What enrollment patterns are required for premiums to meet or exceed claims paid on the health insurance exchanges (HIEs), and can we reasonably expect these patterns to emerge?

We showed in our last note that, before subsidies and penalties, average gross premiums for a given level of deductible will rise in the market for non-group health insurance in 2014

On a net basis (adding in the effect of subsidies and penalties), for some market participants with lower incomes a given level of coverage nevertheless becomes more affordable. For others with incomes at the top of, or just above the ranges of subsidy eligibility, a given level of coverage becomes less affordable. For still others with incomes well above the subsidy eligible ranges, because of penalties coverage becomes more affordable[1] in and beyond 2014

Now that we know what various age / income groups are paying for individual coverage, and what they can generally expect in terms of the amount of coverage gained for a given net premium, the immediate question becomes whether the age and health adjusted mix of enrollees is sufficient to ‘balance’ the market, i.e. whether premiums paid by health insurance exchange (HIE) enrollees will meet or exceed claims paid on behalf of HIE enrollees

We framed our analysis in simple terms of price and value. Specifically, we compared the price a given age / income group of beneficiaries would pay for coverage (premiums, net of subsidy and penalty effects) to the value received (coverage of allowable health costs above a given threshold). For any given age / income group the price of coverage is the same, but because health costs are highly varied within an age / income group, the value of coverage also is highly varied[2]

For example, if a person buys coverage but has no health costs, then (ignoring the value of reduced risks) the policy has a negative value equal to the net premiums paid. Very simply, you paid for coverage but didn’t use any of it. Alternatively, if a person buys coverage and has covered health costs equal to the sum of net premiums paid plus the plan’s deductible[3], the coverage has zero value – you paid the premiums, and the plan gave the money back to you in the form of paid claims equal to the net premiums you paid. Finally, if a person buys coverage and has covered health costs in excess of the net premiums paid, the coverage has a positive value equal to the difference between total health costs and the sum of net premiums and deductibles, because the amount of health costs paid by the plan exceeds the net premiums you paid to be covered

To estimate who purchases coverage, we divide the candidate pool[4] of beneficiaries into three groups: those to whom coverage is essentially free, those to whom coverage is not free and whose covered health costs are likely to exceed their net premiums, and those to whom coverage is not free and whose covered health costs are not likely to exceed their net premiums

We assumed everyone who chooses coverage buys the least expensive bronze plan available to them. Because individual plans in 2014 all have out of pocket maximums no higher than $6,350, the range of out of pocket maxima in the individual market has been compressed, making these limits fairly similar across plans of all metal tiers. It follows that, all else equal (e.g. ignoring narrow networks in cheaper plans) the optimal plan choice for most persons – including those with very high expected costs — is likely to be the bronze plan[5]. We also assumed that penalties for not having coverage, which rise gradually from 2014 to 2016, do not rise gradually but are instead immediately at their maximum statutory values in 2014. This has the effect of lowering the marginal cost of buying coverage on the health insurance exchanges (HIEs), which maximizes enrollment, and thus gives the simulation the best chance of finding that premiums collected meet or exceed claims paid

Group 1: We assume that persons who get coverage for free accept the coverage – regardless of whether they think they need it. We defined ‘free’ as net premium costs at or below 1pct of income. For reference, Exhibit 1 shows the distribution of net premium costs as a percent of income for various age groups. Group 1 contains about 21.3 million persons, distributed by age as shown in Exhibit 2. Because coverage is free, we assume the decision to enroll is made without regard to whether Group 1 beneficiaries anticipate having health costs. Because of this, we expect the persons in Group 1 will have the same distribution of health costs (on an age and income adjusted basis) as those in the general population. By extension, this implies Group 1 will pay (or have paid on their behalf) more in premiums than is claimed in covered health costs. In the interest of a conservative analysis, we assume all Group 1 members enroll, as this assumption maximizes the likelihood the market will ‘balance’, i.e. that HIE premiums will meet or exceed claims

Group 2: For those whose net premium costs exceed 1pct of income, we assume the decision to buy coverage is based on an expectation that covered health costs are likely to exceed net premiums paid. We recognize this bends the traditional ‘random risk’ basis of an insurance market, but in the context of health insurance we believe this is reasonable. Some health costs are random, but a very large proportion (we believe a majority) of health costs are not random: the presence or absence of insurance influences total costs; and, a large percentage of persons with significant health costs are able to anticipate these costs. Also, the structure of non-group insurance markets under the ACA ‘enables’ (by mandating coverage of pre-existing conditions and eliminating medical underwriting) our assumption that beneficiaries will be more likely to enroll if they anticipate high health costs than if they do not[6].

