Health Services Research
Volume 58, Issue 5 p. 1077-1088
RESEARCH ARTICLE
Open Access

Effects of state reinsurance programs on health insurance exchange premiums and insurer participation

Onyinye Oyeka PhD, 

Corresponding Author

Onyinye Oyeka PhD

Department of Health Management and Policy, University of Iowa, Iowa City, Iowa, USA

Correspondence

Onyinye Oyeka, Department of Health Management and Policy, College of Public Health, University of Iowa, 145 N. Riverside Drive, Iowa City, IA 52242-2007, USA.

Email: oioyeka@uiowa.edu

Search for more papers by this author
George L. Wehby PhD, 

George L. Wehby PhD

Department of Health Management and Policy, University of Iowa, Iowa City, Iowa, USA

Department of Economics, University of Iowa, Iowa City, Iowa, USA

Department of Preventive & Community Dentistry, University of Iowa, Iowa City, Iowa, USA

Public Policy Center, University of Iowa, Iowa City, Iowa, USA

National Bureau of Economic Research, Cambridge, Massachusetts, USA

Search for more papers by this author
First published: 24 July 2023
https://doi.org/10.1111/1475-6773.14205

Abstract

Objective

The aim of the study was to estimate the effect of the state-based reinsurance programs through the section 1332 State Innovation Waivers on health insurance marketplace premiums and insurer participation.

Data Source

2015 to 2022 Robert Wood Johnson Foundation Health Insurance Exchange Compare Datasets.

Study Design

An event study difference-in-differences (DD) model separately for each year of implementation and a synthetic control method (SCM) are used to estimate year-by-year effects following program implementation.

Data Collection/Extraction Methods

Not applicable.

Principal Findings

Reinsurance programs were associated with a decline in premiums in the first year of implementation by 10%–13%, 5%–19%, and 11%–17% for bronze, silver, and gold plans (p < 0.05). There is a trend of sustained declines especially for states that implemented their programs in 2019 and 2020. The SCM analyses suggest some effect heterogeneity across states but also premium declines across most states. There is no evidence that reinsurance programs affected insurer participation.

Conclusion

State-based reinsurance programs have the potential to improve the affordability of health insurance coverage. However, reinsurance programs do not appear to have had an effect on insurer participation, highlighting the need for policy makers to consider complementary strategies to encourage insurer participation.

What is known on this topic?

  • The Affordable Care Act (ACA) has expanded access to coverage; however, affordability of coverage continues to be an issue.
  • Some states have employed ACA section 1332 State Innovation Waivers to implement a state-based reinsurance program in efforts to stabilize their individual health insurance programs, both in and outside the marketplace.
  • The evidence on effects of these state-based reinsurance programs on premiums and insurer participation is limited.

What this study adds?

  • The study provides new evidence on effects of state-based reinsurance programs on health insurance marketplace premiums and insurer participation.
  • State-based reinsurance programs between 2018 and 2021 are associated with meaningful declines in premiums.
  • There is no evidence of an effect on insurer participation.

1 INTRODUCTION

The primary goal of the Affordable Care Act (ACA) was to expand access to affordable insurance coverage. Through a set of health insurance reforms, the ACA has expanded access to coverage; however, affordability of coverage continues to be an issue. Premiums in the health insurance exchanges (HIXs) have increased each year from 2014 to 2018.1 Average premiums for the benchmark plans increased by 2% in 2015 reaching an increase of 37% by 2018, with considerable variation among states.2-6 In contrast, average benchmark premiums declined by 2.2% annually between 2019 and 2022.7 However, benchmark premiums increased by an average of 3.4% in 2023, largely due to inflation, increased health care utilization, higher health care prices, and uncertainty regarding subsidy extension under the American Rescue Plan (ARP) Act of 2021.8 One of the factors contributing to the rise in premiums is insurers exiting the HIX as a result of actuarial and political uncertainties. In the early years of the ACA, insurers exited the HIX because they experienced financial losses due to inaccuracies predicting enrollee costs, elimination of the risk corridor program, and the end of the ACA transitional reinsurance program.9-11 Insurer exit from the HIX led to reduced competition and caused the remaining insurers to increase premiums.12 Political uncertainties regarding repealing all or some parts of the ACA such as the individual mandate and enacting a replacement, termination of the cost-sharing reduction (CSR) payments,13 actual elimination of the federal individual mandate,14 and sale of noncompliant ACA plans also resulted in insurers increasing premiums.15, 16

