets. We build a panel of monthly ZCTA-level temperature exposure and a panel of monthly
credit market measures. Our findings indicate that extreme heat days lead to more payday
loan inquiries, higher delinquency rates, and higher default rates. However, we do not see a
corresponding increase in the number of accounts open or credit issued. This indicates that
extreme heat increases demand for payday loan borrowing but reduces supply, presumably
because of the increased risk of delinquency and default.
Our study sheds light on the relationship between environmental shocks and household
finance, particularly regarding the use of non-traditional credit products by lower-income
households under extreme temperature shocks. Our findings highlight the heightened finan-
cial vulnerability of low-income households during extreme weather events and underscore
the urgent need for targeted interventions and policies. Developing strategies to mitigate
the adverse impacts of climate change on vulnerable individuals and assisting low-income
households in building resilience against these challenges is essential. Our research con-
tributes to the ongoing discourse on climate change adaptation and the financial well-being
of low-income households.
References
Addoum, Jawad M, David T Ng, and Ariel Ortiz-Bobea. 2020. “Temperature shocks and
establishment sales.” The Review of Financial Studies 33 (3):1331–1366.
Adhvaryu, Achyuta, Namrata Kala, and Anant Nyshadham. 2020. “The light and the heat:
Productivity co-benefits of energy-saving technology.” Review of Economics and Statistics
102 (4):779–792.
24
Allcott, Hunt, Joshua Kim, Dmitry Taubinsky, and Jonathan Zinman. 2022. “Are high-
interest loans predatory? Theory and evidence from payday lending.” The Review of
Economic Studies 89 (3):1041–1084.
Barreca, Alan, Karen Clay, Olivier Deschenes, Michael Greenstone, and Joseph S Shapiro.
2016. “Adapting to climate change: The remarkable decline in the US temperature-
mortality relationship over the twentieth century.” Journal of Political Economy
124 (1):105–159.
Barreca, Alan, R Jisung Park, and Paul Stainier. 2022. “High temperatures and electricity
disconnections for low-income homes in California.” Nature Energy 7 (11):1052–1064.
Benz, Susanne Amelie and Jennifer Anne Burney. 2021. “Widespread race and class
disparities in surface urban heat extremes across the United States.” Earth’s Future
9 (7):e2021EF002016.
Bhutta, Neil, Paige Marta Skiba, and Jeremy Tobacman. 2015. “Payday loan choices and
consequences.” Journal of Money, Credit and Banking 47 (2-3):223–260.
Blonz, Josh, Brigitte Roth Tran, and Erin E Troland. 2023. “The Canary in the Coal
Decline: Appalachian Household Finance and the Transition from Fossil Fuels.” Tech.
rep., National Bureau of Economic Research.
Burke, Kathleen, Jonathan Lanning, Jesse Leary, and Jialan Wang. 2014. “CFPB Data
Point: Payday Lending.” Tech. rep., The CFPB Office of Research. URL https://
files.consumerfinance.gov/f/201403_cfpb_report_payday-lending.pdf.
Cohen, Fran¸cois and Antoine Dechezleprˆetre. 2022. “Mortality, temperature, and public
health provision: evidence from Mexico.” American Economic Journal: Economic Policy
14 (2):161–192.
Colmer, Jonathan. 2021. “Temperature, labor reallocation, and industrial production: Evi-
dence from India.” American Economic Journal: Applied Economics 13 (4):101–124.
Correia, Filipe, Peter Han, and Jialan Wang. 2022. “The Online Payday Loan Premium.”
Working Paper .
Del Valle, Alejandro, Therese C Scharlemann, and Stephen H Shore. 2022. “Household
Financial Decision-Making After Natural Disasters: Evidence from Hurricane Harvey.” .
Dell, Melissa, Benjamin F Jones, and Benjamin A Olken. 2014. “What do we learn from the
weather? The new climate-economy literature.” Journal of Economic literature 52 (3):740–
798.
Deschenes, Olivier. 2022. “The impact of climate change on mortality in the United States:
Benefits and costs of adaptation.” Canadian Journal of Economics/Revue canadienne
d’´economique 55 (3):1227–1249.
Deschˆenes, Olivier and Michael Greenstone. 2007. “The economic impacts of climate change:
Evidence from agricultural output and random fluctuations in weather.” American Eco-
nomic Review 97 (1):354–385.
———. 2011. “Climate change, mortality, and adaptation: Evidence from annual fluctua-
25
tions in weather in the US.” American Economic Journal: Applied Economics 3 (4):152–
185.
Di Maggio, Marco, Angela Ma, and Emily Williams. 2021. “In the Red: Overdrafts, Payday
Lending and the Underbanked.” Payday Lending and the Underbanked (September 14,
2021) .
Dobridge, Christine L. 2018. “High-cost credit and consumption smoothing.” Journal of
Money, Credit and Banking 50 (2-3):407–433.
EPA. 2022. “Climate Change Indicators: Heat Waves.” https://www.epa.gov/climate-
indicators/climate-change-indicators-heat-waves .
Fonseca, Julia. 2023. “Less Mainstream Credit, More Payday Borrowing? Evidence from
Debt Collection Restrictions.” The Journal of Finance 78 (1):63–103.
Gallagher, Emily, Stephen B Billings, and Lowell Ricketts. 2020. “Human capital investment
after the storm.” The Review of Financial Studies .
Gallagher, Justin and Daniel Hartley. 2017. “Household finance after a natural disaster: The
case of Hurricane Katrina.” American Economic Journal: Economic Policy 9 (3):199–228.
Gathergood, John, Benedict Guttman-Kenney, and Stefan Hunt. 2019. “How do payday
loans affect borrowers? Evidence from the UK market.” The Review of Financial Studies
32 (2):496–523.
Graff Zivin, Joshua and Matthew Neidell. 2014. “Temperature and the allocation of time:
Implications for climate change.” Journal of Labor Economics 32 (1):1–26.
Hancock, Peter A, Jennifer M Ross, and James L Szalma. 2007. “A meta-analysis of perfor-
mance response under thermal stressors.” Human Factors 49 (5):851–877.
Heal, Geoffrey and Jisung Park. 2016. “Reflections—temperature stress and the direct im-
pact of climate change: A review of an emerging literature.” Review of Environmental
Economics and Policy .
———. 2020. “Reflections—temperature stress and the direct impact of climate change: A
review of an emerging literature.” Review of Environmental Economics and Policy .
Melzer, Brian T. 2011. “The real costs of credit access: Evidence from the payday lending
market.” The Quarterly Journal of Economics 126 (1):517–555.
———. 2018. “Spillovers from costly credit.” The Review of Financial Studies 31 (9):3568–
3594.
Miller, Sarah and Cindy K Soo. 2021. “Do neighborhoods affect the credit market decisions
of low-income borrowers? Evidence from the moving to opportunity experiment.” The
Review of Financial Studies 34 (2):827–863.
