Tutorials › Exam ALTAM - Advanced Long-Term Actuarial Mathematics › Common Sign and Boundary Errors in Thiele Calculations

Exam ALTAM · Thiele Differential Equations · Tutorial 558 of 999

Common Sign and Boundary Errors in Thiele Calculations

Use componentwise and matrix checks to diagnose sign, orientation, time-direction, and boundary-condition errors in Thiele reserve calculations.

Advanced 12 min read

What You'll Learn

  • Apply a consistent sign convention to premiums, ongoing benefits, and transition benefits.
  • Check each transition term using the destination reserve minus the origin reserve.
  • Translate componentwise Thiele equations into a matrix equation with a stated vector orientation.
  • Distinguish explicit Euler steps backward in time from implicit steps.
  • Set terminal reserves according to the timing of maturity benefits and the contract boundary.
  • Use limiting and arithmetic checks to locate errors in reserve derivatives.

Why Thiele Calculations Need an Error Check

A Thiele equation combines interest accumulation, premium income, ongoing outgo, transition benefits, and changes in reserve when the insured moves between states. A calculation can look plausible and still have the wrong sign or boundary condition. The safest approach is to establish a cash-flow convention, write the equation state by state, and check how each transition changes the insurer’s liability.

This tutorial uses reserves defined as the prospective value of future benefits and expenses less future premium income. Positive premiums are therefore income to the insurer, while positive benefits and expenses are outgo. Time \(t\) runs forward from issue, even when a numerical solution is calculated backward from a known terminal reserve. These checks are aligned with the SOA Exam ALTAM (Advanced Long-Term Actuarial Mathematics) syllabus.

Definition: In state \(i\), let \(V_i(t)\) be the reserve at time \(t\), \(\delta\) the force of interest, \(p_i(t)\) the premium income rate, and \(c_i(t)\) the ongoing benefit and expense outgo rate. For a possible transition from state \(i\) to state \(j\), let \(\mu_{ij}(t)\) be its transition intensity and \(b_{ij}(t)\) the outgo paid at that transition. With reserves evaluated immediately after time-\(t\) cash flows, the Thiele equation is

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

All cash-flow rates and reserves must use consistent units. The equation applies on intervals where the stated intensities, cash flows, and reserve definitions apply; a scheduled boundary payment must be handled through the appropriate boundary condition.

How to Read the Transition Term

If the insured moves from \(i\) to \(j\), the insurer pays \(b_{ij}\) at the transition and then holds reserve \(V_j\), instead of \(V_i\). The transition changes the liability by \(b_{ij}+V_j-V_i\). The intensity-weighted sum is the expected rate of this change; it is subtracted in the forward reserve equation. The reserve difference is destination minus origin, not the reverse.

The other signs follow the same prospective-reserve convention. Interest contributes \(+\delta V_i\). Premium income contributes \(+p_i\) to the time derivative: as time advances, fewer future premiums remain to reduce the prospective liability. Ongoing outgo contributes \(-c_i\).

Justification: Over a short interval of length \(h\), a transition \(i\to j\) occurs with probability \(\mu_{ij}(t)h+o(h)\). If it occurs, the immediate outgo plus the new reserve is \(b_{ij}+V_j(t+h)\); if it does not, the reserve is \(V_i(t+h)\). Discounting the future value at force \(\delta\), and including ongoing net outgo \(c_i-p_i\), gives to first order

\(\displaystyle V_i(t)=V_i(t+h)-\delta V_i(t)h+(c_i-p_i)h+\sum_{j\ne i}\mu_{ij}\bigl(b_{ij}+V_j-V_i\bigr)h+o(h).\)

Rearranging, dividing by \(h\), and taking \(h\to0\) yields the stated Thiele equation. The approximation assumes transition intensities are finite and that the probability of two or more transitions in the short interval is \(o(h)\).

