The effect of opioid abuse on credit union performance and risk Open Access

Opioid abuse in the U.S. has reached an unprecedented level. According to the Centers for Disease Control and Prevention (CDC), by 2016, 11.4 million Americans, 3.5% of the population, had misused opioids. Opioid abuse is currently the leading cause of injury-related deaths in the country. In 2017, the number of opioid overdose deaths was six times higher than in 2009 ^[1]. In 2017, the White House declared a nationwide public health emergency to raise awareness of the opioid epidemic. Moreover, the opioid crisis intensified during the COVID-19 pandemic with the CDC recording over 81,000 opioid overdose fatalities in the 12 months ending in May 2020, the highest ever recorded in a single year ^[2]. Given the growing severity of opioid abuse in the U.S., various studies have examined its impact on the local economy, such as the labor (Harris et al., 2020; Cho et al., 2021; Langford and Feldman, 2021; Ouimet et al., 2021; Boubaker et al., 2023), real estate (Custodio et al., 2021; Ho and Jiang, 2021; D’Lima and Thibodeau, 2022), and municipal bond (Li and Zhu, 2019; Cornaggia et al., 2022) markets. The adverse shock of the opioid crisis on the local economy could ultimately affect credit unions, as they are closely tied to the economic health of their regions (McKillop et al., 2020). Understanding how opioid abuse impacts local credit unions — key players in facilitating local economic development — is of significant interest to regulators and entrepreneurs.

Banking literature has investigated the effect of opioid abuse on the performance and risk of banks. Agarwal et al. (2023) examine the impact of the opioid epidemic on consumer finance, finding that banks with significant reliance on depositors in areas severely affected by the epidemic experience notable deposit losses and respond with higher interest rates and lower loan supply. Li and Ye (2024) reinforce that the adverse impact of opioid abuse propagates through banking networks. Jansen (2023) highlights that the opioid epidemic increases defaults on subprime auto loans. While these studies focus on banks, our research explores another type of financial institution that plays a vital role in supporting local businesses — credit unions. Extending these banking studies, we posit that opioid abuse has a detrimental impact on credit union performance and risk.

Unlike banks, which are for-profit institutions owned by shareholders, credit unions operate as not-for-profit entities owned by their members, allowing them to enjoy tax exemption. While tax benefits may help credit unions cushion negative shocks like the opioid epidemic, they remain vulnerable due to structural inefficiencies. Credit unions often lack economies of scale, have less diversified loan portfolios, and cannot access equity markets. In contrast, banks are relatively large — the top ten national banks combined are larger than the entire credit union sector, better diversified geographically, and more flexible in their products and services. Banks also face fewer client restrictions, further enhancing their resilience to economic shocks. These factors suggest that credit unions’ smaller size, limited diversification, and localized focus can make them more susceptible to economic disruptions.

The economic shock from the opioid crisis can have significant implications for credit unions, given their reliance on the local economy (McKillop et al., 2020). We hypothesize that credit unions in counties with higher opioid death rates are less profitable. Using opioid overdose death data from the CDC, which covers over 3,000 U.S. counties from 2003 to 2018, and credit union data from the SNL Bank Regulatory dataset provided by S&P Market Intelligence, we find that county-level opioid death rates — defined as the number of opioid overdose deaths per 100,000 population — negatively affect credit union profitability, as measured by return on assets (ROA) and return on equity (ROE). Specifically, a unit increase in opioid death rate leads to a decrease of 2.25 standard deviations in credit union ROA and 0.24 standard deviation in credit union ROE, respectively. These results remain qualitatively unchanged when credit union and county characteristics are controlled, year fixed effects are included, and standard errors are clustered at the credit union or county level. Our evidence implies that opioid abuse adversely affects credit unions, impeding their capabilities to facilitate local economic growth.