While most marketplace enrollees receive subsidies that may shield against premium hikes, unsubsidized enrollees and those who purchase coverage off-marketplace face the total cost of coverage.4 Rising premiums have particularly reduced the enrollment of individuals not eligible for subsidies. Specifically, enrollment data from the Centers for Medicare & Medicaid Services (CMS) indicate a 45% decline in unsubsidized enrollment decline among unsubsidized enrollees between 2016 and 2019; in contrast, subsidized enrollment increased slightly during the same period.17 Even among subsidy-eligible enrollees, younger and moderate-income enrollees faced relatively higher after-subsidy premiums with rising premiums.18-20

In efforts to stabilize the HIX, some states have employed the ACA section 1332 waivers21 (also referred to as the State Innovation Waivers) to implement a state-based reinsurance program. These waivers became available to states beginning in 2017.22 To date, 17 states have implemented reinsurance programs between 2017 and 2021. Reinsurance reduces the financial risk to insurers from covering costlier enrollees. Reinsurance works by reimbursing insurers who enroll high-cost enrollees some portion of enrollees' medical expenditures when it reaches a prespecified threshold (the attachment point) up to the point at which no more reinsurance payments are made (reinsurance cap). The ACA included a temporary federal reinsurance program that was in effect from 2014 to 20169 and has been credited for keeping premiums low and encouraging insurer participation in the HIX.23 Evidence of its effect on HIX premiums suggests that the program reduced premium rates by a range of 6%–14% on average.24, 25 Because the reinsurance program protects insurers from catastrophic expenditures, it may incentivize insurers to charge lower premiums, encourage entry into the HIX, and expand health plan offerings into new markets.21, 26 As insurer participation increases, competition in the HIX increases which may also lower premiums.12, 27 Despite its theoretical appeal, the empirical evidence on the effectiveness of state reinsurance programs is slim. In this paper, we provide evidence on the effects of state reinsurance programs on HIX premiums and insurer participation.

States have autonomy on how to structure and fund their reinsurance programs. States decide the program type, attachment point, coinsurance rate, and reinsurance cap. Most state reinsurance programs, except for Alaska and Maine, follow a claims cost-based approach that reimburses insurers a percentage of high-cost claims that exceed a specified threshold. Alaska employs a conditions-based approach that reimburses insurers the cost for specific health conditions, and Maine utilizes a hybrid approach that reimburses claims costs for specific health conditions.21

Emerging evidence on the reinsurance programs in early adopting states-Alaska, Minnesota, and Oregon, from an evaluation by the CMS shows mixed effects. Marketplace premiums declined by 26%–37% and by 22%–34% in Alaska and Minnesota, respectively, across metal levels. However, in Oregon, even though premiums declined following implementation of the reinsurance program, estimates were not statistically significant, so the researchers were unable to attribute premium decline to the reinsurance program.28, 29 Despite its importance, the generalizability to other states that have implemented reinsurance programs is unclear considering potential heterogeneity in program design and local insurance market factors that could modify the program effects. Outside of the HIX, reinsurance has been federally implemented in Medicare Part D, with growing evidence that it has led to competitive markets and stable premiums.30-32

This study examines the effects of the reinsurance programs on premiums across states that have implemented a reinsurance program through 2021. We also examine the effects of the reinsurance programs on insurer participation. In doing so, the study contributes to the growing literature on effects of state implemented reinsurance programs in the HIX.

For setting premium rates in a given state, the ACA requires insurers to consider all enrollees in ACA-compliant plans offered by the insurer inside and outside the HIX in the state as members of a single risk pool.33 State-based reinsurance programs allow states to waive the single risk pool requirement, which permits insurers to consider the reinsurance payments when setting market-wide index rates, possibly leading to a decline in premiums or slowing the growth of premiums. However, the change in premiums will also depend on other local market conditions, such as market competition, provider networks, and profitability.15, 27, 34 Therefore, it is important to evaluate potential heterogeneity in the reinsurance effects across states and year of program implementation. Further, the states' decisions to implement the reinsurance program may indirectly affect insurer participation. Because reinsurance reduces insurers' exposure to risk, more insurers may be incentivized to enter the HIX. Reinsurance may also encourage insurers already participating in the HIX to expand into counties with relatively riskier health pools. However, the decision to enter the HIX or expand into more geographic areas may be tempered by market competition, profitability, perceived health status of the risk pool, and provider network,35, 36 further highlighting the need for empirical evaluation considering potential heterogeneity across states and time.