Worked Example: Auditing the Transition Signs

At time \(t\), a hypothetical policy is in state \(A\). Use thousands of dollars for reserves and benefits. The data are \(V_A=50\), \(V_B=80\), and \(V_D=0\), where \(D\) is a death state. The intensities are \(\mu_{AB}=0.02\) and \(\mu_{AD}=0.01\) per year. A transition from \(A\) to \(B\) pays 5, and death pays 100. Assume \(\delta=0.04\), premium income \(p_A=3.4\) thousand dollars per year, and no ongoing outgo. Find \(dV_A/dt\).

Method. Apply the componentwise equation, calculating each destination-minus-origin transition change before multiplying by its intensity.

For \(A\to B\), the liability change is \(5+80-50=35\) thousand dollars. Its intensity-weighted contribution is \(0.02(35)=0.70\) thousand dollars per year.

For \(A\to D\), the liability change is \(100+0-50=50\) thousand dollars. Its contribution is \(0.01(50)=0.50\) thousand dollars per year. The total transition contribution is \(0.70+0.50=1.20\).

The interest and premium terms are \(\delta V_A=0.04(50)=2.00\) and \(p_A=3.40\), both in thousands of dollars per year. Therefore

$$ \frac{dV_A}{dt}=2.00+3.40-1.20=4.20 \quad\text{thousand dollars per year.} $$

The derivative is positive: under these assumptions the reserve in state \(A\) is increasing at this instant. Common mistake avoided: reversing \(V_B-V_A\) in the first transition gives \(5+50-80=-25\), rather than \(35\). Reversing both reserve differences would produce a transition sum of \(0.02(-25)+0.01(150)=1.00\), and an incorrect derivative of \(5.40-1.00=4.40\) thousand dollars per year. Checking each transition separately reveals the error.

Matrix Orientation: State the Convention First

Matrix notation is a frequent source of sign and orientation errors. Use a row generator \(Q(t)\): its off-diagonal entry \(q_{ij}=\mu_{ij}\), and its diagonal entry is \(q_{ii}=-\sum_{j\ne i}\mu_{ij}\). Let \(\mathbf V\) be a column vector of reserves, and let \(\mathbf p\) and \(\mathbf c\) be column vectors of premium and ongoing outgo rates. Define \(r_i=\sum_{j\ne i}\mu_{ij}b_{ij}\), the expected rate of transition-benefit outgo in state \(i\). Then the component equations can be written as

$$ \mathbf V'=\delta\mathbf V+\mathbf p-\mathbf c-\mathbf r-Q\mathbf V. $$

The term \((Q\mathbf V)_i=\sum_{j\ne i}\mu_{ij}(V_j-V_i)\) contains the reserve changes but not the transition benefits. Thus subtracting both \(\mathbf r\) and \(Q\mathbf V\) reproduces the componentwise transition sum. This displayed form assumes a common force of interest for the states shown; if forces differ by state, the interest term must be formed state by state.

Worked Example: Checking a Column-Vector Calculation

A hypothetical contract has transient states \(A\) and \(B\) and a death state \(D\). Use thousands of dollars and annual rates. The transition intensities are \(\mu_{AB}=0.1\), \(\mu_{AD}=0.2\), \(\mu_{BA}=0.3\), and \(\mu_{BD}=0.1\). At the time considered, \(\mathbf V=(40,70,0)^\mathsf{T}\), \(\delta=0.05\), and \(\mathbf p=(0.8,0.4,0)^\mathsf{T}\); ongoing outgo is zero. Transition benefits are \(b_{AB}=2\), \(b_{AD}=100\), \(b_{BA}=1\), and \(b_{BD}=60\). Find the reserve derivatives in states \(A\) and \(B\).

Method. Build \(Q\) with rows representing the state of origin and multiply it by the column vector of reserves. Then calculate transition-benefit outgo separately.

