Extreme temperatures and low-income household finance: evidence from payday loans

Abstract

Extreme temperature shocks, intensified by climate change, are becoming increasingly frequent and severe, yet their effects on household credit conditions remain underexplored. Using loan-level payday lending data, we show that extreme temperature shocks increase demand for high-cost credit, reduce payday loan credit availability, and worsen loan performance through higher delinquency and default rates. We explore potential mechanisms using borrower income information and differences between online and storefront lenders. Our findings reveal how extreme temperature events can amplify financial vulnerability among credit-constrained households, particularly in the absence of formal disaster assistance for such events.

This study finds that extreme heat increases demand for payday loans, while reducing access to credit and raising default rates. This suggests that heat shocks can deepen financial strain among low-income households.

Introduction

Extreme temperature shocks, intensified by climate change, have become increasingly frequent and severe^1,2. In a 2024 survey, seven in ten Americans reported having been personally affected by extremely hot weather during the preceding five years, making it the most commonly experienced type of extreme weather event³. Yet, despite their prevalence, extreme temperature events have not been federally declared natural disasters and thus remain ineligible for federal emergency disaster assistance. Lower-income households are particularly vulnerable to potential weather-related financial challenges, such as lost income and rising energy and healthcare expenses. So far, little is known about how extreme heat affects household finances.

Understanding the impact of extreme temperatures on household financial behavior is crucial for designing effective policy interventions. This involves examining how borrowers respond to unexpected expenses by changing their demand for credit, and how lenders adjust credit supply and loan pricing in response. Examining these behavioral responses sheds light on the broader economic and social consequences of extreme temperatures, particularly for vulnerable populations. In this study, we focus on the payday loan market, a sector that reveals both the immediate financial needs of low- and middle-income households and the constraints they face in accessing traditional credit during periods of distress.

We combine spatially and temporally granular inquiry- and loan-level payday loan data from a nationwide consumer reporting agency specializing in alternative financial services with satellite-derived gridded weather data. We then implement a fixed-effects temperature-bin approach⁴ to estimate how extreme heat and cold affect credit demand, credit supply, and loan performance. This approach allows us to flexibly identify the nonlinear impact of extreme temperature days on payday loan market outcomes, while accounting for a rich set of confounding factors.

The results reveal that extreme heat is associated with increased payday loan demand. A 1-standard-deviation increase in the number of extreme heat days per month, defined as days with daytime mean temperatures above 33 °C, is associated with approximately a 0.4 percent increase in total inquiries relative to the baseline. Consistent with the income channel, we find that extreme heat days are associated with an application pool with lower average incomes. Furthermore, the results persist after controlling for borrower income, indicating the potential role of additional channels, such as rising energy or health costs. In contrast, extreme cold exposures are associated with fewer payday loan inquiries, consistent with institutional features such as winter utility shutoff protections that alleviate short-run liquidity constraints⁵.

We also show that extreme heat days are associated with reduced credit supply and a deterioration in the performance of existing payday loans, marked by higher default rates. Despite the increase in demand, lenders may tighten credit supply during extreme heat days due to concerns about borrower income and heightened risks of default and delinquency. Consistent with this, we find that a 1-standard-deviation increase in the number of extreme heat days per month is associated with about a 3 percent increase in default rates relative to the baseline. These effects are concentrated in the online payday loan market. Online inquiries, delinquency rates, and default rates increase with extreme heat, while accounts opened and credit issued decline. Storefront payday loan outcomes are less responsive to extreme temperatures.

Overall, our findings suggest that extreme heat shocks contribute to household financial distress in the payday loan market, driven by heightened demand, reduced credit access, and rising default rates. This reveals a deterioration in credit conditions in alternative credit markets that has not been previously documented. Previous work has identified modest effects of natural disasters on credit card borrowing⁶, reductions in student loans⁷, and increased lender screening and mortgage arrears⁸. These studies focus on federally declared natural disasters. Notably, Ge et al.⁹ studies household adaptation to rising insurance premiums due to climate risk. In contrast, we examine the effects of an overlooked yet increasingly pertinent source of shock: extreme temperature shocks. This distinction is important because previous research has emphasized the effectiveness of federal aid in mitigating post-disaster financial distress¹⁰. However, no study has focused on the impacts of extreme temperatures, which typically do not receive Federal Emergency Management Agency (FEMA) disaster declarations. Most existing studies also examine traditional credit products^11,12. By focusing on alternative credit products, a short-term source of liquidity used by lower-income households, our study highlights the financial vulnerabilities of households with limited access to mainstream credit.

