Initial-State Mixtures in Benefit Valuations

Why the Initial State Matters

A population may contain members who begin a contract in different states: for example, active and disabled, or in different levels of long-term care. Even when everyone has the same benefit rules, the expected present value can differ by starting state because the possible future benefits differ. The population value must therefore reflect both the conditional value for each starting state and the proportion of the population in that state.

The essential operation is a weighted average. First calculate the expected present value conditional on each possible initial state. Then weight each conditional value by the probability of starting in that state. The weights describe the population at the valuation date; they are not transition probabilities or benefit amounts.

Definition: Let $J$ denote an individual’s initial state, taking values in $1,\ldots,m$, and let $\pi_j=P(J=j)$, where $\pi_j\geq 0$ and $\sum_{j=1}^{m}\pi_j=1$. Let $Z$ be the present value of the benefits for that individual, and define $V_j=E[Z\mid J=j]$. The expected present value for a randomly selected member of the population is $$ E[Z]=\sum_{j=1}^{m}\pi_j V_j. $$

The conditional value $V_j$ must use the benefit and transition assumptions applicable to someone who starts in state $j$. If benefits are paid only after a transition, the conditional value includes the probability of that transition given the initial state. If benefits depend on occupancy of a benefit state at specified times, the conditional value uses the corresponding occupancy probabilities. The mixture formula does not replace those state-specific calculations; it combines their results.

Why the Weighting Formula Holds

The initial states form a partition: exactly one of the possible states is the individual’s initial state. The expected value can therefore be split into the contribution from each event $J=j$. On that event, the conditional expected value is $V_j$, and the event has probability $\pi_j$.

\begin{aligned} E[Z] &=\sum_{j=1}^{m}E[Z\,\mathbf{1}_{\{J=j\}}]\\ &=\sum_{j=1}^{m}P(J=j)E[Z\mid J=j]\\ &=\sum_{j=1}^{m}\pi_jV_j. \end{aligned}

Here $\mathbf{1}_{\{J=j\}}$ equals 1 when the initial state is $j$ and 0 otherwise. In the second line, $E[Z\,\mathbf{1}_{\{J=j\}}]=P(J=j)E[Z\mid J=j]$, which is the conditional-expectation identity for an event of positive probability. If $\pi_j=0$, that state contributes nothing to the population value, so its conditional value need not be specified for this calculation.

This argument requires finite expected present values and a correctly specified initial-state distribution. It does not require the state-specific benefit values to be equal, nor does it require every member to face identical conditional transition probabilities. Any differences relevant to the conditional valuation must, however, be captured in $V_j$ or in a finer classification of the initial groups.

Worked Example: A Benefit Paid on a Transition

A hypothetical one-year contract has two possible initial states, $A$ and $B$. At issue, 60% of members are in $A$ and 40% are in $B$. A benefit is paid at the end of the year only if the member makes a specified transition: the payment is $1,000 for a member starting in $A$, and $600 for a member starting in $B$. Conditional on the initial state, the probabilities of that transition during the year are 0.20 and 0.25, respectively. The effective annual interest rate is 5%. Find the expected present value per member.

Method. Calculate the conditional expected present value for each initial state, discounting the end-of-year payment once, and then weight by the initial-state proportions. The calculation avoids treating transition probabilities as population proportions.

For state $A$, the expected end-of-year payment conditional on starting in $A$ is $0.20(1{,}000)=\$200$. With $v=1/(1.05)=0.95238095$, its conditional present value is

V_A=0.20(1{,}000)v =\frac{200}{1.05} =\$190.47619\ldots

For state $B$, the expected end-of-year payment conditional on starting in $B$ is $0.25(600)=\$150$. Thus

V_B=0.25(600)v =\frac{150}{1.05} =\$142.85714\ldots

The initial-state weights are $\pi_A=0.60$ and $\pi_B=0.40$. Therefore

E[Z]=0.60(190.47619\ldots)+0.40(142.85714\ldots) =114.28571\ldots+57.14286\ldots =\$171.42857\ldots

Rounded to the nearest cent, the expected present value is $171.43 per member. As a check, the weights sum to 1, and the result lies between the two conditional values, as a weighted average with positive weights should. A common mistake is to average the two conditional values equally, obtaining $(190.48+142.86)/2=\$166.67$. That calculation incorrectly assumes equal initial-state proportions.

