Insurance by BogdanovaThe Encyclopedia of Insurance and Risk Science

RESEARCH DOSSIER

Policyholder Lapse and Surrender Models

Abstract. Policyholder lapse and surrender models translate contractual continuation and exit decisions into the distribution and timing of insurance cash flows. Their central actuarial object is a set of transitions among active, altered, surrendered, expired, deceased, and other legally defined policy states, conditioned on economic circumstances, policy design, and the surviving population. This monograph establishes a comprehensive analytical treatment of these transitions in life insurance and annuities. It distinguishes lapse without a surrender payment, full surrender for value, partial withdrawals, nonpayment and premium persistency, paid-up transitions, and the exercise of other embedded options. It derives discrete-time rates, competing-risk intensities, multistate transition systems, simple surrender-value comparisons, and discounted liability sensitivities; examines dynamic specifications, empirical identification, guarantees, premiums, interest rates, and liquidity; and reconciles these with IFRS 17, U.S. principle-based reserving and European supervisory stress practice. The primary evidence consists of the published SOA/LIMRA experience programs, an empirical lapse study, actuarial standards and official regulatory material. Transparent numerical examples show why increasing the probability of surrender can either increase or decrease a measured liability while increasing immediate cash requirements. All calculations are illustrative rather than calibrated estimates of an individual insurer. The integrated conclusion is that defensible assumptions require a joint model of event identity, conditional incidence, state-dependent amount, survivor composition, and assets available when payments fall due.

Keywords: Policyholder lapse; cash surrender value; lapse and surrender; persistency; premium persistency; competing risks; multiple decrements; dynamic lapse; life insurance; deferred annuities; policyholder options; lapse-supported contracts; mortality selection; contractual service margin; IFRS 17; VM-20; liquidity stress; actuarial experience studies; model validation.

Introduction

The value of a long-duration insurance contract changes with both insured contingencies and policyholder decisions. Coverage may continue after premium interruption, cash value may be withdrawn while some benefits persist, or the entire contract may be surrendered before death or maturity. Each path alters the premiums received, guarantees retained, benefits paid, expenses incurred, and assets required to discharge obligations. A lapse assumption is therefore simultaneously a probabilistic input, a contract interpretation, and a statement about future insurer cash flows. Its statistical meaning depends on the exposure and event coding from which it was derived.

The Society of Actuaries' policyholder-behavior literature review identifies the diversity of modeling approaches and insurer practices in life insurance and annuities (Campbell et al., 2014). An expanding collection of term and whole-life experience, universal-life lapse studies, and annuity experience reports makes the need for precise event distinctions particularly visible (SOA Research Institute and LIMRA, 2024). A single observed percentage can represent a count-based contract exit rate, a face-amount rate, a value-weighted surrender measure, or a calendar-year administrative ratio. These quantities do not have identical probabilistic interpretations or valuation consequences.

The research question is how to specify, estimate, validate, and use policyholder transition models so that their legal meaning, observed experience, economic drivers, reserve impact, and liquidity implications remain coherent. The method combines primary-source comparison, actuarial derivation, transparent numerical calculations, and critical analysis of identification and applicability. Six chapters follow the approved dossier architecture. Their formulas identify assumptions before drawing inferences; empirical estimates are attributed to the actual study populations; and reporting standards are treated according to their stated scope and effective dates.

The monograph covers life insurance, permanent and universal-life contracts, fixed deferred and indexed annuities, and variable annuities where exercise rights determine meaningful long-term cash flows. It addresses internationally relevant actuarial questions using U.S. and European evidence without implying worldwide uniformity in taxation, contractual rights, policyholder protections, or supervisory treatment. It offers original analytical synthesis and reproducible demonstrations. No claim of a new proprietary theory, universal behavioral law, or newly estimated market coefficient is made.

Policyholder Options and Termination

Event taxonomy and insurance contract states

A model must start with the legal contract and the insurer's administration system. In a traditional term policy, failure to pay a required premium after any grace period can terminate coverage with no cash payment. In a permanent-life contract, voluntary surrender may produce a cash value net of charges, loans, or other obligations. A maturing endowment can terminate by scheduled benefit payment rather than by surrender. A death claim, contract expiry, conversion, transfer, replacement, and policyholder-initiated cancellation may also leave the in-force ledger, but they represent different decrements.

Industry reporting sometimes groups lapse without cash value and surrender for value under the word lapse. The 2015–2022 term/whole-life study uses a combined observational construct for these terminations (SOA Research Institute and LIMRA, 2024). Such grouping may be appropriate when estimating a particular persistency table, provided the projected cash flow treatment retains the payment distinction. A surrender with a high payout and a lapse without payout can have the same effect on policy count but opposite effects on immediate cash demands.

Flexible-premium universal life requires separate attention. A period without premium payment can coexist with active coverage if account value, a contractual no-lapse guarantee, or other financing continues to support the policy. Premium persistency is the pattern of contributions, while policy persistency describes whether the contract remains in force. The SOA/LIMRA universal-life persistency program explicitly addresses both. Conflating them can create spurious sensitivity to premium interruption, and it may lead the projection to terminate benefits earlier than permitted by the actual contract.