Payday loans are a controversial source of liquidity for low- to middle-income customers, with concerns over higher interest rates and potential persistent debt cycles^13–15. Previous research finds mixed results on the effects of payday loans on consumers^16–25. Focusing on a smaller form of temporary credit access, Lodermeier⁵ shows that winter utility shutoff protections reduce eviction filings among low-income households. We advance this literature by directly assessing how extreme temperatures affect the demand for, supply of, and performance of payday loans. Our loan-level data allow us to distinguish between online and storefront loans. Our paper also contributes to the literature examining the diverse impacts of extreme temperature exposures in the United States^26–31. A smaller body of literature emphasizes the critical role of credit access in facilitating climate adaptation^32–34. Our findings indicate that extreme heat days in the United States simultaneously drive increased demand and reduced supply for high-interest alternative credit products. In the absence of federal aid for extreme temperature events, these findings highlight households’ unmet credit demand, which may hinder climate change adaptation.

Results

Descriptive statistics

Payday loans are a short-term source of liquidity used by low- to middle-income individuals. Typically, the lender advances the borrower $100–$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–$320 per $100 principal balance, equivalent to an annual percentage rate (APR) of 400–600%. About 3% of respondents in the Survey of Consumer Finances indicated that they had indeed taken payday loans. As shown in Supplementary Fig. 1, the top reasons for choosing payday loans include emergency needs, convenience, and paying other bills or loans. A substantial fraction of respondents take payday loans to pay medical bills or utilities. Further details are provided in Supplementary Note 1.

We measure payday loan activity at the ZIP Code Tabulation Area (ZCTA) level using three sets of outcomes. First, we construct inquiry-based measures: total inquiries, defined as the monthly count of all inquiries, including multiple inquiries by the same individual, and unique inquiries, defined as the monthly count of distinct individuals making inquiries. We also observe self-reported income when applicants make inquiries. The inquiry sample covers the period of 2012–2019. Second, to evaluate loan performance, we construct delinquency and default rates. A loan is classified as delinquent if it has a non-missing delinquency date and as in default if the amount past due is greater than zero. Third, we measure realized borrowing using new accounts opened and total credit extended. The tradeline sample covers the period of 2013–2019.

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. 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 new accounts are opened in one ZCTA each month. The median loan amount is $475, with $500 from storefront lenders and $400 from online lenders, respectively. On average, the delinquency rate is 8.7 percent and the default rate is 7.3 percent.

Table 1.

Summary statistics of the payday loan dataset

	Full sample
	Mean	sd	p25	p50	p75	N
new accounts	1.73	1.38	1	1	2	115,925
total amounts of credit issued	637.10	703.90	255	475	765	115,925
delinquency rate	8.69	26.51	0	0	0	115,925
default rate	7.33	24.57	0	0	0	115,925
inquiries made	4.66	5.43	1	3	6	476,928
unique inquiries	1.76	1.37	1	1	2	476,928
average monthly income	2661	1343	1742	2500	3261	456,435
	Storefront sample
	Mean	sd	p25	p50	p75	N
new accounts	1.94	1.62	1	1	2	30,853
total amounts of credit issued	882.90	1075.00	300	500	1000	30,853
delinquency rate	6.38	23.00	0	0	0	30,853
default rate	4.42	19.44	0	0	0	30,853
inquiries made	2.05	2.22	1	1	2	54,497
unique inquiries	1.46	0.98	1	1	2	54,497
average monthly income	1758	1193	921	1494	2272	33,230
	Online sample
	Mean	sd	p25	p50	p75	N
new accounts	1.57	1.15	1	1	2	89,921
total amounts of credit issued	518.4	421.3	255	400	605	89,921
delinquency rate	9.62	28.1	0	0	0	89,921
default rate	8.42	26.46	0	0	0	89,921
inquiries made	4.64	5.36	1	3	6	454,759
unique inquiries	1.69	1.26	1	1	2	454,759
average monthly income	2706	1345	1751	2500	3333	443,623

This table provides the summary statistics for the full payday loan sample, the storefront payday loan subsample, and the online payday loan subsample aggregated at the ZCTA level. New accounts opened, inquiries made, and unique inquiries are monthly counts at the ZCTA level. Total amount of credit issued and average monthly income are measured in U.S. dollars. Delinquency and default rates are reported in percentage points.

Our key independent variable is a vector of temperature and precipitation measures. Data on temperature and precipitation outcomes come from ERA5-Land, which includes atmospheric variables with enhanced spatial resolution based on the ERA5 climate reanalysis data, provided by the European Center 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. In terms of spatial granularity, our payday sample covers 15,373 ERA5 grid points and 20,655 ZCTAs. Allowing for nonlinear effects, we adopt the fixed-effect temperature bin estimation strategy, where we calculate the number of days in a month in which the daytime (between 8 a.m. and 8 p.m. local time) average temperature falls within each bin. In our baseline specification, we first construct 14 temperature bins: below –3 °C, above 33 °C, and twelve bins for every 3 °C in between. We include monthly average precipitation data as controls, where daily precipitation is measured as the depth in meters of cumulative precipitation in a day.

