RESEARCH DOSSIER
Duration and Convexity in Insurance Asset-Liability Management
Abstract. Duration and convexity are central measures of the response of fixed-income assets and insurance liabilities to changes in discount rates, yet their conventional interpretation becomes incomplete when insurance cash flows depend on mortality, longevity, policyholder behavior, guarantees, inflation, credit events, and supervisory valuation bases. This monograph develops the first- and second-order mathematics of insurance asset-liability management (ALM) from discounted cash-flow valuation, distinguishing Macaulay duration, modified duration, effective duration, convexity, dollar sensitivity, key-rate exposures, and the economic surplus derivatives that govern a matched balance sheet. It reconstructs the intellectual trajectory from Macaulay's bond analysis through Redington's local immunization conditions and subsequent work on term structures; analyzes why equal average duration cannot guarantee stability under nonparallel shocks; and examines the circumstances in which convexity helps or harms an insurer's surplus. Original, explicitly stylized and independently recalculated examples demonstrate equal-value/equal-duration portfolios with systematically different convexities, the consequences of mismatched dollar duration in the presence of a surplus, and the failure of a parallel-shift hedge under segmented yield-curve movements. The comparative analysis connects mathematical measures to international insurance supervision, U.S. actuarial cash-flow standards, European prudential valuation, the U.K. matching-adjustment regime, and IFRS insurance accounting. Contemporary European stress tests are interpreted as conditional evidence rather than universal causal estimates. The synthesis establishes a practical order of scientific interpretation: define the liability and asset valuation bases, identify the curve and cash-flow shock, calculate the surplus derivatives, distinguish local approximation from path-dependent performance, and test liquidity and capital consequences separately. The investigation proposes no new theory, model, doctrine, or formal principle. Its contribution is an integrated, traceable interpretation of established financial and actuarial techniques for insurance obligations.
Keywords: duration; convexity; insurance asset-liability management; Macaulay duration; modified duration; effective duration; option-adjusted duration; dollar duration; surplus sensitivity; immunization; Redington; Fisher and Weil; interest-rate risk; liability discount curves; key-rate duration; term-structure risk; cash-flow matching; annuities; surrender options; guarantees; Solvency II; matching adjustment; Solvency UK; IFRS 17; IFRS 9; IAIS; EIOPA; cash-flow testing; actuarial model risk.
Introduction
Life insurers and annuity writers transform premiums and investment income into promises payable over many years. These promises are often contingent on survival, death, policyholder choice, legal settlement terms, or market-linked guarantees. Assets are usually held in tradable securities, loans, derivatives, cash, and other investments whose market values change when interest-rate expectations change. A comparison of asset duration and liability duration is therefore an economical initial description of a particular risk, but the scientifically relevant quantity is the change in the economic value and payment capacity of the combined asset-liability position.
International supervision explicitly recognizes the problem. The IAIS ALM policy requirement in ICP 16.5 calls for management of asset-liability interactions, cash-flow coverage, and changes in economic value under different scenarios. The IAIS states that matching should manage mismatches effectively rather than seek the greatest feasible resemblance between asset and liability portfolios. This wording is important because a liability duration target is meaningful only after defining liabilities, the economic factors measured, the available assets, and competing demands on capital and liquidity.
Duration measures a first derivative of value with respect to a defined rate or set of rates. Convexity measures a second derivative. Taken together, they describe the local geometry of a valuation function. Insurance ALM extends this geometry across two valuation functions—assets and liabilities—and then assesses the difference. For a smooth one-factor parallel shift, matching value-weighted duration removes the first-order rate effect on economic surplus; the remaining second-order effect depends on value-weighted convexity. An insurer can therefore have no first-order exposure and still suffer a predictable local second-order loss from movements in either direction.
The technical precision of that statement is essential. A duration of eight years obtained from cash flows discounted at a flat continuously compounded yield differs in definition from an effective duration obtained by revaluing an insurance balance sheet under a prescribed regulatory risk-free curve and adjusted behavioral assumptions. Neither number automatically measures liquidity needs, asset credit defaults, lapse losses, accounting volatility, or the profitability of holding the assets to maturity. Insurers can use related measures for different decisions, but comparison requires consistent shock definitions, currency, valuation time, inclusion of options, and cash-flow boundaries.