An annuity permits further distinctions. A withdrawal of 10% of account value may reduce the account and future fees, alter a guaranteed benefit base, or trigger a contractual charge while preserving the annuity itself. Full surrender eliminates the continuing account and changes future guarantee exposure. Annuitization converts an accumulation contract into an income-payment structure. Death benefits and owner-versus-annuitant status add further event identities. The relevant state space must include every option that alters future material payments: active, premium-paying, active without premium, reduced paid-up, partially withdrawn, surrendered, lapsed without value, deceased, matured, converted, or other documented status.

A transition table should preserve effective event dates, administration dates, premium due dates, charge periods, guaranteed crediting rates, account and surrender values, loan balances, insured age, policy duration, and the original exposure class. Reinstatement after lapse requires its own treatment because the policy may leave and later reenter an active state. The state design is adequate when it reproduces material contractual payments and the probabilistic exposure needed for the stated valuation task.

Economic option value and contractual exercise

The right to surrender gives the owner a discretionary cash flow. Let S(t,x) be the net surrender proceeds at time t under contract and economic state x, and C(t,x) the value to the owner of continuing the contract, incorporating expected future premiums, insurance protection, income, guarantees, taxes, costs, and liquidity. An elementary economic comparison asks whether S(t,x) exceeds C(t,x). This comparison is a useful benchmark for how contractual features shape incentives, although owners may exercise because of emergency liquidity needs, imperfect information, advice, credit constraints, or other circumstances even when a financial continuation model favors remaining.

A stylized optimal-stopping equation is

V(t,x)=max⁡{S(t,x), Et,xQ[e−∫tt+Δru duV(t+Δ,Xt+Δ)+Bt,t+Δ]},V(t,x)=\max\left\{S(t,x),\ \mathbb{E}_{t,x}^{\mathbb{Q}}\left[e^{-\int_t^{t+\Delta}r_u\,du}V(t+\Delta,X_{t+\Delta})+B_{t,t+\Delta}\right]\right\},

where B is the net value of interim contractual benefits and payments expressed consistently at time t+Delta. Q is a valuation measure suitable for the modeled financial risks; it is not the real-world frequency with which owners exercise their options. An estimated surrender hazard belongs under the observed behavioral probability measure. The distinction matters whenever option moneyness is translated into actual exits. Observed surrender intensity cannot be equated to optimal stopping under frictionless assumptions, and a derivative price cannot be used directly as an empirical lapse probability.

Contract design determines the payoff comparison. Surrender charges reduce exercise proceeds before scheduled expiry, while market-value adjustments may modify the cash settlement when market rates move. Guaranteed investment rates can increase continuation value when reinvestment returns are low. A variable-annuity living benefit may remain valuable even if the account declines, making an apparent account-value comparison incomplete. The SOA/LIMRA fixed indexed annuity behavior research observes both withdrawal and termination behavior across these complex products (SOA Research Institute and LIMRA, 2024). Its evidence supports distinguishing contractual types without justifying one universal surrender-response coefficient.

Accounting contract boundaries and state survival

A contract boundary governs which premiums and cash obligations enter an applicable accounting measurement. IFRS 17 Insurance Contracts provides a current-fulfillment-cash-flow approach, risk adjustment, and contractual service margin as relevant components of measurement (IASB, 2017, as amended). The IFRS 17 application guidance expressly includes expected policyholder behavior in surrender, renewal, conversion, and cessation of premiums (XRB, 2026, paragraphs B61–B65). Expected behavior and the uncertainty around it enter the measurement according to the applicable reporting model; a long statistical forecasting horizon does not automatically extend the contract boundary.

Survival and exit are also selection mechanisms. Let two insured subpopulations A and B have original weights w_A and w_B, death probabilities q_d,A and q_d,B, and termination probabilities q_l,A and q_l,B, where causes are mutually exclusive over the year. After the year, their relative surviving weights become proportional to w_A(1-q_d,A-q_l,A) and w_B(1-q_d,B-q_l,B). Differential exits alter later mortality and guaranteed-benefit exposure even if the insurer has not changed its original underwriting. This connects lapse modeling directly to multiple-decrement mortality projection.

The SOA's post-level-term lapse and mortality report analyzes lapse and mortality around the expiration of level premiums (SOA, 2014). When premiums rise sharply, some owners may seek replacement insurance; those who cannot obtain equivalent protection can have different incentives to remain. Actual selection magnitude must be established from relevant experience rather than automatically inferred from the existence of a premium shock.

Experience-Based Lapse Rates

Exposure as a scientific definition

Consider a portfolio of contracts active at the beginning of an observation interval. Let N_t count those contracts, D_s the number fully surrendered, D_l lapses without surrender payment, and D_d deaths. For a complete fixed cohort followed over one year, a simple combined termination fraction is

q^term,t=Ds,t+Dl,tNt.\widehat q_{\mathrm{term},t}=\frac{D_{s,t}+D_{l,t}}{N_t}.

The denominator makes this a cohort event proportion. It differs from a rate derived from earned policy-years when policies enter observation partway through the year, are renewed on staggered anniversaries, or depart through administrative transfers. It also differs from a payout-weighted or face-amount-weighted measure. For analysis, the denominator must state precisely which contracts were eligible for a given exit and for what duration.

