Hospital system access to capital and the COVID-19 pandemic Authors: Sung J. Choi, Jeffrey S. McCullough Published: 2026 ABSTRACT This study examines the impact of COVID-19 on U.S. hospital systems’ access to capital through the municipal bond market. Using hospital bond issuance data from the Electronic Municipal Market Access system matched to the American Hospital Association Annual Survey (April 2019 to February 2021), we analyze changes in the volume, cost, and composition of hospital borrowing. The total quantity of hospital bond issuances declined by 34%, with the number of issuances dropping by 13% and individual bond sizes shrinking by 23%. Despite these contractions, borrowing costs remained stable, likely due to Federal Reserve policies and COVID relief funds. The composition of borrowers shifted notably: smaller and rural hospital systems increasingly relied on external capital, while larger systems demonstrated greater financial resilience. State-level COVID-19 incidence was not significantly associated with bond market outcomes, and selection correction models confirmed that findings were not driven by selection bias. Policy interventions helped stabilize hospital capital markets, but concerns persist regarding long-term debt sustainability, particularly for smaller and rural hospital systems that may require continued financial support. Introduction The Pandemic had profound and far-reaching financial consequences for the United States economy and healthcare institutions, particularly health systems. In regions where COVID-19 cases surged, healthcare organizations saw overwhelming demands, leading to substantial capacity constraints1–3. Furthermore, the pandemic triggered significant disruptions in global supply chains for essential medical supplies and equipment, exacerbating the challenges faced by healthcare organizations4,5. As health systems prioritized their resources to care for COVID-19 patients, they saw the loss of revenue from the cancellation of profitable elective procedures 2,6. Throughout 2020, hospitals exhibited lower year-over-year operating margins7. Bonds are the most important source of external financing for nonprofit health systems and hospitals. In contrast to investor-owned organizations, nonprofit organizations cannot access equity markets and thus rely on debt financing through the bond market. Bonds, therefore, serve as critical instruments for funding capital investment projects. They are essential for purchasing and maintaining capital equipment and facilities. The financial shock encountered by hospitals during the pandemic had a detrimental impact on their access to debt. Key credit rating agencies, such as Standard & Poor’s (S&P) and Moody’s, cast negative outlooks for the not-for-profit hospital and health sector in 2020 and 2021, constraining their access to the bond market8–10. Healthcare industry bond indices showed sharp declines in March 2020 as states implemented lockdowns, beginning with California on the 19th (Fig. 1).Fig. 1Trends in the S&P Municipal Bond Indices.The S&P Municipal Bond Hospital Index consists of bonds in the S&P Municipal Bond Health Care Index from the Hospital sector. Trends in the S&P Municipal Bond Indices. The S&P Municipal Bond Hospital Index consists of bonds in the S&P Municipal Bond Health Care Index from the Hospital sector. Financial shocks and the ensuing constraints to accessing debt present significant challenges to health systems and hospitals’ long-run sustainability in terms of their capital investments11,12. Limited access to capital may hinder investments in aging infrastructure, leaving hospitals unable to maintain their facilities, which could lead to closures. COVID relief funds may have temporarily offset the financial losses, especially for government, rural, and smaller hospitals7. COVID relief funds helped most hospitals temporarily increase total margins and avoid financial distress during the pandemic13,14. Critical Access Hospitals (CAH) experienced an improvement in their ability to cover debt service requirements and the overall capital structure remained relatively stable, most likely due to the federal Public Health Emergency (PHE) funds15. However, as relief funds wane and significant financial challenges persist, the healthcare industry is witnessing an elevated risk of bankruptcies and closures, notably among rural hospitals and health systems16–18. Amidst these looming financial challenges, we aim to provide a better understanding of the impact of the pandemic on health system capital access. We explore the diverse facets of hospitals and health systems’ access to capital. This includes an examination of the direct effects of the pandemic on the cost and quantity of capital raised. Additionally, we examine potential shifts in credit ratings, changes in the overall credit market influenced by factors such as federal reserve action and interest rates, changes in types of bond issuances such as bond maturity, and shifts in the mix of borrowers, encompassing both larger and smaller health systems, as well as borrower mix related to CARES Act funding. This analysis sheds light on how the healthcare industry can navigate these financial challenges effectively. COVID’s impact may not have been the same for all hospital systems. Anecdotal evidence suggests that hospitals serving lower-income communities have faced significant financial hardships during the pandemic, whereas wealthy hospitals maintained profitability19,20. For-profit systems were prepared to control costs to weather the pandemic, but safety-net hospitals were unprofitable and liquidity-constrained even before the pandemic. Policy interventions came quickly to alleviate the pandemic’s financial burden on hospital systems. The Coronavirus Aid, Relief, and Economic Security (CARES) Act was signed into law on March 27, which added a 20% increase in inpatient payments for COVID-19 patients21, temporarily suspended the Medicare sequestration from May 1, 2020, through December 31, 202022, and