Using the distribution of health costs within each age group (Exhibits 3a, 3b), we assume that all persons with costs above a threshold (net premiums plus deductible) purchase coverage. By definition, persons in Group 2 will have covered health costs that exceed their premiums. We estimate 2.0 million persons in Group 2, distributed by age as shown in Exhibit 4

If we stop here, the combined market of Groups 1 and 2 is clearly imbalanced – despite the fact Group 1 has more than 10 times the enrollees of Group 2, the excess of premiums over costs in Group 1 is insufficient to cover the excess of costs over premiums for Group 2 (Exhibit 5)

Group 3: This leaves a third and final set of potential enrollees – upon whom the market is wholly reliant for ‘balance’, i.e. for premiums to meet or exceed claims paid. Persons in this third potential set of enrollees are those for whom net premium costs are non-trivial (≥ 1pct of income), and whose ‘expected’ health costs are too low to result in a powerful expectation that claim payments will exceed net premiums. The only reason we should expect members of Group 3 to enroll is that they have some level of risk aversion – i.e. they don’t expect their costs to exceed the threshold in the plan year, but prefer paying premiums over accepting the risk of unexpected health costs

Within Group 3 different age groups should – and apparently do – have differing levels of risk aversion. Objectively, we define the risk premium these enrollees pay as the expected value of the difference between net premiums and claims paid. The greater the risk premium, the less likely it is to be paid. The risk premium varies by age, such that the average risk premium paid by the young (21-29 y.o.) is about twice the average risk premium paid by those aged 64. The risk premium changes by age in linear manner, such that if we index the relative likelihood of paying the risk premium by age to 1.0 for age 64, we get the linear relationship shown in Exhibit 6 (grey columns). Alternatively, we can estimate risk appetites subjectively, by simply examining respondents’ self-characterization of risk aversion by age group[7]. Again indexing to 1.0 at age 64, we find that the young are about half as risk averse as the old, but that the relationship is logarithmic from young to old – i.e. the young are very risk tolerant, but this risk tolerance fades rapidly as the young reach adulthood (Exhibit 6, again, green columns)

For the purposes of scaling enrollment by age in Group 3, we use this second, logarithmic curve of risk tolerance by age – both because we believe it’s more reflective of true risk aversion, and because it results in a more conservative estimate (i.e. that overall enrollment of Group 3 will be greater, and that the average age or Group 3 enrollees will be lower)

We then determine how many Group 3 members would have to enroll in order for the market to be ‘balanced’, i.e. to avoid adverse selection. Specifically, we ‘titrate’ Group 3 potential enrollees into the market in proportion to their relative risk aversion[8], until expected premiums paid in the overall market equal expected claims paid in the overall market

Using this approach, we find that in aggregate 37pct of Group 3’s potential beneficiaries must enroll in order for gross premiums to match claims paid (Exhibit 7). Exhibit 8 illustrates the age mix of the ‘balanced’ market; the grey columns illustrate the percentage of total enrollees (left y-axis) by age category (x-axis). The solid green line illustrates the percentage of all potential enrollees in a given age category that are enrolled, regardless of whether they are in Group 1 (coverage is free), Group 2 (expected claims exceed net premiums), or Group 3 (net premiums exceed expected claims). The dashed green line represents the percentage of all Group 3 potential enrollees that actually enroll, by age category. Note that younger participants are a larger percentage of total enrollees (grey columns) but have lower rates of participation (green lines); this occurs because the younger age categories account for a disproportionately large share of potential enrollees

Realistically however, matching premiums collected and claims paid is insufficient; premiums collected will have to exceed claims paid by a margin sufficient to cover the operating costs of the insurers. If we instead seek to force the market to a 92 medical loss ratio (MLR, ratio of claims to premiums), consistent with recent observations from non-profit plans in the Massachusetts market, we have to push the Group 3 enrollment rate to 64pct. Note that to achieve a 92MLR the necessary Group 3 enrollment rate among potential enrollees in the 21-29 y.o. age category rises to 43pct ( Exhibit 9). Notably, we cannot push the market to an 85MLR; at 100pct enrollment of all Group 3 age categories, the expected MLR is 85.5