2 METHODS

2.1 Data

The primary data source is the publicly available Robert Wood Johnson Foundation (RWJF) HIX Compare Datasets from 2015 to 2022. The RWJF HIX Compare datasets include information on all individual and small group fully insured plans in the HIX for all 50 states and the District of Columbia. The datasets include detailed information on monthly premiums (for a child aged 0–14; an individual aged 30 and two children aged 0–14; and individuals aged 27 and 50) across plans, metal levels (Bronze, Catastrophic, Gold, Platinum, Standard Silver, and Silver Variants) in each year, geographic rating area (GRA), and state. Because insurers do not always offer plans across an entire GRA, the RWJF HIX Compare dataset provides information on plans offered on-and off-marketplace in each year and county. The data also contain a county-GRA crosswalk.

2.2 Outcome variables

The primary outcomes are monthly premiums for individual coverage for a 27-year-old in bronze, silver, and gold plans at the GRA level. We focus on this age because, in most states, premiums for individuals at other ages are the same multiple of the premium for the 27-year-old, making analyses for other ages unnecessary. Catastrophic and platinum plans are excluded because they account for a small number of plans offered in the HIX and are not offered in all GRAs and years. Another outcome is the number of insurers participating in the county (since insurers may choose to participate in some counties and not others in a GRA).

2.3 Statistical analysis

To estimate the effects of the 1332 waiver reinsurance program, we employ two designs, an event study difference-in-differences (DD) model and the synthetic control method (SCM).37 Because states implemented their reinsurance programs at different times, estimating DD models that compare changes in outcomes (monthly premiums and insurer participation) in the treatment states to outcome changes in the control states may be biased in the presence of heterogeneous effects by timing of implementation. To circumvent this issue, we estimate separate DD models by year of implementation (Appendix Table A1, supporting information lists reinsurance states and their effective dates), specifically for each group of states that implemented their programs in 2018 (Alaska, Minnesota, and Oregon), 2019 (Maine, Maryland, New Jersey, and Wisconsin), 2020 (Colorado, Delaware, Montana, North Dakota, and Rhode Island), and 2021 (New Hampshire and Pennsylvania). Idaho and Virginia have also received approvals to implement a reinsurance program, but their programs were not effective during the study period, so we include them in the control groups. The control group comprises only the 35 states that have not implemented a reinsurance program by 2022. For each reinsurance state group defined by the implementation year, states that implemented reinsurance in other years are excluded from the model. For example, for states that implemented their reinsurance in 2018, only those states and the control states as defined earlier are included in the model for 2018 implemented reinsurance programs; states that implemented in other years are excluded from that model and so on for the separate models for states expanding in subsequent years.

2.4 Premiums

Using the 2015–2022 RWJF HIX Compare datasets, we estimate the effect of state reinsurance programs on monthly premiums at the GRA level. For each group defined by year of implementation, we estimate the reinsurance effects year by year after implementation compared with the year before implementation as the reference year. We also examine pre-trends in outcomes during the period before implementation in the same regression model. We estimate the following DD regression:
Log Y gst = α 0 + ∑ t ≠ − 1 T β Reinsurance s * Year t + X gst + State s + Year t + ε st , (1)
where Y gst represents the log monthly premiums for a 27-year-old across metal levels in GRA g at time t. The primary model uses log-transformed premiums because premiums are right skewed, and the log transformation provides percentage estimates relative to premium level. As a robustness check, we also estimate the effect of the reinsurance program on the original (not logged) monthly premiums. Reinsurance s is a binary indicator equal to 1 for states that have implemented the reinsurance program for a given period (for example, Alaska, Minnesota, and Oregon for 2018 reinsurance states) and 0 for the control states that have not implemented a reinsurance program throughout the study period. Year t is a vector of Year fixed effects with the year before the implementation as the reference category. The model includes the following time-varying GRA or state-level covariates in X gst : demographic variables (percent of the population under 18-year-old, above 65-year-old, uninsured, and below poverty level), economic variables (median household income and unemployment rate), and population health status indicators (percent of the population in poor physical and mental health, and rates of smoking, diabetes, and fair or poor health) to account for the size, sociodemographic, and risk profile of the local market. We also control for state-level covariates, state-level health care supply variables (primary care physician, specialty physician, and hospital beds per 10,000 population) to control for potential competition on the supply side which might impact premiums or insurer entry, and policy covariates including whether the state expanded Medicaid, silver loading strategy and the state allowed the sale of short-term limited duration (STLD) health plans, the presence of Medicaid managed care organizations (MMCO), and the marketplace type (state-based marketplace [SBM] or federally-facilitated marketplace [FFM]) to control for legal and policy decisions at the state level that may influence insurers' decisions to enter or expand in the local market and set premiums. The control variables are derived from multiple sources (Appendix Table A2, supporting information provides the geographic level of the covariates and their data sources). Because GRAs may consist of single or multiple counties, variables are constructed by aggregating the data to the GRA level using the county-GRA crosswalk data provided by RWJF HIX Compare and county population as weights. State s and Year t are vectors of state and year fixed effects respectively to account for the time-invariant state heterogeneity as well as the national secular trend and common shocks related to the outcomes. In a robustness study, we exclude Minnesota from the model because Minnesota has a Basic Health Plan (BHP) program also known as MinnesotaCare that covers eligible enrollees with incomes up to 200% of the federal poverty level (FPL). The existence of the BHP program may influence the HIX and the state's decision to implement the reinsurance program in ways that may not be captured by our data.38, 39 We also estimate a model similar to Equation (1) that includes plan-level fixed effects to account for plan sample composition and also because including the plan fixed effects could explain more of the error term and reduce the estimates' variances. The model is estimated using ordinary least squares (OLS). Standard errors are clustered at the state level.