With actual exposure time T_i in years for policy i and a surrender count D_s, the person-time estimator of a constant cause-specific surrender intensity is

λ^s=Ds∑iTirisk.\widehat\lambda_s=\frac{D_s}{\sum_i T_i^{\mathrm{risk}}}.

Its unit is surrender events per exposed policy-year. The estimator is a measure of intensity, whereas q_s is a probability over an interval. An intensity of 0.12 per year implies an annual exit probability 1-exp(-0.12)=0.11308 under a single constant hazard and no other cause. An annual event probability of 0.12 instead corresponds to intensity -log(0.88)=0.12783 in the same simple model. Substituting intensity for probability can systematically misstate multi-year survival and the timing of expected surrender cash flows.

An industry experience table may use start-of-period count, exposure adjusted for exact entry and exit timing, or a defined approximation such as Balducci. A model using a published rate must preserve the denominator and underlying event convention. When raw observations are available, exposure should be reconstructed from actual dates and reconciled against financial and administrative records. The final report should identify whether contract or account count, face amount, account value, premium, or policy-year exposure is being used.

Competing decrements and cumulative incidence

An active policy can exit through mutually exclusive causes. For cause k, define the intensity

λk(t∣x)=lim⁡Δ↓0Pr⁡(t≤T<t+Δ, J=k∣T≥t, Xt=x)Δ.\lambda_k(t\mid x)=\lim_{\Delta\downarrow0} \frac{\Pr(t\leq T<t+\Delta,\ J=k\mid T\geq t,\ X_t=x)}{\Delta}.

If the K cause-specific intensities are constant over a year, overall active survival is exp(-sum_k lambda_k). For a death intensity lambda_d and surrender intensity lambda_s, the one-year surrender probability is

qs=λsλs+λd[1−exp⁡{−(λs+λd)}],q_s=\frac{\lambda_s}{\lambda_s+\lambda_d} \left[1-\exp\{-(\lambda_s+\lambda_d)\}\right],

and death probability is the analogous formula with lambda_d in the numerator. Continuation has probability exp[-(lambda_s+lambda_d)]. These three probabilities sum to one; if lapse without surrender is modeled separately its intensity belongs in the total and has its own event probability.

For varying intensities, define S(t)=exp[-integral_0^t sum_k lambda_k(u)du]. Cumulative incidence of surrender is F_s(t)=integral_0^t S(u-)lambda_s(u)du. This representation demonstrates why treating death as ordinary independent censoring and estimating one minus the Kaplan–Meier survival for surrender can overstate the cumulative probability of surrender: death permanently removes a contract from exposure to its original surrender event. Cause-specific intensity and cumulative incidence answer distinct questions and must not be substituted casually.

Multi-state contracts require a transition matrix. In a time-homogeneous continuous-time Markov approximation, let generator Q have nonnegative off-diagonal intensities and diagonal entries equal to the negative sum of outgoing intensities. Transition probabilities over interval h are P(h)=exp(Qh). A row of P(h) must sum to one. States such as active, paid-up, reduced coverage, partially withdrawn, surrendered and deceased can carry different premium, guarantee, and mortality cash flows. Whether Markov memorylessness is plausible should be tested; elapsed duration since charge expiration, account depletion, or prior withdrawals may require expanded states or a semi-Markov formulation.

Duration, cohort, calendar and data quality

Three time dimensions must be distinguished: policy duration since issue, issue cohort or vintage, and observation calendar time. Early-duration surrender may depend heavily on commission and surrender-charge design. Issue cohorts can reflect product redesign or historical guarantee rates. Calendar time captures shared interest rates, unemployment and market returns. A portfolio-wide rise in observed surrender can result from any combination of these factors, including changes in the mixture of participating companies. A trend is not identified by two unadjusted aggregate ratios when their underlying risk sets differ.

Credibility requires careful pooling. ASOP 25 on credibility procedures describes relevant professional considerations in selecting and combining actuarial experience (ASB, 2013). A simple demonstration blends an internal estimated probability q_i with an external reference q_ref by q*=Zq_i+(1-Z)q_ref, where 0≤Z≤1. The actuarial judgment lies in defining an appropriate Z and establishing comparability of the two experiences. Hierarchical pooling of log hazards or regression random effects may be preferable when cohorts have different durations or heterogeneous risk profiles. Abundant observations from normal interest-rate conditions do not provide direct evidence of rare systemic withdrawal responses.

Administrative transfers, block acquisitions, death, maturity, observation-period end, and lost records must be classified separately. Right-censoring methods depend on assumptions about why observation ended. An insurer that migrates poor-performing policies off a source system can create informative censoring. Apparent surrender spikes can arise from late posting of transaction dates, product conversion rules, or incorrect grace-period treatment. Source reconciliation is a scientific prerequisite because greater model sophistication cannot recover true causes from systematically misclassified events.

Interpretation of large observational studies

The SOA/LIMRA 2015–2022 term/whole-life report summarizes 135.9 million policy-count exposures and 5.4 million lapses or surrenders from 28 contributing companies (SOA Research Institute and LIMRA, 2024). This is a large observed sample, yet it is not a census of global insurance or a randomized test of economic causes. Contract type, issue age, exposure convention, and the contributing companies affect comparability. Its aggregates support development of credible reference experience within their stated domain.