State-Occupancy Benefits at Several Payment Dates

The same rule applies when a benefit may be paid more than once. For each initial state, find the expected discounted benefit over all payment dates conditional on that state. Then mix those conditional present values. In particular, a benefit paid at the end of year $t$ while the member occupies a specified state has conditional expected value equal to the payment amount multiplied by the conditional probability of occupying that state at time $t$, and by $v^t$.

For a fixed benefit amount $b_t$ payable at the end of each year $t$ in which a designated state is occupied, the state-specific value is the sum of the discounted expected payments. The population value is the weighted sum of those state-specific values. This follows from additivity of expectation: each payment date can be valued separately before the initial-state mixture is applied.

Worked Example: Two Annual Payments for a Benefit-State Member

A hypothetical two-year benefit pays $500 at the end of each year if the member is in benefit state $B$ at that year-end. Members begin in either state $A$ or $B$, with initial proportions 60% and 40%. Under the stipulated state model, the probabilities of occupying $B$ at the ends of years 1 and 2, conditional on starting in $A$, are 0.20 and 0.30. Conditional on starting in $B$, they are 0.70 and 0.55. The effective annual interest rate is 4%. Find the expected present value per member.

Method. Value each year’s payment conditional on each starting state, add the two discounted amounts, and then apply the initial-state weights. The occupancy probabilities are given conditional on the initial state; they are not the initial-state mixture weights.

The discount factors are $v=1/1.04=0.96153846\ldots$ and $v^2=1/1.04^2=0.92455621\ldots$. For someone starting in $A$, the conditional value is

\begin{aligned} V_A &=500(0.20)v+500(0.30)v^2\\ &=100(0.96153846\ldots)+150(0.92455621\ldots)\\ &=96.15385\ldots+138.68343\ldots\\ &=\$234.83728\ldots \end{aligned}

For someone starting in $B$, the calculation is

\begin{aligned} V_B &=500(0.70)v+500(0.55)v^2\\ &=350(0.96153846\ldots)+275(0.92455621\ldots)\\ &=336.53846\ldots+254.25296\ldots\\ &=\$590.79142\ldots \end{aligned}

Now weight those conditional values by 0.60 and 0.40:

E[Z] =0.60(234.83728\ldots)+0.40(590.79142\ldots) =140.90237\ldots+236.31657\ldots =\$377.21894\ldots

Rounded to the nearest cent, the answer is $377.22 per member. The value is between $V_A$ and $V_B$, and is closer to $V_A$ because more members start in $A$. A common error is to multiply each payment by the overall probability of being in $B$ at that date without conditioning on the initial state. That discards the very distinction the mixture calculation is intended to preserve.

Benefits Across Living and Dead States

An initial-state mixture can also include state-specific future outcomes such as death. The conditional valuation must include each possible payment under the model, including a death benefit where applicable. The population weighting is then applied to the completed conditional values. It is not applied separately to each transition probability unless the calculation is explicitly being organized as a direct sum over initial states and outcomes.

Worked Example: End-of-Year Payments in Three Possible States

A hypothetical one-year contract begins with a member in either active state $A$ or disabled state $D$. At issue, 70% of members are active and 30% are disabled. At year-end, the member is in exactly one of the states $A$, $D$, or dead $X$. Conditional on starting in $A$, the probabilities of these outcomes are 0.80, 0.15, and 0.05. Conditional on starting in $D$, they are 0.20, 0.70, and 0.10. The year-end benefit is $100 in $A$, $400 in $D$, and $2,000 in $X$. The effective annual interest rate is 5%. Find the expected present value per member.