2.5 Insurer participation

For the analyses of insurer participation, we use the 2015–2022 RWJF HIX Compare data on insurers participating in the marketplace at the county level and estimate a model similar to Equation (1).
Y cst = α 0 + ∑ t T β Reinsurance s * Year t + X cst + State s + Year t + ε st , (2)
where Y cst represents the number of insurers participating in county c at time t; Reinsurance s and Year t are as defined above in Equation (1) and includes year fixed effects with the year of the implementation as the reference category. X cst represents time-varying county and state-level covariates similar to the covariates in the previous model. We also use the Urban Influence Code (UIC) to assign a county as either metropolitan or urban (UIC 1–2), micropolitan (UIC 3–8), and small and remote rural (UIC 9–12) to account for insurers' decisions to enter or expand into the market given the size of some these markets especially in rural areas. State s and Year t are vectors of state and year fixed effects. Equation (2) is estimated by OLS with standard errors clustered at the state level.

2.6 State-by-state analyses

Because reinsurance programs are regulated at the state level, and states decide how to structure their reinsurance program, including whether to employ a condition-based or traditional reinsurance model and how to fund their respective programs, and because states differ in other ways that modify the reinsurance market effect such as marketplace type, we implement a SCM developed by Abadie et al.37 to evaluate the reinsurance program impact separately for each treated state. SCM generates a counterfactual synthetic control by taking the weighted average of pretreatment outcomes from selected donor states from the control group that have not implemented a reinsurance program. The counterfactual synthetic control simulates what the outcome path for a treated state would be if it did not implement the program. Similar to the DD model, states that have implemented a reinsurance program do not contribute to the pool of donor states. The synthetic control states are selected based on their pre-reinsurance program trends in covariate and outcome values. The effect of the reinsurance program on the outcomes is the difference in outcomes between a treated state and its synthetic control states after program implementation. For inference, we generate a distribution of placebo estimates assuming each control state becomes treated, one at a time. The p-value of the estimated reinsurance effect is the percentile rank of the reinsurance state's estimate relative to the placebo estimates.

3 RESULTS

Appendix Figure B1, supporting information presents the descriptive trends in average premiums for a 27-year-old by metal tier in reinsurance states separately by year of program implementation and non-reinsurance states from 2015 to 2022. Overall, there is a decline in premiums across metal tiers after reinsurance enactment and in most cases more so than states that did not enact reinsurance. Appendix Figure B2, supporting information presents the descriptive trends in the number of insurers offering plans in the HIX in reinsurance and non-reinsurance states from 2015 to 2022. Overall, those trends indicate some increase in insurer participation for both reinsurance and control states, but this increase starts before the reinsurance implementation for states that started the program in 2019–2021, suggesting potential pre-trends for that outcome.

3.1 Premiums

Figures 1 and 2 presents the event study DD estimates of the reinsurance program effects on log premiums by metal level and program implementation year; detailed estimates are in Appendix Table C1, supporting information. Results that exclude Minnesota and plan-level fixed effects with and without log transformation are similar to the estimates in the main model (shown in Appendix: Figures C2 and C4, Tables C3 and C5, supporting information).