allowed to defer employer payroll taxes23. Rapid government intervention appears to have eased the market, as stocks and bond indices quickly rebounded after the initial plunge. However, the hospital municipal bond index lagged the overall bond index from March 2020 onward (Fig. 1). The COVID-19 pandemic was a staggering disruption to hospital operations. As researchers uncover the pandemic’s financial impact using emerging data, the implications for capital access and investment remain unclear, especially in the long run. We examined how the pandemic impacted long-term capital access. Specifically, we examined (i) trends in hospital system bond issuances and cost of capital and (ii) the relationship between hospital system characteristics and access to capital with respect to credit ratings, system affiliation, scale, and local COVID-19 incidence. Results Summary statistics The bond issuances are summarized for the pre- and post-COVID pandemic periods in Table 1. The number of hospital bond issuances fell by more than 13% following the pandemic, from an average of 161 per month to an average of 139 per month during the pandemic. Furthermore, the average principal amount per issuance fell from $14.3 million to $11 million during the pandemic. The total debt issued fell from $2.3 billion per month to $1.5 billion per month, a 34% reduction.Table 1Summary statistics of bond issuances preceding and during the COVID pandemicPre COVIDPost COVIDp-valueN = 1770N = 1528Real principal amount ($1 M)14.26 (28.17)10.97 (27.25)<0.001Total principal in $1 M per day478.87 (362.66)309.01 (300.38)<0.001Treasury rate matched to bond maturity (%)1.92 (0.40)0.94 (0.47)<0.001Coupon rate %a3.93 (1.37)3.86 (1.16)<0.001Maturity of bond in yearsa21.03 (9.00)22.66 (9.45)<0.001Median Rating (larger is better rating)a8.22 (1.79)7.69 (1.63)<0.001Total number of hospitals in systema12.37 (23.04)7.60 (7.48)<0.001Total number of beds in systema2940.79 (5503.63)1564.68 (1384.17)<0.001Rural beds in system %a7.63 (18.07)8.49 (17.40)<0.001Total admissions in system 1000sa140.14 (283.23)72.58 (64.59)<0.001Medicare discharges in system %a44.40 (11.38)44.35 (11.51)<0.001Medicaid discharges in system %a21.64 (9.16)23.61 (11.65)<0.001COVID state casesa–294,233.60 (381,132.27)<0.001COVID state deathsa–8063.16 (9967.41)<0.001Data are presented as mean (SD) or percent (N).aWeighted by principal amount. Dollars adjusted to 2020 dollars. COVID event is 19Mar2020. Table 1Summary statistics of bond issuances preceding and during the COVID pandemicPre COVIDPost COVIDp-valueN = 1770N = 1528Real principal amount ($1 M)14.26 (28.17)10.97 (27.25)<0.001Total principal in $1 M per day478.87 (362.66)309.01 (300.38)<0.001Treasury rate matched to bond maturity (%)1.92 (0.40)0.94 (0.47)<0.001Coupon rate %a3.93 (1.37)3.86 (1.16)<0.001Maturity of bond in yearsa21.03 (9.00)22.66 (9.45)<0.001Median Rating (larger is better rating)a8.22 (1.79)7.69 (1.63)<0.001Total number of hospitals in systema12.37 (23.04)7.60 (7.48)<0.001Total number of beds in systema2940.79 (5503.63)1564.68 (1384.17)<0.001Rural beds in system %a7.63 (18.07)8.49 (17.40)<0.001Total admissions in system 1000sa140.14 (283.23)72.58 (64.59)<0.001Medicare discharges in system %a44.40 (11.38)44.35 (11.51)<0.001Medicaid discharges in system %a21.64 (9.16)23.61 (11.65)<0.001COVID state casesa–294,233.60 (381,132.27)<0.001COVID state deathsa–8063.16 (9967.41)<0.001Data are presented as mean (SD) or percent (N).aWeighted by principal amount. Dollars adjusted to 2020 dollars. COVID event is 19Mar2020. Summary statistics of bond issuances preceding and during the COVID pandemic Data are presented as mean (SD) or percent (N). aWeighted by principal amount. Dollars adjusted to 2020 dollars. COVID event is 19Mar2020. Bond issuance characteristics in Table 1 are weighted by principal amount. The coupon rates remained practically unchanged (3.93% pre-COVID vs. 3.86% post-COVID). Similarly, the maturity of bonds was almost unchanged (21.0 years pre-COVID vs. 22.7 years post-COVID). The median credit rating of borrowing firms fell slightly by 0.53 units. This corresponds to a fall from about AA- to A + . This may reflect changes in the distribution of borrowers as well as changes in the market. While it is plausible that the market’s cost of capital remained unchanged and the credit ratings were nearly constant, the reality may have been more complicated given the observed shift in secondary market bond valuations (Fig. 1). This apparent stability may reflect: (i) Market shifts in the costs of capital (e.g., a falling treasury rate), (ii) Changes in the mix of issuers (e.g., observed hospital system characteristics), and (iii) Greater selectivity in the market, only issuing debt to better-rated companies (e.g., unobserved issuer characteristics). We examine these differences empirically. Borrower characteristics were weighted by principal amount. Overall, borrower characteristics shifted to smaller organizations in post-COVID-19, with smaller systems (average of 12.4 hospitals per system pre-COVID-19 vs. 7.6 hospitals per system post-COVID-19), fewer beds in the system (2940.8 beds pre-COVID-19 vs. 1564.7 beds post-COVID-19), and fewer admissions (140.1 thousand admissions pre-COVID-19 vs. 72.6 thousand admissions post-COVID-19). The proportion of rural beds in the system increased slightly from 7.63% pre-COVID-19 to 8.49% post-COVID-19. The proportion of Medicare was all but unchanged at 44%, but the Medicaid share increased from 22% to 24%. Figure 2 shows the distribution of credit ratings by COVID. The proportion in each category was weighted by the principal amount. The proportion of borrowers rated BBB + , A, A+ increased post-COVID-19, while the proportion of borrowers rated AA-, AA, AA+ decreased post-COVID-19. This is consistent with a modest drop in hospitals’ average credit rating.Fig. 2Credit Rating distributions of hospital bond issuances preceding and during the COVID pandemic.Percentages in each category were weighted by principal amounts ($). Credit