Method. First calculate the expected year-end benefit conditional on each initial state, discount it one year, and then weight the conditional present values. This keeps the initial-state distribution separate from the conditional year-end outcome probabilities.

For a member starting in $A$, the expected year-end benefit is

0.80(100)+0.15(400)+0.05(2{,}000) =80+60+100 =\$240.

Discounting at 5% gives

V_A=\frac{240}{1.05}=\$228.57143\ldots

For a member starting in $D$, the expected year-end benefit is

0.20(100)+0.70(400)+0.10(2{,}000) =20+280+200 =\$500,

V_D=\frac{500}{1.05}=\$476.19048\ldots

The population expected present value is

E[Z]=0.70(228.57143\ldots)+0.30(476.19048\ldots) =160+142.85714\ldots =\$302.85714\ldots

Rounded to the nearest cent, the answer is $302.86 per member. The outcome probabilities sum to 1 for each initial state, and the calculation includes a death benefit only through the probability of the dead state. A common mistake is to use the death probability for the whole population without conditioning on the initial state; the death probabilities here are 0.05 and 0.10, not one common probability.

Practical Checks and Common Traps

Initial-state mixtures are simple to write but easy to misapply when several probabilities appear in the same problem. Keep the roles of the quantities explicit:

Initial-state weights describe the proportion or probability of beginning in each state. They must be nonnegative and sum to 1.
Conditional outcome probabilities describe future transitions or occupancy given a particular initial state. For each relevant conditional distribution, its outcome probabilities must sum to 1.
Conditional present values include the benefit amounts, their timing, discounting, and the relevant probabilities conditional on the initial state.
Population present value is the weighted sum of the conditional present values, not an unweighted average unless all initial-state weights are equal.

For a population of $N$ members with the same initial-state proportions and assumptions, the expected aggregate present value is $N\sum_j\pi_jV_j$. This is an expected aggregate, not a guarantee of the realized total: actual outcomes may vary. If the question asks for a per-member value, do not multiply by $N$; if it asks for an aggregate value, state clearly whether the result is per member or for the whole population.

A useful check is to verify that the mixture weights sum to 1 and that the final per-member value lies between the smallest and largest conditional values when all weights are nonnegative. If it does not, inspect for an omitted weight, a duplicated weight, or a calculation that has mixed conditional and unconditional probabilities. Also keep payment timing consistent: a year-end payment is discounted by $v$, while a payment at time 0 is not discounted.

Key takeaway: Value each initial state on its own conditional assumptions, then take the probability-weighted average. Do not replace state-specific assumptions with population averages before valuation, and do not apply the initial-state weights twice.

Check Your Understanding

For each problem, calculate the requested value using the initial-state mixture. Round dollar answers to the nearest cent where applicable.

A one-year hypothetical benefit has conditional present values of $320 for initial state $R$ and $80 for initial state $S$. The initial-state probabilities are 0.25 and 0.75. Find the expected present value per member. (Answer: $140.00)
A population begins in one of three states with probabilities 0.40, 0.35, and 0.25. The conditional present values are $100, $220, and $400, respectively. Find the population expected present value per member. (Answer: $217.00)
A one-year hypothetical contract has initial-state probabilities 0.80 for active and 0.20 for disabled. At year-end, active members receive $500 with probability 0.20 and $100 otherwise; disabled members receive $700 with probability 0.50 and $300 otherwise. The effective annual interest rate is 5%. Find the expected present value per member. (Answer: $232.38)
A hypothetical group has 1,000 members distributed across states $A,B,C$ in proportions 0.65, 0.25, and 0.10. Their state-specific expected present values are $240, $500, and $1,200 per member. Find the expected aggregate present value. (Answer: $401,000.00)
A population has initial-state probabilities 0.50, 0.30, and 0.20, with conditional present values $400, $100, and $250. A revised population mix moves 0.05 of the total population from the first state to the second, leaving the third unchanged. Find the revised value per member and the decrease from the original value. (Answer: revised value $265.00; decrease $15.00)

What You'll Learn