Details are in the caption following the image
FIGURE 1
Open in figure viewerPowerPoint
Event study estimates of the reinsurance program effect on health insurance marketplace premiums by metal level in states that implemented in 2018 and 2019.

Source: 2015–2022 premiums in the health insurance marketplace for a 27-year-old by metal level (bronze, silver, and gold) before application of the advanced premium tax credit (APTC) using the Robert Wood Johnson Foundation Health Insurance Exchange Compare dataset

. 2018 reinsurance states include Alaska, Minnesota, and Oregon; 2019 reinsurance states include Maine, Maryland, New Jersey, and Wisconsin. Regression models control for demographic variables (percent of the population under 18 years of age, above 65 years of age, uninsured, and below poverty level); economic variables (median household income and unemployment rate); population health status indicators (percent of the population in poor physical and mental health, and rates of smoking, diabetes, and fair or poor health); state-level health care supply variables (primary care physician, specialty physician, and hospital beds per 10,000 population); number of insurers in the marketplace; state policy covariates (Medicaid expansion status; silver loading strategy; sale of short-term limited duration [STLD] health plans; and the presence of Medicaid managed care organizations [MMCO]); and marketplace type (state-based marketplace [SBM] or federally facilitated marketplace [FFM]). State fixed effects included for all models and standard errors are clustered at the state-level. *p < 0.05; **p < 0.01; ***p < 0.001. [Color figure can be viewed at wileyonlinelibrary.com]
Details are in the caption following the image
FIGURE 2
Open in figure viewerPowerPoint
Event study estimates of the reinsurance program effect on health insurance marketplace premiums by metal level in states that implemented in 2020–2021.

Source: 2015–2022 premiums in the health insurance marketplace for a 27-year-old by metal level (bronze, silver, and gold) before application of the advanced premium tax credit (APTC) using the Robert Wood Johnson Foundation Health Insurance Exchange Compare dataset

. 2020 reinsurance states include Colorado, Delaware, Montana, North Dakota, and Rhode Island; and 2021 reinsurance states include Pennsylvania and New Hampshire. Regression models control for demographic variables (percent of the population under 18 year of age, above 65 years of age, uninsured, and below poverty level); economic variables (median household income and unemployment rate); and population health status indicators (percent of the population in poor physical and mental health, and rates of smoking, diabetes, and fair or poor health); state-level health care supply variables (primary care physician, specialty physician, and hospital beds per 10,000 population number of insurers in the marketplace); state policy covariates (Medicaid expansion status; silver loading strategy; sale of short-term limited duration [STLD] health plans; and the presence of Medicaid managed care organizations [MMCO]); and marketplace type (state-based marketplace [SBM] or federally facilitated marketplace [FFM]). State fixed effects included for all models and standard errors are clustered at the state level. *p < 0.05; **p < 0.01; ***p < 0.001. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 1 Panel 1 shows the estimates for states that implemented their reinsurance programs in 2018 (Alaska, Minnesota, and Oregon). Relative to 2017, reinsurance in those states was associated with a 10%, 19%, and 13% decline in premiums in the first implementation year (2018) for a 27-year-old bronze, silver, and gold plans, respectively (p < 0.05). Premiums were also lower in subsequent years than 2017. There was a pre-trend of faster rising premiums in reinsurance states before enactment (a trend opposite to that after implementation), suggesting that the reinsurance effect on lowering premiums could be larger than estimated (in absolute value).

Figure 1 Panel 2 shows the event study estimates for states that implemented their reinsurance programs in 2019 (Maine, Maryland, New Jersey, Wisconsin). Similar to the 2018 implemented programs, the reinsurance programs implemented in 2019 were associated with lower premiums, with a larger decline after 4 years (2021) than the first year of implementation. By 2022, there is a decline by 15%, 12%, and 11% in bronze, silver, and gold plans, respectively (all statistically significant at p < 0.05). Unlike the 2018 implemented programs, there is no evidence of pre-trends that would bias the post-implementation estimates.

