Why Consistency Checks Matter
A state-specific reserve is a conditional value: it measures the expected present value of future net outgo given the policy’s current state and the valuation basis. A reserve calculation can be arithmetically correct at each line and still be inconsistent overall if it uses the wrong state probabilities, omits a terminal value, or discounts a cash flow at the wrong time. Independent checks are therefore useful both when building a reserve calculation and when reviewing one.
This tutorial develops a compact set of checks for discrete-time multi-state reserves. The examples use hypothetical contracts and a simple finite-state model; the same logic applies when a model has more states or more valuation periods. The emphasis is on reconciling the state recursion with the cash flows it represents, not on developing a new reserve method. The treatment is aligned with the SOA Exam ALTAM (Advanced Long-Term Actuarial Mathematics) syllabus.
This convention places both the transition-contingent cash flow and the continuation reserve at time \(t+1\). If a contract instead pays a premium or benefit at the start of a year, that payment must be placed at its actual time; it cannot simply be inserted into this end-of-year formula without adjustment. Terminal reserves also need a clear interpretation. A zero terminal reserve means there are no future net cash flows beyond the model horizon; a nonzero terminal reserve represents their value at that horizon.
Deriving the Recursion and Setting Up the Checks
The recursion follows by conditioning on the state reached at the end of the year. Given current state \(s\), the contract reaches state \(j\) with probability \(p_{sj}\). Conditional on that transition, the insurer pays \(C_{t+1}(s,j)\) at time \(t+1\), and the value at that time of later net outgo is \(V_{t+1}(j)\). Their sum is a time-\(t+1\) value, so it is discounted once by \(v\). Taking the probability-weighted average over all possible destination states gives the stated equation.
The equation gives a practical audit sequence. First verify that the transition probabilities from each current state sum to one. Then confirm that the sign and timing of every cash flow match the definition of \(C\). Check that each probability multiplies the cash flow and continuation reserve for the same destination state. Finally, compare the result with a direct expected-present-value calculation from the original state, using the same horizon and assumptions.
For every current state, verify that the listed transition probabilities are nonnegative and sum to one. Confirm the interest rate and the time unit used for each period.
Record outgo as positive and income as negative, and attach each amount to the transition and payment time specified by the contract.
Use the reserve for the destination state at the next valuation time, including any nonzero terminal reserve.
Calculate the expected present value of the future net cash flows directly and compare it with the reserve recursion.
Worked Examples of State-Reserve Checks
Worked Example: Reconciling a Two-Year State Recursion
A hypothetical contract has active state \(A\), disabled state \(D\), and terminal state \(T\). The terminal state is absorbing and has no future reserve. At the end of each year, the transition probabilities are \(P(A\to A)=0.8\), \(P(A\to D)=0.2\), \(P(D\to D)=0.5\), and \(P(D\to T)=0.5\). There are no other possible transitions. Each year’s net cash flows are \(-\$100\) for \(A\to A\), \(\$200\) for \(A\to D\), \(\$120\) for \(D\to D\), and \(\$50\) for \(D\to T\), all paid at year-end. The horizon is two years, with \(V_2(A)=V_2(D)=V_2(T)=0\). The annual effective interest rate is \(i=10\%\). Find \(V_0(A)\), then check it by direct expected present value.
Method. Work backward using the state recursion, then calculate the expected net cash flow in each year conditional on starting in \(A\). The principal trap is treating the negative active-state cash flow as positive premium income under the insurer’s net-outgo convention.
First, \(v=1/1.10=0.9090909\). The probabilities from \(A\) sum to \(0.8+0.2=1\), and those from \(D\) sum to \(0.5+0.5=1\). At time 1, the time-2 continuation reserves are zero, so
Now include the year-1 cash flow and the appropriate time-1 reserve for each destination state:
For the independent check, the expected year-1 cash flow from \(A\) is \(0.8(-100)+0.2(200)=-40\). At time 1, the contract is in \(A\) with probability \(0.8\) and in \(D\) with probability \(0.2\). The expected year-2 cash flow is therefore \(0.8(-40)+0.2(85)=-15\), where \(-40\) and \(85\) are the one-year expected cash flows from \(A\) and \(D\), respectively. Thus the direct value is
Both methods give \(V_0(A)=-\$48.76\), rounded to the nearest cent. The negative reserve is possible because, on this valuation basis, the expected present value of future premium income exceeds that of future outgo. Agreement between the recursion and direct calculation confirms the transition weights, cash-flow signs, and discounting.
Worked Example: Checking a Population-Weighted Reserve
Use the same hypothetical contract and assumptions as in the preceding example. At time 0, a group has \(60\%\) of contracts in state \(A\) and \(40\%\) in state \(D\). Calculate the average time-0 reserve and verify it using the group’s expected cash flows. Keep full precision in intermediate calculations and round the final dollar amount to cents.
Method. First calculate the reserve conditional on each initial state; then weight those reserves by the initial state proportions. The trap is weighting by the transition probabilities from one state alone instead of by the group’s actual time-0 state distribution.