For fixed deferred annuities, the 2015–2022 experience study reports 13.4 million contract-count surrender exposures and over 900,000 surrenders from 23 companies (SOA Research Institute and LIMRA, 2024). A newer 2023–2024 fixed deferred study describes 4.8 million contract-count exposures, over 567,000 surrenders and $612 billion in value exposure from 24 contributors (SOA Research Institute and LIMRA, 2025). The ratio of total events to total exposures in either publication should be interpreted under that study's definition and mix. Their apparent difference alone cannot establish an effect of higher market interest rates because vintage, charge schedule, product and company composition also change.

Similarly, the 2022–2025 universal-life data request solicits updated information but is not itself a published experience result (SOA Research Institute and LIMRA, 2026). Research completeness requires distinguishing a planned study from an available outcome. These distinctions guide both the responsible use of industry benchmarks and the scope of claims possible from public sources.

Dynamic Surrender Models

Conditional behavioral specifications

A static lapse table specifies rates by duration, product or cohort. Its principal strengths are transparency, simple implementation and potentially robust prediction when economic incentives remain stable. A dynamic model specifies how exit probability changes with time-varying contract and economic information. Let q_s,i,t be the conditional one-period surrender probability. An illustrative logit representation is

logit⁡(qs,i,t)=αd(i,t),p(i)+βrΔri,t+βcci,t+βggi,t+βuut+γ⊤zi,t.\operatorname{logit}(q_{s,i,t})=\alpha_{d(i,t),p(i)} +\beta_r \Delta r_{i,t} +\beta_c c_{i,t} +\beta_g g_{i,t} +\beta_u u_t+\gamma^\top z_{i,t}.

Here d is duration, p is product class, Delta r is the market-rate minus the contract crediting rate, c is the surrender-charge proportion, g is a defined measure of guarantee moneyness, u is an observable macroeconomic stress measure and z covers additional properly specified characteristics. The sign of beta_r may be positive for contracts that become relatively unattractive when alternative rates rise. A higher surrender charge often reduces immediate exercise incentives. Neither sign must be assumed identical across products, ages, guarantees or stressed periods.

If q_s is fitted in isolation while mortality, maturity and other decrements also act, summing separately estimated event probabilities may exceed one. A multinomial model can specify competing one-period states directly. A cause-specific intensity model can instead estimate lambda_k(t|x) for each decrement and convert those intensities into mutually exclusive transition probabilities. A complementary log-log specification q=1-exp[-exp(eta)] has an intensity interpretation for a fixed unit interval, but duration length must be explicit. Annual, quarterly and monthly outputs cannot be compared without conversion.

Dynamic surrender is usefully linked to dynamic amounts. For a withdrawal, one model estimates Pr(W>0|x) and a second estimates E(W/A|W>0,x), with A the account value. The expected cash amount is A times the probability of positive withdrawal times its conditional proportional severity when the identity and conditioning are consistently defined. Guarantee changes generally depend on the entire conditional severity distribution and account path, so replacing partial withdrawals with an assumed full lapse destroys relevant option states. Premium persistency similarly has an incidence and amount dimension; keeping the contract active is a third, distinct result.

Empirical fit and the identification problem

The econometric cointegration analysis by Kuo, Tsai, and Chen studies relations between life insurance lapse rates, unemployment and interest rates and distinguishes short-run from long-run associations (Kuo et al., 2003). It helps define hypotheses concerning reinvestment incentives and emergency liquidity needs. The study does not experimentally identify a universal causal coefficient for current annuities. Financial products, income conditions, surrender charges, owner expectations and distributions have changed since the observation period.

A practical identification difficulty arises when all policyholders in a company experience nearly the same rate history and when unemployment, rates, asset prices and insurer crediting decisions move together. A model might fit historical macro conditions while misassigning influence to the spread variable. Calendar effects can absorb time variation; vintage effects can mimic duration effects; simultaneous product redesign can mimic changing behavioral preference. Credible estimation requires sufficient contrasting observations, appropriate controls, and an explicit assessment of extrapolation outside the observed economic regime.

Covariate units must be recorded. A rate spread represented as 0.02 in decimal units is 200 basis points; a spread represented as 2.0 in percentage points has the same economic meaning but changes fitted coefficient scale by a factor of 100. A regression with a nonlinearity at a 100-basis-point threshold must transform both representations consistently. Report parameter uncertainty, rate regime, included product classes, censoring method, prediction horizon, and statistical target. Predictive discrimination in the training sample is weaker evidence than out-of-time calibration of financially material exits.

Charges, guarantees and contract-event discontinuities

Surrender charge expiry can create clustering near a known anniversary. A smooth age-duration regression can miss a discrete jump that is economically clear from the contract. Piecewise terms for months to charge expiry, the rate of remaining charge, and post-expiry duration may provide a better description, subject to evidence. Some deferred annuity contracts also have market-value adjustments or free-withdrawal windows. An appropriately specified exposure table reflects the exact contractual date of each provision.