Figure 2 Panel 1 shows the event study estimates for states that implemented their reinsurance programs in 2020 (Colorado, Delaware, Montana, North Dakota, and Rhode Island). Similar to the 2018 and 2019 implemented programs, there is a decline in the first year of implementation. The declines are also statistically significant in each of the 3 years compared with the year before implementation, although there is less of a trend of a growing decline compared with the 2019 implemented programs. By 2022, there is a decline in premiums by 20%, 16%, and 20% for bronze, silver, and gold plans, respectively. Similar to the 2019 implemented programs, there is no evidence of pre-trend differences that bias the reinsurance program effect estimates.

Finally, Figure 2 Panel 2 shows the estimates for states that implemented their reinsurance programs in 2021 (Pennsylvania and New Hampshire). For these states, there was a decline in premium rates for bronze plans in 2021 and 2022, but not for silver and gold plans. Compared with 2020, premiums were lower in 2022 by 6% for bronze; however, estimates are not statistically significant. There was a pre-trend of faster rising premiums in the earlier years before implementation but not so closer to implementation.

3.2 Insurer participation

Figure 3 presents the event study DD estimates of the reinsurance program effects on insurer participation by year of implementation. Detailed estimates are in Appendix Table F1, supporting information. There is no evidence of an effect for states that implemented their reinsurance programs in 2018 or 2019. For states that implemented in 2020 and 2021, there is a decline in insurer participation after implementation, but that decline appears to reflect a prior trend before. Taken as a whole, these results suggest no evidence of an effect on insurer participation.

Details are in the caption following the image
FIGURE 3
Open in figure viewerPowerPoint
Event study estimates of the reinsurance program effect on insurer participation in the health insurance marketplace, 2015–2022.

Source: Source is the 2015–2022 Robert Wood Johnson Foundation Health Insurance Exchange Compare Insurer by County report dataset

. 2018 reinsurance states include Alaska, Minnesota, and Oregon; 2019 reinsurance states include Maine, Maryland, New Jersey, and Wisconsin; 2020 reinsurance states include Colorado, Delaware, Montana, North Dakota, and Rhode Island; and 2021 reinsurance states include Pennsylvania and New Hampshire. Regression models control for demographic variables (percent of the population under 18 years of age, above 65 years of age, uninsured, and below poverty level); economic variables (median household income and unemployment rate); population health status indicators (percent of the population in poor physical and mental health, and rates of smoking, diabetes, and fair or poor health); state-level health care supply variables (primary care physician, specialty physician, and hospital beds per 10,000 population) number of insurers in the marketplace; state policy covariates (Medicaid expansion status; silver loading strategy; sale of short-term limited duration [STLD] health plans); marketplace type (state-based marketplace [SBM] or federally facilitated marketplace [FFM]); and Urban Influence Code (UIC) to assign a county as either urban, micropolitan, and small and remote rural areas. State fixed effects included for all models and standard errors are clustered at the state-level. *p < 0.05; **p < 0.01; ***p < 0.001. [Color figure can be viewed at wileyonlinelibrary.com]

3.3 Synthetic control estimates

Figure 4 summarizes the SCM estimates of the reinsurance program's effects on premiums averaging the estimates over all post-implementation years separately for each state; Appendix: Tables D1–D3, G1 and Figures E1–E4, supporting information show the detailed results and SCM graphs year by year after implementation. Pretreatment trends are similar for most states except Alaska, New Jersey (bronze plans), North Dakota, Rhode Island, and New Hampshire, so the SCM estimates for these states should be interpreted with caution. In most states, the SCM analyses indicate a decline in premiums following program implementation across metal types, although several point estimates are not statistically significant. Among declines that are statistically significant (p < 0.05), the range is between $16 in Wisconsin for a bronze plan and $143 in Minnesota for a silver plan. One exception is an increase in silver plan premiums in Pennsylvania because of a slower decline compared with the synthetic control (although not for the bronze and gold plans), which explains the DD estimates of no effect on silver plan premiums for the states that implemented in 2021 (Pennsylvania being one of those two states). Taken as a whole, the SCM analyses echo the findings from the DD estimates of a decline in premiums following implementation but highlight some potential heterogeneity in the magnitude of effects across states.

Details are in the caption following the image
FIGURE 4
Open in figure viewerPowerPoint
Health insurance marketplace premium average treatment effect by metal level using synthetic control states, 2015–2022.

Source: 2015–2022 premiums in the health insurance marketplace for a 27-year-old by metal level (Bronze, Silver, and Gold) before application of the advanced premium tax credit (APTC) using the Robert Wood Johnson Foundation Health Insurance Exchange Compare dataset.