The generator and matrix product are

$$ Q= \begin{pmatrix} -0.3&0.1&0.2\\ 0.3&-0.4&0.1\\ 0&0&0 \end{pmatrix}, \qquad Q\mathbf V= \begin{pmatrix} -0.3(40)+0.1(70)+0.2(0)\\ 0.3(40)-0.4(70)+0.1(0)\\ 0 \end{pmatrix} = \begin{pmatrix} -5\\-16\\0 \end{pmatrix}. $$

The transition-benefit rates are \(r_A=0.1(2)+0.2(100)=20.2\) and \(r_B=0.3(1)+0.1(60)=6.3\), in thousands of dollars per year. In state \(A\), interest plus premium is \(0.05(40)+0.8=2.8\), so

$$ V_A'=2.8-20.2-(-5)=-12.4. $$

In state \(B\), interest plus premium is \(0.05(70)+0.4=3.9\), so

$$ V_B'=3.9-6.3-(-16)=13.6. $$

Thus \(V_A'=-12.4\) and \(V_B'=13.6\) thousand dollars per year. As a componentwise check, the transition terms are \(0.1(2+70-40)+0.2(100+0-40)=3+12=15\) in state \(A\), giving \(2.8-15=-12.2\). This does not match the matrix calculation, so it signals an arithmetic issue: \(0.1(32)=3.2\), not \(3\). The corrected sum is \(3.2+12=15.2\), and \(2.8-15.2=-12.4\), as above. For state \(B\), the sum is \(0.3(1+40-70)+0.1(60+0-70)=-8.7-1=-9.7\); hence \(3.9-(-9.7)=13.6\). Both checks agree.

Common mistake avoided: with this convention, \(\mathbf V\) is a column vector and \(Q\mathbf V\) is the relevant product. Replacing it with \(\mathbf VQ\) is not valid: the dimensions and interpretation differ.

Backward Time and Terminal Reserves

Thiele equations are written with the derivative in forward time, \(V_i'(t)=f_i(t,\mathbf V(t))\). A known reserve at the end of the contract or interval is a boundary value; it does not change the direction in which the derivative is defined. To approximate one step backward from \(t\) to \(t-h\), an explicit Euler step uses the derivative at the known endpoint:

$$ \mathbf V(t-h)\approx \mathbf V(t)-h\,\mathbf f(t,\mathbf V(t)). $$

This is an explicit Euler step backward in time. An implicit step instead uses the derivative at the unknown earlier endpoint and satisfies \(\mathbf V(t-h)=\mathbf V(t)-h\,\mathbf f(t-h,\mathbf V(t-h))\). Confusing these updates can lead to a numerical method being described or implemented incorrectly.

The terminal condition must match the timing of the final cash flow. If a maturity benefit is paid at time \(T\), the reserve immediately before payment includes that benefit; immediately after payment, the benefit has left the insurer and must not still be included. State-contingent maturity benefits can therefore give different terminal reserves by state. Writing down whether \(V_i(T)\) is before or after the boundary payment is essential.

Worked Example: Integrating Backward From a Zero Terminal Reserve

A hypothetical contract pays a continuous benefit at a rate of 100 dollars per year while its insured remains in a single active state. There are no premiums or transitions. The force of interest is \(\delta=0.10\), and the contract ends at \(T=1\) year with \(V(1)=0\). Use two explicit Euler steps of \(h=0.5\) year to estimate \(V(0.5)\) and \(V(0)\), then compare with the exact prospective reserve.

Method. With ongoing outgo \(c=100\), the equation is \(V'=0.10V-100\). Start at the stated terminal reserve and step backward, evaluating the derivative at each known endpoint.

At \(t=1\), \(V(1)=0\), so \(V'(1)=0.10(0)-100=-100\). The first backward step gives \(V(0.5)\approx0-0.5(-100)=50\) dollars.

At \(t=0.5\), the derivative evaluated using this approximation is \(V'(0.5)\approx0.10(50)-100=-95\). The second step gives \(V(0)\approx50-0.5(-95)=97.50\) dollars.

