Korea Insurance Actuary common mistakes
Korea Insurance Actuary · Understand the traps and how to avoid them.
Using the force of interest δ in place of the annual effective rate i (or vice versa) when computing a(t)=e^{∫δ ds} versus (1+i)^t, especially in annuity-immediate and annuity-due present values where the timing convention (v^k at k=0 vs k=1) also shifts the result.
Candidates memorize the symbol list but not the conversion δ=ln(1+i) and d=i/(1+i), and under time pressure they grab whichever rate appears in the prompt without checking whether the annuity is due or immediate, or whether payments are continuous.
What to do instead
Before any calculation, write a one-line timeline: payment dates, first payment date, and whether the product is due/immediate/continuous. Then convert all rates to one basis (use i and v=1/(1+i)) and re-derive the annuity factor from the timeline rather than recalling a formula. Practice 20 mixed-timing annuity problems with a checklist.
Mixing up (and failing to state) the survival-probability direction in life-contingency notation: using ₜpₓ where ₜqₓ is required, or computing ₜqₓ as 1−ₜpₓ without adjusting for the correct starting age in multi-life or deferred products.
The symbols look similar and candidates are rushed; they also forget that deferred insurance requires the joint event of surviving the deferment and dying in the next interval, so the factor is ₜ|qₓ = ₜpₓ·qₓ₊ₜ.
What to do instead
For each problem, write the event in words first ('survives t years, then dies within the next year'), then translate to notation. Build a one-page notation card and drill deferred/term/whole-life combinations until the translation is automatic.
Computing reserves prospectively without reconciling to the retrospective formula, and ignoring the recursive relationship V_t = (V_{t−1}+P)(1+i) − q_{x+t−1}(S − V_t) when asked to verify a reserve at a specific duration.
Candidates learn one formula and apply it mechanically; the recursion requires careful indexing of the death benefit timing, and small index errors produce plausible but wrong numbers that pass a sanity check.
What to do instead
Always compute the reserve two ways (prospective and recursive) and require agreement to the cent before moving on. Draw the recursion as a one-year cash-flow box (opening reserve + premium, interest, expected death cost) and label the index t−1 vs t explicitly.
Applying a deductible or policy limit incorrectly in aggregate loss models: e.g., treating a per-loss deductible as if it were a per-payment deductible, or forgetting that the limited expected value E[min(X,d)] differs from E[(X−d)₊].
The distinction between per-loss and per-payment deductibles, and between the deductible and the limit, is subtle; candidates default to the individual-risk formula without checking the collective-risk assumption.
What to do instead
Classify every problem first as individual risk or collective risk, then write the payment function Y = min((X−d)₊, u) explicitly. Compute E[Y] via the survival function (E[Y]=∫ S_Y(y) dy) as a cross-check. Practice separating deductible and limit effects on frequency and severity.
Assuming independence when aggregating risks (summing variances of individual losses) in a collective-risk or portfolio problem where the shared parameter or common shock induces correlation, and thus understating the variance of the aggregate.
The standard variance formula Var(S)=E[N]Var(X)+Var(N)E[X]^2 assumes independence between frequency and severity and across risks; candidates apply it without checking whether a common random parameter (e.g., a shared λ) is present.
What to do instead
Use conditional variance: Var(S)=E[Var(S|Θ)]+Var(E[S|Θ]). Whenever a problem mentions a random or shared parameter, flag it and apply the law of total variance. Drill problems with a common λ and verify the extra variance term appears.
Confusing the risk-neutral pricing measure with the real-world measure in derivative or embedded-option valuation, and discounting expected payoffs at the risk-free rate while using real-world growth assumptions for the underlying.
Candidates learn the Black–Scholes formula as a black box and forget that the drift under the risk-neutral measure is r (or r−q), not the expected return μ; the two measures produce different but superficially similar numbers.
What to do instead
Before valuing any option, state the measure explicitly and the drift used. Re-derive the risk-neutral expectation for a one-period binomial as a sanity check (p*=(e^{rΔt}−d)/(u−d)). Never mix μ from a real-world scenario with discounting at r.
Miscalculating the after-tax cost of debt or the weighted average cost of capital by using the coupon rate instead of the yield to maturity, or by applying the tax shield to equity rather than debt.
Candidates grab the coupon from the bond description because it is a round number, and they forget that the tax shield applies only to interest on debt (r_d(1−T)), producing a WACC that is off by several percentage points.
What to do instead
Always compute r_d from the bond's yield (solve for YTM from price) and apply the (1−T) factor only to the debt component. Write the WACC formula with each component labeled and its weight summing to 1 before substituting numbers.
Treating a regulatory reserve or solvency capital requirement as if it were an accounting provision, or applying the wrong valuation basis (best-estimate vs prudent) when reconciling statutory and IFRS figures.
Korean insurance regulation (Insurance Business Act and supervisory standards) and IFRS 17 use different measurement bases; candidates conflate the two and produce reserves that satisfy neither.
What to do instead
Maintain a two-column comparison sheet: statutory basis vs IFRS 17 (fulfilment cash flows plus contractual service margin), with the key differences in discounting, risk adjustment, and profit recognition. For any reconciliation problem, label each line by basis before computing.
Misclassifying a product feature or regulatory requirement: e.g., treating a variable/unit-linked product's investment risk as borne by the insurer, or ignoring the cooling-off (청약철회) period and disclosure obligations in a product-management question.
Product-management questions test the link between design, risk-bearing, and consumer-protection rules; candidates answer from general knowledge and overlook Korea-specific disclosure and withdrawal rights.
What to do instead
Build a product matrix: product type × who bears investment risk × guaranteed features × disclosure/withdrawal requirements. Review the Insurance Business Act provisions on disclosure, suitability, and cooling-off before the exam and map each product to its obligations.
Using a normal approximation for a small-sample or heavily skewed statistic (e.g., aggregate claims with few policies, or a binomial proportion with np<5), and reporting a confidence interval without checking the approximation's validity.
The normal approximation is the default in candidates' toolkits; they skip the validity check and the continuity correction, producing intervals that are too narrow or shifted.
What to do instead
State the sample size and skewness before choosing a method; use exact (binomial/Poisson) or bootstrap methods when np or n(1−p) is small. Apply a continuity correction for discrete data and always report the method used. Drill 10 problems where the normal approximation fails and compare with the exact answer.