Boundary Conditions for Transient and Absorbing States

Why Boundary Conditions Need State-Specific Care

A multi-state reserve is not a single number at the contract endpoint: its value can depend on the state occupied at that time. A contract might pay different maturity amounts to a policyholder who is active or receiving care, while a death benefit may be paid earlier when a transition into death occurs. These payments belong at different points in the valuation: maturity amounts determine terminal boundary values, whereas death benefits on transitions enter the reserve equation before the endpoint.

A second distinction is equally important. An absorbing state is one that cannot be left in the model. That property alone does not determine its reserve. A death state after all death benefits have been paid often has zero reserve, but an absorbing claim state can still carry future benefit obligations.

Definition: Let $V_i(t)$ denote the reserve at time $t$ conditional on the contract being in state $i$, under the stated reserve convention. A terminal boundary condition specifies $V_i(T^-)$, the reserve immediately before terminal cash flows at time $T$. If a state-specific maturity amount $M_i$ is paid at $T$ and no liability remains afterward, then $V_i(T^-)=M_i$. The value immediately after that payment is zero. A transition payment at a time before $T$ is instead included in the transition benefit and is not also added to $V_i(T^-)$.

The left-limit convention matters: $T^-$ means just before the terminal payment. If a model instead defines its reserve immediately after the payment at $T$, its boundary value is zero when the contract has ended and no other liability remains. Either convention can be used, but the cash flows included in the reserve must match the chosen timing.

Terminal Values in a Multi-State System

Suppose the states at time $T$ are active, disabled, and dead. If the contract pays maturity amounts only to living policyholders, the terminal values might be $M_A$ in the active state, $M_C$ in the disabled state, and zero in the dead state. The exact amounts depend on the contract; the state labels do not determine them.

If death occurs before $T$, the payment made on that transition is denoted $b_{ij}(t)$ for a transition from state $i$ to state $j$. The reserve after the transition, $V_j(t)$, is a separate quantity. For example, if a death benefit is paid immediately and the contract then ends, the post-payment death-state reserve is zero. The transition changes the liability by $b_{ij}(t)+V_j(t)-V_i(t)$, where $V_i(t)$ is the liability just before the transition.

Using the course convention in which $c_i(t)$ is continuous net income in state $i$, $\delta$ is the force of interest, and $\lambda_{ij}(t)$ is the transition intensity, the multi-state Thiele equation is

\frac{dV_i(t)}{dt} = \delta V_i(t)+c_i(t) - \sum_{j\ne i}\lambda_{ij}(t) \bigl(b_{ij}(t)+V_j(t)-V_i(t)\bigr).

In this equation, net income $c_i(t)$ is positive when it is income under the stated convention; expenses can reduce it or make it negative. The transition term is subtracted. A transition causes a change in liability equal to the payment plus the new-state reserve minus the old-state reserve, and the backward reserve derivative includes the negative of the intensity-weighted change. Do not reverse this sign when calculating the reserve derivative.

Boundary rule: First specify whether the terminal reserve is measured immediately before or after the time-$T$ cash flow. Then assign a boundary value separately in every state. Include a payment on a transition in $b_{ij}(t)$, not again as a terminal benefit, unless the contract explicitly provides both payments.

Why the Transition Term Has This Sign

The sign can be checked by considering one possible transition over a short interval of length $h$. In state $i$, a transition to $j$ occurs with probability $\lambda_{ij}(t)h+o(h)$. If it occurs, the insurer pays $b_{ij}(t)$ and the liability changes from $V_i(t)$ to $V_j(t)$. Thus the change in liability associated with that event is

b_{ij}(t)+V_j(t)-V_i(t).

The contribution to the backward reserve derivative is the negative of this change, multiplied by the transition intensity. Summing over possible destination states gives

-\sum_{j\ne i}\lambda_{ij}(t) \bigl(b_{ij}(t)+V_j(t)-V_i(t)\bigr).

This is why a transition with a large payment can make the derivative smaller, not larger. It is also why a transition can have the opposite effect when the destination-state liability is much lower than the current-state liability.

Worked Examples

Worked Example: Assigning Maturity Boundaries by State

Consider a hypothetical three-year contract with states active $A$, receiving care $C$, and dead $D$. At time three, the contract pays \$1,200 to a policyholder in state $A$, \$4,500 to a policyholder in state $C$, and nothing to a policyholder in state $D$. The contract ends at maturity. Find the terminal reserves immediately before the maturity payments.