Morse, Adair. 2011. “Payday lenders: Heroes or villains?” Journal of Financial Economics
102 (1):28–44.
Niemel¨a, Raimo, Mika Hannula, Sari Rautio, Kari Reijula, and Jorma Railio. 2002. “The
effect of air temperature on labour productivity in call centres—a case study.” Energy and
26
Buildings 34 (8):759–764.
Park, Jisung, Nora Pankratz, and Arnold Behrer. 2021. “Temperature, workplace safety,
and labor market inequality.” .
Park, R Jisung, Joshua Goodman, Michael Hurwitz, and Jonathan Smith. 2020. “Heat and
learning.” American Economic Journal: Economic Policy 12 (2):306–39.
Roth Tran, Brigitte. 2023. “Sellin’ in the Rain: Weather, Climate, and Retail Sales.” Man-
agement Science .
Skiba, Paige Marta and Jeremy Tobacman. 2019. “Do payday loans cause bankruptcy?”
The Journal of Law and Economics 62 (3):485–519.
Wilson, Daniel J. 2019. “Clearing the fog: The predictive power of weather for employment
reports and their asset price responses.” American Economic Review: Insights 1 (3):373–
388.
Woolf, Steven, Joseph Morina, Evan French, Adam Funk, Roy Sabo, Stephen Fong, Jeremy
Hoffman, Derek Chapman, and Alex Krist. 2023. “The Health Care Costs of Extreme
Heat.” Tech. rep., Center for American Progress. URL https://www.americanprogress.
org/article/the-health-care-costs-of-extreme-heat/.
Zivin, Joshua Graff and Jeffrey Shrader. 2016. “Temperature extremes, health, and human
capital.” The Future of Children :31–50.
27
Figure 1: Distribution of daytime mean temperatures, 2012–2019
0 10 20 30 40
Number of Days per Year
<-9
-9~-6
-6~-3
-3~0
0~3
3~6
6~9
9~12
12~15
15~18
18~21
21~24
24~27
27~30
30~33
>33
All ZCTAs Clarity payday sample
Notes: The figure shows the historical average distribution of daytime mean temperatures across
16 temperature-day bins. Each bar represents the average number of days per year in each
temperature bin from 2012–2019. The “All ZCTAs” dataset represents ZCTAs of the continental
U.S., summing to 365 days across all bins. The “Clarity payday sample” covers a subset of all
ZCTAs, since not every ZCTA reports payday loans every month. In total, there are 159 days
across the bins.
28
Figure 2: Geographical distribution of daytime temperature bins
(a) Average monthly number of days with daytime mean temperature below -9
◦
C
(b) Average monthly number of days with daytime mean temperature above 33
◦
C
Notes: The figure shows the spatial distribution across ZCTAs of the monthly number of days
that had daytime (between 8am and 8pm) mean temperatures below -9
◦
C in panel (a) and above
33
◦
C in panel (b).
29
Figure 3: Impact on total inquiries and unique inquiries
Notes: The figure provides the estimated impact of a day in six different daytime mean temperature bins on number of total inquiries
(left) and unique inquiries (right), relative to a day in the -3
◦
C to 27
◦
C bins. Responses are estimated using specification (1). The
95% confidence intervals are based on standard errors clustered at the ZCTA level.
30
Figure 4: Impact on account open and total credits
Notes: The figure provides the estimated impact of a day in six different daytime mean temperature bins on number of accounts open
(left) and log of total credits (right), relative to a day in the -3
◦
C to 27
◦
C bins. Responses are estimated using specification (1). The
95% confidence intervals are based on standard errors clustered at the ZCTA level.
31
Figure 5: Impact on delinquency rate and default rate
Notes: The figure provides the estimated impact of a day in six different daytime mean temperature bins on delinquency rate (left) and
default rate (right), relative to a day in the -3
◦
C to 27
◦
C bins. Responses are estimated using specification (1). The 95% confidence
intervals are based on standard errors clustered at the ZCTA level.
32
Figure 6: Impact on average income
Notes: The figure provides the estimated impact of a day in six different daytime mean tem-
perature bins on log of average income, relative to a day in the -3
◦
C to 27
◦
C bins. Responses
are estimated using specification (1). The 95% confidence interval is based on standard errors
clustered at the ZCTA level.
33
Figure 7: Storefront payday loans
-3
-2
-1
0
1
2
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Total inquiries*100
-1
-.5
0
.5
1
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Unique inquiries*100
-2
-1
0
1
2
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Account open*100
-1
0
1
2
percent
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Log(total credits)*100
-.4
-.2
0
.2
.4
percent
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Delinquency rate
-.2
0
.2
.4
percent
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Default rate
Point estimate CI 95%
Notes: The figure provides the estimated impact of a day in six different daytime mean temperature bins
relative to a day in the -3
◦
C to 27
◦
C bins on the storefront payday loan market. Dependent variables
considered are number of total inquiries (top left), unique inquiries (top right), number of accounts open
(middle left), log of total credits (middle right), delinquency rate (bottom left), and default rate (bottom
right). Responses are estimated using specification (1). The 95% confidence interval is based on standard
errors clustered at the ZCTA level.
34
Figure 8: Online payday loans
-2
0
2
4
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Total inquiries*100
-.5
0
.5
1
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Unique inquiries*100
-1
-.5
0
.5
1
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Account open*100
-1
-.5
0
.5
1
percent
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Log(total credits)*100
-.5
0
.5
percent
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Delinquency rate
-.4
-.2
0
.2
.4
.6
percent
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Default rate
Point estimate CI 95%
Notes: The figure provides the estimated impact of a day in six different daytime mean temperature bins
relative to a day in the -3
◦
C to 27
◦
C bins on the online payday loan market. Dependent variables considered
are number of total inquiries (top left), unique inquiries (top right), number of accounts open (middle left),
log of total credits (middle right), delinquency rate (bottom left), and default rate (bottom right). Responses
are estimated using specification (1). The 95% confidence interval is based on standard errors clustered at
the ZCTA level.
35
Figure 9: Alternative subprime credit: installment loans
0
2
4
6
8
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Total inquiries*100
-1
-.5
0
.5
1
1.5
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Unique inquiries*100
-.5
0
.5
1
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Account open*100
-1
-.5
0
.5
1
percent
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Log(total credits)*100
-.4
-.2
0
.2
.4
percent
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Delinquency rate
-.4
-.2
0
.2
.4
percent
<-9 -9~-6 -6~-3 -3~27 27~30 30~33 >33
Default rate
Point estimate CI 95%
Notes: The figure provides the estimated impact of a day in six different daytime mean temperature bins
relative to a day in the -3
◦
C to 27
◦
C bins on installment loans. Dependent variables considered are number
of total inquiries (top left), unique inquiries (top right), number of accounts open (middle left), log of
total credits (middle right), delinquency rate (bottom left), and default rate (bottom right). Responses are
estimated using specification (1). The 95% confidence interval is based on standard errors clustered at the
ZCTA level.