This Research Dossier concentrates on the mathematical meaning and insurance use of duration and convexity. The broader subjects of asset-liability portfolio construction, stochastic ALM optimization, hedge execution, and collateral operations are examined insofar as they establish the validity and limitations of those sensitivities. Its canonical Science home is Insurance Finance and Asset-Liability Management; additional approved Science connections are Actuarial Science, Annuities and Retirement Insurance, and Solvency and Capital. Each connection uses the same research identity and canonical target.
The analysis has an information cutoff of November 14, 2026. Historical intellectual attributions are verified against publisher or institutional bibliographic records. Applicable supervisory rules are read in their dated jurisdictional settings. The worked figures are constructed illustrations rather than insurer observations or investment recommendations.
Research Methodology, Notation, and Evidentiary Boundaries
The method combines formal derivation, comparative primary-source interpretation, transparent deterministic calculations, and critical reading of empirical supervisory publications. Equations follow explicitly stated conventions. A symbol representing duration is not assumed to have the same definition in all regimes. Whenever a valuation function can change its projected cash flows in response to the rate being shocked, its derivative must include that dependence.
Let time zero be the valuation date. Let A denote the current economic value of a specified asset portfolio and L the current economic value of specified insurance liabilities on a consistent economic basis. Define economic surplus S = A − L. A, L, and S are currency amounts. A rate change Δy is measured as a decimal, so 100 basis points equals 0.01. The sign convention for a positive liability is a future payment obligation. Unless otherwise stated, formulas initially assume fixed nonnegative future cash flows, deterministic discounting, continuously compounded annual rates, no default, and a common scalar curve shift. These assumptions are then relaxed deliberately.
This single-number surplus is not a universal statutory capital metric. An accounting liability under IFRS 17, a Solvency II technical provision, and a U.S. statutory reserve use materially different purposes and components. Even when the insurer computes the duration of each, the resulting durations refer to the valuation bases actually used. Comparability follows the measurement contract, not the label.
The primary-source hierarchy is as follows. The NBER record of Macaulay's book supports the historical date and title. The publisher record of Redington's article, the Fisher–Weil journal record, and the Boyle stochastic-immunization record establish scholarly provenance. The full original Fisher–Weil text was not accessed in this research session; its abstract-level subject and bibliographic attribution are not treated as a verified replication of every result. Modern legal, accounting, actuarial, and insurance supervision statements are taken directly from primary institutional documents, within their stated scope. EIOPA monitoring and stress testing are evidence about defined reporting populations and modeled shocks, not universal laws.
The mathematical examples are recalculated from formulas in this monograph and reported to a sensible number of decimals. The original analytical synthesis concerns the interaction of those results with contingent insurance liabilities and different institutional valuation requirements. No originality is claimed for standard duration formulas, immunization mathematics, or the official regulatory provisions.
Value Derivatives — Concepts, History, and Scientific Object
Economic value as a function of rates and cash flows
A valuation process takes a dated and possibly contingent set of payments and produces a present value using a specified curve, discounting mechanism, and risk treatment. In a deterministic illustration with dated payments c(t_i), a continuously compounded zero-rate curve z(t) gives
A uniform increment of all zero rates by y changes the value to
When the projected payments are independent of y, its first two derivatives at zero are
For positive cash flows, the first derivative is negative and the second derivative is positive. The geometry arises from discounting and does not require a security to be a conventional bond. Deterministic insurance benefit payments can be studied using the same derivatives, although mortality-adjusted benefit streams and net cash-flow obligations require additional care.
The economic value of a life insurance obligation often includes expected benefits, expenses, premiums, and options. An effective value may include a risk adjustment, guarantee costs, or prudential components that are not expressible by a single list of fixed, positive payments. Its value derivative should then be computed from the actual specified liability valuation function. The identity connecting sensitivity to average payment time holds for simple fixed cash flows, while a reprojected insurance liability requires a more general derivative.