Summary, Risks and Conclusions

We see almost no likelihood that health plans sold on the HIEs will meet the prevailing +/- 85MLR standard of for-profit health plans. And, we believe the HIEs are more likely than not to fall short of even the prevailing +/- 92MLR of non-profit health plans

Our model assumes that everyone for whom coverage is essentially free (net premiums < 1pct of income) will enroll – this is 21.3 million persons, with an average MLR of around 85, meaning the ‘free’ group provides a large absolute dollar amount of premiums in excess of health costs. In reality, not everyone for whom coverage costs less than 1pct of income will enroll; and, those that do are on average likely to have higher health costs than those that do not. Thus the contribution of the ‘free’ group to balancing the overall market is likely to be smaller than what we’ve modeled

Our model also assumes that everyone buys bronze plans – including those below age 30, who are eligible to buy cheaper catastrophic plans. If a large percentage of the young who enroll do so under catastrophic coverage – as is highly likely — the excess premiums from this group will be smaller than we’ve modeled

The only group that we can rely on to do almost exactly what we’ve assumed are those whose health costs are likely to exceed their net premiums (Group 2). Thus on net, our structuring of the three groups of potential enrollees, and our inferences regarding their behavior, appear more likely to underestimate per-beneficiary costs and to overestimate per-beneficiary premiums paid

There are two further risks to our method that could lead us to over-estimate the likelihood of adverse selection. By assuming enrollment in the cheapest bronze plan, we could be assuming that the prices offered by smaller carriers with limited enrollment capacity apply to the entire market. To control for this, rather than using the single lowest bronze premium in a given area, we’ve used an average of the lowest bronze premiums in each area. Separately, we use total health cost data from the Medical Expenditure Panel Survey (MEPS) database, and our method inherently assumes that these health cost values apply to costs that would fall within the scope of coverage for plans sold on the HIEs. We believe this is realistic, particularly given the expansion of scope that occurs in 2014, but we cannot rule out that covered health costs are smaller than the MEPS value of health costs we used for modeling, which would result in a greater likelihood of premiums covering health costs

On net, our view is that the model is a conservative estimate of adverse selection pressures on the HIEs, and that our findings point to a very real likelihood of premiums being insufficient to cover both claims costs and the operating costs of insurers offering plans on the HIEs

It follows that we expect significant premium inflation in and beyond 2015, and further expect that the legislation and/or regulations governing the HIEs will have to be modified if the HIEs are to remain viable. Policy options include: allowing greater differences in premiums based on age, allowing significantly higher out-of-pocket maximums, reducing the scope of benefits contained within the definition of minimum essential coverage, increasing subsidies, and/or increasing penalties

Addendum: Evidence that persons with high potential costs are currently excluded from the non-group market

We established an apparent price elasticity of demand for non-group health insurance in 2010 (by age and income), to use as a metric for estimating enrollment by age and income group in 2014. Ultimately we chose not to rely on this approach for several reasons. Broadly speaking, the 2010 market for non-group insurance is like a credit market with over-tight lending standards – the more you need health insurance (credit) the less likely you are to get it. In the ‘pre-ACA’ non-group market, 18 percent of applicants were rejected – presumably because of pre-existing conditions / poor health. The effect of these rejections on skewing who’s covered is enormous — 20.7pct of persons[9] with a body mass index (BMI) of <25 are covered, as compared to only 13.2pct of persons with BMI’s >=25 (Exhibit 10). Higher BMI’s correlate with poorer health, i.e. it’s clear that less healthy persons have in the recent past been excluded from the non-group market. In contrast, the 2014 non-group health insurance market (essentially the health insurance exchanges, or ‘HIEs’) looks like a credit market with too-liberal lending standards – anyone can get a policy at a standardized price, even when it’s clear receipts (premiums) from a specific applicant won’t cover outlays (claims) to that specific applicant

Convinced that measured price elasticities in 2010 wouldn’t translate to 2014, we shifted to the price / value framework used as the primary basis for this resource note. Nevertheless the evidence that 18 percent of applicants are denied in the ‘current’ market – paired with the presumption that these applicants arguably have higher average medical costs than those currently enrolled – further indicates that the non-group market in and beyond 2014 is likely to suffer adverse selection