Figure 1 highlights that daytime mean temperatures below –3 °C or above 33 °C represent extreme temperatures in our sample. The bars represent the average annual distribution of daytime mean temperatures across 14 temperature bins over the 2012–2019 period, measured as the fraction of days in each bin (in percent) and averaged across ZCTAs. For reference, Supplementary Fig. 3 presents the same distribution in terms of the average number of days per year. Using all ZCTAs in the continental U.S., the average ZCTA experiences about 24.4 days annually with daytime temperatures below –3 °C (6.7 percent) and 4.3 days above 33 °C (1.2 percent). The standard deviation of extreme heat days where daytime temperatures exceed 33 °C is around 2 days. Within ZCTA-months in the payday sample, the observed temperature distribution corresponds to 159 days annually; the average ZCTA experiences about 5.4 days per year with daytime temperatures below –3 °C (3.4 percent) and 3.2 days above 33 °C (2 percent).

Fig. 1. Distribution of daytime mean temperatures, 2012–2019.

The figure shows the historical average distribution of daytime mean temperatures across 14 temperature-day bins. Each bar represents the average fraction of days in each temperature bin over the period 2012--2019, expressed in percent. The “All ZCTAs” dataset represents ZCTAs of the continental U.S., with the fractions summing to 100 percent across all bins. The “Payday sample” covers a subset of all ZCTAs, since not every ZCTA reports payday loans every month. In this sample, the fractions across bins also sum to 100 percent, corresponding to 159 observed days. Blue bars represent all ZIP Code Tabulation Areas (ZCTAs) in the continental United States, and magenta bars represent ZCTAs in the payday-loan sample.

Our primary goal is to identify the effects of temperature exposures on the payday loan market at the ZCTA level. We adopt a nonlinear temperature bin approach using the form specified in Eq. (1). Specifically, we regress ZCTA-level payday loan outcomes on a vector of weather variables measuring the number of days within different temperature bins in a given month. We choose 3 °C–27 °C as the omitted baseline daytime (between 8 a.m. and 8 p.m.) mean temperature bins. The estimated coefficients therefore measure the effects of additional days in a given temperature bin relative to days in the omitted 3 °C–27 °C baseline bin. The specification controls for precipitation, year-month, county-year, and state-quarter fixed effects, as well as additional covariates such as lagged outcomes and local income.

Throughout this paper, we adopt the term “temperature shocks”, as commonly used in the literature⁴. This approach captures the nonlinear effects of temperature variations, defined as deviations from the mean within our panel fixed-effects framework. Year-month fixed effects absorb aggregate trends and national business cycle conditions that may coincide with a general temperature warming trend. County-year fixed effects account for local economic conditions that may simultaneously affect temperature exposures and payday loan market outcomes. In addition, state-quarter fixed effects flexibly control for regional seasonality. Our identifying variations are therefore interpreted as temperature shocks, exploiting ZCTA-month level temperature deviations from the county-year, state-quarter, year-month averages.

Do temperature exposures drive demand for payday loans?

We first examine whether temperature exposures affect borrowers’ demand. We use two measures of payday loan demand, total inquiries and unique inquiries, as outcome variables in the regression. Table 2 and Fig. 2 provide the estimated impact of an extra day in each of the six extreme daytime temperature bins on the number of total inquiries, relative to a day with daytime mean temperature falling in the 3–27 °C bins. We find that total inquiries significantly increase with more extreme heat, when daytime mean temperature goes above 30 °C. In particular, one extra day with daytime mean temperature above 33 °C per month is associated with a 0.009 increase in total inquiries. This translates to a 1-standard-deviation increase in the number of extreme heat days per month, resulting in approximately a 0.4 percent increase in total inquiries relative to the baseline.

Table 2.