We address endogeneity using 2SLS regression, propensity score matching, and entropy balancing. In 2SLS, we use the average opioid death rate in surrounding counties and same-quartile credit unions as instruments, confirming the causal link between opioid deaths and credit union profitability. Propensity score matching ensures results are not driven by covariate imbalance, while entropy balancing further validates the findings. Robustness checks — including heteroskedasticity-robust and Fama-MacBeth regressions, as well as alternative sample constructions — support our results. Causal mediation analysis reveals that opioid abuse reduces profitability by slowing asset, loan, and deposit growth, lowering interest income, and increasing operating expenses and nonperforming loans, raising credit risk. Besides, opioid abuse effects are stronger for state-chartered, community, and industry credit unions.

This study contributes to the growing body of research on the financial implications of opioid abuse (Li and Zhu, 2019; Harris et al., 2020; Cho et al., 2021; Custodio et al., 2021; Ho and Jiang, 2021; Langford and Feldman, 2021; Ouimet et al., 2021; Cornaggia et al., 2022; D’Lima and Thibodeau, 2022; Boubaker et al., 2023; Agarwal et al., 2023; Jansen, 2023; Li and Ye, 2024). While prior research primarily studies the crisis impact on banks (Agarwal et al., 2023; Jansen, 2023; Li and Ye, 2024), the current study is, to our knowledge, the first to examine its effects on credit unions. This study extends the literature on credit unions (Smith, 1988; Li and Van Rijn, 2024) and other financial institutions (Demetriades and Law, 2006; Bolt et al., 2012; Zhang et al., 2018; McKillop et al., 2020) by identifying the negative impact of opioid abuse on credit unions.

The number of drug overdose deaths rose by nearly 30% from 2019 to 2020 and has quintupled since 1999. Nearly 75% of the 91,799 drug overdose deaths in 2020 involved opioids. During 1999–2020, more than 564,000 Americans died from prescription or illicit opioids. The U.S. opioid crisis can be outlined in three waves of opioid overdose deaths, according to the CDC ^[3]. The first wave began in the 1990s and is characterized by abuse of natural and semi-synthetic opioids, such as oxycodone and methadone. The second began in 2010 and witnessed a rapid rise in heroin addiction. The third began in 2013 and centered on abuse of synthetic opioids like fentanyl.

Multiple factors contributed to the rising incidences of opioid abuse in the first wave of the opioid crisis. First, the medical community advocated for aggressive pain treatment by claiming pain as “the fifth vital sign” ^[4]. However, the notion of opioid-based pain treatment was rarely backed by evidence on its long-term efficacy (Mularski et al., 2006). Second, the Food and Drug Administration (FDA) hastily approved new types of prescription opioids, such as Morphine Sulfate Contin in 1987. The lax regulation by the Drug Enforcement Agency (DEA) ^[5] was another culprit ^[6]. Opioid prescriptions increased gradually throughout the 1980s and early 1990s. In 1995, pharmaceutical firms such as Purdue Pharma introduced oxycodone, the so-called third sustained-release opioid ^[7]. It was quickly approved by the FDA in 1995 and aggressively marketed to physicians by medical sales representatives ^[8]. OxyContin, the brand name of oxycodone, became one of the most abused opioid-based pain relievers ^[9].

Heroin abuse is a major cause of the second wave of the opioid crisis, whereas the third wave is primarily propelled by synthetic opioids. Nearly 75% of the drug overdose deaths in 2021 involving heroin also involved synthetic opioids, such as fentanyl ^[10]. According to the CDC, more than 56,000 Americans died from synthetic opioids in 2020, 56% higher than in 2019 and 17 times higher than in 2013. More than 100,000 died from drug overdoses in the 12-month period ending in January 2022, and 67% of these cases involved synthetic opioids like fentanyl. Fentanyl is considered the deadliest drug case in the U.S. ^[11]. The latest drug overdose death counts through June 2021 suggest an exacerbation of synthetic opioid abuse during the pandemic ^[12].