36
Table 1: Summary statistics of the payday loan dataset
Full sample
mean sd p25 p50 p75 N
accounts open 1.73 1.38 1 1 2 115925
total highest credit 637.10 703.90 255 475 765 115925
delinquency rate 8.69 26.51 0 0 0 115925
default rate 7.33 24.57 0 0 0 115925
inquiries made 4.66 5.43 1 3 6 476928
unique inquiries 1.76 1.37 1 1 2 476928
average monthly income 2661 1343 1742 2500 3261 456435
Storefront sample
mean sd p25 p50 p75 N
accounts open 1.94 1.62 1 1 2 30853
total highest credit 882.90 1075.00 300 500 1000 30853
delinquency rate 6.38 23.00 0 0 0 30853
default rate 4.42 19.44 0 0 0 30853
inquiries made 2.05 2.22 1 1 2 54497
unique inquiries 1.46 0.98 1 1 2 54497
average monthly income 1758 1193 921 1494 2272 33230
Online sample
mean sd p25 p50 p75 N
accounts open 1.57 1.15 1 1 2 89921
total highest credit 518.4 421.3 255 400 605 89921
delinquency rate 9.62 28.1 0 0 0 89921
default rate 8.42 26.46 0 0 0 89921
inquiries made 4.64 5.36 1 3 6 454759
unique inquiries 1.69 1.26 1 1 2 454759
average monthly income 2706 1345 1751 2500 3333 443623
Notes: This table provides the summary statistics for the full payday loan sample, the storefront
payday loan, and the online payday loan subsamples.
37
Table 2: Summary statistics of ZCTA daily temperatures
All ZCTAs
mean sd p99 p95 p90 p50 p10 p5 p1
Daytime mean 14.80 11.18 33.28 29.95 28.07 16.39 -0.56 -4.79 -13.38
24-hour mean 12.92 10.72 30.59 27.56 25.78 14.34 -1.61 -5.79 -14.49
Maximum 18.19 11.02 36.90 33.14 31.11 19.95 2.40 -1.23 -9.07
ZCTAs with payday loan inquiries
mean sd p99 p95 p90 p50 p10 p5 p1
Daytime mean 17.55 10.32 34.38 31.04 29.28 19.45 2.72 -1.04 -9.18
24-hour mean 15.65 9.91 31.64 28.58 27.08 17.34 1.50 -2.01 -10.19
Maximum 20.90 10.20 38.17 34.36 32.34 22.89 5.92 1.81 -5.34
Notes: This table provides the summary statistics for daily temperatures of all ZCTAs (upper
panel) and ZCTAs that have payday loan inquiries in that month (lower panel) during 2012–2019.
Daytime mean refers to the average temperature between 8am and 8pm local time.
38
Table 3: Impacts of extreme temperatures on payday loan markets
(1) (2) (3) (4) (5) (6) (7)
log(income)*100 total inquiries *100 unique inquiries*100 accounts open *100 log(credits)*100 delinquency rate default rate
Days below -9
◦
C 0.008 1.091** 0.207** 0.539* 0.006 -0.029 -0.072
(0.065) (0.500) (0.101) (0.300) (0.194) (0.068) (0.059)
Days between -9
◦
C and -6
◦
C 0.008 1.862** 0.396** 0.471 0.754** -0.139 0.010
(0.123) (0.918) (0.199) (0.565) (0.367) (0.137) (0.123)
Days between -6
◦
C and -3
◦
C 0.007 -0.284 -0.227 0.721* 0.174 0.192* 0.081
(0.082) (0.679) (0.150) (0.420) (0.256) (0.101) (0.090)
Days between 27
◦
C and 30
◦
C -0.095*** -0.504** 0.038 -0.235** -0.070 -0.007 0.014
(0.022) (0.220) (0.055) (0.108) (0.063) (0.024) (0.023)
Days between 30
◦
C and 33
◦
C -0.090*** 1.509*** 0.506*** -0.083 0.008 0.034 0.032
(0.026) (0.327) (0.076) (0.129) (0.073) (0.029) (0.028)
Days above 33
◦
C -0.135*** 0.864* 0.270** -0.603*** -0.419*** 0.091** 0.114***
(0.035) (0.462) (0.127) (0.191) (0.109) (0.042) (0.040)
Precipitation -53.220 1,294.401*** 195.912* -97.662 -84.979 163.910*** 126.838***
(47.473) (439.052) (103.660) (182.022) (119.894) (46.061) (43.275)
Observations 439,568 461,993 461,993 110,246 109,424 110,246 110,246
R-squared 0.156 0.195 0.241 0.225 0.283 0.179 0.159
Year-month FE X X X X X X X
◦
County*Year FE X X X X X X X
Notes: The estimation results for Equation (1) are presented for seven different outcome variables. The independent variables are the number of days
in a month with daytime mean temperature within a specific range. The “Days between -3
◦
C and 27
◦
C” bin is the omitted category. The coefficient
β
k
is interpreted as the estimated impact of one additional day with daytime mean temperature within each respective temperature bin, relative to
the impact of a day with daytime mean temperature between -3
◦
C and 27
◦
C. Standard errors clustered at the ZCTA level are reported in parentheses.
***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.
39
Table 4: The effects of LIHEAP on payday loan credits
(1) (2) (3) (4) (5)
total inquiries*100 days inquired*100 log(credits)*100 delinquency rate default rate
Treat -17.49*** -3.235** -5.749 1.171 2.789
(5.564) 1.477 (4.208) (1.940) (1.838)
Observations 32,555 32,555 2,156 2,170 2,170
R-squared 0.221 0.178 0.5 0.402 0.351
Year-month FE X X X X X
County*Year FE X X X X X
Renter FE X X X X X
Age group FE X X X X X
Notes: The estimation results for Equation (3) are presented for five different outcome variables. Treat is
equal to 1 if an individual’s income is within the $1,000 bandwidth and below the LIHEAP eligibility cutoff.
We also include the temperature bins and the precipitation variables as controls. Standard errors clustered
at the ZCTA level are reported in parentheses. ***, **, and * represent statistical significance at the 1%,
5%, and 10% levels, respectively.
40
Internet Appendix
Appendix A. Background information
Appendix A.1. Reasons for taking payday loans
In 2013, 2016, and 2019 waves of Survey of Consumer Finances
21
respondents were asked
two specific questions related to their use of payday loans and the reasons behind their
decision to use them. A total of 2,745 participants answered “YES” to the first question,
indicating that they had indeed taken out payday loans. That is about 3% of all survey
participants. Among these individuals who had used payday loans, Figure A.1 illustrates
the distribution of various reasons for taking them. The most prevalent reason reported was
for “Emergency” needs, with covering bills and utilities being other commonly mentioned
reasons. Additionally, nearly 13% of the respondents stated that they perceived payday
loans as their only available option.