For indexed annuities, exercise incentives depend on future crediting options, surrender charges, indexed participation, guaranteed floors and owner-specific circumstances. The SOA/LIMRA 2021–2022 indexed annuity study reports 12 companies, 4.8 million surrender exposures by count, over 227,000 surrenders and $15.9 billion withdrawals (SOA Research Institute and LIMRA, 2024). Its 2023 indexed annuity successor reports 17 companies, around 2.7 million surrender exposures and more than 208,000 surrenders (SOA Research Institute and LIMRA, 2025). These study populations are useful empirical references but should be normalized before making across-year behavioral comparisons.

Variable annuity policyholders may choose surrender, continuing withdrawals, or retaining a rider guarantee that is more valuable than the account balance. The SOA/LIMRA variable annuity behavior study describes 11.5 million contract-count exposures and more than 625,000 surrenders, while reporting $56.7 billion in withdrawals among the studied business (SOA Research Institute and LIMRA, 2025). The distinction is essential: full exit removes future contract exposure, whereas an owner who withdraws funds may preserve a significant insurance guarantee. Both event probability and remaining benefit-base amount must be carried forward.

Scenario paths and measure consistency

In a stochastic projection, a common economic scenario determines interest rates, asset prices, crediting terms and guarantee values. The surrender model calculates the conditional transition probabilities on that path; contract cash flows are generated from those transitions; the asset model supplies cash, investment returns and asset-sale constraints. Repeating the process allows future policyholder decisions to respond to evolving states. The sequence matters: using a surrender intensity calculated before rate changes, while discounting obligations after the changes, would introduce inconsistent assumptions into the same simulated financial world.

The 2026 NAIC Valuation Manual addresses anticipated policyholder behavior in VM-20 section 9.D, including dynamic or scenario-dependent assumptions unless a static approach adequately reflects them, as applicable (NAIC, 2026). It calls for testing and margins within its statutory context. The rule is not a universal directive that all policyholder behavior be simulated under every calculation, and it does not mean a best-estimate fitted parameter is automatically an adequate adverse-risk assumption.

Mathematical option pricing may use risk-neutral financial valuation, whereas empirical lapse models use real-world probabilities. A combined projection has to state its measure, the treatment of nonhedgeable behavior risk, and how assumptions enter reporting or capital calculations. Risk-neutral financial drift cannot simply be inserted into a regression estimated under historical investment returns without considering the conceptual difference. The same contractual option may be priced, experienced and prudentially stressed under distinct probability and measurement frameworks.

Economic and Contract Drivers

Rate spreads and alternative investments

The opportunity cost of continuing a contract increases when a readily available substitute offers greater net economic value, after taxes, switching costs, mortality protection and guarantees. For a liquid, relatively simple fixed-rate account, a rise in external yields above a legacy credited rate can strengthen surrender incentives. The connection weakens when a contract pays a materially valuable guaranteed death benefit, provides protection unavailable on comparable terms, or incurs high charges on exit. A market-rate spread alone is therefore an incomplete measure of contractual moneyness.

An owner may face a competing liquidity motive. Unemployment, disability, debt service or major expenditure can generate cash demand regardless of whether an insurance contract offers an attractive long-run yield. The Kuo–Tsai–Chen lapse study provides historical evidence that unemployment and rate variables may have different dynamic relationships with aggregate lapse (Kuo et al., 2003). A scientifically specified model distinguishes these channels because their timing and interactions differ. For example, high rates may simultaneously raise investment opportunities and borrowing costs, and recession may alter both financial stress and the availability of substitutes.

Policyholder expectations can matter as much as the observed spot yield. A temporary rate increase might induce less exit than an expected persistent rise; conversely, credible expectations of future credited-rate increases can support continuation. There is no automatic reason to impose a linear, instantaneous response. Event-history models can include lagged spreads, nonlinear thresholds, duration since a rate shock, and information interactions where policy-level data support their identification.

Charges, guarantees, advice and information

Surrender penalties affect net proceeds and produce known expiry dates. A tax-deferral benefit or tax penalty may shape the same decision, subject to jurisdictional law. Policy loans and partial withdrawals can supply liquidity without full termination. Distribution advisers may influence replacements, and sales campaigns can produce correlated exits within a channel. An insurer's credit rating, solvency news or communication quality can alter perceived safety of a long-duration guarantee; such factors are harder to quantify from ordinary policy transactions.

These variables are contract-specific and jurisdiction-specific. Whole-life paid-up options, universal-life secondary guarantees, fixed annuity market-value adjustments, variable annuity income riders and health-related replacement constraints cannot be represented by the same economic spread without losing information. The term and whole-life industry evidence is relevant to traditional life products, whereas the fixed annuity experience program samples different contractual economics (SOA Research Institute and LIMRA, 2024; 2025).

Mortality selection and lapse-supported guarantees

Term products can exhibit selective lapse when premiums increase after the guaranteed level period. Policyholders with better replacement insurability may leave while others retain contracts even at substantial cost. This can increase the conditional mortality of survivors relative to the original population. The post-level-term SOA investigation examines the interaction; any quantitative selection assumption must still be calibrated to a relevant block and product design (SOA, 2014).