Predictor variables include pretreatment premiums for each metal level. [Color figure can be viewed at wileyonlinelibrary.com]

Figure 5 summarizes the SCM estimates for the insurance program effects on insurer participation for each state averaging across post-implementation years with detailed SCM results in Appendix: Figure F2 and Tables F3 and G2, supporting information. With some exceptions (Wisconsin, New Jersey, Delaware, North Dakota, and Pennsylvania), the pre-trends were overall well aligned between the treatment states and their synthetic controls. Overall, there is no evidence of widespread effects on insurer participation from the SCM analyses and most estimates are relatively small and not statistically significant.

Details are in the caption following the image
FIGURE 5
Open in figure viewerPowerPoint
Insurer participation average treatment effect using synthetic control states, 2015–2022.

Source: Source is the 2015–2022 Robert Wood Johnson Foundation Health Insurance Exchange Compare Insurer aggregated at the state level.

Predictor variables include pretreatment insurer participation. [Color figure can be viewed at wileyonlinelibrary.com]

4 DISCUSSION

This study examines the effects of state-based reinsurance programs through the ACA section 1332 State Innovation Waivers on premiums and insurer participation in the HIX. Overall, there is evidence of a decline in HIX premiums beginning in the first year of the implementation of these reinsurance programs across metal levels. Moreover, there is a trend of sustained declines especially for states that implemented their programs in 2019 and 2020. At the same time, there is no evidence of an effect on insurer participation following these programs. The declines in premiums are meaningful, with the DD estimates for the first implementation year that are statistically significant ranging between 10% and 13%, 5% and 19%, and 11% and 17% for bronze, silver, and gold plans, respectively (p < 0.05). The state-by-state analyses suggest some heterogeneity across states but also premium declines across most states (although many estimates are not statistically significant in these state specific analyses). For example, among the earliest implementing states in 2018, there is a larger decline in Minnesota than Oregon (for which most estimates are not statistically significant) consistent with another study.29

The findings from this study suggest that states seeking to curtail and stabilize rising premiums and have not yet implemented a reinsurance program would benefit from doing so. Particularly important is the potential to increase access to affordable coverage for unsubsidized or not fully subsidized enrollees who receive limited financial assistance. However, there is significant cost to implement a state-based reinsurance program, which may prevent states with more constrained budgets from applying for the state innovation waivers.40, 41 Considering that, reconsidering a federal reinsurance program to stabilize and potentially lower premiums nationally, not only in specific states, seems an appropriate policy direction. In addition, a federal reinsurance program removes the uncertainty that states face considering their budgets whether to renew or extend their reinsurance program.21 Other alternatives are the implementation of a budget-neutral reinsurance program that applies a per capita amount predicated on projected reinsurance disbursements,41 or a state-operated reinsurance program financed by the federal government. In addition, the findings also show that there are differences in premium declines depending on the reinsurance implementation year. Premiums increased between 2015 and 20182-6 but moderated between 2019 and 2022, which is similar to what was once the 6-year underwriting cycle.7, 42, 43 Therefore, it is possible that earlier reinsurance programs may have been more impactful in countering the trends of rising premiums than later periods when marketplace premiums and conditions were somewhat stabilized.7 Our results generally support this premise. The results show that for late reinsurance adopters (Pennsylvania and New Hampshire), plan premiums did not change when compared with control states following implementation of the reinsurance program and the estimates are generally smaller. Therefore, it might be that earlier adoption was more effective, although other reinsurance program parameters other than timing may also play a role. At the same time, late reinsurance adopters may have also been looking ahead and taking into consideration the pandemic, potential increase in health care utilization, and uncertainties around ARP subsidies and how these factors may impact their health insurance market when deciding to adopt a reinsurance program.

Affordability of health care coverage continues to be an area of concern for individuals, families, and policy makers. It is also particularly important for unsubsidized enrollees who are ineligible to receive advanced premium tax credits to lower their coverage costs. The ARP has made coverage more affordable and available to marketplace enrollees by enhancing the amount of tax credits and extending eligibility to individuals who have incomes over 400% of the federal poverty level (FPL). Despite that expansion, however, a Kaiser Family Foundation survey poll in 2022 reported that 33% of insured adults are concerned about affording their monthly premiums.44 Because enrollees are sensitive to monthly premiums,45, 46 premium declines such as those associated with reinsurance programs in this study may play a critical role in decisions to obtain and maintain coverage.