Credit union is a member-owned financial institution with the mission of maximizing member benefits. In the U.S., credit unions are not-for-profit, tax-exempt (under section 501(c)(3) of the Internal Revenue Code) organizations established with the Federal Credit Union Act of 1934 ^[13]. Credit unions need a charter from either the National Credit Union Administration (NCUA), or a state credit union regulator. Federally chartered credit unions must have an NCUA-approved field of membership, which is the legal description of the persons, organizations, and other entities the credit union will serve ^[14]. However, some state-chartered credit unions are also regulated by NCUA due to the fact that the National Credit Union Share Insurance Fund (NCUSIF) may also cover those state-chartered credit unions ^[15]. According to the NCUA, as of December 2020, there are 5,099 federally insured credit unions, with assets totaling over $1.85 trillion and net loans near $1.16 trillion. In total, 51.6% of their assets are real estate loans, 32.7% are auto loans, and 5.3% are unsecured credit card loans.

To maintain a tax-exempt status, most credit unions are limited to providing common bond memberships and financial services to certain specific segments of the local community, such as educational institutions, labor unions, and religious groups ^[16]. These are referred to as occupational credit unions ^[17]. Examples include trade-union-specific credit unions, employer-specific credit unions, and university credit unions. The five largest U.S. credit unions are Navy Federal Credit Union, State Employees’ Credit Union, PenFed Credit Union, SchoolsFirst Federal Credit Union, and Boeing Employees Credit Union.

Another way to classify credit unions is by specific segments of borrower-saver populations. Some credit unions attempt to benefit savers, borrowers only, or both sides of the table. Conflict may emerge because saving members want as high a return on savings as possible, while borrowing members want as low a loan rate as possible. Neutral-oriented credit unions try to balance benefits between savers and borrowers, attract members who share the same goal, and therefore help to maintain the vitality of the institution (Smith, 1988).

Credit union plays an essential role in stabilizing the local economy, especially during dire times. During the 2008 financial crisis, compared with banks, credit unions engaged in approximately 80% less subprime lending and were nearly 70% less likely to fail (Li and Van Rijn, 2024). U.S. credit unions more than doubled their lending to small businesses from $30 billion in 2008 to $60 billion in 2016, whereas the total lending to small businesses during this period declined by almost $100 billion. Moreover, small businesses are 80% less likely to be dissatisfied with a credit union than a commercial bank ^[18]. Credit unions are also more likely to serve low-income and racially diverse area and provide retail and small business loans (Petersen and Rajan, 1994), which are vital for local economic growth and job creation ^[19].

The opioid crisis negatively impacts the local economy in three major ways, affecting credit unions that rely on local economic stability (McKillop et al., 2020). First, it weakens the labor market. Opioid abuse deteriorates workforce health, reducing labor supply and productivity (He et al., 2019; Harris et al., 2020; Langford and Feldman, 2021). Nearly half of working-age men not in the labor force regularly take pain medication (Krueger, 2017), with stronger effects among white men and certain minority groups (Aliprantis et al., 2023). Exposure to oxycodone increases unemployment, particularly among younger and less-educated individuals (Cho et al., 2021), while opioid prescriptions negatively affect long-term employment prospects (Ouimet et al., 2021). Firms respond to labor shortages by investing in automation, further reshaping local economies. Additionally, rising labor costs increase financial risk for local firms, prompting them to substitute share repurchases for dividends to maintain flexibility (Boubaker et al., 2023).

Second, the opioid crisis depresses the real estate market. Higher opioid prescription rates lead to declining property values, with a one standard deviation increase in opioid prescriptions reducing home prices by 1.36% over five years (Custodio et al., 2021). Reducing opioid prescriptions through assistance programs raises median home prices (Ho and Jiang, 2021), while properties near opioid dispensaries experience price declines, with the effect diminishing with distance (D’Lima and Thibodeau, 2022).

Third, it disrupts the municipal bond market by increasing credit risk for local governments. Opioid overdose deaths reduce tax revenues while raising expenditures on law enforcement and criminal justice (Cornaggia et al., 2022), leading to lower credit ratings, higher borrowing costs, and reduced bond issuances (Li and Zhu, 2019). These effects make it more expensive for municipalities to finance economic initiatives, though prescription drug monitoring programs help mitigate borrowing costs (Cornaggia et al., 2022).

These disruptions in labor markets, real estate, and municipal finance have direct consequences for credit unions. Lower labor force participation reduces household income and deposits (Bolt et al., 2012), declining property values increase nonperforming loans (Zhang et al., 2018), and higher municipal borrowing costs limit public investments that support local economies.