Impact of extreme temperatures on payday loan markets

	(1)	(2)	(3)	(4)	(5)	(6)	(7)	(8)
	total inquiries  × 100	unique inquiries  × 100	delinquency rate	default rate	new accounts  × 100	$\log \left(\mathrm{credit}\right)\times 100$	APR	$\log \left(\mathrm{income}\right)\times 100$
T  < –3°C	–0.786**	–0.324***	–0.080	–0.073	–0.048	0.183	–0.006*	0.000
	(0.312)	(0.089)	(0.063)	(0.065)	(0.159)	(0.150)	(0.003)	(0.043)
–3°C  < T  < 0°C	–0.688**	–0.177**	0.092	0.024	0.146	0.303	–0.007	–0.090
	(0.282)	(0.076)	(0.089)	(0.084)	(0.149)	(0.187)	(0.005)	(0.075)
0°C  < T  < 3°C	–0.640***	–0.276***	0.005	0.057	–0.024	0.111	0.001	–0.007
	(0.217)	(0.067)	(0.065)	(0.062)	(0.105)	(0.128)	(0.004)	(0.057)
27°C  < T  < 30°C	0.151	0.179***	–0.048*	–0.022	–0.117*	–0.089	–0.000	–0.038
	(0.197)	(0.053)	(0.025)	(0.023)	(0.062)	(0.075)	(0.002)	(0.028)
30°C  < T  < 33°C	1.061***	0.351***	0.033	0.042	0.082	0.095	–0.001	–0.082**
	(0.220)	(0.057)	(0.034)	(0.030)	(0.078)	(0.073)	(0.002)	(0.037)
T  > 33°C	0.940**	0.324***	0.077*	0.096**	-0.179*	-0.323***	–0.004	–0.148***
	(0.421)	(0.121)	(0.040)	(0.038)	(0.100)	(0.105)	(0.003)	(0.045)
Precipitation	251.457	76.568	166.224***	134.873**	40.137	-35.434	7.407*	–33.196
	(260.785)	(78.322)	(59.267)	(57.574)	(128.870)	(133.229)	(4.338)	(56.519)
Observations	1,982,880	1,982,880	110,244	110,244	485,810	109,422	90,748	439,133
R-squared	0.205	0.267	0.182	0.162	0.318	0.284	0.301	0.163
Year-month FE	✓	✓	✓	✓	✓	✓	✓	✓
County*Year FE	✓	✓	✓	✓	✓	✓	✓	✓
State*Quarter FE	✓	✓	✓	✓	✓	✓	✓	✓

The estimation results for Eq. (1) are presented for eight different outcome variables. The independent variables are the number of days in a month with daytime mean temperature within a specific range. The “3°C < T < 27°C” bin is the omitted category. The coefficient 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 two-way clustered at the ERA5 grid and year-month levels are reported in parentheses. Statistical significance is based on two-sided t tests. ***, **, and * represent statistical significance at p < 1%, p < 5%, and p < 10%, respectively.

Fig. 2. Impact on payday loan market outcomes.

The figure presents the estimated impact of one additional day in six daytime mean temperature bins, relative to a day in the 3–27 ^∘C bin, on payday loan market outcomes. Dependent variables considered are total inquiries (top left), unique inquiries (top right), delinquency rate (2nd row left), default rate (2nd row right), number of new accounts opened (3rd row left), log of total amount of credit issued (3rd row right), APR (bottom left), and log of income (bottom right). The sample sizes are n = 1,982,880 ZCTA-month observations for total inquiries and unique inquiries, n = 110,244 for delinquency rate and default rate, n = 485,810 for new accounts, n = 109,422 for total credit issued, n = 90,748 for APR, and n = 439,133 for income. Estimates are obtained using specification (1). The 95 percent confidence interval is based on standard errors two-way clustered at the ERA5 grid and year-month levels. Statistical significance is based on two-sided t-tests. Blue solid lines with circular markers represent point estimates, dark-blue dashed lines represent 95% confidence intervals, and black horizontal lines denote zero. CI denotes confidence interval; APR denotes annual percentage rate; ZCTA denotes ZIP Code Tabulation Area; and ERA5 denotes the fifth-generation atmospheric reanalysis produced by the European Center for Medium-Range Weather Forecasts.

Our findings on total inquiries may partly reflect an increase in borrowers making multiple inquiries on extreme temperature days, potentially driven by impulsive decisions, loan stacking practices, desperation, or loan denials. To disentangle the effects of such behavior, we analyze the impact of temperature exposures on unique inquiries, which represent the number of applicants instead of total applications. The estimation results are shown in column 2 of Table 2 and in the right panel in the first row of Fig. 2. Consistent with the findings on total inquiries, we observe similar effects. The number of unique inquiries increases as the number of days with a daytime mean temperature exceeding 30 °C increases. Specifically, the figure shows that one extra day above 33 °C is associated with a 0.003 increase in unique inquiries. Taken together, these results indicate that extreme heat exposures lead to higher demand for payday loans.