Assume an annual effective interest rate of $i=0.03$, so $\delta=\ln(1.03)$. Before maturity, transitions have constant intensities $\lambda_{AC}=0.08$, $\lambda_{AD}=0.02$, and $\lambda_{CD}=0.10$ per year; these correspond to constant-force, exponential waiting-time assumptions for the relevant transitions. These transition assumptions do not change the terminal payment amounts.

Step 1: Identify what is payable at time three. In state $A$, \$1,200 is due. In state $C$, \$4,500 is due. In state $D$, no maturity payment is due.

Step 2: Apply the pre-payment boundary convention. The reserve immediately before the payment includes the amount payable at that time:

V_A(3^-)=1{,}200,\qquad V_C(3^-)=4{,}500,\qquad V_D(3^-)=0.

Step 3: Check the post-payment interpretation. Because the contract ends when maturity payments are made, each post-payment reserve is zero. That does not change the pre-payment boundary values just calculated. The common mistake is to set all three pre-payment reserves to zero merely because the contract terminates at time three.

Check. Each living state’s pre-payment boundary equals its stated maturity amount, and the dead state’s boundary is zero because no payment or continuing liability is specified.

Worked Example: A Death Transition Before Maturity

At a time before maturity, a hypothetical policy is in a living state $L$. Let $\delta=0.04$, $V_L(t)=\$3{,}000$, and $V_D(t)=0$, where $D$ is an absorbing death state after settlement. The continuous net income in state $L$ is $c_L(t)=\$100$ per year. Death occurs with constant force $\lambda_{LD}=0.08$ per year, and the death payment is $b_{LD}(t)=\$7{,}000$. Find $dV_L(t)/dt$.

The exponential waiting-time assumption for death has survival probability $e^{-0.08s}$ over $s$ years, and the force of interest is constant. The transition benefit is paid on death; the zero value $V_D(t)$ is the liability after that payment.

Step 1: Calculate the interest and net-income terms.

\delta V_L(t)=0.04(3{,}000)=120, \qquad c_L(t)=100.

Step 2: Calculate the change in liability on death.

b_{LD}(t)+V_D(t)-V_L(t) =7{,}000+0-3{,}000=4{,}000.

Step 3: Apply the negative transition contribution.

\frac{dV_L(t)}{dt} =120+100-0.08(4{,}000) =220-320 =-100.

Therefore, $dV_L(t)/dt=-\$100$ per year. The derivative is negative because the intensity-weighted increase in liability on death exceeds the interest and net-income terms. A common error is to add $0.08(4{,}000)$, which would give $+\$540$ per year; that reverses the transition-term sign.

Check. Substitution into the stated Thiele equation gives $0.04(3{,}000)+100-0.08(7{,}000+0-3{,}000)=-100$, as required.

Worked Example: An Absorbing State With a Nonzero Terminal Liability

A hypothetical contract has states working $W$, receiving care $C$, and dead $D$, and matures at time $T$. It pays \$1,000 at maturity in state $W$, \$4,000 in state $C$, and nothing in state $D$. State $C$ is absorbing: once the policyholder enters care, the model has no transition out of that state. Find all terminal boundaries. Then, immediately before maturity, calculate $dV_W(t)/dt$ using $V_W(t)=\$1{,}000$, $V_C(t)=\$4{,}000$, and $V_D(t)=0$.

Assume constant force of interest $\delta=0.03$, constant transition intensities $\lambda_{WC}=0.20$ and $\lambda_{WD}=0.05$ per year, and no other transitions. Transition waiting times from $W$ are exponential under these constant intensities. In state $W$, continuous net income is $c_W(t)=\$300$ per year. A transition from $W$ to $C$ pays \$200, and a transition from $W$ to $D$ pays \$9,000.

Step 1: Set the terminal boundaries. The terminal payments are state-dependent, so

V_W(T^-)=1{,}000,\qquad V_C(T^-)=4{,}000,\qquad V_D(T^-)=0.

State $C$ is absorbing, but its terminal reserve is \$4,000 because the contract pays that maturity amount there. Absorbing does not mean zero reserve.

Step 2: Calculate the interest and income terms in state $W$.