Rating distributions of hospital bond issuances preceding and during the COVID pandemic. Percentages in each category were weighted by principal amounts ($). Summary statistics of hospitals The study hospital characteristics are summarized for the pre- and post-COVID pandemic periods (Supplementary Table 1). There was a modest drop in the number of hospitals issuing new bonds from 124 to 106. Hospital characteristics between the two periods remained stable, with most hospitals being non-public, with a bed size of about 360 and admissions of about 15,000. Unweighted summary statistics are also provided in Supplementary Table 2. OLS estimates of cost of debt Table 2 measures changes in the cost of debt (e.g., bond coupon rate) during the pandemic and their associations with both borrower and market characteristics. There were, however, significant associations between issuance and borrower characteristics. The costs of debt were, for example, slightly lower for organizations with a larger Medicare or rural case mix. While Medicare patients are not necessarily the most profitable, Medicare volume is less sensitive to economic downturns because eligibility and reimbursement rates are set by the federal government and do not depend on employment or economic performance. This revenue stability may reduce the perceived credit risk for organizations with a larger Medicare share. The parameter estimates in Table 2 suggest that coupon rates would have been 0.2 percentage points lower if the mix of borrowing hospital systems remained unchanged. This reinforces the impression that the overall cost of capital did not appreciably change.Table 2OLS estimates of cost of debt1234Post-COVID Post COVID0.22930.23960.43050.4611[−1.265 1.723][−1.220 1.699][−0.930 1.791][−0.786 1.708]0.7630.7470.5340.468Treasury rate matched to bond maturity in %−0.19480.3225*0.2327[−0.410 0.020][0.046 0.599][-0.024 0.490]0.0760.0220.076Real principal amount ($1M adj to 2020)−0.0024*−0.0029**[−0.005 −0.000][-0.005 −0.001]0.0480.010Maturity of bond in years−0.0305***−0.0276**[−0.044 −0.017][−0.040 −0.015]<.001<.001Median Rating (larger number is better rating)−0.0962***[−0.150 −0.043]<.001Total admissions in system 1000s demeaned0.0004[−0.000 0.001]0.121Rural beds in system %−0.0057*[−0.010 −0.001]0.010Medicare discharges in system %−0.0153*[−0.025 −0.005]0.002Medicaid discharges in system %−0.0070[−0.018 0.004]0.231Month Fixed EffectsYesYesYesYesYear Fixed EffectsYesYesYesYesNumber of observations3298329832983298For each variable, the table reports the coefficient estimate, confidence interval, and p-value. Pre-COVID: 07apr2019-18mar2020, Post-COVID:19mar2020-28feb2021. Models include month and year fixed effects to control for time trends and seasonality.***p < 0.001, **p < 0.01, *p < 0.05. Table 2OLS estimates of cost of debt1234Post-COVID Post COVID0.22930.23960.43050.4611[−1.265 1.723][−1.220 1.699][−0.930 1.791][−0.786 1.708]0.7630.7470.5340.468Treasury rate matched to bond maturity in %−0.19480.3225*0.2327[−0.410 0.020][0.046 0.599][-0.024 0.490]0.0760.0220.076Real principal amount ($1M adj to 2020)−0.0024*−0.0029**[−0.005 −0.000][-0.005 −0.001]0.0480.010Maturity of bond in years−0.0305***−0.0276**[−0.044 −0.017][−0.040 −0.015]<.001<.001Median Rating (larger number is better rating)−0.0962***[−0.150 −0.043]<.001Total admissions in system 1000s demeaned0.0004[−0.000 0.001]0.121Rural beds in system %−0.0057*[−0.010 −0.001]0.010Medicare discharges in system %−0.0153*[−0.025 −0.005]0.002Medicaid discharges in system %−0.0070[−0.018 0.004]0.231Month Fixed EffectsYesYesYesYesYear Fixed EffectsYesYesYesYesNumber of observations3298329832983298For each variable, the table reports the coefficient estimate, confidence interval, and p-value. Pre-COVID: 07apr2019-18mar2020, Post-COVID:19mar2020-28feb2021. Models include month and year fixed effects to control for time trends and seasonality.***p < 0.001, **p < 0.01, *p < 0.05. OLS estimates of cost of debt Post-COVID Post COVID For each variable, the table reports the coefficient estimate, confidence interval, and p-value. Pre-COVID: 07apr2019-18mar2020, Post-COVID:19mar2020-28feb2021. Models include month and year fixed effects to control for time trends and seasonality. ***p < 0.001, **p < 0.01, *p < 0.05. Treasury rates fell by 1 percentage point (Table 1), associated with a 0.23 percentage point increase in coupon rates, though this was not statistically significant (P = 0.076) (Table 2). Parameter estimates for the principal amount and bond maturity were statistically significant (P = 0.010, P < 0.001), but these amounted to small changes in the cost of debt at the mean. Issuer ratings fell by 0.53 units (larger is a better rating) (Table 1), which was associated with a 0.10 percentage point decrease in coupon rates per unit (P < 0.001). Borrower characteristics were also significantly associated with the cost of capital: rurality (P = 0.010) and Medicare share (P = 0.002) were both associated with lower coupon rates. In the na¨ıve model, post-COVID was not significantly associated with the coupon rate (P = 0.763) and remained non-significant in the full model (P = 0.468). OLS estimates of quantity of debt Table 3 measures changes in the quantity of debt (e.g., principal amount) during the pandemic and their associations with both borrower and market characteristics.Table 3OLS estimates of quantity of debt123Post-COVIDPost COVID6747−5573−5687[−318 13,812][−32,183 21,037][−31,267 19,893]0.0610.6810.662Coupon rate %−1501*−1746**[−2646 −356][−2892 −600]0.0100.003Maturity of bond in years1909***1847***[1687 2131][1631 2064]<0.001<0.001Median Rating (larger number is better rating)840**[267 1413]0.004Total admissions in system 1000 s demeaned26*[3 49]0.027Rural beds in system %−56**[−96 −16]0.006Medicare discharges in system %−45[−186 97]0.535Medicaid discharges in system %−38[−185 110]0.616Month Fixed EffectsYesYesYesYear Fixed EffectsYesYesYesNumber of observations329832983298For each variable, the table reports the coefficient estimate, confidence interval, and p-value. Pre-COVID: 07apr2019-18mar2020, Post-COVID:19mar2020-28feb2021. Principal amount in $1000. Models include month and year fixed effects to control for time trends and seasonality.