Extreme temperature days could affect the payday loan market through various potential channels, including changes in labor income^26,35,36, increased energy and healthcare costs^37–39, and disruptions to transportation⁴⁰. The increase in payday loan demand could be explained by either income declines or other channels. In Section 2.5, we offer a more detailed discussion on how extreme temperatures affect the applicant income distribution. Before that, we first explore the role of the income channel by comparing specifications with and without income controls, using comparable samples to address the fact that income is only observed at inquiry. Columns 1 and 2 in Supplementary Table 4 present results without controlling for income, while Supplementary Table 3 includes income as a control. We find that for the most extreme heat bin (T>33°C), the estimated effects become insignificant when income is included, while estimates for less extreme bins (30 °C <T< 33 °C) remain robust and significant. This pattern suggests that the income channel explains part, but not all, of the observed effects. To the extent that self-reported income is measured with error and may be overstated, these estimates likely understate the contribution of the income channel. Complementary evidence from variation in eligibility for the Low Income Home Energy Assistance Program (LIHEAP) is consistent with energy assistance alleviating liquidity constraints during extreme heat. Further details on LIHEAP are provided in Supplementary Fig. 4, Supplementary Table 2, and Supplementary Note 3. Table 2 and Fig. 2 show that extreme cold exposures reduce both total and unique inquiries when daytime mean temperatures fall below 0 °C. This effect becomes insignificant for total inquiries once lagged outcomes are included, but remains robust for unique inquiries, as shown in Columns 3 and 4 of Supplementary Table 4. In contrast, the effects of extreme heat on payday loan inquiries remain statistically significant when lagged outcomes are included.

These asymmetries are consistent with differences in underlying mechanisms. Extreme heat appears to generate more immediate financial distress, through increased worker injuries³⁸, reduced work hours, lower productivity, and declines in labor income^26,41,42, leading to higher payday loan demand. In contrast, one plausible explanation for the patterns observed during extreme cold is that individuals may delay non-urgent search and application decisions, reflecting reduced economic activity and lower short-run borrowing demand.

Institutional features specific to cold weather may further contribute to this pattern. Winter utility shutoff protections can relieve short-run liquidity pressure by allowing households to delay utility payments. For example, Lodermeier⁵ shows that such protections effectively act as a temporary line of credit, enabling households to reallocate resources toward rent and smooth consumption, which is consistent with the observed reduction in borrowing demand during extreme cold periods.

Do temperature exposures affect the performance of existing accounts?

After demonstrating the rise in payday loan demand on the extensive margin, we now turn to analyzing delinquency and default rates of existing loans in response to extreme temperature shocks. Columns 3 and 4 of Table 2 and the second row of Fig. 2 show that one extra day with a temperature above 33 °C is linked to a marginally significant 0.08 percentage-point increase in the delinquency rate (left panel) and a significant 0.1 percentage-point increase in the default rate (right panel). In other words, a 1-standard-deviation increase in the number of extreme heat days per month leads to about a 3 percent increase in default rates relative to the baseline. Our analysis also indicates that the effects of temperature exposure on the default and delinquency rates are asymmetric between extreme cold and hot days. Specifically, the performance of existing accounts deteriorates, with default and delinquency rates increasing on hot days but not during cold days.

Do temperature exposures affect credit supply?

Finally, we provide evidence on how extreme temperatures affect payday loan supply. If demand for payday loans increases but the total credit issued decreases, it suggests a reduction in supply from lenders. While we directly observe payday loan demand, we infer the behavior of credit providers by analyzing equilibrium outcomes in payday loan markets. Columns 5 and 6 of Table 2 and the third row of Fig. 2 present results using the total number of new accounts opened (left panel) and (log) total amount of credit issued (right panel) as outcome variables in Eq. (1). The number of new accounts declines, with marginal statistical significance, as the number of days with daytime mean temperatures exceeding 33 °C increases. In magnitude, one additional day above 33 °C is associated with a 0.002 decrease in the total number of new accounts for the average ZCTA. Similarly, we find a statistically significant negative relationship between extreme heat and total credit issued: one additional day above 33 °C is associated with a 0.32 percent decline in credit issued. In contrast, in specifications that control for average income, log credit increases with extreme cold exposure.

The total number of new accounts opened and the total amount of credit issued are jointly determined by credit supply and credit demand. While payday loan inquiries rise significantly with more extreme heat days, equilibrium outcomes suggest a contraction in credit supply during these periods. Credit supply contracts for two possible reasons. First, payday lenders screen out very low-income applicants. We therefore expect credit supply to fall as borrowers’ incomes decline during extreme heat days. Second, increases in default or delinquency rates may induce payday lenders to reduce credit supply. Overall, our findings suggest that credit supply responds asymmetrically to extreme temperatures due to differing impacts on applicant income, as well as default and delinquency rates.