\delta V_W(t)+c_W(t)=0.03(1{,}000)+300=30+300=330.

Step 3: Calculate each transition change. For the transition to care,

b_{WC}+V_C-V_W=200+4{,}000-1{,}000=3{,}200, \qquad \lambda_{WC}(b_{WC}+V_C-V_W)=0.20(3{,}200)=640.

For the transition to death,

b_{WD}+V_D-V_W=9{,}000+0-1{,}000=8{,}000, \qquad \lambda_{WD}(b_{WD}+V_D-V_W)=0.05(8{,}000)=400.

Step 4: Combine terms with the required sign.

\frac{dV_W(t)}{dt} =330-640-400=-710.

The reserve derivative is $-\$710$ per year. The care transition contributes through both the \$200 payment and the \$3,000 increase from the current liability to the care-state liability. The calculation does not add the maturity amount a second time as a transition payment.

Check. The two intensity-weighted transition changes total $640+400=1{,}040$, and $330-1{,}040=-710$. This confirms the sign and arithmetic.

Timing Choices and Common Traps

Use the same timing convention for all states and cash flows. If the reserve is a pre-maturity value, include the maturity amount due at $T$. If it is post-maturity, exclude that payment and use zero only when no obligation remains. Do not switch conventions for one state because its label is “absorbing.”

A death benefit payable on a transition before $T$ belongs in $b_{iD}(t)$. If death settlement ends the contract, $V_D(t)=0$ after settlement. If a specified liability remains after entering an absorbing claim state, its reserve is not zero merely because no further transitions are possible. The benefit description and timing, not the state label, determine the boundary.

Setting every terminal reserve to zero. This is incorrect under a pre-payment convention when maturity amounts are due at $T$.
Setting every absorbing-state reserve to zero. Absorption means no exit transitions; it does not rule out future contractual payments.
Counting a payment twice. A transition payment is part of $b_{ij}(t)$; it is not also a maturity amount unless the contract explicitly specifies both.
Adding the transition term. Under the stated convention, the derivative contains $-\lambda_{ij}(b_{ij}+V_j-V_i)$, not a positive version of that term.
Ignoring payment timing at $T$. A payment due at the endpoint changes the pre-payment boundary but not the post-payment value when the contract has ended.

Key takeaway: Specify the instant at which the reserve is measured, assign a terminal value in each state from the contract’s cash flows, and keep transition payments separate from terminal payments. In Thiele’s equation, subtract each intensity-weighted change in liability.

Check Your Understanding

For each problem, use a reserve immediately before terminal payments unless stated otherwise. Assume continuous-time transition intensities and the multi-state Thiele convention given above.

A two-year contract pays \$2,500 at maturity in state $A$, \$5,000 in state $C$, and nothing in state $D$. The contract ends at maturity. Find $V_A(2^-)$, $V_C(2^-)$, and $V_D(2^-)$. (Answer: \$2,500, \$5,000, and \$0, respectively.)
In state $L$, $\delta=0.05$, $V_L=\$2,000$, $c_L=\$80$ per year, and death has intensity $0.10$. A death payment of \$6,000 is made, after which $V_D=0$. Find $dV_L/dt$. (Answer: $0.05(2{,}000)+80-0.10(6{,}000-2{,}000)=-\$220$ per year.)
An absorbing care state pays \$3,200 at maturity, and the contract ends then. What is its pre-payment terminal reserve? What is its post-payment reserve if no liability remains? (Answer: \$3,200 pre-payment; \$0 post-payment.)
In state $W$, $\delta=0.02$, $V_W=\$1,500$, and $c_W=\$100$ per year. A transition to absorbing state $C$ has intensity $0.10$, payment \$300, and $V_C=\$2,500$. Find the derivative contribution from this transition and the full derivative if there are no other transitions. (Answer: Transition contribution $-0.10(300+2{,}500-1{,}500)=-\$130$ per year; full derivative $30+100-130=\$0$ per year.)
A death benefit of \$8,000 is paid when a policyholder transitions from $L$ to $D$. The contract ends on death, and there is no other liability in $D$. State the correct values to use for the transition payment and post-transition reserve. (Answer: $b_{LD}=\$8{,}000$ and $V_D=0$; the payment is included once, in the transition term.)

What You'll Learn