Valuation With an Uncertain Initial State

Why the Initial State Matters

A policy may be issued when the insured's state is not known with certainty to the person performing the valuation. For example, an assessor may know the probabilities that an insured is in each of two health states, while the future cash flows depend on which state actually applies. The issue-date expected present value must account for both sources of uncertainty: the initial state and the subsequent cash flows conditional on that state.

The essential procedure is to value the contract separately under each possible initial state, using the same valuation date and financial assumptions, and then take the probability-weighted average. This is not the same as choosing an “average state.” State labels are categories, not numerical quantities to average.

Definition: Let $Z_0$ denote the state at issue, with possible values $1,\ldots,m$, and let $p_i=P(Z_0=i)$. The conditional expected present value $V_i$ is the expected present value, measured at issue, of the contract cash flows conditional on $Z_0=i$. If the initial-state probabilities are exhaustive and sum to one, the unconditional issue-date expected present value is $V_0=\sum_{i=1}^m p_iV_i$.

The conditional values $V_i$ can themselves involve future transitions, survival, or other uncertainty. The conditioning fixes only the state at issue; it does not generally fix the state at later times. For cash flows $C_t$ at discrete times $t$, discounted at effective annual interest rate $i$, the definition can be written as

V_i=E\left[\sum_t v^t C_t\mid Z_0=i\right], \qquad V_0=\sum_{i=1}^m p_iV_i, \qquad v=\frac{1}{1+i}.

For continuous cash flows discounted at force of interest $\delta$, the corresponding present-value factor at time $t$ is $e^{-\delta t}$. The value conditional on an initial state is then calculated using the conditional distribution of future state occupancy or transitions. In either setting, every cash flow included must be valued from the issue date if the result sought is an issue-date value.

Why the Probability-Weighted Average Holds

The averaging rule follows from the law of total expectation. Let $Y$ be the random present value of all contract cash flows, measured at issue. The events $Z_0=1,\ldots,Z_0=m$ are mutually exclusive and collectively exhaustive. Partitioning the expectation of $Y$ over those events gives

\begin{aligned} E[Y] &=\sum_{i=1}^{m}E\left[Y\,\mathbf{1}_{\{Z_0=i\}}\right]\\ &=\sum_{i=1}^{m}P(Z_0=i)E[Y\mid Z_0=i]\\ &=\sum_{i=1}^{m}p_iV_i. \end{aligned}

Here $\mathbf{1}_{\{Z_0=i\}}$ equals one when the initial state is $i$ and zero otherwise. The second equality uses the definition of conditional expectation. Thus, the weighting rule does not require the future cash flows to be independent of the initial state. Their dependence is captured by calculating a separate $V_i$ for each initial state.

Valuation check: The initial-state probabilities must satisfy $p_i\geq 0$ and $\sum_i p_i=1$. Each conditional value must be measured at the same valuation date and under the same interest and contract assumptions as the others. If the probabilities describe a population mix, the result is the average value per policy under that mix; it is not automatically the value for a particular individual whose initial state is known.

Fixed-Time Cash Flows Conditional on the Initial State

The first example has cash flows determined by the state at issue. There are no future state changes in this simplified contract. This makes the conditional values direct to calculate and highlights the separate weighting step.

Worked Example: Issue-State-Dependent Payments

A hypothetical one-year contract has two possible states at issue, Standard and Enhanced. The probabilities are $P(Z_0=\text{Standard})=0.75$ and $P(Z_0=\text{Enhanced})=0.25$. A payment is made immediately at issue, followed by a payment at the end of one year. The Standard-state payments are $600 and $1,200; the Enhanced-state payments are $200 and $2,000. Use an effective annual interest rate of $i=4\%$. Assume no mortality or other decrements during the year and no other cash flows. Calculate the issue-date expected present value.

Method. First calculate the present value conditional on each issue state. Then weight those values by the issue-state probabilities. The immediate payments are not discounted; the one-year payments are multiplied by $v=1/1.04=25/26$.

Standard state. The conditional value is

V_{\text{Standard}} =600+1{,}200\left(\frac{25}{26}\right) =600+\frac{30{,}000}{26} =\frac{22{,}800}{13} \approx \$1{,}753.85.

Enhanced state. Its conditional value is

V_{\text{Enhanced}} =200+2{,}000\left(\frac{25}{26}\right) =200+\frac{50{,}000}{26} =\frac{27{,}600}{13} \approx \$2{,}123.08.