Lapse-supported pricing and reserving may depend on termination releasing future guarantee obligations. Lower-than-expected lapse can then raise benefit costs, but it may also preserve premium or fee revenue. Higher-than-expected lapse can release guarantees while producing immediate surrender payments or eroding future profits. The financial sign depends on the full set of discounted amounts, not on whether a product is colloquially described as lapse-supported.

The relevant population changes every period. A mortality table derived for an original issue cohort may cease to represent the active population once lapse varies by underwriting class, health status or socioeconomic factors. Joint models or state-conditioned adjustments can accommodate this composition effect, with credibility and sensitivity checks. Treating mortality and lapse as independent probabilities without considering their common determinants is an assumption that should be made visible.

International and regulatory environment

EIOPA's June 2024 Financial Stability Report discusses increases in life-insurance lapses and potential implications for liquidity as financial conditions change (EIOPA, 2024). Its October 2026 Insurance Risk Dashboard characterizes European-sector liquidity and funding risks as medium, with median cash about 0.7% and liquid assets about 46% of total assets for the Q2 2026 indicators (EIOPA, 2026). Sector medians describe an aggregate supervisory picture, not the liquid resources or dynamic lapse propensity of an individual insurer.

Regulatory rules influence product design, available surrender values and consumer rights; differences in law and market practice affect observed behavior. An insurer's transfer of a U.S. annuity lapse model into a European or Asian market therefore requires verification of contract terms, taxes, distribution practices, macroeconomic history and applicable supervision. The same model form may remain useful, but its inputs, effects and validation evidence can differ substantially.

Reserve and Liquidity Effects

Direct liability sensitivity

A contingent insurance liability is a weighted present value of future payments and receipts under a specified measurement basis. The dependence on lapse becomes transparent in one period. Let q_d be death probability, q_s surrender probability, and 1-q_d-q_s continuation probability. Let C_d denote the death payment, S the surrender payment, V_c an end-period continuation value, and i the discount rate. Ignoring premiums and other expenses solely for this example, define

L(qs)=qdCd+qsS+(1−qd−qs)Vc1+i.L(q_s)=\frac{q_d C_d+q_s S+(1-q_d-q_s)V_c}{1+i}.

If q_d, S, V_c and i are held constant, direct differentiation gives

∂L∂qs=S−Vc1+i.\frac{\partial L}{\partial q_s}=\frac{S-V_c}{1+i}.

The identity determines the sign of the direct effect in this simplified model. A surrender payment larger than the value assigned to continuing the policy raises the expected payment when surrender becomes more likely. A surrender payment smaller than the continuation value reduces it. In a complete insurer model, premiums, expenses, death selection, guarantees, future profit margins, changes in surrender amounts, discounting and regulatory margins also respond; those effects can modify or reverse the simple result.

Worked numerical cases

Take an illustrative portfolio unit for which q_d=0.01, q_s=0.06, continuation probability=0.93, C_d=10,000, V_c=1,000, and i=0.04. Let S=1,200. The undiscounted expected end-year payment is 0.01×10,000+0.06×1,200+0.93×1,000=1,102. Dividing by 1.04 gives L=1,059.6154. Raising surrender probability to 0.12 and reducing continuation to 0.87 gives an expectation of 100+144+870=1,114 and L=1,071.1538. The modeled liability increases by 11.5385 units.

Now retain the mortality, continuation value, discount factor and probability change, but reduce S to 600. The first expected payment is 100+36+930=1,066 and L=1,025.0000. At q_s=0.12 it becomes 100+72+870=1,042 and L=1,001.9231. The liability decreases by 23.0769 units. In each case the death, surrender and continuation probabilities sum to one. These are reproducible algebraic examples in hypothetical currency units, not fitted SOA coefficients or estimates of a real insurer's balance sheet.

The two directions establish a practical discipline: sensitivity analysis must define the payout and continuation economics and state what is held fixed. A financial report stating that a ten-percentage-point increase in lapse releases a certain reserve without describing payments, conditional mortality or future revenue leaves the essential causal calculation unspecified.

Independent cash-liquidity exposure

Insurance liquidity is driven by payments falling due, asset maturities, sale proceeds and access to financing. With N contracts at risk, an extra surrender incidence Delta q_s in a specified window, and average additional cash payout S_cash, an initial approximation is

ΔCsurrender≈N Δqs Scash.\Delta C_{\mathrm{surrender}}\approx N\,\Delta q_s\,S_{\mathrm{cash}}.

For 10,000 contracts, a ten-percentage-point additional surrender incidence and a 5,000 cash payment per incremental exit, the extra immediate cash requirement is 5,000,000 units. This result does not depend on whether the long-run measured insurance liability increases or decreases. In practice, payout amounts may vary by policy size, rate sensitivity and guarantee status, so a granular projection sums policy-level expected cash flows rather than multiplying overall averages.

A rising-rate scenario can simultaneously increase the attraction of higher-yield substitutes and reduce the market value of existing fixed-income securities. Cash needed for surrender payments may then require selling assets at losses or borrowing at elevated cost. If statutory surrender proceeds remain insensitive to the contemporaneous sale price of insurer assets, the insurer can face economic basis risk between required payout and liquidatable asset value. Interest-rate duration, credit spreads, asset-liability matching, collateral calls, asset maturities and surrender charges determine the impact. EIOPA's financial-stability discussion of lapses supports treating surrender pressure as a distinct liquidity concern (EIOPA, 2024).