Rate risk is multidimensional. One might alter a government yield curve, a swap curve, a risk-free curve extrapolated beyond liquid maturities, a credit spread, an inflation curve, or a regulatory adjustment. Labeling all such movements “interest rates” conceals economically different changes. Duration should therefore be understood as the derivative of a specified valuation function with respect to a specified variable.
Development of duration and immunization
Frederick R. Macaulay's 1938 investigation of interest rates and bond yields is the historical anchor for the weighted-average-time measure that bears his name. It arises naturally from the time distribution of discounted payments. That historical relationship should not be confused with a claim that the modern option-adjusted practices of insurers were already fully established in 1938.
F. M. Redington's 1952 life-office valuations study provides the classical actuarial point of reference for local immunization of an insurer's surplus against small interest-rate movements. Redington's setting focused attention on the first and second derivatives of asset and liability value and on the need to consider both sides of an insurer's balance sheet. The precise local conditions are derived below, with all assumptions made explicit.
Fisher and Weil's 1971 interest-rate fluctuation study addressed bondholder strategies and interest-rate risk in a framework that moved beyond a naive single-yield intuition. P. P. Boyle's 1978 stochastic term-structure investigation makes explicit that rate dynamics, model parameters, and uncertainty about investment conditions affect what immunization can reasonably achieve. This intellectual succession explains why local deterministic protection remains useful while failing to amount to an unconditional, permanent hedge.
There is a practical historical lesson in these developments. The mathematical idea of removing an exposure is always relative to a chosen family of shocks. A solution found for an infinitesimal parallel movement of a curve can be a correct solution to that problem and an incomplete description of the insurer's total risk. Expanding from single-rate duration to term-structure, stochastic, and behavioral analysis does not discredit the original measure; it identifies the scope of what it measures.
Conventional derivative measures and measurement choices
For a positive value P and scalar rate y, define continuous modified duration D = −P'(y)/P(y), with units of years for a unit change in an annually quoted continuously compounded rate. Define convexity C = P''(y)/P(y), with units of years squared. Dollar duration is P D. Dollar convexity is P C. A positive duration corresponds to a value decline after a small rate increase under the specified derivative.
Effective duration is commonly computed by repricing under two nearby shocks:
Effective convexity is obtained from
The effectiveness of these approximations depends on a symmetric shock convention, the bump h, valuation precision, and whether optional cash flows are reprojected. In insurance applications, a reported effective convexity may be negative where a conventional optionless positive-cash-flow bond has positive convexity. That sign difference is a result about the modeled valuation function; it does not indicate an arithmetic error by itself.
An internal sensitivity measure should be accompanied by its rate variable, basis, bump size, cash-flow inclusion policy, and valuation date. Those items are part of the scientific definition of the reported quantity.
Macaulay and Modified Duration — Formal Financial and Actuarial Mechanisms
Discrete discounting and duration identities
Consider fixed annual cash flows c_t at integer years t, discounted by a flat annual effective yield y:
Let w_t = [c_t/(1+y)^t]/P(y) for positive payments and P>0. The weights sum to one. Macaulay duration is the present-value-weighted average date of payment:
Differentiate directly:
Consequently modified duration under this annual-effective-yield convention is D_mod = D_M/(1+y). This distinction matters whenever a modified-duration number is translated into a currency change. The familiar approximation ΔP/P ≈ −D_mod Δy uses the yield convention that generated D_mod; using the continuously compounded formula without conversion produces a convention error.
For half-yearly compounding, a nominal annual yield with two compounding periods requires a different derivative with respect to that nominal annual yield. Where c_j occurs at period j and yield y is nominal with frequency m, P(y)=Σc_j(1+y/m)^{-j} and the modified sensitivity per unit annual nominal y equals the Macaulay duration measured in years divided by 1+y/m. The analyst must retain the selected convention consistently throughout an asset-liability comparison.
