By April 2010, it became clear that existing actuarial models that had been used before enactment of the CLASS Act (both those already relied on by HHS and those being developed by outside groups such as Boston College) would be insufficient to provide CLASS estimates and new models would have to be developed. Actuarial modeling of the CLASS program was undertaken by staff in ASPE, and reviewed by the CLASS Office. The model development and modeling were largely supported through a long-standing contract between Actuarial Research Corporation (ARC) and ASPE, and a new contract with AvalereHealth that began in September 2010. The rationale for developing two models was to compare premiums and other program dynamics using different methodological approaches and data and to assess the sensitivity of results to varying model assumptions. This is standard practice in the insurance industry when developing new products. Further, the ARC model does not include Medicaid offset estimates, while the Avalere model does. The key economic and demographic/actuarial assumptions are largely the same. This section briefly describes the two actuarial models, and model development and estimation across three phases: early model development and estimation; model refinement and development of preliminary benefit options; and final model development and estimation.
Both models adopted conservative assumptions that would tend to produce higher premiums and lower take-up rates than the best existing empirical evidence might suggest. For example, in modeling adverse selection it was assumed that potential enrollees would sort themselves perfectly by health status and join CLASS in reverse order (the most disabled first). The models used conservative assumptions because the existing empirical evidence is relatively sparse and there is great uncertainty around the existing estimates.
Summary of Actuarial Research CorporationS Long-Term Care Premium Model
The ARC Long-Term Care Premium Model is designed to calculate long-term care insurance premiums for a government-operated, self-financing program and to project cash flow to assist policymakers in understanding program dynamics (see Appendix L for an in-depth description of the model). It can model various CLASS benefit structures under user-selected assumptions related to: program options, economic and demographic/actuarial assumptions (including antiselection/adverse selection), and long-term care utilization. The latter two sets of assumptions do not vary according to the program options, but are parameters used in the formulas to calculate premiums. Input and output are in Microsoft Excel Worksheets with program calculations performed in Visual Basic for Applications (VBA). The computer code underlying these calculations can be viewed by simply opening Excels Visual Basic Editor.
The key program options that can be modeled include alternative formulations of the following provisions of the CLASS program:
- vesting and work requirements
- earnings requirements
- benefit triggers
- daily benefit amount
- duration of benefits
- scheduled increase in premiums (i.e., indexing to a specific percentage increase or none);
- waiver of premium while on claim (full, partial, or none)
- participation rate
- administrative expense load on premiums.
Two approaches to adverse selection are built into the model: a theoretical approach and a first-year assumption regarding additional claims (a.k.a. the first-in method). The theoretical approach is based on a formula that assumes that adverse selection is greatest at the time of issue and declines the longer an individual is enrolled in the program. The second approach is based on observed data and an estimate of the number of people who are immediately eligible to enroll in the program and who also meet the ADL or cognitive requirement to qualify for benefits. This alternative method assumes that 100 percent of the population with limitations in ADLs or severe cognitive impairment would: (1) choose to enroll in the CLASS program the first year policies are offered, (2) survive the 5-year vesting period, (3) meet the work requirements during the vesting period, and (4) file a claim as soon as possible. After the first year in which benefits are paid, incidence rates for policyholders are assumed to be the same as general population incidence.
The model uses the 2011 OASDI Trustees Report, the Current Population Survey (CPS), the National Health Interview Survey (NHIS), the National Long-Term Care Survey (NLTCS), and the National Nursing Home Survey (NNHS). The model uses the 2011 OASDI Report and the CPS to set input parameters related to future inflation, mortality, interest rates, and labor force participation. The model uses the NHIS, NLTCS, and NNHS to estimate initial long-term care utilization by age and sex. Program options, economic and demographic/actuarial assumptions, and utilization can be saved and retrieved so that estimates can be replicated easily, and the impact of individual assumptions, or sets of assumptions, can be determined.