The reserve from state \(A\), already calculated, is \(V_0(A)=-48.760330\ldots\). For an initial state \(D\), use its two possible transitions and the time-1 reserves:
Here \(V_1(T)=0\). The population-weighted reserve is
To check directly, expected year-1 net cash flow for the group is \(0.6(-40)+0.4(85)=-24+34=10\). After one year, the group probabilities are \(P(A)=0.6(0.8)=0.48\), \(P(D)=0.6(0.2)+0.4(0.5)=0.32\), and \(P(T)=0.4(0.5)=0.20\). These sum to one. Expected year-2 cash flow is \(0.48(-40)+0.32(85)+0.20(0)=-19.2+27.2=8\). Therefore,
The average reserve is \(\$15.70\), rounded to the nearest cent, under both calculations. This check depends on weighting state-specific reserves by the proportion in each state at the valuation time.
Worked Example: Detecting an Omitted Terminal Reserve
A hypothetical one-year projection starts in state \(A\). At year-end, the contract moves to \(A\) with probability \(0.7\) and to \(D\) with probability \(0.3\). The year-end net cash flows are \(\$10\) for \(A\to A\) and \(\$70\) for \(A\to D\). Reserves at year-end are \(V_1(A)=\$40\) and \(V_1(D)=\$160\), representing future obligations after the projection year. The annual effective interest rate is \(5\%\). Calculate \(V_0(A)\), and quantify the error from omitting the terminal continuation reserves.
Method. Apply the one-step recursion with both the year-end cash flow and the reserve for the destination state. The common mistake is to treat the end of the projection period as the end of all obligations, despite the supplied nonzero terminal values.
The transition probabilities sum to \(0.7+0.3=1\), and \(v=1/1.05\). The correct calculation is
The contribution from the \(A\to A\) branch is \(35/1.05=\$33.333333\ldots\), and from \(A\to D\) it is \(69/1.05=\$65.714285\ldots\). Their sum is \(\$99.047619\ldots\). If the continuation reserves are wrongly omitted, the calculation becomes
The omission understates the reserve by \(99.047619\ldots-26.666666\ldots=\$72.380952\ldots\). The correct reserve is \(\$99.05\), rounded to the nearest cent. The terminal reserve is not an additional cash flow paid at year-end; it is the value at that time of later net cash flows, which is why it is added to the year-end cash flow before discounting.
Common Consistency Failures
A direct present-value check is meaningful only when both calculations use the same perspective, timing, assumptions, and horizon. In particular, use the same convention for net outgo throughout: benefits and expenses are positive, while premiums and other income are negative. If a calculation instead defines cash flow from the policyholder’s perspective, its signs reverse, and the resulting value must be interpreted accordingly.
| Check | Inconsistent calculation | Correction |
|---|---|---|
| Cash-flow sign | A premium received is added as positive insurer outgo. | With net outgo defined from the insurer’s perspective, enter a premium as a negative amount. |
| Transition weighting | A benefit on \(A\to D\) is multiplied by the probability of staying in \(A\). | Pair each transition cash flow with its own transition probability. |
| Timing | A year-end payment is discounted twice for a one-year step. | Discount a time-\(t+1\) value once to time \(t\); discount a time-\(t+2\) value twice. |
| Terminal value | A nonzero \(V_{t+1}(j)\) is omitted at the projection boundary. | Add the destination-state continuation reserve to the cash flow at that boundary before discounting. |
| Population weighting | State-specific reserves are averaged using unnormalized or unrelated probabilities. | Use the actual proportions in each state at the valuation time; they must sum to one. |
Another useful check is dimensional: a reserve, a benefit, and a premium should all be expressed in the same currency units, while probabilities are dimensionless. Interest must match the period length. For example, a one-year transition model cannot use a monthly effective rate as though it were an annual rate without conversion. These checks do not replace the reserve calculation, but they often locate errors before they are carried into a larger projection.
Check Your Understanding
For each question, use net outgo from the insurer’s perspective and state the reserve to the nearest cent where applicable.
- A one-year model has states \(A\) and \(D\). From \(A\), the contract moves to \(A\) with probability \(0.75\) and \(D\) with probability \(0.25\). The year-end net cash flows are \(-\$40\) on \(A\to A\) and \(\$100\) on \(A\to D\). The continuation reserves are \(V_1(A)=\$20\) and \(V_1(D)=\$60\). If \(i=8\%\), find \(V_0(A)\). (Answer: \(\$23.15\))
- At a valuation time, \(30\%\) of a population is in state \(A\), with reserve \(V(A)=-\$20\), and \(70\%\) is in state \(D\), with reserve \(V(D)=\$90\). Find the population-weighted reserve. (Answer: \(\$57.00\))
- From \(A\), a one-year transition is to \(A\) with probability \(0.6\) or \(D\) with probability \(0.4\). The year-end cash flows are \(\$20\) and \(\$80\), respectively, and the continuation reserves are \(V_1(A)=\$30\) and \(V_1(D)=\$110\). At \(i=10\%\), calculate \(V_0(A)\). (Answer: \(\$96.36\))
- A two-year model has states \(A,D,T\), with \(T\) terminal and all time-2 reserves zero. Each year, \(A\) moves to \(A\) with probability \(0.5\) or \(D\) with probability \(0.5\); \(D\) moves to \(D\) with probability \(0.6\) or \(T\) with probability \(0.4\). Each year’s end-of-year cash flows are \(-\$30\) on \(A\to A\), \(\$90\) on \(A\to D\), \(\$40\) on \(D\to D\), and \(\$20\) on \(D\to T\). Starting from \(A\), find the two-year reserve at \(i=5\%\) and verify it by direct expected present value. (Answer: \(\$56.69\))