The simple present-value-weighted-mean interpretation depends on positive flows. Insurance net cash flows can change sign when premiums and benefits are combined. In that setting, weights defined from signed cash flows are not probabilities; duration can be outside the apparent range of payment dates or become numerically unstable as the net present value approaches zero. Risk management should then emphasize dollar derivatives of separately classified obligations and receipts rather than an apparently precise but ill-conditioned duration ratio.
Continuous discounting and dollar sensitivities
With a flat continuously compounded y,
For a zero-coupon payment at time T, duration equals T and convexity equals T². For a positive portfolio, the convexity is a weighted second moment of payment time. Therefore
The difference between convexity and squared duration is exactly the present-value-weighted variance of payment dates, provided the fixed positive cash-flow and continuous-discount assumptions hold. This identity supplies an unusually clear economic interpretation: among portfolios with the same positive present value and the same continuous duration, more dispersed payment dates produce greater convexity. It is a statement about ordinary probability-style moments of discounted cash flows, not a universal assertion about assets containing options or defaultable cash flows.
The first-order currency response is ΔP ≈ −P D Δy. For risk systems reporting basis-point sensitivities, a one-basis-point rate movement has magnitude approximately P D × 0.0001 in currency. Some institutions call this PV01 or DV01, but sign conventions vary: one desk reports a signed derivative and another reports a positive price loss from a yield rise. The underlying derivative and its sign definition should be recorded explicitly.
A single bond duration quoted without its market value cannot determine the insurer's aggregate sensitivity. A $200 million asset position with duration five has $1,000 million-years of dollar duration; a $150 million liability position with duration seven has $1,050 million-years. The raw numbers five and seven do not reveal the relative first-order surplus exposure without the currency weights.
Liability timing, reserve projections, and present-value weights
An annuity's deterministic expected-payment stream may be written as c_t = b_t p_t, where b_t is a contractual payment conditional on survival and p_t is a survival probability under an identified valuation basis. When p_t and b_t are held fixed under the rate shock, the same duration formulas apply to expected cash flows. If mortality improvement, indexation, expenses, lapse rates, or dynamic management rules are reprojected when rates move, the derivative contains terms beyond discounting.
For a life insurer's net liability, premium inflows can be subtracted from benefit and expense outflows. That net may have large offsetting positive and negative streams. A duration calculated as an average date of net payments is then susceptible to cancellation. Sound sensitivity analysis records gross benefit obligations, premiums where relevant, asset cash flows, and the actual net valuation separately, identifying when the net value is small compared with its component values.
Cash-flow projection is itself a regulated professional task in some settings. The U.S. Actuarial Standards Board's ASOP No. 7 cash-flow standard, revised in December 2025 and effective for work performed on or after June 1, 2026, explicitly addresses life and health cash-flow risks and the scope of projected assets and liabilities. The standard calls attention to optionality, interim values, scenario design, model and data reliance, and disclosures. It supplies professional process requirements rather than a rule that a particular duration-gap target always solves the problem.
Surplus duration and immunization conditions
Let A(y) and L(y) be values recalculated with the same scalar rate shift y, and let S(y)=A(y)−L(y). Under a common valuation basis and differentiability,
At the valuation date, first-order local surplus immunization requires A D_A = L D_L. This is the dollar-duration condition. Equal duration D_A = D_L is sufficient only when A = L, apart from special cases such as zero duration. If an insurer has A>L, and both asset and liability durations are positive, the asset duration that sets the first derivative to zero is D_A=(L/A)D_L, which is shorter than liability duration.
At a point where S'(y)=0, the surplus is locally minimized by small rate changes if S''(y)>0 under the modeled one-factor shock; it is locally maximized if S''(y)<0. The former is the conventional second-order protection sought in classical immunization settings: A C_A > L C_L. With exact equality, the first two derivatives vanish and the next nonzero higher derivative determines local behavior. These are conditions for the chosen valuation function near a particular rate, not a proof that future cash needs are funded under all yield paths.