Summary of Avalere HealthS Revised Long-Term Care Policy Simulator
Avalere Healths Revised Long-Term Care Policy Simulator (LTC-PS) is an Excel-based model that tracks age-specific groups of CLASS program enrollees for 75 years (see Appendix M for an in-depth description of the model). The LTC-PS builds off a long-term care premium calculator originally developed under a grant from The SCAN Foundation. ASPE contracted with Avalere Health in the fall of 2010 to expand the capacity of the original model to incorporate key features of the CLASS program and a wider set of assumptions.
The basic approach to estimating premiums is similar to ARCs Long-Term Care Premium Model in that the present value of total expected costs of the program (including administrative costs) must equal the present value of total expected income (premiums plus interest on accumulated reserves). The estimated premium represents the average premium required in the initial year for each age of enrollment to accomplish an actuarially balanced model. The model estimates the impact on premiums of different benefit triggers and benefit amounts, program enrollment rates, low-income premium subsidies, and various benefit structures (including cash vs. service reimbursement).
The model incorporates adverse selection through an approach that is a hybrid of that used in the ARC model. Specifically, the LTC-PS estimates the number of people by age that will develop a severe disability over the next five years, and given a rate of assumed overall participation in the program, compares the number of people that would enroll in the program against the total estimated incidence of disability for the entire eligible population over the next five years. Under a pure adverse selection scenario, the model assumes that all people who would develop a severe disability will enroll in the program; this is similar to the ARC first-in method. However, because perfect adverse selection is unlikely to occur, the model builds in several factors that dampen the impact of adverse selection at initial enrollment and over time.
The LTC-PS uses many of the same sources of data as the ARC Long-Term Care Premium Model. For example, the model bases key economic and demographic assumptions on the 2011 OASDI Trustees Report, and uses data from the NNHS and NLTCS on the older population with disabilities, both living in the community and institutions. However, unlike the ARC model that relies on the CPS and NHIS for labor force participation and core disability data, the LTC-PS uses the American Community Survey and the Survey of Income and Program Participation.
Phase I. Early Model Development and Estimation (March 23, 2010 to September 22, 2010)
EARLY WORK ON THE ARC LONG-TERM CARE PREMIUM MODEL
ARC began preliminary modeling of CLASS in late 2009 to help HHS and other federal staff understand how premiums would vary based on different levels of participation and program options that were being considered by Congress at that time. Most of this work relied on an existing premium calculator that ARC had previously developed for a different purpose, and had quickly revised to model the major program features of the CLASS Act. Following passage of the ACA, ARC began to systematically review previous assumptions and premium calculations for accuracy. Major revisions to the model were undertaken through early summer 2010 to incorporate several aspects of the program that were not previously modeled in-depth, most notably the impact of the nominal premiums for low income persons and full-time students. In addition, staff at ARC began to update program parameters, the approach to adverse selection, demographic and actuarial assumptions, and input data. The ARC staff made these revisions with the goal of having independent technical experts review the methodology, assumptions, and data used in the model during the summer of 2010.
The first draft describing the model was produced in early April 2010 and preliminary premium estimates were completed in late August. At that point, the estimated baseline average premium at 2 percent program participation for a $50/day benefit based on a 2+ ADL trigger (or similar level of cognitive impairment) with full waiver of premium while in claim was $354/month. Changing the parameters of the program to increase the work and earning requirements during the vesting period, and indexing the premium reduced the premium to an average of $134/month. The impact on premiums of other changes to the program (e.g., increasing the vesting period to seven years; having enrollees continue to pay premiums if receiving home care; changing the duration of the benefit to seven years; and adding a 90 day elimination period) were also explored.