While the literature has examined banks extensively, credit unions possess certain unique characteristics distinguishing them from banks. At credit unions, depositors are referred to as members, and members own credit unions (Lawrence et al., 2024). Each member, regardless of her deposit amount, has one vote in board member elections. Members can also run for election to the board.

In terms of loan portfolio risk, credit unions generally issue loans that are relatively less risky (Li and Van Rijn, 2024). This feature makes it easier for credit unions to navigate recessions, as they are less exposed to widespread loan defaults. Furthermore, credit union members may face reduced financial hardship due to lower debt burdens (Berger, 2013). This structural resilience can reduce the likelihood of credit unions requiring government bailouts during crises, thereby lessening the burden on taxpayers. Li and Van Rijn (2024) attribute this advantage to the idea that credit unions “internalize the welfare of their customers”, who are simultaneously their owners.

Previous research has primarily focused on the economic consequences of opioid abuse on banks. Agarwal et al. (2023) examine the impact of opioid abuse on consumer finance, finding that banks with significant reliance on depositors in opioid-affected areas experience notable deposit losses, offer higher interest rates, and reduce loan supply. Li and Ye (2024) report similar findings. Jansen (2023) highlights that the opioid epidemic contributes to county-level defaults on subprime auto loans issued by banks and deteriorating credit scores.

Credit unions, being relatively smaller in size and more focused on consumer-oriented loans, are more likely to be impacted by external shocks, than are larger financial institutions ^[20]. Unlike banks, which typically operate under a corporate structure characterized by the separation of shareholders and customers (“duality”), credit unions operate under a “singularity” structure where owners are also customers. This distinction limits credit unions to serving their members and emphasizing local community involvement, potentially heightening their exposure to localized external shocks, such as opioid abuse.

Building on existing banking studies (Agarwal et al., 2023; Jansen, 2023; Li and Ye, 2024), we posit that health crises, such as the opioid epidemic, negatively impact the performance of credit unions. Formally, we hypothesize,

Research links opioid abuse to hyperbolic discounting — a cognitive bias favoring immediate gratification (Kirby et al., 1999). This impulsivity undermines long-term planning, compromising debt repayment capacity and elevating defaults. Opioid abuse also causes neurocognitive impairment, reducing financial decision-making abilities. Together, these factors increase delinquent loans, increasing credit risk for credit unions. Therefore, we further hypothesize,

We collect opioid death data from the National Center for Injury Prevention and Control (NCIPC) of CDC. This data covers the number of opioid overdose deaths for the 3,136 U.S. counties during 2003–2018 based on the Federal Information Processing Standards ^[21]. We use opioid death rate (OPIOID DEATH), defined as the number of opioid overdose deaths per 100,000 population, as a proxy for opioid abuse. The number of opioid overdose deaths are estimated using hierarchical Bayesian models with spatial and temporal random effects ^[22]. Unlike many studies on opioid abuse (e.g. Harris et al., 2020; Cho et al., 2021; Custodio et al., 2021; Ho and Jiang, 2021; Langford and Feldman, 2021; Ouimet et al., 2021), we focus on opioid death rate as opposed to opioid prescription rate ^[23]. The reason is that opioid death rate captures the impact of opioid abuse on the local economy more accurately. First, opioid prescription rate does not account for consumption of illicit opioids sold on the streets. For instance, a county can have a low opioid prescription rate but a high opioid death rate when there are a limited number of pharmacies, and illicit opioids are easily obtainable. Second, a high opioid death rate indicates a large amount of opioid abuse, whereas a high opioid prescription rate could reflect many illnesses or disabilities. For instance, a county with a high population of elderly residents can have a high opioid prescription rate due to regular medical needs. Third, opioid death rate is a local measure and thus easily comparable across counties. In contrast, a considerable proportion of prescription opioids are consumed by travelers.