Weight the conditional values. Using the exact fractions before rounding,

\begin{aligned} V_0 &=0.75\left(\frac{22{,}800}{13}\right) +0.25\left(\frac{27{,}600}{13}\right)\\ &=\frac{17{,}100+6{,}900}{13} =\frac{24{,}000}{13} \approx \$1{,}846.15. \end{aligned}

Check. The probability-weighted immediate payment is $0.75(600)+0.25(200)=500$. The expected end-of-year payment is $0.75(1{,}200)+0.25(2{,}000)=1{,}400$, whose present value is $1{,}400/1.04=1{,}346.1538\ldots$. Their sum is $1,846.15 to the nearest cent, matching the conditional-value calculation. The common mistake avoided is discounting the issue-date payment or averaging the state labels rather than the conditional present values.

Continuous Payments and Future State Changes

When cash flows depend on the state occupied at each future time, the conditional value must reflect transitions after issue. For a continuous payment rate $b_j$ while the process occupies state $j$, the conditional expected present value over a term from time zero to $T$ is

V_i=\int_0^T e^{-\delta t} \left(\sum_j P(Z_t=j\mid Z_0=i)b_j\right)\,dt.

The expression inside the integral is the expected payment rate at time $t$, conditional on the initial state. The discount factor then converts that expected payment rate to issue-date terms. This formula applies to continuous payments over the stated term, with deterministic force of interest and the specified state model. It does not, by itself, include lump-sum transition payments or payments at fixed dates; those require their own cash-flow terms.

Worked Example: Continuous Payments After a Possible State Change

A hypothetical one-year benefit pays continuously at $1,000 per year in state $H$ and $2,000 per year in state $C$. State $H$ changes to $C$ at constant transition intensity $\lambda=\ln 2$ per year; $C$ is absorbing. At issue, $P(Z_0=H)=0.60$ and $P(Z_0=C)=0.40$. Use force of interest $\delta=\ln 2$ per year. Assume the stated transition model applies throughout the year and there are no other benefits or decrements. Calculate the issue-date expected present value, rounding only the final result to cents.

Method. Calculate the discounted expected time in each payment state conditional on each initial state. Starting in $H$, the probability of remaining in $H$ at time $t$ is $e^{-\lambda t}$; the probability of occupying $C$ is $1-e^{-\lambda t}$. Since $\delta=\lambda=\ln 2$, the discounted expected occupancy times starting in $H$ are

\begin{aligned} I_H &=\int_0^1 e^{-\delta t}e^{-\lambda t}\,dt =\frac{1-e^{-2\ln 2}}{2\ln 2} =\frac{3}{8\ln 2} \approx 0.5410106,\\ I_C &=\int_0^1 e^{-\delta t}(1-e^{-\lambda t})\,dt\\ &=\frac{1-e^{-\ln 2}}{\ln 2}-I_H =\frac{1}{2\ln 2}-\frac{3}{8\ln 2} =\frac{1}{8\ln 2} \approx 0.1803369. \end{aligned}

Therefore, starting in $H$, the conditional value is

V_H=1{,}000I_H+2{,}000I_C \approx 541.0106+360.6738 =\$901.6844\ldots.

Starting in $C$, the process remains there for the whole year. Its discounted occupancy time is

I_C\mid Z_0=C =\int_0^1 e^{-\delta t}\,dt =\frac{1-e^{-\ln 2}}{\ln 2} =\frac{1}{2\ln 2} \approx 0.7213475,

so $V_C=2{,}000(0.7213475\ldots)=\$1{,}442.6950\ldots$. Now combine the initial-state values:

\begin{aligned} V_0 &=0.60V_H+0.40V_C\\ &\approx 0.60(901.6844\ldots)+0.40(1{,}442.6950\ldots)\\ &=1{,}118.0886\ldots =\$1{,}118.09. \end{aligned}

Check. Conditional on starting in $H$, the expected undiscounted time in $H$ is $1/(2\ln 2)$, and the expected time in $C$ is the remaining $1-1/(2\ln 2)$; these sum to one year. The common mistake avoided is treating the initial 60% probability of $H$ as the probability of being in $H$ throughout the year. The state can change, so future occupancy probabilities must be used inside the integral.

Using a Transition Matrix Within Each Conditional Value

For discrete-time models, a transition probability matrix can be used to find the distribution of the state at each payment date, conditional on each possible issue state. If $P$ is a one-year transition matrix whose rows represent the current state and columns the next state, the row for initial state $i$ in $P^k$ gives the state probabilities $k$ years later conditional on $Z_0=i$. Those probabilities determine the expected payment at each date. The initial-state distribution is applied after these conditional values have been calculated.

Worked Example: Annual Benefits With a Two-Year State Model

A hypothetical contract has states Active and Disabled. A benefit of $10,000 is paid at the end of each of years one and two if the insured is Disabled at that payment date. The annual transition matrix is

P= \begin{pmatrix} 0.8&0.2\\ 0.3&0.7 \end{pmatrix},

with rows and columns ordered Active, Disabled. The initial-state probabilities are 0.70 Active and 0.30 Disabled. Use effective annual interest $i=5\%$. Assume the Markov transition probabilities are the same in both years, and there are no other payments or decrements. Find the issue-date expected present value.