***p < 0.001, **p < 0.01, *p < 0.05. Table 3OLS estimates of quantity of debt123Post-COVIDPost COVID6747−5573−5687[−318 13,812][−32,183 21,037][−31,267 19,893]0.0610.6810.662Coupon rate %−1501*−1746**[−2646 −356][−2892 −600]0.0100.003Maturity of bond in years1909***1847***[1687 2131][1631 2064]<0.001<0.001Median Rating (larger number is better rating)840**[267 1413]0.004Total admissions in system 1000 s demeaned26*[3 49]0.027Rural beds in system %−56**[−96 −16]0.006Medicare discharges in system %−45[−186 97]0.535Medicaid discharges in system %−38[−185 110]0.616Month Fixed EffectsYesYesYesYear Fixed EffectsYesYesYesNumber of observations329832983298For each variable, the table reports the coefficient estimate, confidence interval, and p-value. Pre-COVID: 07apr2019-18mar2020, Post-COVID:19mar2020-28feb2021. Principal amount in $1000. Models include month and year fixed effects to control for time trends and seasonality.***p < 0.001, **p < 0.01, *p < 0.05. OLS estimates of quantity of debt For each variable, the table reports the coefficient estimate, confidence interval, and p-value. Pre-COVID: 07apr2019-18mar2020, Post-COVID:19mar2020-28feb2021. Principal amount in $1000. Models include month and year fixed effects to control for time trends and seasonality. ***p < 0.001, **p < 0.01, *p < 0.05. The unadjusted quantity of debt per issuance fell by $3.3 M (23%) during the early COVID-19 pandemic (Table 1). Table 3 results suggest that this change is potentially attributable to changes in the mix of borrowers. Controlling for these factors may account for potential confounding due to changes in observed borrower or market characteristics. Parameter estimates for coupon rate and bond maturity were statistically significant (P = 0.003, P < 0.001), amounting to a $1.75 M decrease per percentage point of coupon rate and a $1.85 M increase per year of maturity. Issuer ratings fell by 0.53 units (Table 1), which was associated with an $840 K (P = 0.004) increase in principal amount per rating unit. In the na¨ıve model, post-COVID-19 was associated with a $6.75 M increase in principal amount (P = 0.061), but this became non-significant in the full model (−$5.69 M, P = 0.662), suggesting the unadjusted decline is attributable to changes in borrower mix rather than a direct pandemic effect. Probit estimates of bond issuance We estimated four probit models to examine the relationship between the COVID-19 period and the likelihood of hospital bond issuance. Marginal effects for the probit models are summarized in Table 4. The first model included only a post-COVID main effect and excluded state fixed effects. The second model added an interaction between the post-COVID indicator and hospital characteristics, again without state fixed effects. The third model mirrored the first but included state fixed effects, and the fourth model included both the interaction terms and state fixed effects.Table 4Hospital bond issuances, marginal effects (dy/dx) of probit estimates(1)(2)(3)(4)Post COVID−0.0084−0.0084−0.0087−0.0089[−0.0237, 0.0069][−0.0238, 0.0069][−0.0248, 0.0075][−0.0251, 0.0074]0.28210.28170.29230.2854Non-Public (Ref: Public)0.0231 **0.0263 **[0.0072, 0.0391][0.0076, 0.0451]0.00450.0059Total Beds in System1.74e-05 ***1.68e-05 ***[1.3e-05, 2.2e-05]3.27e-13[1.2e-05, 2.2e-05]1.48e-11% Rural Beds−0.0008 ***−0.0009 ***[−9.8e-04, −5.6e-04]6.23e-13[−0.0012, −6.7e-04]1.70e-13Medicare %−0.0008 *−0.0006[−0.0015, −9.9e-05]0.0246[−0.0014, 1.1e-04]0.0946Medicaid %0.00020.0004[−6.0e-04.001]0.6105[−4.8e-04.0014]0.344Post COVID at beds = 1500−0.022 *−0.0219 *[−0.043, −6.5e-04]0.0434[−0.044, −3.0e-05]0.0497Post COVID at beds = 2000−0.0325 *−0.0321 *[−0.0598 −0.0052][−0.0598 −0.0044]0.01980.0229Post COVID at beds = 2500−0.0447 *−0.0438 *[−0.08, −0.0094][−0.079, −0.0084]0.01310.0153Post COVID at beds = 3000−0.0587 *−0.0571 *[−0.1039, −0.0134][−0.102, −0.0122]0.0110.0128Number of observations3558355833353335For each variable, the table reports the marginal effect, confidence interval, and p-value.***p < 0.001, **p < 0.01, *p < 0.05. Table 4Hospital bond issuances, marginal effects (dy/dx) of probit estimates(1)(2)(3)(4)Post COVID−0.0084−0.0084−0.0087−0.0089[−0.0237, 0.0069][−0.0238, 0.0069][−0.0248, 0.0075][−0.0251, 0.0074]0.28210.28170.29230.2854Non-Public (Ref: Public)0.0231 **0.0263 **[0.0072, 0.0391][0.0076, 0.0451]0.00450.0059Total Beds in System1.74e-05 ***1.68e-05 ***[1.3e-05, 2.2e-05]3.27e-13[1.2e-05, 2.2e-05]1.48e-11% Rural Beds−0.0008 ***−0.0009 ***[−9.8e-04, −5.6e-04]6.23e-13[−0.0012, −6.7e-04]1.70e-13Medicare %−0.0008 *−0.0006[−0.0015, −9.9e-05]0.0246[−0.0014, 1.1e-04]0.0946Medicaid %0.00020.0004[−6.0e-04.001]0.6105[−4.8e-04.0014]0.344Post COVID at beds = 1500−0.022 *−0.0219 *[−0.043, −6.5e-04]0.0434[−0.044, −3.0e-05]0.0497Post COVID at beds = 2000−0.0325 *−0.0321 *[−0.0598 −0.0052][−0.0598 −0.0044]0.01980.0229Post COVID at beds = 2500−0.0447 *−0.0438 *[−0.08, −0.0094][−0.079, −0.0084]0.01310.0153Post COVID at beds = 3000−0.0587 *−0.0571 *[−0.1039, −0.0134][−0.102, −0.0122]0.0110.0128Number of observations3558355833353335For each variable, the table reports the marginal effect, confidence interval, and p-value.