We obtain an initial sample of U.S. credit unions from the SNL Bank Regulatory dataset of S&P Market Intelligence. The dataset is based on financial reports that banks and credit unions are mandated to file with regulatory authorities, such as the Federal Reserve, the Federal Deposit Insurance Corporation (FDIC), the Office of the Comptroller of the Currency (OCC), and NCUA. SNL compiles credit union financial information from the Credit Union Call Report Form 5300 and 5310 filed with NCUA. Our final sample consists of the intersection of the CDC opioid death data and the SNL credit union data and covers 93,108 credit union-year observations over 2003–2018. Specifically, credit unions are matched to counties based on the location of their headquarters.

Our study employs a comprehensive panel dataset that captures annual variations across all key variables. The county-level demographic variables are also collected annually, sourced from the U.S. Census Bureau and Bureau of Labor Statistics. Importantly, we do not interpolate between Census years but instead use actual annual estimates provided by these agencies.

This complete panel structure is crucial for our analysis as it allows us to capture year-to-year changes in opioid death rates, annual variations in local economic conditions, changes in credit union performance over time, and the temporal relationship between opioid deaths and credit union outcomes. We specify regression model as follows:

where:

1. CU_PERFORMANCE_i,t represents the performance measures (ROA or ROE) for credit union i in year t

2. OPIOID_DEATH_c,t is the opioid mortality rate in county c in year t

3. X_i,t is a vector of credit union-specific control variables including:

   • LOG(FIRM AGE)

   • CAPITAL RATIO

   • LN(ASSET)

   • FEDERAL CHARTERED

   • COMMUNITY

   • INDUSTRY

4. Z_c,t is a vector of county-level demographic controls including:

   • EDUCATION

   • POPULATION

   • INCOME

   • UNEMPLOYMENT

   • MARRIED

   • HPICHG

   • MALE

   • MINORITY

   • POPU_AGE

5. α_t represents year fixed effects

6. ε_i,t is the error term

Table A2 reports the sample distribution of U.S. credit unions by year and by state/territory. The number of credit unions has declined nearly monotonically over time from 7,662 in 2003 to 4,095 in 2018, reflecting a consolidation trend in the industry. Texas and California have the most credit unions, whereas Alaska and Nevada have the fewest.

In this study, we examine the effect of opioid abuse on credit union profitability measured by return on assets (ROA) and return on equity (ROE). We also examine how opioid abuse affects various contributing factors of credit union profitability. The positive contributors are asset growth rate (ASSET GROWTH), loan growth rate (LOAN GROWTH), deposit growth rate (DEPOSIT GROWTH), interest income on loans scaled by total assets (INTINC), and interest spread scaled by total assets (INTSPREAD). The negative contributors are interest expense on deposits scaled by total assets (INTEXP), operating expense scaled by total assets (OPEREXP), and nonperforming loan ratio (NONPERFLOAN).

In our regressions, we control for multiple credit union characteristics. ASSET denotes total assets. CAPITAL RATIO denotes the ratio of total capital to total assets. FEDERAL CHARTERED is a dummy variable equal to one for federal-chartered credit unions and zero otherwise ^[24]. COMMUNITY is a dummy variable equal to one for community credit unions and zero otherwise. INDUSTRY is a dummy variable equal to one for industry credit unions and zero otherwise. CU AGE denotes credit union age.

We also control for multiple geographic variables at the county level. EDUCATION denotes the percentage of adults over 25 years old with a bachelor’s degree. POPULATION denotes the logarithm of population. INCOME denotes the logarithm of the median household income. UNEMPLOYMENT denotes the unemployment rate. MARRIED denotes the percentage of married population. HPICHG denotes the percentage change of house price index. MALE denotes the percentage of male population. MINORITY denotes the percentage of non-white population. POPULATION AGE denotes the logarithm of the population weighted average age. Variable definitions are also listed in Table A1.

Table 1 reports summary statistics of the key variables. The typical county in our sample has nearly 14 opioid overdose deaths per 100,000 population during 2003–2018. About 26% of the adults over 25 years old have bachelor’s degrees. The average unemployment rate is 6.2%. Nearly 7% of the population are married, and minorities account for about one fourth of the population ^[25]. The average ROA and ROE of the sample credit unions are 0.4 and 2.9%, respectively. The average book value of their total assets is about $156 million, and the total assets, on average, grow at an annual rate of 3.7%. About 60% of the sample credit unions are federal-chartered, and the average capital ratio is 13.8%. The average interest spread is 5.5%, and the average nonperforming loan ratio is 1.7%.