In terms of the cost of credit, we do not directly observe the APR on payday loans from the data. Instead, we follow the approach in Correia et al.⁴³ and infer APR from the observed loan duration, the maximum credit amount, and the total repayment:

$$\mathrm{APR}=\frac{365}{\mathrm{LoanMaturity}}\times \frac{\mathrm{Repayment}-\mathrm{LoanAmount}}{\mathrm{LoanAmount}}.$$

The bottom left panel of Fig. 2 indicates that the APR appears unaffected by extreme temperature days. Supplementary Fig. 5 illustrates the credit-rationing mechanism that could account for a contraction in credit supply without an increase in APR, and Supplementary Note 4 discusses this interpretation in greater detail.

Do temperatures affect the applicant income distribution?

Before presenting the results, it is important to clarify that our income measure is observed only for individuals who submit payday-loan inquiries. As a result, changes in average applicant income may reflect either changes in the incomes of existing applicants or changes in the composition of applicants entering the sample. Accordingly, the estimates should be interpreted as capturing changes in the income distribution of applicants rather than necessarily changes in individuals’ incomes.

Borrower income plays an important role in both the supply and demand for payday loan lending during extreme temperature days. If extreme temperatures negatively affect borrower income, we could observe an increase in demand for payday loans. Meanwhile, payday loan lenders may screen out borrowers from lower-income backgrounds, resulting in a contraction in payday loan supply. From our dataset, we observe the self-reported income of the applicants from their inquiries and compute the monthly average income among these applicants at the ZCTA level. We regress the log of this monthly ZCTA average income on the weather exposure measures and other control variables as specified in Eq. (1). Column 8 of Table 2 and the bottom right panel of Fig. 2 indicate that while extreme heat shocks increase the number of applications, the resulting pool has a lower average income. In particular, one extra day with daytime mean temperature above 33 °C is associated with a 0.15 percent decrease in monthly income averaged across applications at the ZCTA level. In contrast, we find no significant evidence that extreme cold days impact the mean reported incomes of applicants.

Given the structure of our data, the observed effects on income may stem either from the direct treatment effects of temperature shocks or from composition effects driven by changes in the types of applicants. For example, in many regions, construction activity is concentrated during the summer months due to favorable weather, longer daylight hours, and optimal conditions for materials like concrete and asphalt. Specifically, Graff Zivin and Neidell²⁶ show that labor hours decrease on days with extreme heat, whereas they are less sensitive to lower temperatures. Extreme heat can also lead to employment adjustments, including layoffs in climate-exposed sectors^44,45, which may further contribute to declines in household income. Additionally, the observed income effects at the ZCTA level may reflect a compositional effect, with an increase in payday loan applications among lower-income households.

We further leverage the panel structure of our data by constructing a ZCTA-by-year-month panel in which individual income is carried forward across periods without new inquiries. This approach allows us to hold the sample of individuals fixed after entry and examine variation in average income at the local level. As shown in Supplementary Fig. 6, one additional extreme heat day leads to a statistically significant decline in income. However, the magnitude of this decline is smaller than in the baseline specification that uses only applicants’ income. This attenuation suggests that the observed income decline reflects a combination of both channels: a direct effect of temperature on income and a compositional shift toward lower-income applicants.

Robustness checks

Our main results remain robust across different model specifications.

First, our main results are based on the balanced payday loan dataset. For each ZCTA, we fill missing ZCTA-month observations for total inquiries and unique inquiries with zeros between the first and last months in which the ZCTA appears in the sample. Columns (1) and (2) of Supplementary Table 4 present results without this balancing procedure. The estimates remain robust.

Second, we use alternative methods to define temperature bins. Supplementary Table 5 presents the results in which we define the temperature bins using the 24-h daily mean temperature: below –6 °C, above 30 °C, and 3 °C-wide bins in between. Again, the main results are similar. An additional day in a month with a 24-h mean temperature above 30 °C is associated with higher demand, reduced credit issuance, and increased default rates.

Third, households may smooth consumption in the short run and only adjust borrowing with a delay following income shocks induced by extreme temperatures. To account for this possibility, we augment our baseline specification by including temperature bins averaged over the previous three months, alongside contemporaneous temperature measures. Supplementary Table 6 reports the results. We find that our main results remain robust. In addition, lagged extreme heat exposure has economically meaningful effects, including reductions in account openings, total credit, and the incomes reported in inquiries. These findings suggest that the impact of extreme temperature shocks persists beyond the contemporaneous period and is consistent with delayed adjustments in borrowing behavior.

Finally, Supplementary Table 7 reports results where we use a much narrower omitted category (18–21 °C). The comparisons are made relative to a more localized temperature range. Our results remain robust.