***p < 0.001, **p < 0.01, *p < 0.05. Hospital bond issuances, marginal effects (dy/dx) of probit estimates [1.3e-05, 2.2e-05] 3.27e-13 [1.2e-05, 2.2e-05] 1.48e-11 [−9.8e-04, −5.6e-04] 6.23e-13 [−0.0012, −6.7e-04] 1.70e-13 [−0.0015, −9.9e-05] 0.0246 [−0.0014, 1.1e-04] 0.0946 [−6.0e-04.001] 0.6105 [−4.8e-04.0014] 0.344 [−0.043, −6.5e-04] 0.0434 [−0.044, −3.0e-05] 0.0497 For each variable, the table reports the marginal effect, confidence interval, and p-value. ***p < 0.001, **p < 0.01, *p < 0.05. Across the four model specifications, the post-COVID main term was associated with a lower probability of bond issuance, but the estimate was not statistically significant. In the interaction specifications (Table 4 Column 2) and (Table 4 Column 4), post-COVID was associated with a lower probability of bond issuance (about 2.2 to 5.9 percentage points) as the total number of beds in the system increased. The sample size is smaller in models with state fixed (Table 4 Columns 3, 4) because no bonds were issued in some states during the sample period and such cases were dropped from probit estimation. Full probit coefficients and the corresponding marginal effects across all four specifications are in Supplementary Tables 3, 4. Robustness Stratified regression analyses for the cost of debt suggest that the cost of debt was more sensitive to borrower characteristics post-COVID-19. After COVID-19, the association between median rating and coupon rate increased by 36% (from −0.0908 to −0.1235), and the estimate for hospital size increased substantially (from 0.0004 to 0.002). In analyses of the quantity of debt issued, results indicate that principal amounts became slightly less sensitive to coupon rates and bond maturity post-COVID-19. After COVID-19, the association between coupon rate and the principal amount decreased by 3.1% (from −1482 to −1436), the estimate for bond maturity decreased by 5.8% (from 1902 to 1792), and the estimate for median rating decreased by 58% (from 1213 to 507). Full regression results for the stratified models are in Supplementary Tables 5, 6. Selection correction The Heckman selection models found no statistically significant evidence of selection bias in either the cost or quantity of debt. The inverse Mills ratio (lambda) was not significant for the coupon rate model (λ = −0.295, SE = 0.265, P = 0.266) or the principal amount model (λ = -42,173, SE = 59,306, P = 0.477). These results suggest that the OLS estimates are not substantially biased by selection into bond issuance, although the weak instrument problem (first-stage F = 0.58) limits the statistical power of this test. The outcome coefficients in the Heckman models were consistent with the OLS estimates: credit rating remained the strongest predictor of both coupon rates and principal amounts, and the post-COVID indicator was not significantly associated with either outcome after controlling for borrower characteristics. We re-estimated the models using only the two strongest instruments (total assets and total liabilities, with GMM prediction correlations of 0.94 and 0.76, respectively), and the results were substantively unchanged (λ = −0.292, SE = 0.260 for coupon; λ = -63,166, SE = 58,023 for principal). Full results are in Supplementary Tables 10–14. COVID incidence State-level COVID-19 incidence was not significantly associated with either the cost of debt (coefficient = −0.445 per million cases, P = 0.127) or the quantity of debt (coefficient = $2340 per million cases, P = 0.315). Full incidence regression results are in Supplementary Tables 7, 8. These results suggest that the pandemic’s effects on hospital bond markets were broad-based rather than concentrated in high-incidence areas, consistent with the interpretation that market-wide uncertainty and policy responses – rather than local disease severity – drove the observed changes in hospital borrowing. Discussion Bonds are a crucial source of financing for hospitals to maintain and improve their plant, property, and equipment. Ensuring stable financing is essential for hospitals to maintain their facilities and provide high-quality care. The COVID-19 pandemic has had significant implications for hospital bond issuances and raised concerns about the long-term stability of healthcare institutions. Nonprofit hospitals raised significantly less capital following the COVID-19 pandemic. The number of new issuances fell by about 13%, while the average size of issuances decreased by 23%. These changes resulted in a 34 percent reduction in total borrowing in the 11 months following the pandemic. The S&P Municipal Bond Hospital Index, which tracks the value of hospital debt, had a pre-pandemic high of 320 on Mar 19, 2020, then fell about 12% in 1 week. The index only returned to its pre-pandemic high after five months and then remained stable for the rest of 202024. To contextualize these pandemic-era findings, it is useful to compare them with hospital financing trends during previous U.S. recessions, particularly the 2008 financial crisis, which offers important context for hospital financing behavior during COVID-19. During the Great Recession, nine out of ten hospitals reported difficulty accessing capital, prompting widespread delays or cancellations of construction, technology, and infrastructure investments25,26. Hospitals increasingly relied on internal cash, as credit markets froze and borrowing became costlier11. The 2008 downturn also exposed vulnerabilities in financing strategies, such as auction-rate and variable-rate debt structures11. These challenges mirror trends seen during the early pandemic period, such as contracted borrowing and reliance on internal reserves. However, a key distinction is that extensive fiscal relief and policy interventions may have helped stabilize credit markets more quickly in 2020 compared to the 2008 crisis. These interventions likely buffered some of the immediate financial shocks and maintained market liquidity in ways not seen in previous purely economic recessions. We observed a substantial contraction in borrowing across the healthcare sector, with a modest drop in the number of hospitals issuing new loans, a significant decrease in the number of issuances per hospital, and a substantial reduction in the size of these issuances. Meanwhile, the cost of capital remained relatively stable, with the post-COVID indicator not significantly associated with coupon rates in any specification. This near stability in capital costs contrasts sharply with the marked reductions in borrowing activities, suggesting that while access to capital was relatively affordable, hospitals were reluctant to take on debt. During the COVID pandemic, treasury rates did not