EARLY WORK ON THE AVALERE HEALTH MODEL
In early 2009, ASPE staff learned that Avalere Health was developing a long-term care premium calculator under a grant from The SCAN Foundation. Although the calculator was not intended to directly model the CLASS program, many of the components could be adapted to develop a more robust model that could more fully analyze aspects of CLASS. This was recognized by ASPE staff following the passage of the ACA as an opportunity to develop an alternative model to compare output from ARC, and to better understand how differences in methodology, assumptions, and input data affected premium estimates and CLASS program dynamics. After briefly considering revising the model in-house, ASPE decided to pursue a contract with Avalere Health directly. ASPE developed the Statement of Work and other contract documents over the summer of 2010; the contract was formally awarded to Avalere Health on September 17, 2010. The first contract activity was for staff at Avalere Health to attend a meeting five days later at HHS on actuarial modeling of the CLASS Act.
CLASS ACT MODELS MEETING
A half-day meeting of technical experts was held on September 22, 2010 to discuss progress on modeling the CLASS program (see Appendix N for the agenda, list of participants, and presentations). Participants included actuaries, economists, and analysts in health and long-term care in HHS, and members of several outside organizations, both public and private. The purpose of the meeting was threefold: (1) to describe the updated ARC Long-Term Care Premium Model and critically review the methods, assumptions and data underlying the model; (2) to describe and review Avalere Healths Long-Term Care Policy Simulator developed for The SCAN Foundation and plans for its revision to better model the CLASS program; and (3) to discuss outstanding technical issues and get feedback on such critical questions as:
Do the models incorporate realistic assumptions related to incidence/continuance of functional limitations and trends in disability? Are the assumptions related to the prevalence and trends in cognitive impairment reasonable?
Are there alternative approaches to modeling the relationship between CLASS participation and premiums?
Is potential adverse selection adequately incorporated into the models?
Because the models were still being developed, the preliminary premiums that ARC produced in late August were not presented; the discussion was focused exclusively on how the two models could be improved going forward. Several suggestions that were raised in the meeting led to substantive changes in the models. For example, ARC expanded its approach to adverse selection, adding a second approach that eventually became the first-in method. Staff at ARC also further revised and updated key assumptions and data on long-term care utilization. Suggestions for ways to improve the Avalere Health models estimation of age-specific participation were also eventually incorporated.
Phase II. Model Refinement and Development of Preliminary Benefit Options (September 23, 2010 to June 22, 2011)
Actuarial work over this period focused on further revisions and testing of the ARC model, and the completion of a revised LTC-PS that could more completely model the CLASS program. HHS sought to have both models in near final condition (with extensive documentation) so that preliminary benefit options could be developed and tested, and a Technical Expert Panel (TEP) could thoroughly vet both models in spring 2011.
During this time period, the CLASS Office hired its Chief Actuary, also known as the Director of Actuarial Integrity and Benefit Design. He began developing potential plan designs to mitigate the effects of adverse selection. He also worked to review and understand the ARC and Avalere models and provided his perspective on those. He focused on program provisions having a significant influence on the benefit design, including that: (1) participation in CLASS is voluntary; (2) actuarial soundness is a requirement; (3) any successful benefit design must present a clear value proposition to attract enrollees; (4) no underwriting other than age can be used to set premiums or prevent enrollment; and, (5) CLASS is not an entitlement program. The CLASS Office brought in an actuary from the U.S. Office of Personnel Management on a temporary detail. He and the Chief Actuary coauthored a report, Actuarial Report on the Development of CLASS Benefit Plans (see Appendix O), which discusses their analyses and findings, and describes benefit plans that have the potential of being actuarially sound.
PROGRESS ON THE ARC LONG-TERM CARE PREMIUM MODEL
Staff at ARC continued to revise the model and update the input data based on suggestions made by the CLASS Chief Actuary and the participants in the CLASS Act Models Meeting in September 2010. In early January 2011, another set of baseline premiums was estimated along with several benefit options with various work and earnings requirements during the vesting period (at this point almost all estimation assumed that premiums would be indexed, i.e., increase according to a fixed schedule such as CPI-U). The estimated baseline average (indexed) premium was now slightly lower: $339/month assuming 2 percent program participation for a $50/day lifetime benefit that used a 2+ ADL trigger (or similar level of cognitive impairment) with full waiver of premium while in claim. Increasing both the work requirement (to five years instead of three of five years) and earning requirement (to $12,000 per year instead of $1,120 per year) during the vesting period produced a slightly lower average premium compared to the $134 estimate from April 2010: $127/month. These premium estimates were forwarded to the Chief Actuary at the CLASS Office on February 17, 2011. Contemporaneously, staff in ASPE and the CLASS Office began to explore alternative benefit options that might lead to reduced premiums. Several of these were formally modeled (discussed further below and in the Actuarial Report on the Development of CLASS Benefit Plans which can be found in Appendix O) and eventually presented at the TEP meeting. Analyses of the implications of changes to key economic and demographic/actuarial assumptions continued in preparation for the TEP meeting as well as the development of final estimates to present to the TEP.