This study includes 2 years into the COVID-19 pandemic that included employment and earnings loss, and also the ARP with expanded premium subsidies. These changes may have influenced insurer decision to participate in the HIX or enrollees' decisions to continue or drop coverage. To the extent that such effects were comparable between states with and without reinsurance programs, there may not be a change in effects of reinsurance programs. Our estimates, which are year-specific and cover the pandemic years separately, suggest that effects on lowering premiums are still observed among early reinsurance adopting states.

The Inflation Reduction Act of 2022 has extended the ARP subsidies through 2025, thus providing enrollees with continuous access to generous subsidies and preventing premium hikes.47-49 Whether and how the ARP would affect states' decisions to implement a reinsurance program is not clear a priori. On the one hand, enhancement and expansion of the ARP subsidies may reduce motivation for states to adopt a reinsurance program because the ARP subsidies may achieve the same goals and objectives for implementing a reinsurance program, which include improving affordability,50 increasing enrollment,51, 52 and encouraging insurer participation,8 all without the added cost of standing up a reinsurance program and the concern that reinsurance reduces the size of the premium tax credit for subsidized enrollees.53 On the other hand, because the number of enrollees eligible for subsidies has increased, the amount of pass-through dollars going to states have also increased, which increased state resources to run the reinsurance program or to direct additional resources to other innovations and initiatives such as a cost-sharing assistance program if allowed by legislation.54 However, the ARP subsidies are not permanent and may complicate states' decisions to either extend current reinsurance programs or apply for new state innovation waivers.55 So far, all states coming up on the end of their state innovation waivers have either extended their programs or put in applications to amend and extend.21 Idaho and Virginia have received approval to implement a new reinsurance program in 2023. Therefore, future assessment of the ARP effects on reinsurance take-up and on outcomes of previously enacted reinsurance programs is important.

In addition to the potential heterogeneity among states which our study explores, there are also potential differences in the reinsurance program effects across more local market conditions related to competition. Prior research suggests that increasing competition in the HIX helps to drive down premiums.12, 27, 56 Though reinsurance programs may contribute to lower premiums, the magnitude of premium reduction may vary across rating areas due to differences in market structure. For example, descriptive analyses of premium changes in the lowest silver plan in Oregon in 2017–2020 show significant variation, with premiums increasing in some rating areas but decreasing in others depending on the insurer and rating area.57 Therefore, it is important to examine these potential interactions with local insurance market structures in future research.

The lack of an effect from the reinsurance programs on insurer participation might reflect the multiple factors that insurers consider in their decisions to enter, reenter, expand, or exit the HIX beyond whether states have a reinsurance program or not. Insurers may especially consider local market conditions such competition, risk pool, and the provider market. Research suggests that network adequacy requirements and the health status of the risk pools are key factors for HIX entry.58, 59 Financial performance and the expected revenue from premiums may also impact entry and exit decisions.34 Therefore, policy makers may need to consider complementary strategies to encourage insurer participation and expand market competition and consumer choice specifically in small and remote rural areas by redefining the size of the local market.60, 61

This study is subject to limitations. The SCM creates a synthetic control group using pre-reinsurance program trends in covariate and outcome values; there may be other factors that we may not have accounted for in our analyses. Studies suggest that provider network and network adequacy play a role in insurers' decisions to enter or expand a market as it may affect their ability to negotiate agreeable payment rates and thus premiums,7, 15, 46 we are unable to account for this in our model due to data availability. However, it is important to note that these variables in our model could also be a potential outcome of the reinsurance program, in which case controlling for it (if data were available) might mask some of the full effect on premiums.

5 CONCLUSION

The state-based reinsurance programs created through the section 1332 State Innovation Waivers between 2018 and 2021 are associated with a decline in premiums across metal levels. Therefore, they show a potential to improve the affordability of and access to health insurance coverage. At the same time, there appears to be no effect of reinsurance programs on insurer participation. Future research should continue to examine reinsurance effects as more states obtain approvals to implement these programs and to evaluate the potential heterogeneity in their effects across local market conditions.

ACKNOWLEDGMENTS

The authors would like to thank two anonymous reviewers and participants in seminars in the department of health management and policy at the University of Iowa for their helpful comments and suggestions.

    FUNDING INFORMATION

    No funding was received for this study.

    CONFLICT OF INTEREST STATEMENT

    The authors declare no conflicts of interest.

      The full text of this article hosted at iucr.org is unavailable due to technical difficulties.