appear to have directly affected the issuance of hospital bonds. The variation in Treasury rates had only a modest impact on the cost of debt for hospitals. This subdued effect could be attributable to the federal relief programs that were implemented during the pandemic. These programs likely played a critical role in sustaining hospital systems’ access to debt7,15. Furthermore, economic policy changes may have been anticipated by financial markets. However, we only observed the cost of debt for those hospital systems that borrowed during the study period. Changes in the borrower and issuer characteristics accounted for about a 0.2 percentage point reduction in the cost of debt. The overall credit rating of borrowers fell, although modestly, from an average of AA- to A + , a one-notch downgrade. Additionally, there was a noticeable shift in the distribution of systems that chose to borrow. The summary statistics (Table 1) show that the average borrowing system was substantially smaller post-COVID (7.6 vs. 12.4 hospitals per system; 1565 vs. 2941 beds), with a slightly higher proportion of rural beds (8.5% vs. 7.6%). This shift reflects the extensive margin: smaller and more rural systems were more likely to enter the bond market during the pandemic, even though, conditional on borrowing, their individual issuances were smaller (Table 3). In other words, the pandemic drove smaller systems to seek external capital – likely because they lacked the internal reserves to absorb financial shocks – while larger systems, with greater reserves, could self-finance11,12. A downgrade from AA- to A + , while still within investment grade, increases borrowing costs through higher interest rates demanded by investors. A ”weak or deteriorating position” tends to limit a hospital’s ability to raise debt due to stricter bond covenants, lower debt capacity, and a more challenging borrowing process27. A downgraded hospital may still access the bond market, but often only for smaller issuances, shorter maturities, or with credit enhancements that add cost. Over time, credit downgrades can erode a hospital’s long-term sustainability. Higher debt costs reduce capital investments for facility upgrades and maintenance. Hospitals unable to invest risk falling behind in care quality and competitiveness, especially smaller or rural systems with fewer financial reserves. Persistent credit weakness may lead to service cuts or closure. Therefore, maintaining strong credit is thus essential not only for capital access, but also for delivering on a hospital’s long-term mission27. Smaller hospital systems, often with less financial reserves, limited revenue streams, and fewer resources compared to larger systems, were likely harder hit by the pandemic, making them more vulnerable to operational and financial disruptions. This vulnerability likely drove them to seek additional capital to manage COVID-related costs and stabilize operations. On the other hand, larger systems generally have larger reserves, allowing them to absorb economic shocks without substantially increasing their debt11,12. The findings highlight the financial challenges small rural hospital systems faced during the pandemic and underscore the importance of targeted financial support to these more vulnerable providers. Initial response to the pandemic might have been predicated on the assumption that COVID-19 would be an acute disruption. However, ongoing concerns about recessionary issues in the healthcare industry indicate a more prolonged impact. The pandemic has led to a contraction in healthcare labor supply and rising costs, further compounding the challenges for healthcare systems. Some COVID-19 Public Health Emergency policies became permanent, reflecting a policy shift acknowledging that the pandemic may have enduring effects28. These factors suggest that hospital systems’ financial and operational pressures, particularly smaller and rural systems, are likely to persist in the coming years. One of this study’s primary limitations was distinguishing between market dynamics and policy responses. Both knowledge of and reactions to the COVID-19 pandemic evolved rapidly during our study period. Hospital bond issuances during this time were varied and sparse. Additionally, the quantity and price of debt are jointly determined; thus, we cannot disentangle the causal effects of policy interventions from other market forces. Our study period ends in February 2021, providing approximately 11 months of post-pandemic data. A longer post-period would enable analysis of longer-run dynamics, including the recovery trajectory and the effects of waning federal relief. However, extending the study window faces two binding constraints. First, bond issuance data from the Electronic Municipal Market Access (EMMA) system are labor-intensive to clean: hospitals issue through conduits, so a single system’s bonds may appear under different county or authority names across issuances, and matching requires reading individual PDF offering documents. Bond credit ratings are often embedded as images rather than structured data. Second, AHA survey data availability constrains the hospital-level covariates. Additionally, the sharp decline in bond issuance volume during 2021 introduces substantial noise that complicates longer-horizon inference. Further research with emerging data is needed to examine longer-term responses that will shape hospital capital markets. We also explored regression discontinuity and interrupted time series designs. These results were highly sensitive to specification, an almost mechanical consequence of data sparsity: there are relatively few issuances around any given event window, and the wide variation in issuance sizes exacerbates instability. These data characteristics motivated the descriptive framing supplemented by the selection correction approach described above. The selection correction models, while finding no evidence of selection bias, were limited by weak instruments (first-stage F = 0.58). The exclusion