PROGRESS ON AVALERE HEALTHS LTC-PS
Because the original LTC-PS was not designed to model CLASS, work by staff at Avalere Health concentrated on developing a thorough CLASS baseline prior to the planned TEP meeting and building in as much flexibility to model alternatives as possible. The first preliminary estimates were produced in late January 2011. The average premiums were very similar to those being estimated by the ARC model, although the distribution of premiums by age was different. Actuarial work in late winter and the spring, as well as drafting documentation, focused on preparation for the TEP Meeting which was scheduled for June 2011.
TECHNICAL EXPERT PANEL MEETING ON ACTUARIAL MODELING OF THE CLASS PROGRAM
The full-day meeting of the TEP took place on June 22, 2011 (see Appendix P for the agenda and meeting materials). As with the previous meeting, participants included actuaries, economists, experts in disability data, and analysts in health and long-term care; none of the formal members of the TEP were federal employees, although participants included the CLASS Chief Actuary, actuaries from SSA and CMS, and other technical experts. Time during the morning was devoted to presentations on the two models and a review of methods, assumptions, and data. The agenda in the afternoon consisted of a review of the premiums produced by each model under different sets of assumptions and alternative benefit designs. The TEP reached consensus that the models methods and demographic/actuarial assumptions were credible and that the estimates were plausible. There was some debate as to whether the incidence rates in the ARC model were too high, and thus premium estimates also too high. The TEP also extensively discussed issues of adverse selection and suggested follow up work to improve the models handling of adverse selection; however, TEP members reiterated that there was no definitive way to determine the impact of participation and adverse selection a priori because CLASS is such a unique program, and CLASS modeling would thus be inherently uncertain. The discussion of alternative benefit designs was brief and there were no strong opinions voiced one way or the other about specific options.
Phase III. Final Model Development and Estimation (June 23, 2011 to Present)
Model development at this stage has focused on further improvements to the ARC Long-Term Care Premium Model to take into account situations of extremely low enrollment (e.g., under 1 percent), modeling of an alternative benefit design contained in the Actuarial Report on the Development of CLASS Benefit Plans (Appendix O and further described in the next section), and additional reviews of both models calculations and assumptions. Two independent actuaries are undertaking the latter effort as part of ASPEs ongoing contract with Avalere Health. The CLASS Actuary also explored an alternative approach based on information derived from Genworths net premium rates, with adjustments (see page 14 of the Actuarial Report on the Development of CLASS Benefit Plans, Appendix O).
FEDERAL ACTUARIES MEETING
The CLASS Actuary convened a meeting of government actuaries on June 28, 2011 to discuss actuarial modeling on CLASS and alternative plan options. Attendees included actuaries from the CMS Office of the Actuary and the Center for Consumer and Insurance Oversight, the Social Security Administration, and the Office of Personnel Management. Additional attendees included CLASS and ASPE staff members. The group discussed plans outlined in detail in the next section of this report. The consensus was that some benefit options under consideration could theoretically reduce adverse selection and have the potential to be actuarially sound. However, concerns were raised about: how to interpret the three percent administrative cost provision contained in the law; the policy and administrative complexities associated with some of the options; the unique marketing challenges of offering a federal benefit to large employer groups; and the very high level of uncertainty around assumptions in the actuarial models.