Question: During the past year, have you (or anyone in your family living here) taken out
a “payday loan,” that is, borrowed money that was supposed to be repaid in full out of your
next paycheck?
IF YES: Please do not include personal loans from family members or friends.
1. YES
2. NO
Question: Why did you choose this type of loan?
1. Buy food
2. Buy gas
3. Buy medicine/medical payments
4. Pay utilities
5. Pay rent
6. Vehicle expenses other than gas
7. Pay other bills/loans
21
https://www.federalreserve.gov/econres/aboutscf.htm
1
Appendix Figure A.1: Reasons for taking payday loans
Notes: The figure presents the percentages representing individual reasons for taking a payday loan, calcu-
lated based on responses from the Survey of Consumer Finances for the years 2013, 2016, and 2019. The
reasons are ordered and plotted in descending order from top to bottom. The percentage indicates how
many respondents, out of the total responses, selected each specific reason. The percentages add up to 100.
8. “Christmas”
9. Help family
10. “Emergency”/”needed quick money”
11. “Convenient”
12. “Only option”
Appendix A.2. LIHEAP program
Since its inception in 1981, LIHEAP
22
has been playing a crucial role in providing vital
support to eligible low-income households. LIHEAP offers a range of assistance programs,
including addressing heating and cooling energy costs, providing bill payment assistance,
offering aid during energy crises, facilitating weatherization efforts, and supporting energy-
related home repairs. By helping these households manage their utility bills during the
crucial cold or hot months of the year, LIHEAP ensures that their energy needs are met.
Furthermore, the program offers “crisis” funds to promptly restore utility services for eligi-
ble households that have experienced service interruption or are at risk of it. In addition,
22
See https://www.liheap.org/.
2
LIHEAP extends support to eligible households by allocating funds for weatherization mea-
sures and energy-related home repairs. Notably, the heating and cooling payment assistance
program stands as the largest component of LIHEAP. Each year, nearly two-thirds of funding
was used for heating and cooling assistance.
The implementation of LIHEAP also differs across states, resulting in varying program
structures and offerings. For example, in Arizona and California, LIHEAP provides one-time
financial assistance to help eligible households manage their utility bills. In Wisconsin, it is
a once-a-year benefit available only between October 1 and May 15. Meanwhile, Delaware
offers bills and/or energy assistance twice a year. Florida allows applications up to three
times a year. The processing time of applications varies across states as well. For example,
in Ohio there is a 12-week application processing time. Oklahoma’s non-emergency cooling
and winter heating assistance may take up to 60 days for processing.
In most cases, approved households won’t receive payment directly. LIHEAP almost
always pays grants directly to the energy utility. In Minnesota, initial benefits average $500
per household and can be up to $1,400.
According to the latest White paper, cold weather states traditionally spend more than
70% of the LIHEAP funds during the first two quarters of the federal fiscal year.
3
Appendix B. Additional Figures and Tables
Appendix Figure B.1: Geographic Distribution of loans per capita
(a) Storefront
(b) Online
Notes: This figure plots the geographic distribution of loans per capita for storefront (upper) and online
(lower) payday loan subsamples, respectively.
4
Appendix Figure B.2: Average daytime mean temperature
Notes: The figure shows the spatial distribution of the daytime (between 8am and 8pm) mean
temperatures across ZCTAs, averaged from 2012–2019. Around 5% ZCTA-day level observations
have a daytime mean temperature above the 30
◦
C threshold, and around 5% ZCTA-day level
observations have a daytime mean temperature below the -5
◦
C threshold.
5
Appendix Figure B.3: Geographic distribution of income and LIHEAP eligibility
(a) LIHEAP Eligibility
(b) Income of Payday Loan Borrowers
Notes: This figure plots the geographic distribution of LIHEAP eligibility cutoff (upper) and income of
payday loan borrowers (lower), respectively.
6
Appendix Table B.1: Robustness check: controlling for average income
(1) (2) (3) (4) (5) (6)
total inquiries *100 unique inquiries*100 accounts open *100 log(credits)*100 delinquency rate default rate
Days below -9
◦
C 1.256** 0.260** 0.557 -0.336 -0.024 -0.098
(0.519) (0.105) (0.411) (0.249) (0.097) (0.083)
Days between -9
◦
C and -6
◦
C 2.061** 0.399* 1.073 0.702 -0.033 0.180
(0.961) (0.207) (0.802) (0.471) (0.190) (0.168)
Days between -6
◦
C and -3
◦
C -0.431 -0.249 0.887 0.278 0.108 -0.032
(0.718) (0.157) (0.567) (0.317) (0.137) (0.120)
Days between 27
◦
C and 30
◦
C -0.632*** 0.029 -0.296** -0.115 -0.028 0.003
(0.227) (0.056) (0.133) (0.074) (0.029) (0.027)
Days between 30
◦
C and 33
◦
C 1.377*** 0.510*** -0.210 0.011 0.022 0.016
(0.335) (0.077) (0.154) (0.084) (0.035) (0.033)
Days above 33
◦
C 0.610 0.220* -0.664*** -0.501*** 0.102** 0.127***
(0.471) (0.129) (0.216) (0.121) (0.050) (0.048)
Precipitation 1,109.098** 197.105* 50.061 -92.023 200.117*** 152.962***
(466.878) (107.568) (232.759) (140.525) (57.151) (53.556)
log(avg income) -0.493*** 0.059*** 0.001 0.125*** -0.025*** -0.021***
(0.021) (0.006) (0.014) (0.008) (0.002) (0.002)
Observations 439,568 439,568 82,114 81,521 82,114 82,114
R-squared 0.200 0.254 0.268 0.271 0.186 0.166
Year-month FE X X X X X X
◦
County*Year FE X X X X X X
Notes: The estimation results for Equation (1) are presented for six different outcome variables. Average income is included as a control variable.
The independent variables are the number of days in a month with daytime mean temperature within a specific range. The “Days between -3
◦
C and
27
◦
C” bin is the omitted category. The coefficient β
k
is interpreted as the estimated impact of one additional day with daytime mean temperature
within each respective temperature bin, relative to the impact of a day with daytime mean temperature between -3
◦
C and 27
◦
C. Standard errors
clustered at the ZCTA level are reported in parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.