restrictions – predicted financial characteristics and predicted interest rates – were jointly insignificant in the selection equation, likely because the sharp decline in issuance volume during 2021 overwhelmed cross-sectional variation in hospital financial characteristics. The insignificant lambda could therefore reflect either genuine absence of selection bias or insufficient power to detect it. Methods Analytics sample In this study, we merged data on hospital municipal bond issuances with demographic information on healthcare institutions. This comprehensive dataset was subsequently merged with general economic and market performance indicators. Electronic Municipal Market Access (EMMA) provided hospital municipal bonds data and disclosure documents. EMMA, operated by the Municipal Securities Rulemaking Board (MSRB), serves as the official repository for municipal securities disclosures and trade data. Our systematic data collection and validation process began by using EMMA’s Advanced Search feature to filter issuances by Sector: Health between April 7, 2019, and February 28, 2021. This search provided detailed information for each issuance, including coupon rates, maturity dates, principal amounts, and bond ratings. Each bond’s credit ratings from different agencies (S&P, Moody’s, and Fitch) were standardized to S&P-equivalent ratings (AAA, AA + … BBB-, Junk) in Fig. 2 and converted to an index where each rating category is represented by an integer (e.g., 1 for Junk and 11 for AAA). Following MSRB’s EMMA Reporting Guidelines, we conducted a thorough manual inspection of each issuance’s documents, examining issuer names and characteristics to identify hospital and health system bonds. To ensure accuracy in identifying individual issuances, we addressed the challenge that municipalities often serve as issuing entities, resulting in variations across issuances even within a single hospital. Each issuance underwent manual verification using official statements and disclosure documents to confirm the borrowing entity’s identity and characteristics. We excluded a few extreme outliers like a large state of California university bond and a state bond fund. For each issuing entity, we identified the corresponding CMS Certification Number (CCN), which enabled linkage to the American Hospital Association (AHA) database. The American Hospital Association (AHA) Annual Survey includes detailed hospital characteristics of not-for-profit community hospitals (non-federal, short-term, general, and specialty hospitals) such as demographics, system affiliation, utilization, and financials29. Hospital bond issuances were matched to the AHA data, yielding a sample of 3298 observations. We could not, however, match 311 bond issuances to the AHA data. We carefully examined the unmatched bond issuances to address potential bias concerns. Through manual inspection of EMMA’s database, we identified several systematic reasons for these unmatched cases. First, some municipal bonds were issued by healthcare facilities such as assisted-living centers and hospice facilities, which fall outside our study’s focus on acute care hospitals. Second, we found two primary scenarios where healthcare facilities lack CCNs: (1) non-Medicare-participating specialty hospitals (including rehabilitation, psychiatric, and religious institutions), and (2) new or under-construction facilities that issue bonds before receiving their CCN. The systematic nature of these exclusions suggests minimal risk of selection bias, as the unmatched cases primarily represent either non-hospital healthcare facilities outside our study scope or future facilities not yet operational during our study period. Other variables were merged with the bond issuance data, including state COVID incidence rates30, daily treasury rates31, and daily S&P index32. Treasury rates were matched to bond maturity. Lastly, principal amounts were adjusted to 2020 dollars using the Consumer Price Index for Hospital and Related Services33. In addition, we created a health-system (where some systems are standalone-hospital consisting of a single facility) level data using the AHA data and merged it with the bond EMMA data for the years 2019 and 2020, yielding a sample of 3558 observations. Health system data was used to analyze the relationship between system characteristics and bond issuance. Empirical models We aim to understand how the COVID-19 pandemic affected the cost and quantity of capital raised by hospital bonds. Pandemic-related changes could work directly, but they could also work indirectly through changes in the overall credit market (e.g., interest rates), the types of debt being issued, and the mix of hospitals borrowing. We employ a series of simple empirical models to better understand these relationships. We begin by measuring the association between hospital characteristics and the issuance of bonds in the periods before and during the pandemic. We then measured changes in capital costs following the COVID-19 pandemic. We modeled the change in capital costs (p) before and after the COVID-19 pandemic. We further control for the treasury rate (r), issuance characteristics (W), and the characteristics of the issuing hospital system (X).1\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${p}_{{jt}}={\alpha {COV}{ID}}_{{jt}}+{\beta r}_{{jt}}+\gamma {W}_{j}+{\delta X}_{{jt}}+{\epsilon }_{{jt}}$$\end{document}pjt=αCOVIDjt+βrjt+γWj+δXjt+ϵjt The terms {α, β, γ, δ} are parameters to be estimated. The term α measures the association between the pandemic and the cost of capital conditional upon observed characteristics. The β parameter captures changes in the market cost of capital and Federal Reserve interventions in the credit market, while γ and δ capture the effects of issuance (e.g., principal amount and maturity) and health system characteristics, respectively. The empirical model was estimated using ordinary least squares (OLS), and the errors were clustered by issuing systems. We also measure changes in the amount of