Method. Compute the probability of being Disabled at each payment date, conditional on each initial state. At year one, the probabilities are the Disabled entries in the corresponding rows of $P$: 0.20 from Active and 0.70 from Disabled.

At year two, the Active-start probability is obtained from the Disabled column of the Active row of $P^2$:

P(Z_2=\text{Disabled}\mid Z_0=\text{Active}) =0.8(0.2)+0.2(0.7)=0.30.

For an initially Disabled insured, the year-two probability is

P(Z_2=\text{Disabled}\mid Z_0=\text{Disabled}) =0.3(0.2)+0.7(0.7)=0.55.

With $v=1/1.05=20/21$, the conditional issue-date values are

\begin{aligned} V_{\text{Active}} &=10{,}000\left(0.20\frac{20}{21} +0.30\frac{400}{441}\right)\\ &=\frac{2{,}040{,}000}{441} \approx \$4{,}625.85,\\[4pt] V_{\text{Disabled}} &=10{,}000\left(0.70\frac{20}{21} +0.55\frac{400}{441}\right)\\ &=\frac{5{,}140{,}000}{441} \approx \$11{,}655.33. \end{aligned}

Finally, average over the initial states:

\begin{aligned} V_0 &=0.70V_{\text{Active}}+0.30V_{\text{Disabled}}\\ &=\frac{0.70(2{,}040{,}000)+0.30(5{,}140{,}000)}{441}\\ &=\frac{2{,}970{,}000}{441} \approx \$6{,}734.69. \end{aligned}

Check. The year-one unconditional probability of Disabled is $0.70(0.20)+0.30(0.70)=0.35$. The year-two unconditional probability is $0.70(0.30)+0.30(0.55)=0.375$. Thus the expected discounted benefits directly are $10{,}000(0.35v+0.375v^2)=\$6{,}734.69$ to the nearest cent. The common mistake avoided is using the initial probabilities as though they were the state probabilities at both future payment dates.

Common Errors and a Practical Sequence

The method is compact, but errors often occur in the conditioning or timing rather than in the final weighted average. In particular, a conditional value is not yet the value for the uncertain initial state. It is one component of that value. Likewise, a probability distribution at issue is not generally the state distribution at later times.

Specify the initial states and probabilities.
List mutually exclusive, exhaustive states and verify that their probabilities sum to one.

Value cash flows conditional on each initial state.
Include future state transitions where relevant, and discount every payment from its actual payment time to issue.

Average the conditional values.
Multiply each $V_i$ by $p_i$, add the products, and check against any direct calculation using unconditional cash-flow expectations.

A frequent wrong calculation is to calculate one state-specific value, such as $V_H$, and report it as the issue-date value even though the initial state is uncertain. The correction is to calculate every $V_i$ and use $V_0=\sum_i p_iV_i$. Another error is to apply the initial-state probabilities at every later date without evolving them through the transition model. The correction is to use probabilities conditional on the initial state for each date, as in the matrix example, and only then perform the initial-state average.

This conditioning-and-averaging approach is aligned with the SOA Exam ALTAM (Advanced Long-Term Actuarial Mathematics) syllabus. Its main practical value is that it separates two tasks that should not be conflated: modeling cash flows given a particular issue state, and representing uncertainty about which issue state applies.

Check Your Understanding

For each problem, calculate the requested issue-date value using the stated assumptions.

A hypothetical contract has initial-state probabilities 0.40 and 0.60. Its conditional issue-date expected present values are $800 and $1,250, respectively. What is the unconditional issue-date expected present value? (Answer: $1,070.00)
A one-year continuous benefit pays $500 per year in state $H$ and $1,500 per year in state $C$. State $H$ changes to absorbing state $C$ at intensity $\lambda=\ln 2$; the force of interest is $\delta=\ln 2$. The initial probabilities are 0.25 in $H$ and 0.75 in $C$. Assuming no other decrements or benefits, find the issue-date expected present value to cents. (Answer: $946.77)
A two-year contract pays $5,000 at each year-end if Disabled. Its annual transition matrix, ordered Active and Disabled, is $\begin{pmatrix}0.9&0.1\\0.2&0.8\end{pmatrix}$. The initial probabilities are 0.40 Active and 0.60 Disabled, and $i=4\%$. Assuming the same transition matrix each year and no other cash flows, find the issue-date expected present value to cents. (Answer: $4,644.97)
A hypothetical two-state contract has conditional issue-date expected present values of $2,400 in state A and $6,800 in state B. The initial probability of state A is 0.35, with the remaining probability assigned to state B. Calculate the unconditional issue-date value and state the probability used for state B. (Answer: State B probability 0.65; issue-date value $5,260.00)

What You'll Learn