Health insurance in the United States is increasingly delivered through markets where the government subsidizes enrollees to choose among privately provided health insurance plans (Gruber, 2017). Fifty-four million Medicaid beneficiaries in managed care plans, for instance, select a health plan from a set of choices offered by private insurers (Hinton et al., 2020). Premiums for these plans are paid by the government and are typically set at zero dollars for the enrollee. Medicare Advantage enrollees also select subsidized private health plans, and they often have several zero-premium plan options (Douven et al., 2020). The Health Insurance Marketplaces created by the Affordable Care Act (ACA) work similarly, though zero-dollar premium plans are less common (Branham and DeLeire, 2019). Despite the similarities between Medicaid managed care and the Marketplaces, take-up of subsidized Marketplace coverage is at least 36 percentage points lower than Medicaid take-up (McDermott et al., 2021, Decker et al., 2022, Bhanja et al., 2021). Can coverage take-up be improved by fully subsidizing premiums?

In this paper, we show that setting premiums to zero, rather than at a low but slightly positive level, increases coverage duration but not enrollment. Our analysis, based on a regression discontinuity design and 2016–2019 administrative data from Colorado’s Marketplace, exploits discontinuities in the availability of zero-premium plans. Our primary finding is that people who are just eligible to sign up for a zero-premium plan end up being covered for 2.3 to 4.8 days more than people who are just ineligible. Scaling this estimate by the take-up rate of zero-premium plans, we find that enrolling in a zero-premium plan increases coverage by 29–47 days. This extra duration is driven by lower-income enrollees and comes from both reduced drop-out and, especially, increased early coverage start dates. The main mechanism responsible for the increased duration in zero-premium plans is that enrollees in positive-premium plans must make a small initial payment after signing up for coverage. The required payment creates a transaction cost which zero-premium plans eliminate. Overall, our results suggest that transaction costs can be a meaningful barrier to coverage and benefit take-up, especially for lower-income households.

Our context is the Health Insurance Marketplaces, which have had zero-premium plans since their implementation in 2014. For eligible households, premiums in the Marketplaces are subsidized by premium tax credits (PTC). Premium tax credit generosity varies across households and is set so that the net premium an eligible household pays for a “benchmark” plan is equal to a sliding-scale percentage of its income. If a household’s PTC exceeds the premium of the lowest-premium plan available to them, then the household is able to purchase a zero-premium plan (Branham and DeLeire, 2019, Drake and Anderson, 2020).1 Zero-premium plan availability has expanded greatly since 2014. With the passage of the American Rescue Plan Act of 2021, roughly 12 million Americans had the option to purchase a zero-premium Marketplace plan in 2021 (Rae et al., 2021, Branham et al., 2021).

Most enrollees do not select a health plan until or shortly before December 15th, the deadline for January 1st coverage. Applicants that selected plans with non-zero premiums must then make an initial premium payment by December 31st to begin coverage on January 1st. For the 36 states that used the federal healthcare.gov Marketplace from 2017 through 2021, failure to make this initial payment by December 31st meant that applicants were ineligible to purchase subsidized coverage until the following year (Anderson et al., 2019). The Biden Administration reversed this policy for 2022 coverage in healthcare.gov states, allowing applicants a “second chance” to obtain February 1st coverage by signing up by January 15th and making an initial premium payment by January 31st. Colorado, which does not use healthcare.gov, has maintained at least a January 15th enrollment deadline since 2014.

The transaction costs of making the initial premium payment can be minimal for relatively higher-income applicants, who simply can make a credit card payment at the time of enrollment. This is not the case, however, for many lower-income eligible applicants, who often lack access to credit cards and are unbanked (Federal Deposit Insurance Corporation, 2020).2 These applicants must write a check or obtain a money order, mail it, and confirm that the insurer receives it over the last two weeks of the year. This process must be completed during a two-week period that is often highly financially stressful for low-income Americans (Swartz and Graves, 2014). That we only find coverage duration and not enrollment effects likely reflects Colorado’s policy of allowing applicants a second chance to enroll in January (i.e., Colorado provides applicants two chances to address these financial transaction costs).

Our analysis uses regression discontinuity (RD) designs and 2016–2019 administrative data from Colorado. The data include plan selections, plan availability, and, crucially, PTCs awarded for all enrollees. Our RD models leverage discontinuities in the availability of zero-premium plans to identify how zero-premium plan availability affects coverage choices. Our running variable is the premium of the lowest-premium plan available to an enrollee, less their awarded PTC. As a function of the running variable, post-subsidy premiums are continuous but kinked. However, when the running variable becomes negative, zero-premium plans become available, which is our key discontinuity. The running variable is a complex function of household income, size, and premiums available; it does not appear to be manipulated.

We investigate the effects of zero-premium plan availability and enrollment on Marketplace enrollment rates, plan choice, and coverage duration. Our primary finding is that enrollment durations are 2.3 to 4.8 days higher among enrollees just eligible for zero-premium plans. Scaling this estimate by the discontinuity in take-up of these plans, about eight percent, implies that enrollment in a zero-premium plan increases coverage by 29–47 days. This increased duration is driven entirely by enrollees with household incomes below 200% of the federal poverty level (FPL). Most of this duration effect comes from earlier start dates—enrollees are more likely to begin coverage on January 1st—although a sizable minority of the effect comes from reduced drop-out. We do not find economically significant responses on margins beyond duration. Enrollment in Marketplace coverage and generosity of selected plans are not responsive to zero-premium plan availability.