If the insurer adjusts its assets dynamically as rates move, A(y) is the value of a different strategy from that of an unchanged asset portfolio. The distinction between a static portfolio repricing and a rule for rebalancing must remain explicit. A rebalancing rule involves market liquidity, trading costs, turnover, regulatory investment restrictions, and the possibility that liability estimates change simultaneously.
Asset and Liability Convexity — Quantitative Methods and Assumptions
Second derivatives and quadratic approximations
Taylor expansion supplies the standard relationship:
The insurer's surplus change is correspondingly
The two coefficients have different economic meanings. The first is the slope of surplus with respect to a specified rate; it governs the sign of a sufficiently small, nonzero shock if it is nonzero. The second measures curvature and becomes the leading term when first-order sensitivity is removed. Reporting only first-order duration thus omits the dominant term for a locally duration-neutral but convexity-mismatched insurer.
A second derivative is not itself a “guarantee.” It is a local approximation that can deteriorate rapidly for large shocks or sharply changing cash flows. The neglected third and higher derivatives matter as the shock increases. An insurer's investment decision may also create first-order or second-order exposures to other market factors. Those factors must be treated as separate coordinates, not silently absorbed into convexity.
Convexity measured on a common continuously compounded rate convention can be compared across fixed-cash-flow assets and liabilities. Convexity measured from one-sided shocks, mixed yield bases, or endogenous assumptions may be inappropriate for simple aggregation. A portfolio manager should reconcile the calculation method before treating a numerical surplus-convexity gap as hedgeable.
Conditional and option-adjusted liability sensitivities
Suppose the cash flow at time t is a differentiable function c_t(y) of the shocked rate. Then
has derivatives
The additional c_t' and c_t'' terms are direct evidence that liability duration and convexity can be functions of policyholder behavior and product design, not simply weighted averages of benefit dates.
For example, a surrender feature can bring a future cash flow forward when alternative market yields rise above a guaranteed crediting rate. The conditional amount and timing of the payment then depend on rates. A fixed-projection duration might report substantial long-term exposure, while an effective duration under economically responsive surrender assumptions might be shorter. At some shock sizes, the relationship can be nonlinear or discontinuous. The modeled rate–surrender relationship must be justified with experience data, policy terms, and limitations, not assumed from a generic statement that all policyholders always surrender when rates rise.
Embedded interest guarantees, annual ratchets, indexed annuity options, minimum-return guarantees, or management actions can produce liability convexity that varies with the starting rate and shock direction. Effective convexity can be reduced, zero, or negative after optionality is modeled. Some assets—callable bonds, mortgage-backed securities, prepayable loans—also exhibit endogenous cash flows. Combining them may produce offsetting risks in one scenario and amplifying risks in another. An insurer should calculate the entire balance sheet under common scenarios rather than infer full matching from labels such as “callable asset” or “surrenderable liability.”
Convexity asymmetry and hedging instruments
A long option-free fixed-rate bond generally benefits from positive price convexity: for equal magnitude parallel yield moves, the price gain after a fall in rates exceeds the price loss after a rise, measured around the same initial price. If the insurer's liabilities are more convex than its assets, the same positive bond convexity is insufficient to preserve surplus. The relevant comparison is A C_A versus L C_L, not whether asset convexity is positive in isolation.
Plain-vanilla interest-rate swaps can change the balance sheet's first-order term-structure exposure. Swaptions and other options can add nonlinear exposure. Their usefulness depends on market conventions, collateral posting, counterparty credit, exercise behavior, accounting classification, and the exact liability risk being hedged. A swap may lower a modeled dollar-duration gap while creating material margin calls during a market shock. An option hedge may provide convexity while becoming expensive or behaving differently from a policyholder guarantee during volatility changes.
The actuarial distinction between economic value sensitivity and the ability to fund contractual payments remains fundamental. A hedge with attractive model convexity can create an early cash requirement larger than immediately available cash. ALM therefore tracks rate-sensitivity reduction and liquidity-at-risk as separate, connected dimensions.