Discussion

The storefront and online payday loan markets have distinct institutional features that may produce heterogeneous responses to temperature exposure. Most existing studies of payday lending rely on geographic variation in state regulations governing payday loan access. In our setting, this approach could raise concerns because extreme temperatures may be geographically correlated with restrictive or prohibitive state lending laws. Our loan-level data allow us to distinguish between storefront and online payday loans and directly examine how extreme temperatures relate to credit demand, supply, and loan performance in each market. Because online lending is less constrained by state-level prohibitions and allows borrowers to apply regardless of their location, the online results help alleviate concerns that the estimated effects reflect geographic sorting in payday-lending regulations. The distinction is also economically relevant: storefront loans require in-person applications, whereas online lenders acquire borrowers through lead generators, advertisements, landing pages, and the sale or resale of consumer information. To explore the heterogeneity between the two markets, we repeat our earlier analysis separately for storefront (Supplementary Fig. 7 and Supplementary Table 8) and online (Supplementary Fig. 8 and Supplementary Table 9) payday loan markets.

For online payday loans, we observe significant increases in unique inquiries, delinquency, and default rates, with marginally significant decreases in accounts opened and credit issued under more extreme heat days. Online payday loan inquiries decrease when there are more extreme cold days. Compared to online borrowing, storefront loan borrowers’ inquiries and account openings are less responsive to extreme temperatures. We also find that storefront delinquency and default rates respond less strongly to extreme temperatures than those for online loans.

Overall, 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, can make both the supply and demand sides more sensitive to extreme temperature shocks. This pattern is also consistent with limited substitution toward other credit products among payday borrowers, who typically have constrained access to mainstream credit¹⁹.

Furthermore, our findings raise the possibility that payday loans may contribute to cycles of repeated borrowing. To explore this issue, we use exposure to extreme heat as an instrumental variable for payday borrowing and examine whether current borrowing affects subsequent payday loan outcomes. The first-stage results indicate that extreme heat is a strong predictor of payday credit issuance. The IV estimates suggest that a 1% increase in current-month payday borrowing is associated with a marginally significant 3.1% increase in total payday borrowing over the following three months, providing suggestive evidence of persistent borrowing behavior. We do not find significant effects on subsequent delinquency or default rates. These results are consistent with the existence of debt cycles in payday lending, although alternative channels, including the lagged effects of heat shocks on household economic conditions, cannot be fully ruled out. Full details of the empirical specification and results are reported in Supplementary Note 5 and Supplementary Table 10.

Methods

Data sources

Our analysis combines two datasets from a major credit bureau—inquiry-level and loan-level payday loan data—obtained from a data provider specializing in alternative financial services data, with satellite-derived gridded weather data.

Our credit data come from the Gies Consumer and Small Business Credit Panel (GCCP), which links mainstream credit data with alternative credit data. The alternative credit data provider covers about 70% of non-prime consumers in the U.S.⁴⁶. Existing studies have consistently considered the alternative credit data provider as the best available source for understanding payday borrowing behavior. Fonseca⁴⁷ compare the characteristics of payday loans in GCCP with those of other studies with individual- or loan-level data on these products (see Supplementary Note 2 for more details about GCCP).

We use two samples of alternative credit data. The first is an inquiry-level dataset consisting of consumers from 2012 to 2019. Inquiries are requests made by prospective borrowers to prospective lenders. Each applicant comes with a unique applicant ID. For each inquiry, we observe the date of the application and self-reported information obtained during the application process, including the state and ZIP code of residence, income, pay frequency, age, homeowner indicator, months of residence in the current address, and whether the application takes place online or at a storefront. Though alternative credit 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 verified⁴³. The second is a loan-level dataset known as the tradeline sample, which consists of approved loans from 2013 to 2019. Each loan corresponds to a newly opened account, and we observe loan characteristics including origination, maturation, delinquency dates, loan size, and amount past due. The applicant/borrower IDs from two datasets are one-to-one linked. However, we do not have a link between the application and the approved loan at the record level. Supplementary Fig. 2 shows the geographical distribution of online and storefront borrowers across the U.S. by state.

Of the 621K borrowers who submitted an inquiry for any type of subprime credit, 223K borrowers inquired about payday loans, while the remaining borrowers inquired about other products such as installment loans. We focus on payday lending in this study. We observe 840K payday loan inquiries and 200K payday loan accounts. We convert ZIP codes to ZIP Code Tabulation Area (ZCTA) using crosswalks from the U.S. Census. A ZCTA is a geographic unit created by the U.S. Census Bureau to approximate a 5-digit ZIP Code service area for census data analysis. ZCTAs are constructed from census blocks.

