Actuarial Projection

An estimate of future events, such as mortality, morbidity, or claims costs, based on statistical analysis of historical data patterns.

Source: Bowers et al. 1997

The quantities projected are the rates at which defined events occur in a population and the amounts attached to them, so a projection combines a frequency component with a severity component. Frequency is estimated from historical experience expressed against exposure, typically as rates per person-year within groups defined by age, sex and other characteristics that predict risk, and assembled into life tables or transition rates between states. Severity is estimated from the distribution of amounts observed per event. Multiplying the two across the projected population gives expected cost, and the distribution of outcomes around it is what determines the reserves required.

Source: Bowers et al. 1997

For long horizons the assumption about how rates will change over time usually matters more than the current rates, since a small annual improvement in mortality compounds substantially over decades. The discount or investment return assumption is the other dominant input wherever payments fall in the future, because the present value of a long stream is highly sensitive to it. Inflation applied to costs, and the rate at which people leave the covered population, follow. Sensitivity analysis is therefore concentrated on these rather than distributed evenly across every input.

Source: healtheconomics.wiki

A best estimate is normally accompanied by an explicit allowance for adverse deviation, so that the figure carried is deliberately prudent rather than central, and the size of that allowance is itself a stated assumption. Deterministic sensitivity analysis varies each dominant assumption in turn to show the effect on the result. Stochastic simulation generates many possible futures from assumed distributions, which produces a distribution of outcomes rather than a single figure and is what allows statements about the probability of a shortfall to be made.

Source: healtheconomics.wiki

They are used to price and reserve for health insurance, where premiums must cover claims not yet incurred, and to set capitation rates where a payer transfers risk to a provider organisation for a defined population. They support estimation of long term care liabilities and of the funding required for schemes providing benefits over decades. Public health systems use the same techniques to project demand and cost for services whose users can be tracked over time, and to value liabilities that fall outside the annual budget cycle.

Source: healtheconomics.wiki

The method assumes that the relations observed historically will continue, so it handles gradual change well and structural change badly, and a new treatment, a change in eligibility or a shift in clinical practice can invalidate the base experience. Small populations produce unstable rates, which is why experience from a limited group is normally blended with wider data rather than used alone. Selection is a further problem wherever people choose whether to join or remain covered, since those choices are correlated with the risks being projected.

Source: Bowers et al. 1997