7
Appendix Table B.2: Robustness check: daily max temperature
(1) (2) (3) (4) (5) (6) (7)
log(income)*100 total inquiries *100 unique inquiries*100 accounts open *100 log(credits)*100 delinquency rate default rate
Days below -6
◦
C 0.055 1.450** 0.292** 0.887** 0.185 -0.094 -0.117
(0.076) (0.583) (0.117) (0.355) (0.225) (0.090) (0.077)
Days between -6
◦
C and -3
◦
C -0.129 -1.360 -0.577*** -0.865* -0.011 0.080 0.115
(0.118) (0.940) (0.196) (0.492) (0.331) (0.130) (0.112)
Days between -3
◦
C and 0
◦
C 0.049 0.701 0.260* 1.115*** 0.392* 0.079 0.021
(0.077) (0.643) (0.146) (0.344) (0.224) (0.084) (0.075)
Days between 30
◦
C and 33
◦
C -0.092*** -0.064 0.082 -0.245* -0.068 -0.009 0.013
(0.024) (0.252) (0.069) (0.125) (0.069) (0.025) (0.024)
Days between 33
◦
C and 36
◦
C -0.070*** 1.366*** 0.477*** -0.023 0.017 0.028 0.021
(0.027) (0.338) (0.083) (0.139) (0.076) (0.030) (0.028)
Days above 36
◦
C -0.118*** 1.099*** 0.350*** -0.483*** -0.325*** 0.082** 0.099***
(0.032) (0.405) (0.113) (0.173) (0.094) (0.034) (0.033)
Precipitation -62.570 1,099.071*** 199.937** -114.348 -92.196 153.490*** 119.788***
(46.157) (420.493) (100.351) (172.860) (113.071) (43.948) (41.439)
Observations 441,842 464,452 464,452 111,124 110,302 111,124 111,124
R-squared 0.156 0.195 0.241 0.226 0.283 0.179 0.160
Year-month FE X X X X X X X
◦
County*Year FE X X X X X X X
Notes: The estimation results for Equation (1) are presented for seven different outcome variables. The independent variables are the number of
days in a month with daily maximum temperature within a specific range. We use 16 temperature bins: below -6
◦
C, above 36
◦
C, and 14 3
◦
C bins in
between. The “Days between 0
◦
C and 30
◦
C” bin is the omitted category. The coefficient β
k
is interpreted as the estimated impact of one additional
day with daytime mean temperature within each respective temperature bin, relative to the impact of a day with daytime mean temperature between
0
◦
C and 30
◦
C. Standard errors clustered at the ZCTA level are reported in parentheses. ***, **, and * represent statistical significance at the 1%,
5%, and 10% levels, respectively.
8
Appendix Table B.3: Robustness check: 24-hour mean temperature
(1) (2) (3) (4) (5) (6) (7)
log(income)*100 total inquiries *100 unique inquiries*100 accounts open *100 log(credits)*100 delinquency rate default rate
Days below -12
◦
C 0.007 0.878 0.230* 0.475 0.165 -0.111 -0.141*
(0.087) (0.621) (0.129) (0.385) (0.253) (0.090) (0.078)
Days between -12
◦
C and -9
◦
C -0.039 2.029* 0.174 1.196* 0.350 0.211 0.184
(0.147) (1.160) (0.250) (0.698) (0.429) (0.152) (0.127)
Days between -9
◦
C and -6
◦
C 0.061 0.692 -0.015 0.445 0.289 0.008 0.016
(0.102) (0.815) (0.173) (0.478) (0.299) (0.120) (0.108)
Days between 24
◦
C and 27
◦
C -0.052** -0.529*** 0.007 -0.214** -0.049 -0.037 -0.017
(0.020) (0.198) (0.048) (0.102) (0.057) (0.023) (0.022)
Days between 27
◦
C and 30
◦
C -0.118*** 0.918*** 0.405*** -0.089 0.028 0.041* 0.039*
(0.021) (0.237) (0.055) (0.103) (0.061) (0.023) (0.021)
Days above 30
◦
C -0.115*** 1.183*** 0.333*** -0.582*** -0.377*** 0.068* 0.090**
(0.030) (0.391) (0.103) (0.158) (0.096) (0.038) (0.036)
Precipitation -27.028 775.340* 27.222 -73.425 -88.748 137.286*** 106.046**
(45.881) (417.775) (100.134) (176.575) (113.472) (43.957) (41.441)
Observations 441,842 464,452 464,452 111,124 110,302 111,124 111,124
R-squared 0.156 0.195 0.241 0.226 0.283 0.179 0.160
Year-month FE X X X X X X X
◦
County*Year FE X X X X X X X
Notes: The estimation results for Equation (1) are presented for seven different outcome variables. The independent variables are the number of days
in a month with 24-hour daily mean temperature within a specific range. We use 16 temperature bins: below -12
◦
C, above 30
◦
C, and 14 2
◦
C bins in
between. The “Days between -6
◦
C and 24
◦
C” bin is the omitted category. The coefficient β
k
is interpreted as the estimated impact of one additional
day with daytime mean temperature within each respective temperature bin, relative to the impact of a day with daytime mean temperature between
-6
◦
C and 24
◦
C. Standard errors clustered at the ZCTA level are reported in parentheses. ***, **, and * represent statistical significance at the 1%,
5%, and 10% levels, respectively.
9
Appendix Table B.4: Robustness check: local extreme temperature
(1) (2) (3) (4) (5) (6) (7)
log(income)*100 total inquiries *100 unique inquiries*100 accounts open *100 log(credits)*100 delinquency rate default rate
Days below min(-6
◦
C, 5th percentile) -0.037 1.799*** 0.348*** 1.154*** 0.449*** -0.084* -0.087**
(0.054) (0.463) (0.101) (0.247) (0.143) (0.045) (0.040)
Days above max(33
◦
C, 95th percentile) -0.112*** 1.411*** 0.664*** -0.105 -0.014 0.104** 0.093**
(0.035) (0.368) (0.084) (0.189) (0.105) (0.041) (0.039)
Precipitation -10.407 1,119.272*** 37.628 -139.382 -107.111 122.955*** 93.978**
(45.950) (423.157) (101.039) (178.874) (118.450) (44.795) (42.199)
Observations 441,393 463,795 463,795 111,701 110,822 111,701 111,701
R-squared 0.107 0.150 0.201 0.169 0.216 0.112 0.097
Year-month FE X X X X X X X
◦
County FE X X X X X X X
Notes: The estimation results for Equation (1) are presented for seven different outcome variables. The independent variables are the number of days in a
month with daytime mean temperature within a specific range. We use three temperature bins: below the minimum of -6
◦
C and below the 5th percentile
temperature in the ZCTA’s history, above the maximum of 33
◦
C and above 95th percentile temperature in the ZCTA’s history, and temperature in between.
The last bin is the omitted category. The coefficient β
k
is interpreted as the estimated impact of one additional day with daytime mean temperature within
each respective temperature bin, relative to the impact of a day with daytime mean temperature between the two extreme values. Standard errors clustered
at the ZCTA level are reported in parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.