Besides, our sample covers counties with wide demographic variation, from small rural communities (minimum 970 residents) to major metropolitan areas (maximum 2.8 million residents). The median county population of 9,946 indicates that many credit unions in our sample serve smaller communities, which aligns with the traditional role of credit unions in serving local markets. The median household income of $55,048 with a range from $38,397 to $87,909 suggests our sample covers economically diverse communities. The relatively tight standard deviation ($8,421) indicates that most credit unions operate in middle-income areas.

Table A3 reports the correlations between the key variables. OPIOID DEATH is negatively correlated with ROA and ROE, significant at the 5% level, providing preliminary support for our hypothesis of the negative association between opioid death rate and credit union profitability. As for the contributing factors of credit union profitability, OPIOD DEATH is negatively correlated with ASSET GROWTH, INTEXP, INTINC, and INTSPREAD but positively correlated with OPEREXP and NONPERFLOAN, providing motivating evidence on the potential channels through which the opioid crisis affects credit union profitability.

In this subsection, we discuss baseline regression results in Table 2. The regressions test the effect of county-level opioid death rate (OPIOID DEATH) on credit union profitability measured by ROA (Panel A) and ROE (Panel B). The regressions control for credit union characteristics but vary in control of county characteristics. Year fixed effects and standard errors clustered at the credit union or county level are used.

Overall, we find that county-level opioid death rate is negatively associated with credit union profitability. The estimated coefficient of OPIOID DEATH is significantly negative for all the eight models in Table 2. For ROA (Model 1), the coefficient of −0.018 indicates that a one standard deviation increase in log(1+OPIOID DEATH) [0.435] is associated with a decrease of 0.018 standard deviations in ROA. Given ROA’s standard deviation of 0.008, this translates to a decrease of 0.0144% points (−0.018 × 0.008). Relative to the mean ROA of 0.004 (0.4%), this represents a 3.6% decrease in ROA. When county characteristics are controlled and clustered standard errors are used, the results remain qualitatively unchanged, as shown in Models 2–4. For ROE (Model 5), the coefficient of −0.018 indicates that a one standard deviation increase in log(1+OPIOID DEATH) is associated with a decrease of 0.018 standard deviations in ROE. Given ROE’s standard deviation of 0.075, this translates to a decrease of 0.135% points (−0.018 × 0.075). Relative to the mean ROE of 0.029 (2.9%), this represents a 4.66% decrease in ROE. As shown in Models 6–8, when county characteristics are controlled and clustered standard errors are used, the results remain consistent with Model 5. The evidence in Table 2 suggests that credit unions in counties with higher opioid death rates are associated with lower profitability, supporting our hypothesis. As credit unions closely depend on the local economy (McKillop et al., 2020), the adverse shock of the opioid crisis to the local economy eventually transmits to them, hurting their performance.

Additionally, larger credit unions and those with higher capital ratios are associated with higher profitability, highlighting the positive role of capital adequacy in financial performance. Moreover, credit unions in counties with older populations, lower unemployment rates, and higher housing prices tend to have higher profitability. This evidence appears to suggest that elderly residents, a group with a low unemployment rate and high savings rate, and a prosperous real estate market contribute to credit union profitability.

To better understand the relationship between opioid abuse and credit union profitability, we tested whether local variations in opioid abuse affect our results using credit union fixed effects but found no significant relationship between opioid deaths and credit union performance. However, credit union fixed effects may not be the most appropriate specification for three key reasons: (1) opioid abuse impacts credit unions gradually through economic and behavioral shifts rather than sharp year-to-year changes, making within-institution variation insufficient for identification; (2) credit unions adapt policies over time in response to opioid-related challenges, complicating efforts to isolate the direct effects of opioid deaths; and (3) most variation in opioid impact occurs across communities, but fixed effects remove this key cross-sectional information. Instead, our preferred approach examines how opioid exposure differences across communities affect credit union performance while controlling for local economic and demographic factors, better capturing the institutional and economic context of the issue.