For comparison, the exact prospective value at time \(t\) is the present value of the remaining continuous benefit payments:

$$ V(t)=\int_t^1 100e^{-0.10(s-t)}\,ds =\frac{100}{0.10}\left(1-e^{-0.10(1-t)}\right). $$

Therefore \(V(0.5)=1000(1-e^{-0.05})\approx48.7706\) dollars and \(V(0)=1000(1-e^{-0.10})\approx95.1626\) dollars. The Euler estimates are above these exact values, as expected here from the stepwise approximation. Rounded to the nearest cent, the exact values are \(\$48.77\) and \(\$95.16\); the two-step estimates are \(\$50.00\) and \(\$97.50\). Common mistake avoided: adding \(hV'(t)\) instead of subtracting it reverses the direction of the time step and would give \(V(0.5)=-50\), contrary to the positive prospective value.

A Practical Error-Checking Sequence

When auditing a calculation, do not begin by changing signs until the convention is clear. Use the following sequence to isolate the source of a discrepancy.

1
State the reserve convention.
Confirm that reserve means future outgo less future income, and identify which cash flows are paid at transitions or continuously.
2
Write one component equation.
For each transition, record its intensity, benefit, origin reserve, and destination reserve before summing.
3
Check the units and signs.
Verify that reserves and benefits use the same currency units, rates use a consistent time unit, premiums are income, and benefits are outgo.
4
Check the vector convention.
State whether \(Q\) is a row generator and whether \(\mathbf V\) is a column vector. Verify a matrix result against at least one component equation.
5
Check time direction and boundaries.
Use forward-time derivatives even when stepping backward, and specify whether a terminal reserve is before or after a terminal payment.

A derivative check can also reveal a sign error. For example, if an expected transition usually pays a large benefit and moves to a state with a larger reserve, then \(b_{ij}+V_j-V_i\) is positive and its contribution reduces \(V_i'\). If it is instead a transition to a lower-reserve state with little benefit, the liability change can be negative; subtracting it increases \(V_i'\). A negative transition term is not automatically an error.

Key takeaway: In the componentwise Thiele equation, each transition contributes \(-\mu_{ij}(b_{ij}+V_j-V_i)\). Keep the forward-time derivative, state the matrix orientation, and set boundary reserves according to the timing of boundary cash flows. These checks are complementary, not interchangeable.

Check Your Understanding

For each problem, use the reserve convention and time direction stated in the tutorial.

  1. A state has \(V_i=20{,}000\), \(\delta=0.03\), premium income of \(\$600\) per year, and ongoing outgo of \(\$100\) per year. There are no transitions. Find \(V_i'\). (Answer: \(\$1{,}100\) per year.)
  2. A state has \(V_i=25{,}000\), \(\delta=0.02\), and premium income of \(\$300\) per year. It transitions to a state with reserve \(\$40{,}000\) at intensity \(0.04\), paying \(\$2{,}000\) at transition. There is no other outgo. Find \(V_i'\). (Answer: \(\$120\) per year.)
  3. A hypothetical contract pays a maturity benefit of \(\$30{,}000\) at time \(T\). State whether the reserve immediately before payment includes that benefit, and give the reserve immediately after payment if no other future benefits or premiums remain. (Answer: Immediately before, \(\$30{,}000\); immediately after, \(\$0\).)
  4. For \(V'=0.2V-40\), \(V(2)=0\), take explicit Euler steps backward with \(h=0.5\). Find the estimates of \(V(1.5)\) and \(V(1.0)\). (Answer: \(V(1.5)\approx\$20\), \(V(1.0)\approx\$38\).)
  5. Use \(Q=\begin{pmatrix}-0.5&0.5\\0.25&-0.25\end{pmatrix}\), \(\mathbf V=(10,30)^\mathsf{T}\) thousand dollars, \(\delta=0.04\), and zero premiums, outgo, and transition benefits. With \(\mathbf V'=\delta\mathbf V-Q\mathbf V\), calculate the two derivatives. (Answer: \((-9.6,\ 6.2)^\mathsf{T}\) thousand dollars per year.)
Clinical Biostats