10
Appendix Table B.5: Robustness check: controlling for lags of dependent variable
(1) (2) (3) (4) (5) (6) (7)
log(income)*100 total inquiries *100 unique inquiries*100 accounts open *100 log(credits)*100 delinquency rate default rate
Days below -9
◦
C 0.163* 1.174 0.332* -0.540 -0.221 -0.018 -0.048
(0.084) (0.764) (0.171) (0.399) (0.232) (0.061) (0.049)
Days between -9
◦
C and -6
◦
C -0.133 3.813*** 1.051*** -0.060 0.364 -0.104 0.056
(0.162) (1.414) (0.322) (0.778) (0.439) (0.122) (0.102)
Days between -6
◦
C and -3
◦
C -0.022 -1.516 -0.579** 1.640*** 0.768*** 0.203** 0.114
(0.105) (1.002) (0.237) (0.532) (0.294) (0.092) (0.073)
Days between 27
◦
C and 30
◦
C -0.076*** -0.573** -0.040 -0.236* -0.077 0.013 0.031
(0.026) (0.278) (0.065) (0.130) (0.066) (0.022) (0.021)
Days between 30
◦
C and 33
◦
C -0.070** 1.204*** 0.378*** -0.081 0.068 0.050* 0.030
(0.029) (0.382) (0.081) (0.151) (0.077) (0.029) (0.027)
Days above 33
◦
C -0.100*** 1.112** 0.241** -0.522*** -0.308*** 0.015 0.051
(0.038) (0.502) (0.111) (0.190) (0.108) (0.041) (0.039)
Precipitation 0.204*** 1,424.161** 296.600** -318.021 -136.330 66.932 47.426
(0.004) (591.465) (138.467) (238.527) (133.597) (42.647) (38.280)
lag of dep. var -47.568 0.364*** 0.514*** 0.518*** 0.506*** 0.108*** 0.082***
(53.872) (0.008) (0.008) (0.017) (0.008) (0.009) (0.010)
Observations 261,224 281,151 281,151 70,013 69,711 70,013 70,013
R-squared 0.211 0.309 0.437 0.450 0.504 0.157 0.126
Year-month FE X X X X X X
◦
County*Year FE X X X X X X
Notes: The estimation results for Equation (1) are presented for seven different outcome variables. The lag of dependent variable is included as a
control variable. The independent variables are the number of days in a month with daytime mean temperature within a specific range. The “Days
between -3
◦
C and 27
◦
C” bin is the omitted category. The coefficient β
k
is interpreted as the estimated impact of one additional day with daytime
mean temperature within each respective temperature bin, relative to the impact of a day with daytime mean temperature in between -3
◦
C and 27
◦
C.
Standard errors clustered at the ZCTA level are reported in parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10%
levels, respectively.
11
Appendix Table B.6: Robustness check: daily max temperature and controlling lag of dependent variable
(1) (2) (3) (4) (5) (6) (7)
log(income)*100 total inquiries *100 unique inquiries*100 accounts open *100 log(credits)*100 delinquency rate default rate
Days below -6
◦
C 0.197* 1.351 0.484** 0.189 -0.028 -0.044 -0.068
(0.101) (0.908) (0.199) (0.502) (0.270) (0.088) (0.070)
Days between -6
◦
C and -3
◦
C -0.061 -0.512 -0.400 -2.511*** -0.423 -0.058 0.056
(0.159) (1.456) (0.323) (0.776) (0.418) (0.119) (0.092)
Days between -3
◦
C and 0
◦
C -0.110 0.349 0.242 2.026*** 0.867*** 0.186** 0.126**
(0.097) (0.926) (0.219) (0.465) (0.260) (0.076) (0.063)
Days between 30
◦
C and 33
◦
C -0.059** -0.053 0.013 -0.187 -0.076 0.016 0.036*
(0.028) (0.315) (0.075) (0.148) (0.070) (0.023) (0.022)
Days between 33
◦
C and 36
◦
C -0.045 1.001** 0.319*** -0.039 0.071 0.037 0.013
(0.029) (0.396) (0.087) (0.162) (0.079) (0.029) (0.027)
Days above 36
◦
C -0.082** 1.270*** 0.284*** -0.404** -0.232*** 0.036 0.063**
(0.035) (0.440) (0.098) (0.168) (0.089) (0.033) (0.031)
Precipitation -54.240 1,240.376** 295.072** -328.192 -148.389 64.127 46.888
(52.179) (565.080) (133.128) (228.140) (126.673) (41.593) (37.555)
lag of dep. var 0.204*** 0.364*** 0.514*** 0.518*** 0.505*** 0.109*** 0.083***
(0.004) (0.008) (0.008) (0.017) (0.008) (0.009) (0.010)
Observations 262,442 282,533 282,533 70,489 70,187 70,489 70,489
R-squared 0.212 0.309 0.436 0.450 0.503 0.157 0.126
Year-month FE X X X X X X
◦
County*Year FE X X X X X X
Notes: The estimation results for Equation (1) are presented for seven different outcome variables. The lag of dependent variable is included as a
control variable. The independent variables are the number of days in a month with daily maximum temperature within a specific range. We use
16 temperature bins: below -6
◦
C, above 36
◦
C, and 14 3
◦
C bins in between. The “Days between 0
◦
C and 30
◦
C” bin is the omitted category. The
coefficient β
k
is interpreted as the estimated impact of one additional day with daytime mean temperature within each respective temperature bin,
relative to the impact of a day with daytime mean temperature between 0
◦
C and 30
◦
C. Standard errors clustered at the ZCTA level are reported in
parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.
