Long-Term Care: Insure or Self-Insure, With the Assumptions Visible

Nobody decides "what should a long-term-care premium be?" People decide "should I take the policy I was quoted, or keep the premiums and carry the risk?" The model computes the second question, because that is the one on the table.

Long-term-care planning tools usually try to be actuaries: estimate a care probability, price a hypothetical policy, output a recommendation. The Worthune long-term-care model is deliberately not that. It takes a real quote — the premium, the daily benefit, the benefit period, all as caller inputs — and a care scenario, and runs both paths: buy the policy, or self-insure from retirement savings. The outputs are the two out-of-pocket totals, the damage to savings under each path, a break-even age, and a boolean verdict under the stated scenario.

The ten inputs

The scenario side: current age, the age care begins, how many years it lasts, today's annual care cost, and a care-cost inflation rate between two and eight percent — care inflation gets its own input because care costs have, over most long stretches, outrun general inflation. The balance-sheet side: retirement savings and an annual return. The policy side, straight off a quote sheet: the monthly premium, the daily benefit, and the benefit period in years.

Costs are projected forward: today's annual care cost compounds at the care inflation rate until care begins, which is why a scenario twenty-five years out produces headline numbers that look startling next to today's prices. That is the point of the projection — the sticker shock is the future's, not the model's.

projected_cost = annual_care_cost × (1 + care_inflation)^(care_start_age − current_age), summed over the care years;  insured path: premiums until care starts, then costs net of (daily_benefit × 365 × min(care_years, benefit_period));  self-insured path: full projected costs from savings;  is_insurance_worth_it: true when out-of-pocket with insurance is lower than without

What the comparison charges each path

The insured path pays premiums from now until care begins, then pays whatever the projected care costs exceed the policy's total benefit — the daily benefit annualized across the benefit period, capped by the years of care actually modeled. The self-insured path pays the full projected cost out of savings, which meanwhile grow at the caller's return. The savings-impact outputs show what each path does to the retirement balance; the break-even age and the worth-it boolean summarize which path wins under this scenario and when.

Two disclosed conventions in the spec deserve an integrator's attention. The benefit period caps the policy's payout in the headline totals, but the year-by-year savings trajectory the model charts does not re-apply that cap — the spec's known-issues section names the inconsistency rather than letting a careful reader discover it. And the headline totals stop charging premiums at care start, the common contract design, which the spec states as a convention rather than assuming every policy matches it.

• ✓Care event is a scenario, not a probability — the model is conditional on care happening as described.
• ✓Premiums and benefits are quote inputs — the model never prices a policy.
• ✓Care costs compound at their own inflation rate, two to eight percent.
• ✓Benefit period caps the totals; the charted trajectory's cap behavior is a named known issue.
• ✓No inflation riders on the benefit — a named exclusion; Medicaid and informal family care are simply outside the model's scope.

The probability question, left where it belongs

The model does not ask how likely care is. It computes the consequences of a scenario the caller specifies. That is a deliberate boundary: lifetime care probabilities are population statistics, they vary enormously with the definition of "care," and a single embedded probability would smuggle an actuarial opinion into what is otherwise verifiable arithmetic. A well-built LTC UI presents the scenario inputs as the user's levers — care at eighty for three years, care at seventy-five for eight — and lets the model show the consequences of each, priced consistently. Federal survey research on lifetime care usage is the right place for a UI to source its framing statistics, cited as such, next to the model's scenario outputs.^[1]

Reading the response

A call returns the projected total care cost, total premiums, total policy benefit, out-of-pocket under both paths, the retirement-savings impact of both paths, the break-even age, the worth-it boolean, and a year-by-year series for charting — plus the spec version, an assumptions array naming that version and the model's cross-field constraint (care cannot start before the current age), an empty facts array (no tax-year constants; every dollar figure is a caller input or a projection of one), and the record.sha256 that makes the stored comparison recomputable. For a recommendation a client may act on years later, the stored record is the artifact that proves what was compared, under which assumptions, at which spec version.