A coherent stress projection links asset prices and owner decisions on the same economic path. It measures immediate net cash flow, available liquidity buffers, realized and unrealized asset losses, collateral demands, post-stress liabilities, and solvency capital. A projection that computes only ultimate liability release can miss a near-term inability to settle claims even when the remaining portfolio appears profitable.

Accounting and statutory interpretations

IFRS 17's fulfillment-cash-flow framework requires expected cash flows within the contract boundary and their uncertainty to be treated through the relevant accounting measurement components (IASB, 2017, as amended). Changes in estimates can affect profit recognition and the contractual service margin differently depending on whether the changes relate to future service, other components and the applicable accounting model. The one-period expectation above is therefore a conceptual building block and cannot be treated as an exact IFRS 17 financial statement calculation.

The U.S. VM-20 policyholder behavior requirements address appropriate anticipated experience, margins, sensitivities, and scenario dependency under their applicable statutory formulas (NAIC, 2026, section 9.D). Certain post-level-term provisions can impose a 100% post-level lapse assumption for specified deterministic-reserve circumstances to prevent reliance on projected post-level profits; that is a targeted prudential rule, not an empirical statement that every affected policyholder will leave.

European solvency capital requirements use their own scenario definitions. In the regulatory context covered by EIOPA's mass-lapse Q&A No. 2290, the delegated-regulation provision specifies a 40% discontinuance event (EIOPA, 2021). The figure defines a regulatory stress for the relevant version and class of policies; it does not forecast annual market behavior. Application in 2026 requires checking operative amendments. Expected accounting cash flows, prudent statutory reserves and prescribed solvency stress outcomes are different quantities. An insurer must not replace one with another because both include the word lapse.

Guarantees, margin and capital interactions

A lapse shock may reduce future guarantee payments, remove fee income, change hedging needs, accelerate acquisition-cost recovery or release margins. For variable and indexed annuities, moneyness can correlate owner exit with market movements in the underlying hedging assets. A higher lapse assumption can reduce one guarantee value while increasing immediate cash payments. The aggregate reserve and solvency response is potentially nonlinear and depends on simultaneous product states.

Mortality selection also creates lagged impacts. Even when exits reduce in-force policy count immediately, the insured lives that remain may have different mortality, premium payment capacity, and exercise incentives. Reporting a single first-year cash outflow is insufficient to describe a multi-decade guarantee portfolio. The full analysis separates cash paid now, policyholder options retained later, asset funding capacity, and the measurement basis used to recognize resulting gains or losses.

Behavior Risk Diagnostics

Source data and reproducibility

Validation begins with contract administration and cash-book reconciliation. Each record needs a stable policy identifier, contract type, state history, premium due dates, actual payments, event-effective and posting dates, surrender value, charges, account or benefit base, exposure units and a justified censoring indicator. Reconcile beginning in-force plus new business and reinstatements less deaths, maturities, full surrender, lapse and other transfers to ending in-force. A mismatch can reveal double counts, delayed postings, missing transfers or a definitional inconsistency.

Data quality checks should flag negative time at risk, impossible ages, withdrawals greater than permitted account value, duplicated surrender dates, continuing benefits after true full termination, and use of expired charge schedules. An administration system may report a lapse effective at the end of a grace period rather than on the missed payment date. That detail can move events between calendar years or distort charge-expiry analysis. A stated event-date convention must be applied consistently across historical estimation and projected cash flows.

The term "experience rate" should always carry denominator, weighting, event definition, time period, product scope and observation unit. In aggregate comparisons, differentiate a change in participant mix from a conditional change in propensity. A claim about macroeconomic cause requires stronger identification than an observed year-on-year difference. Access-controlled study supplements were not acquired for this research, and their detailed tabulations are not silently represented as personally checked microdata.

Structural and numerical checks

A competing-risk implementation must produce probabilities in [0,1] that sum to one across mutually exclusive transitions. If death and surrender are absorbing in a given state system, the model must not later assign active-premium income to those same states. Partial withdrawal and paid-up transitions must update balances, fees, benefit bases and future options. Continuous-time transition generators require valid nonnegative off-diagonal intensities and row sums of zero; discretizations should preserve stochastic transition matrices.

The worked liability example provides a useful unit test. Switching S from 1,200 to 600 under the stated probability change reverses the direct reserve effect from approximately +11.5385 to −23.0769. Another test makes S=V_c and confirms zero direct sensitivity while all other terms remain fixed. Setting lambda_s=0 eliminates the surrender event. A single constant hazard lambda=.12 should yield the annual probability 0.11307956. These tests check identities, units and implementation logic rather than predictive accuracy.

The local ASOP 56 modeling standard provides professional context for design, assumptions, reliance, testing, documentation and communication in its relevant U.S. scope (ASB, 2019). International guidance is reflected in the IAA standards catalog, including ISAP 1 and the IFRS 17-specific ISAP 4, subject to actual jurisdictional or professional adoption (IAA, 2026). Compliance with a professional standard and scientific predictive validation remain different questions that both deserve evidence.