capital borrowed (q) during the pandemic. Following Eq. 1 above, we condition on bond issuance and system characteristics. More formally, we estimate:2\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$${q}_{{jt}}=\alpha {COV}\,{{ID}}_{{jt}}+{\beta p}_{{jt}}+\gamma {W}_{j}+{\delta X}_{{jt}}+{\epsilon }_{{jt}}$$\end{document}qjt=αCOVIDjt+βpjt+γWj+δXjt+ϵjtThe treasury rate (r) is excluded as it works through the observed coupon rate (p). This model is estimated by OLS with the standard errors clustered by the health system. This model is identified by short-term differences in the quantity of capital raised. In particular, the 11 months before and 11 months after March 19, 2020, when California issued the first statewide stay-at-home order34,35. This first-differences model is especially appropriate because COVID-19 was an exogenous shock to hospitals, the country, and the world economy. However, there remains a potential endogeneity of capital costs, bond ratings, and selection into offerings. We address this issue by testing for the sensitivity of the results regarding the mixes of firms, credit ratings, and prevailing interest rates in the robustness section. All regression models include month-of-year and year fixed effects to control for potential seasonality and time trends in bond issuance patterns. We also conducted a series of robustness tests for seasonality using the S&P Municipal Hospital Bond Index, which provides a longer time series with a lower signal-to-noise ratio than the bond-level data. We tested for stationarity (Phillips-Perron), serial correlation (Breusch-Godfrey), and month-of-year effects using autoregressive models with linear trends. No evidence of seasonality was found at the monthly level, while the COVID effect remained large and precisely estimated (Supplementary Figs. 1, 2). We were also concerned that we only observe the issuances that occur. Consequently, we examined differences in the numbers of bond issuances per month in the pre and post-pandemic periods. Furthermore, we tested a regression specification where the data were stratified by COVID status where Pre-COVID is Apr 7, 2019-Mar 18, 2020 and Post-COVID is Mar 19, 2020-Feb 28, 2021. Equation 2 follows the same identification and robustness test strategy as Eq. 1 above. Furthermore, we used a probit model to estimate the relationship between COVID and bond issuance, controlling for hospital characteristics and state fixed effects. Specification (1) controlled for hospital characteristics, specification (2) controlled for hospital characteristics and included COVID interaction terms, specification (3) controlled for hospital characteristics and included state fixed effects, specification (4) controlled for hospital characteristics, included COVID interaction terms and included state fixed effects. For all regression analyses, the standard errors were clustered at the hospital level. Statistical analysis was performed using Stata version 1836. Selection correction Our OLS estimates are conditional on bond issuance; hospitals that did not issue bonds are unobserved. If unobserved factors that drive issuance decisions also affect bond terms, our estimates may be biased. We addressed this concern using a Heckman two-step selection correction37. The Heckman model requires exclusion restrictions – variables that predict selection (bond issuance) but do not directly affect the outcome (coupon rate or principal amount). We constructed two sets of instruments. First, we estimated the cost of debt using Hausman-style instruments: coupon rates on contemporaneous issuances by other similar firms. We employed lasso to select among interactions of bond ratings, maturity categories, and other issuance characteristics, as machine learning methods are appropriate for first-stage prediction without introducing bias in the second stage. Mean coupon rates exhibit a clear threshold at AA- across ratings and an inverted term structure across maturity categories, motivating the rating-based instrument and the lasso interaction terms (Supplementary Figs. 3, 4). Second, we constructed instruments for hospitals’ financial characteristics using system GMM38–40. Hospital financial variables – total liabilities, total assets, age of plant, and days cash on hand – may be endogenous because hospital managers observe more about their financial condition than econometricians. However, sufficiently lagged changes in these stock variables serve as valid instruments, as shocks older than the order of autocorrelation affect current stocks while being uncorrelated with current error terms. We determined the lag structure for each variable using Breusch-Godfrey autocorrelation tests and verified instrument validity using Sargan overidentification tests (Supplementary Table 9). The predicted financial characteristics (\documentclass[12pt]{minimal} \usepackage{amsmath} \usepackage{wasysym} \usepackage{amsfonts} \usepackage{amssymb} \usepackage{amsbsy} \usepackage{mathrsfs} \usepackage{upgreek} \setlength{\oddsidemargin}{-69pt} \begin{document}$$\hat{W})$$\end{document}Wˆ) and predicted interest rate (rˆ) entered the selection equation of the Heckman model as exclusion restrictions, alongside hospital characteristics and year fixed effects. We estimated two outcome equations: cost of debt (coupon rate) and quantity of debt (principal amount). As a robustness check, we re-estimated the Heckman models using only the two strongest instruments (total assets and total liabilities), excluding the weaker predictors (age of plant and days cash on hand). We also tested whether state-level COVID-19 incidence (cumulative confirmed cases) was associated with bond market outcomes. Because COVID incidence is mechanically zero in the pre-pandemic period, including it as a regressor is equivalent to interacting it with the post-COVID indicator; the coefficient captures the differential effect of local pandemic severity on bond market outcomes. Supplementary information Supplementary Information Supplementary Information