It is possible that our baseline results are driven by omitted variables, reverse causality, or covariate imbalance. Thus, we address potential endogeneity issues using multiple techniques: 2SLS regressions, regressions based on propensity score matching, and weighted regressions based on entropy balancing.

Table 3 presents the 2SLS regression results. The average OPIOID DEATH in the surrounding counties (excluding the focal county) and counties of credit unions in the same size quartile as the focal credit union (excluding the focal credit union) are used as the instrumental variables. In the first stage, OPIOID DEATH is regressed on the instrumental and control variables ^[26]. In the second stage, credit union ROA and ROE are regressed on the predicted OPIOID DEATH from the first stage. The estimated coefficient of the predicted OPIOID DEATH is negative for Models 1 and 2, significant at the 1% level. Moreover, two identification tests are conducted for the second stage. The Hansen statistic is insignificant for Models 1 and 2, indicating that our instrumental variables are indeed exogenous and therefore supporting instrument validity. This exogeneity is further confirmed by the significant Kleibergen–Paap Wald F-statistic of 444.2, indicating that our instruments are not weak (Stock and Yogo, 2005). Collectively, the results in Table 3 suggest that our baseline findings are not driven by omitted variables or reverse causality.

To further mitigate the endogeneity concerns, we run regressions based on propensity score matching and report the results in Table A4. First, the probability of a credit union being in the highest OPIOID DEATH quartile is regressed on credit union and county characteristics with year fixed effects. We obtain the predicted probabilities from this logit regression and match each treated credit union with a control credit union in the lower OPIOID DEATH quartiles in the same year with the closest predicted probability (highest propensity score). Panel A shows the results of the logit regressions before vs. after the matching procedure. None of the control variables remain significant after matching, and the pseudo R-squared statistic declines from 3.79% before matching to 0.234% after matching, indicating a good covariate balance ^[27]. Panel B reports the regressions of credit union ROA and ROE on OPIOID DEATH using only the treated and control (matched) observations. As shown in Panel B, the estimated coefficients of OPIOID DEATH are significantly negative at the 1% level, suggesting that our baseline findings are not driven by covariate imbalance. The persistence of our results in matched samples provides additional evidence that the relationship is not purely driven by correlations with other local characteristics.

We now run weighted regressions based on entropy balancing and report the results in Table A5. Entropy balancing is a data preprocessing method to achieve covariate balance with binary treatments (Hainmueller, 2012). It does not require propensity scores and instead directly targets balancing the moments of covariates (e.g. mean and variance) between the treated and control groups.

Panel A reports the mean, variance, and skewness of the control variables of the credit unions in the highest OPIOID DEATH quartile vs. those in the lower OPIOID DEATH quartiles after the balancing procedure. The standardized differences in the control variables between these two groups are insignificant, indicating a good covariate balance. Panel B reports the results of weighted regressions of credit union ROA and ROE on OPIOID DEATH where the weights are obtained from the balancing procedure. The estimated coefficients of OPIOID DEATH are significantly negative at the 1% level, further confirming that our baseline findings are not driven by covariate imbalance.

In this subsection, we conduct multiple statistical robustness checks to account for potential biases originated from our regression specifications or sample construction and report the results in Table A6. In Panel A, we run robust regressions to account for heteroskedasticity. In Panel B, we run the Fama and MacBeth (1973) regressions to focus on the cross-sectional association between OPIOID DEATH and credit union ROA and ROE. In Panel C, we exclude credit unions in the four major financial hubs (New York City, Boston, Chicago, and San Francisco) to test whether our findings are driven by local financial development. In Panel D, we exclude credit unions in the top and bottom 5% OPIOID DEATH counties to test whether our findings are driven by extreme observations and outliers. In all four panels, the coefficients of OPIOID DEATH remain significantly negative at the 1% level, and their sizes are consistent with the baseline results in Table 2.