12
Appendix Table B.7: Robustness check: 24-hour mean temperature and controlling lag of dependent variable
(1) (2) (3) (4) (5) (6) (7)
log(income)*100 total inquiries *100 unique inquiries*100 accounts open *100 log(credits)*100 delinquency rate default rate
Days below -12
◦
C 0.231** 0.890 0.391* -0.099 0.193 -0.070 -0.067
(0.114) (0.968) (0.215) (0.538) (0.307) (0.086) (0.069)
Days between -12
◦
C and -9
◦
C -0.120 3.013* 0.500 0.424 0.083 0.016 0.067
(0.197) (1.744) (0.418) (0.922) (0.514) (0.130) (0.098)
Days between -9
◦
C and -6
◦
C -0.069 0.766 -0.010 -0.016 0.148 0.132 0.099
(0.132) (1.216) (0.283) (0.698) (0.359) (0.111) (0.092)
Days between 24
◦
C and 27
◦
C -0.033 -0.398 0.030 -0.095 -0.044 -0.013 0.004
(0.024) (0.258) (0.060) (0.122) (0.062) (0.020) (0.019)
Days between 27
◦
C and 30
◦
C -0.098*** 0.574** 0.252*** -0.117 0.059 0.052** 0.042**
(0.023) (0.280) (0.061) (0.115) (0.061) (0.022) (0.020)
Days above 30
◦
C -0.081** 1.506*** 0.328*** -0.466*** -0.272*** 0.016 0.041
(0.032) (0.429) (0.095) (0.167) (0.100) (0.038) (0.035)
Precipitation -25.211 1,032.087* 166.360 -346.897 -198.784 36.904 24.970
(51.984) (560.747) (132.086) (225.638) (126.011) (41.485) (37.539)
lag of dep. var 0.204*** 0.364*** 0.514*** 0.518*** 0.505*** 0.109*** 0.083***
(0.004) (0.008) (0.008) (0.017) (0.008) (0.009) (0.010)
Observations 262,442 282,533 282,533 70,489 70,187 70,489 70,489
R-squared 0.212 0.309 0.436 0.449 0.503 0.157 0.126
Year-month FE X X X X X X
◦
County*Year FE X X X X X X
Notes: The estimation results for Equation (1) are presented for seven different outcome variables. The lag of dependent variable is included as a
control variable. The independent variables are the number of days in a month with 24-hour daily mean temperature within a specific range. We use
16 temperature bins: below -12
◦
C, above 30
◦
C, and 14 3
◦
C bins in between. The “Days between -6
◦
C and 24
◦
C” bin is the omitted category. The
coefficient β
k
is interpreted as the estimated impact of one additional day with daytime mean temperature within each respective temperature bin,
relative to the impact of a day with daytime mean temperature between -6
◦
C and 24
◦
C. Standard errors clustered at the ZCTA level are reported in
parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.
13
Appendix Table B.8: Robustness check: balanced and partially balanced panels
balanced partially balanced
(1) (2) (3) (4) (5) (6) (7) (8)
total inquiry*100 unique inquiry*100 accounts open*100 log(credits)*100 total inquiry*100 unique inquiry*100 accounts open*100 log(credits)*100
Days below -6
◦
C 0.484*** 0.109*** 0.123*** 0.006 0.586*** 0.109*** 0.470*** 0.006
(0.078) (0.021) (0.036) (0.194) (0.141) (0.038) (0.115) (0.194)
Days between -6
◦
C and -3
◦
C 0.624*** 0.150*** 0.076 0.754** 0.950*** 0.227*** 0.127 0.754**
(0.164) (0.046) (0.075) (0.367) (0.294) (0.081) (0.231) (0.367)
Days between -3
◦
C and 0
◦
C -0.149 -0.121*** 0.104* 0.174 -0.340 -0.196*** 0.070 0.174
(0.143) (0.042) (0.062) (0.256) (0.242) (0.068) (0.172) (0.256)
Days between 30
◦
C and 33
◦
C 0.229*** 0.183*** -0.088*** -0.070 0.092 0.177*** -0.174*** -0.070
(0.082) (0.026) (0.027) (0.063) (0.110) (0.033) (0.053) (0.063)
Days between 33
◦
C and 36
◦
C 1.029*** 0.319*** 0.045 0.008 1.238*** 0.398*** 0.017 0.008
(0.130) (0.037) (0.033) (0.073) (0.172) (0.047) (0.061) (0.073)
Days above 36
◦
C 0.936*** 0.321*** -0.175** -0.419*** 0.908*** 0.307*** -0.313*** -0.419***
(0.244) (0.083) (0.071) (0.109) (0.294) (0.095) (0.113) (0.109)
Precipitation 253.279* 37.464 -8.949 -84.979 461.867** 96.214* -82.862 -84.979
(134.968) (39.466) (40.596) (119.894) (196.540) (55.090) (80.458) (119.894)
Observations 1,983,936 1,983,936 771,960 109,424 1,249,745 1,249,745 321,532 109,424
R-squared 0.203 0.265 0.189 0.283 0.206 0.267 0.214 0.283
Year-month FE X X X X X X X X
◦
County*Year FE X X X X X X X X
Notes: The estimation results for Equation (1) are presented for four different outcome variables using balanced or partially balanced panels. The
independent variables are the number of days in a month with daytime mean temperature within a specific range. The “Days between -3
◦
C and
27
◦
C” bin is the omitted category. The coefficient β
k
is interpreted as the estimated impact of one additional day with daytime mean temperature
within each respective temperature bin, relative to the impact of a day with daytime mean temperature in between -3
◦
C and 27
◦
C. Standard errors
clustered at the ZCTA level are reported in parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.
14
Appendix Table B.9: Storefront payday loan market
(1) (2) (3) (4) (5) (6) (7)
log(income)*100 total inquiries *100 unique inquiries*100 accounts open *100 log(credits)*100 delinquency rate default rate
Days below -9
◦
C -0.720* -0.289 -0.326 1.088* -0.026 0.085 0.057
(0.373) (1.042) (0.321) (0.592) (0.372) (0.095) (0.074)
Days between -9
◦
C and -6
◦
C 0.927* -0.469 0.250 -0.055 0.831 -0.105 -0.064
(0.496) (1.252) (0.585) (1.044) (0.618) (0.158) (0.111)
Days between -6
◦
C and -3
◦
C 0.147 0.338 0.455 0.933 0.164 0.125 0.089
(0.341) (0.773) (0.384) (0.850) (0.471) (0.137) (0.104)
Days between 27
◦
C and 30
◦
C -0.240** 0.941*** 0.076 -0.216 -0.151 -0.042 -0.007
(0.108) (0.295) (0.129) (0.208) (0.157) (0.047) (0.043)
Days between 30
◦
C and 33
◦
C 0.014 0.456 -0.107 -0.347 0.257 0.122* 0.016
(0.124) (0.388) (0.138) (0.250) (0.202) (0.064) (0.052)
Days above 33
◦
C -0.444** 1.338** 0.220 0.008 -0.791*** 0.057 0.184*
(0.179) (0.576) (0.223) (0.261) (0.259) (0.126) (0.106)
Precipitation 238.505 -1,040.247** -330.609 -161.737 -17.555 166.106 24.858
(239.716) (443.716) (246.392) (380.107) (326.824) (114.989) (98.605)
Observations 31,070 52,589 52,589 30,081 30,081 30,081 30,081
R-squared 0.307 0.334 0.242 0.311 0.445 0.244 0.172
Year-month FE X X X X X X X
◦
County*Year FE X X X X X X X
Notes: The estimation results for Equation (1) are presented for seven different outcome variables based on the storefront payday loan market. The
independent variables are the number of days in a month with daytime mean temperature within a specific range. The “Days between -3
◦
C and
27
◦
C” bin is the omitted category. The coefficient β
k
is interpreted as the estimated impact of one additional day with daytime mean temperature
within each respective temperature bin, relative to the impact of a day with daytime mean temperature between -3
◦
C and 27
◦
C. Standard errors
clustered at the ZCTA level are reported in parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.