We argue that our results are most consistent with a financial transaction costs-based mechanism, and that these transaction costs create a welfare loss for a segment of the population. In particular, financial transaction costs explain our finding that zero-premium plans increase coverage through on-time starts and sign-ups. This finding is difficult to rationalize with other mechanisms. Psychological mechanisms appear to be less important in our context since they would imply plan choice effects, which we do not observe. Liquidity constraints are unlikely because our regression discontinuity design estimates the effect of zero-premium relative to $1 or lower. These transaction costs likely have substantial consequences for a share of enrollees. For the compliers—the eight percent of people who select a zero-premium plan when it is available—the presence of financial transaction costs reduce coverage by slightly over one month, causing them to lose several hundred dollars of health insurance subsidies which are obtainable at a premium of $1 or less.

Our results have several broader implications beyond Colorado’s Health Insurance Marketplace. Our finding of no effect of zero-premium plans on overall Marketplace enrollment is arguably surprising, and likely driven by the fact that Colorado’s open enrollment period extends into January. Enrollees that miss the December 31st payment deadline have another month to sign up for coverage. Policies that facilitate late enrollment may mitigate transactions costs. In most states, during the study period, however, enrollees who miss the payment deadline cannot sign up for subsidized coverage until the following year. Thus, zero-premium plans may increase enrollment as well as duration in other states, though we acknowledge that there are other institutional differences across states that make extrapolation difficult. More broadly, policies that ease transaction costs may be helpful in increasing take-up and reducing uninsurance, both of which directly improve health outcomes for middle-aged and lower-income populations (Goldin et al., 2021, Miller et al., 2021). Policies to reduce Marketplace transaction costs include longer open enrollment periods, grace periods for initial premium payments, and subsidies to reduce low premiums to zero.

Our findings contribute to two bodies of literature. Primarily, we contribute to literature examining the causes of low take-up of social insurance programs. A common explanation for this low take-up is that enrollment in social insurance programs often requires overcoming “ordeals”. Examples of ordeals include long wait times, extensive paperwork, application complexity, and opaque eligibility rules. Economists have long recognized that such ordeals likely reduce take-up (e.g., Currie, 2006). Since at least the work of Nichols and Zeckhauser (1982), however, economists have also recognized that ordeals can increase program efficiency by better targeting funds, because they can differentially screen out relatively less needy applicants, such as high-income applicants. Recent work has investigated the screening properties of these ordeals, finding that hassle costs are an important reason that benefits are left on the table, but with mixed results regarding whether ordeals increase or decrease targeting efficiency (Deshpande and Li, 2019, Finkelstein and Notowidigdo, 2019, Goldin et al., 2021, Homonoff and Somerville, 2020). Especially relevant is recent work by  Domurat et al. (2021) who conduct a field experiment in California’s Health Insurance Marketplace, providing information about deadlines and benefit eligibility. Consistent with our transaction cost mechanism, they find that providing information about sign-up deadlines increases coverage. However, they do not examine the transaction costs of making a premium payment. Our primary contribution to this literature, which has largely focused on government-supplied benefits, is to show that market provision of subsidized benefits can also produce an ordeal in the form of transaction costs. In our study context, this ordeal differentially screens out low-income beneficiaries, likely reducing targeting efficiency. The transaction cost that we document may also help explain the apparent low willingness-to-pay (and subsequent low take-up) for subsidized insurance coverage documented by  Finkelstein et al. (2019).

Our findings also contribute to literature focusing on zero-price effects in health care and health insurance. Ching et al. (2022) and Iizuka and Shigeoka (2021) show that demand for health care appears to be discontinuous at a price of zero, pointing to a zero-price effect. In the health insurance context,  Douven et al. (2019),  Newhouse and McGuire (2014), and  Stockley et al. (2014) argue that observed premium patterns are consistent with a zero-price effect, because many plans offer a premium of exactly zero, and survey respondents indicate a zero-premium preference in discrete choice experiments. They do not study actual insurance choices, however. Using aggregate data,  Drake and Anderson (2020) find an association between zero-premium plan availability and Marketplace enrollment in the states using the healthcare.gov platform. However, their ability to identify precise zeroes is limited by their use of aggregate data, and they do not examine other outcomes such as coverage duration or plan choice. Dague (2014) finds that when Wisconsin’s Medicaid program introduces a $10 premium, enrollees were much more likely to drop out.  McIntyre et al. (2021) find that a policy in Massachusetts of automatically switching enrollees to zero-premium plans (when available) if they default on their premiums substantially reduces drop-out.

We make two contributions relative to this literature. First, we investigate zero-premium effects on enrollment, coverage duration, and plan choice, which the previous literature has not investigated. Our findings show that a large share of zero-premium plans’ overall effect comes from earlier start dates, not only reduced drop-out. Second, we show that transaction costs, rather than a literal psychological bias towards zero, likely explain the zero-price effects we document. This distinction is important for policymakers and price setters because transaction costs can be eliminated with alternative payment rules. We show, at least in our context, that transaction costs account for zero-price effects.3