In this subsection, we examine multiple potential channels through which opioid abuse affects credit union profitability: ASSET GROWTH, LOAN GROWTH, DEPOSIT GROWH, INTEXP, INTINC, INTSPREAD, OPERX, and NONPERFLOAN. Particularly, we conduct a causal mediation analysis of these mediators based on the implementation of Imai et al. (2010) and report the results in Table 4. Causal mediation analysis estimates the mediation and direct effects by simulating predicted values of mediator or dependent variable and then computing the appropriate statistics of interest, such as average causal mediation effect (ACME), direct effect, and total effect.

For each panel, in the first stage, a mediator is regressed on OPIOID DEATH. In the second stage, credit union ROA is regressed on OPIOID DEATH and the mediator ^[28]. The percentage of the total effect of OPIOID DEATH on ROA mediated is our statistic of interest.

For Panels A–F, in the first stage, the estimated coefficient of OPIOID DEATH is significantly negative, suggesting that credit unions in counties with higher opioid death rates are associated with lower asset growth rate, loan growth rate, deposit growth rate, interest expense, interest income, and interest spread. In the second stage, ASSET GROWTH, LOAN GROWTH, DEPOSIT GROWH, and INTINC have relatively high percentages (over 5%) of total effect mediated, suggesting that opioid abuse reduces the asset growth rate, loan growth rate, deposit growth rate, and interest income of local credit unions, which in turn reduces their profitability. For Panels G and H, in the first stage, the estimated coefficient of OPIOID DEATH is significantly positive, suggesting that credit unions in counties with higher opioid death rates are associated with higher operating expense and nonperforming loan ratio. In the second stage, OPERX has an exceptionally high percentage (over 40%) of total effect mediated, suggesting that opioid abuse increases the operating expense of local credit unions, thereby hurting their profitability. More importantly, nonperforming loan ratio plays a significant mediating role, confirming our hypothesis that opioid abuse hurts credit unions’ profitability by increasing their credit risk.

Overall, the evidence in Table 4 suggests that the adverse shock of the opioid crisis to the local economy is reflected in many aspects, including credit risk, of local credit unions, resulting in their lower profitability.

In this subsection, we examine heterogeneity in the effect of the opioid crisis on credit union profitability measured by ROA across credit unions of various features and report the results in Table 5 ^[29].

We categorize credit unions into terciles by total assets. Panel A shows that OPIOID DEATH negatively affects ROA across all groups, with the impact strongest for larger credit unions (−0.046 vs. −0.022 for small credit unions), likely due to greater exposure to community effects. Panel B examines charter types, showing a stronger negative impact on state-chartered credit unions, potentially due to their higher solvency risk. Panel C analyzes community vs. non-community credit unions, finding a more pronounced effect on community credit unions, likely due to their local economic dependence. Panel D compares industry vs. non-industry credit unions, showing a stronger effect on industry credit unions, possibly due to less diversified loan portfolios. Panel E contrasts non-crisis vs. crisis periods, indicating a weaker effect during the 2008 financial crisis, as opioid abuse played a secondary role in profitability.

The worsening U.S. opioid crisis harms both public health and the local economy. We hypothesize that credit unions in counties with higher opioid death rates experience lower profitability, as they are more reliant on local economies, lack stockholders to absorb risk, and have limited ability to pass on costs. To our knowledge, this is the first study examining the opioid crisis’s impact on credit unions. Our analysis shows that higher county-level opioid death rates negatively affect credit union profitability, measured by ROA and ROE, with findings confirmed through 2SLS regression, propensity score matching, and entropy balancing. Causal mediation analysis reveals that opioid abuse reduces profitability by slowing asset, loan, and deposit growth, decreasing interest income, and increasing operating expenses and nonperforming loans, raising credit risk. The effect is most pronounced for state-chartered, community, and industry credit unions. This study underscores the financial consequences of opioid abuse, highlighting its impact on credit unions and the broader economy. Policymakers should consider these financial risks when assessing the crisis, and future research could explore how opioid abuse disrupts economic activity through supply chains.

The supplementary material for this article can be found online.