Our payday loan dataset is an unbalanced ZCTA-month panel because ZCTAs are observed only in months with reported payday loan activity. This feature could introduce bias if missing ZCTA-month observations are systematically related to local conditions or weather exposure. To address this concern, we construct a balanced panel. Specifically, for each ZCTA, we fill in missing ZCTA-month observations with zeros between the first and last months in which that ZCTA appears in the sample. This approach preserves intermittently observed ZCTAs and mitigates concerns that the results are driven by changes in panel composition. In the balanced-panel specification, inquiry outcomes are set to zero in filled ZCTA-month cells with no observed payday loan activity. For new accounts opened, we set the variable to zero when inquiries are positive, but no account is opened, and to missing when total inquiries are zero.

Supplementary Table 1 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 99th and 1st percentiles of daytime mean temperatures are 34 °C and −9 °C, respectively. For each ZCTA, we average the monthly number of days that had daytime mean temperatures below ‒3 °C and above 33 °C. Extreme heat days occur in southern ZCTAs along the Sun Belt 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 is concentrated 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. From 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.

Statistical analyses

Our primary goal is to identify the effects of temperature exposures on the payday loan market at the ZCTA level. We adopt a nonlinear temperature bin approach using the following form:

$${\mathrm{outcome}}_{it}={\mathit{\theta }}^{\mathsf{T}}{\mathbf{T}}_{it}+{\mu }_t+{\varphi }_{sq}+{\eta }_{cy}+{\mathrm{controls}}_{it}+{\epsilon }_{it},$$ 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 weather variables that we discuss below; μ_t is a year-month fixed effect; ϕ_sq is a state-quarter fixed effect; η_cy is a county-year fixed effect; and ε_it is the error term. The control variables, controls_it, always include precipitation. Depending on the specification, they may also include lagged values of the outcome variables and average income. Because weather variables are assigned at the ERA5 grid level, multiple ZCTAs may share the same underlying temperature variation. To account for this, we cluster standard errors two-way, at the ERA5 grid level and year-month level, following Cameron et al.⁴⁸.

The main independent variables, T_it, include location- and time-specific temperature and precipitation measures. We use fourteen temperature bins: below –3 °C, above 33 °C, and twelve bins for every 3 °C in between. We choose 3 °C–27 °C as the omitted baseline daytime (between 8 a.m. and 8 p.m.) 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 –3 °C as extreme cold days and days above 33 °C as extreme heat days.

Our adoption of the nonlinear temperature bin specification follows the climate-economy literature⁴, motivated by facts related to thermal stress. High temperatures beyond certain thresholds cause worker fatigue and lower task performance. For example, Hancock et al.⁴⁹ is a meta-analysis and documents that task performance losses start to increase in a nonlinear manner at high temperature ranges. Other studies also indicate that labor productivity drops sharply as temperature increases. These include, but are not limited to, evidence from assembly lines, laboratories, meta-analysis, and self-reported surveys^26,35,41,50.

Reporting summary

Further information on research design is available in the Nature Portfolio Reporting Summary linked to this article.

Author contributions

All authors contributed equally to this work, and their names are listed alphabetically. All authors (V.X., S.X. and X.Z.) contributed to the conceptualization of the study, methodology design, writing, reviewing, and editing of the manuscript. V.X. and S.X. secured funding for the study. V.X. cleaned and processed the temperature data, while S.X. cleaned and processed the payday-loan data. S.X. performed and supervised the formal analysis. All authors contributed to the interpretation and discussion of the findings.

Peer review

Peer review information

Nature Communications thanks Daniel J. and other anonymous reviewers for their contribution to the peer review of this work. A peer review file is available.

Funding

The authors gratefully acknowledge financial support from the Alfred P. Sloan Foundation through the NBER Household Finance Small Grant Program, as well as support from the Sustainable Business Institute at Santa Clara University.

Data availability

All raw data used in this study are publicly available except the payday loan data, which we accessed through the Gies Consumer and Small Business Credit Panel at https://dsrs.illinois.edu/ and are subject to the Panel’s data-use restrictions. ERA5-Land data and codebook are obtained from https://www.ecmwf.int/en/era5-land. ZCTA shapefiles are obtained from the U.S. Census Bureau website https://www.census.gov/programs-surveys/geography/guidance/geo-areas/zctas.html. Data and codebook for the Survey of Consumer Finances are obtained from https://www.federalreserve.gov/econres/scfindex.htm. State median household income and poverty line used in Supplementary Note 3 are obtained from the Office of Community Services website https://acf.gov/ocs/policy-guidance/liheap-information-memoranda.

Code availability

Stata code used for analyses and to create the tables and figures is available in a Zenodo repository at https://doi.org/10.5281/zenodo.21498377. The code was run using Stata 19.5.

Competing interests

The authors declare no competing interests.

Supplementary information

The online version contains supplementary material available at https://doi.org/10.1038/s41467-026-76945-y.