15
Appendix Table B.10: Online payday loan market
(1) (2) (3) (4) (5) (6) (7)
log(income)*100 total inquiries *100 unique inquiries*100 accounts open *100 log(credits)*100 delinquency rate default rate
Days below -9
◦
C 0.046 1.306*** 0.239** -0.114 -0.331 -0.118 -0.221**
(0.065) (0.505) (0.099) (0.217) (0.208) (0.103) (0.092)
Days between -9
◦
C and -6
◦
C -0.098 2.125** 0.367* 0.277 0.211 -0.168 0.160
(0.123) (0.954) (0.187) (0.416) (0.422) (0.217) (0.202)
Days between -6
◦
C and -3
◦
C 0.000 -0.214 -0.287** -0.032 -0.188 0.212 0.106
(0.082) (0.711) (0.143) (0.317) (0.280) (0.149) (0.140)
Days between 27
◦
C and 30
◦
C -0.082*** -0.672*** 0.033 -0.089 -0.030 0.002 0.006
(0.022) (0.220) (0.051) (0.112) (0.067) (0.028) (0.027)
Days between 30
◦
C and 33
◦
C -0.059** 1.356*** 0.464*** -0.027 -0.014 0.037 0.042
(0.026) (0.327) (0.072) (0.131) (0.075) (0.033) (0.032)
Days above 33
◦
C -0.097*** 0.702 0.215** -0.577*** -0.282*** 0.097** 0.101**
(0.035) (0.439) (0.108) (0.201) (0.107) (0.046) (0.044)
Precipitation -92.237* 1,340.747*** 170.709* -190.852 -114.387 199.849*** 171.400***
(47.162) (456.438) (101.488) (186.691) (123.221) (52.285) (50.379)
Observations 427,243 440,260 440,260 84,580 83,689 84,580 84,580
R-squared 0.147 0.190 0.227 0.168 0.190 0.189 0.170
Year-month FE X X X X X X X
◦
County*Year FE X X X X X X X
Notes: The estimation results for Equation (1) are presented for seven different outcome variables based on the online payday loan market. The
independent variables are the number of days in a month with daytime mean temperature within a specific range. The “Days between -3
◦
C and
27
◦
C” bin is the omitted category. The coefficient β
k
is interpreted as the estimated impact of one additional day with daytime mean temperature
within each respective temperature bin, relative to the impact of a day with daytime mean temperature between -3
◦
C and 27
◦
C. Standard errors
clustered at the ZCTA level are reported in parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.
16
Appendix Table B.11: Heterogenous effects
(1) (2) (3) (4) (5) (6) (7)
Hispanic × log(income)*100 total inquiries *100 unique inquiries*100 accounts open *100 log(credits)*100 delinquency rate default rate
Days below -6
◦
C 0.191 -5.211 -1.962 0.525 -5.295 1.581 1.473
(0.876) (4.390) (1.681) (2.383) (3.229) (1.555) (1.526)
Days between -6
◦
C and -3
◦
C -0.796 12.923 3.392* -1.271 -0.839 -1.166 -0.597
(0.899) (8.082) (2.044) (3.305) (3.453) (1.239) (1.221)
Days between -3
◦
C and 0
◦
C -0.317 -1.251 -0.883 0.636 2.180 -1.057 -1.288**
(0.513) (4.631) (1.141) (2.199) (2.149) (0.710) (0.647)
Days between 30
◦
C and 33
◦
C 0.030 -0.654 -0.117 0.027 -0.012 -0.039 -0.056
(0.039) (0.483) (0.108) (0.190) (0.104) (0.042) (0.040)
Days between 33
◦
C and 36
◦
C 0.078 1.668** 0.485*** -0.023 -0.271* 0.067 0.058
(0.049) (0.760) (0.167) (0.243) (0.140) (0.059) (0.056)
Days above 36
◦
C -0.099 0.031 -0.206 -0.369 0.143 -0.124 -0.177**
(0.071) (1.064) (0.284) (0.325) (0.199) (0.079) (0.075)
Observations 439,462 461,881 461,881 110,246 109,424 110,246 110,246
R-squared 0.156 0.195 0.241 0.225 0.283 0.179 0.159
Year-month FE X X X X X X X
◦
County*Year FE X X X X X X X
Notes: The estimation results for Equation (2) are presented for several specifications. Standard errors clustered at the ZCTA level are reported in
parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.
17
Appendix Table B.12: Alternative subprime credit
(1) (2) (3) (4) (5) (6) (7)
log(income)*100 total inquiries *100 unique inquiries*100 accounts open *100 log(credits)*100 delinquency rate default rate
Days below -9
◦
C 0.072 3.746*** 0.756*** -0.022 -0.566* -0.106 -0.008
(0.070) (0.659) (0.127) (0.141) (0.297) (0.108) (0.092)
Days between -9
◦
C and -6
◦
C -0.261** 4.743*** 1.251*** 0.266 0.154 0.087 0.114
(0.128) (1.245) (0.240) (0.264) (0.563) (0.204) (0.178)
Days between -6
◦
C and -3
◦
C -0.001 1.313 -0.720*** -0.320* 0.116 -0.003 -0.143
(0.089) (0.922) (0.181) (0.185) (0.377) (0.128) (0.113)
Days between 27
◦
C and 30
◦
C -0.016 0.697** 0.364*** 0.140** -0.002 0.055* 0.066***
(0.021) (0.332) (0.076) (0.055) (0.081) (0.029) (0.025)
Days between 30
◦
C and 33
◦
C -0.021 2.847*** 0.451*** 0.215*** 0.124 0.029 -0.020
(0.025) (0.470) (0.110) (0.069) (0.099) (0.036) (0.031)
Days above 33
◦
C -0.130*** 2.832*** 0.747*** 0.154 -0.091 0.028 0.070**
(0.033) (0.702) (0.200) (0.096) (0.108) (0.037) (0.033)
Precipitation -33.585 637.916 95.391 -104.090 -24.907 74.216 19.083
(44.244) (665.711) (141.651) (114.752) (178.037) (63.625) (55.048)
Observations 498,199 654,366 654,366 161,322 161,194 161,322 161,322
R-squared 0.119 0.313 0.330 0.158 0.190 0.115 0.101
Year-month FE X X X X X X X
◦
County*Year FE X X X X X X X
Notes: The estimation results for Equation (1) are presented for seven different outcome variables based on alternative subprime credit. The
independent variables are the number of days in a month with daytime mean temperature within a specific range. The “Days between -3
◦
C and
27
◦
C” bin is the omitted category. The coefficient β
k
is interpreted as the estimated impact of one additional day with daytime mean temperature
within each respective temperature bin, relative to the impact of a day with daytime mean temperature between -3
◦
C and 27
◦
C. Standard errors
clustered at the ZCTA level are reported in parentheses. ***, **, and * represent statistical significance at the 1%, 5%, and 10% levels, respectively.
18