Statistical performance and calibration drift

Out-of-sample testing should separate training observations from policy histories occurring in later calendar periods. Product-vintage and issuer holdouts provide additional tests of transferability. A randomly partitioned policy-month dataset can leak almost identical contracts and interest-rate paths into both training and testing, resulting in optimistic validation. For a proper holdout, compare predicted and observed cause-specific cumulative incidence by duration, charge-expiry month, market spread, product, owner age, face amount and distribution channel.

For fully observed fixed intervals, a Brier score and Bernoulli log loss can assess probabilistic quality. Censored competing-risk data require appropriate risk-set adjustments and time-dependent cumulative-incidence validation; a censored policy should not automatically be labeled a non-lapse. Report calibration plots, uncertainty bounds, model stability and explicit segments with weak observed information. A complex machine-learning specification may improve fitting yet become fragile under a new rate regime; statistical flexibility alone does not establish actuarial decision usefulness.

The expected response under severe economic conditions is particularly uncertain. Stress testing should separate observed baseline experience, coefficient sensitivity, externally specified regulatory shocks and internally designed extreme-but-plausible scenarios. An observed historical rate is not proof of an extreme forward-looking probability. Conversely, a regulatory mass-lapse scenario is a solvency test and should not be reinterpreted as a best estimate of policyholder preference.

Integrated stress and uncertainty

A practical sensitivity program varies interest rates, replacement yields, crediting-rate behavior, surrender charges, mortality selection, premium interruption, unemployment, partial withdrawals, asset-sale liquidity and correlation between policyholders. Examine immediate spikes, delayed exit after charge expiry, persistent elevated withdrawal severity and a lower-lapse scenario in which expensive guarantees remain in force. Translate each variation into expected cash flow by date, reserve, solvency measure, liquid asset buffer and possible forced-sale loss. Sensitivities should be repeated across relevant product segments rather than relying on one portfolio average.

The NAIC 2026 VM-20 framework provides an example of prudential attention to relevant scenario-based policyholder behavior and assumption margins (NAIC, 2026). The EIOPA October 2026 risk assessment shows that sector liquidity is monitored with financial indicators alongside other risks (EIOPA, 2026). Neither alone specifies the entire stress governance protocol for each company.

For disclosure, an insurer or research report should identify which behavior rates were estimated empirically, which were pooled by credibility, which were inferred from economic reasoning, and which were imposed as stress assumptions. Such provenance allows future researchers to distinguish observed frequency from judgment. Consistent attribution also prevents false precision when model parameters are transported across products or jurisdictions.

Analytical synthesis and limitations

Six conclusions follow from the integrated analysis. A lapse rate is inseparable from its exposure and exit definition. A dynamic response is inseparable from the legal cash value, charge and guarantee structure. A statistically significant association is inseparable from the historical population and identification conditions supporting it. A reserve change is inseparable from the value of surrender relative to continuation and the future cash-flow path. A liquidity shock is inseparable from payment timing, asset-market conditions and realizable funding. A model approval is inseparable from documented data, assumptions, uncertainty, scenario coherence and an independent implementation check.

These conclusions constitute analytical synthesis of established actuarial methods and published evidence. They are not presented as a new named theory, proprietary principle or newly discovered law. The research's quantitative calculations are reproducible demonstrations, not empirically fitted hazard coefficients. The public pages of the SOA/LIMRA studies provide reported sample sizes, methodologies and highlights; restricted detailed policy-level supplements have not been analyzed. This monograph therefore makes no claim to insurer-specific estimation, randomized evidence, industry-wide causal rate elasticity or direct production-system validation.

The analysis remains valid as a methodological reference when new experience becomes available, because its claims distinguish contractual identities, mathematical identities, observed studies and regulatory requirements. Quantitative calibration, current law and contract-specific behavior must always be rechecked for the actual application and date. External academic peer review is reserved for a separate decision or mandatory requirement.

Conclusion

A policyholder can change the contractual trajectory of an insurance obligation through lapse, surrender, partial withdrawal, premium cessation, reinstatement or exercise of a guarantee. The actuarial model must describe these as legally meaningful and probabilistically coherent state transitions, preserving both the number of contracts at risk and the resulting money flows. Exposure definitions and competing decrements determine how observed experience can be estimated. Dynamic economic and contractual drivers determine how future probabilities should respond to the modeled environment.

The research shows why the financial consequences of lapse cannot be inferred from an increasing or decreasing termination rate alone. In a transparent one-period valuation identity, the direct sign of surrender sensitivity depends on surrender payment minus continuation value; more complete models incorporate future premiums, expenses, mortality selection, guarantees and reporting conventions. Immediate liquidity exposure follows a separate path: concentrated surrender payments can require asset sales even when the insurer's measured long-term liability declines.

SOA/LIMRA experience programs provide substantial product-specific empirical observations with important comparability limits. IFRS 17, VM-20, relevant actuarial practice standards and EIOPA supervisory materials supply distinct measurement and control contexts. Their integration supports a fundamental scientific conclusion: a credible policyholder behavior model is a traceable link among contract rights, observed transitions, conditional probabilities, future financial obligations, and liquid assets available to discharge them. The complete mathematical and editorial foundation is established here for subsequent policy-specific calibration and implementation.

References