Local optimality and limits of immunization
The classical Redington interpretation is a local calculus result. Set the first derivative of surplus to zero and ensure a favorable second derivative near the initial yield. This gives local, not global, control. The first derivative can change after a yield movement even without trading, because duration evolves with the discount curve and with the passage of time. The second derivative can also change, particularly for callable securities and insurance options. Maintaining a hedge becomes a process rather than a one-time achievement.
The core historical Redington immunization reference should therefore be read alongside later recognition of stochastic rate dynamics. Boyle's term-structure immunization discussion identifies dependence on the rate model and parameters and acknowledges that actual insurance contingencies complicate a deterministic liability assumption. This perspective is compatible with deriving exact local derivatives and then testing their fragility under broader conditions.
For a liability with stochastic death or survival benefits, a “guaranteed” immunization of expected value is not equivalent to certainty that realized benefit payments will be funded. Expected present value, variance of benefit cash flows, and the probability of cash shortfall are distinct objects. A portfolio matched to expected liabilities can still be exposed to adverse longevity experience, correlated investment losses, and sudden claims.
Duration Gap Mathematics — Insurance Application and Comparative Evidence
Term-structure gradients and key-rate risk
A market yield curve has many degrees of freedom. A parallel shift is only one direction in this space. Let z_k be an identified set of zero-rate nodes. For each component value P, define the derivative with respect to a specified node shock. If only the cash flows discounted by node k are affected in a segmented curve experiment,
In a more realistic interpolated curve, each cash flow's discount factor has a sensitivity to several nodes. The general derivative includes those interpolation weights. It is necessary to specify the curve construction and interpolation method when publishing key-rate values.
For assets and liabilities, the first-order change in surplus under a vector of node shifts is
A zero sum of all node sensitivities can coexist with large positive and negative individual sensitivities that cancel only for a parallel movement. The second-order expression includes a Hessian matrix of cross-sensitivities. A parallel-shift convexity is the quadratic effect along a single direction; it does not characterize the whole Hessian.
An insurer holding a medium-maturity bullet bond against short- and long-dated benefit obligations can achieve the same scalar duration as the liabilities while retaining substantial exposure to a curve twist. The assets react to movements at the middle tenor and the liabilities react at the ends. If all rates shift equally, scalar duration supplies a good local description; if only the long end moves, that scalar comparison loses its relevance.
The EIOPA risk-free rate resources illustrate the practical importance of specifying curve construction and the shock used in option-adjusted duration reporting. The site provides shifted risk-free term structures for financial-stability reporting, including an option-adjusted duration use case. An insurer's internal key-rate grid may be more detailed than a supervisory reporting grid, but the latter reinforces that “duration” depends on the stated curve and measurement regime.
Currency, spread, inflation, and basis differentiation
Insurance groups and portfolios may receive asset coupons in one currency and pay liabilities in another. A measure derived from a domestic yield curve does not capture currency translation or basis between funding and liability curves. Cross-currency swaps can transform some exposures, but introduce counterparty, liquidity, and basis risk. Each currency should have its own sensitivity vector before exposures are aggregated into reporting currency.
A defaultable corporate bond changes value as a function of risk-free yields and credit spreads. A yield-to-maturity duration computed from a single quoted yield can mechanically mix the effects. For insurance ALM, useful reporting separates risk-free rate duration, spread duration, default and recovery exposures, and sometimes market liquidity premiums. A fall in an insurer's liability value because the prescribed discount curve changes does not offset a credit loss on its assets unless the liability valuation basis changes in the same way.
Inflation-linked liabilities create another discrepancy. Benefits indexed to price levels or wage indices are sensitive to real interest rates and expected inflation, possibly with caps, floors, and contractual lags. Nominal fixed-income duration alone provides little assurance that inflation-driven nominal benefit increases are matched. Sensitivities to nominal rates, real rates, inflation breakevens, and actual indexation rules should be stated separately.
A further distinction is the reinvestment dimension. Fixed cash flows that coincide in present value and first-order duration are not necessarily timed to produce liquidity at the moment benefits are due. The insurer may have to reinvest interim coupons or sell assets before maturity. Local mark-to-market immunization, cash-flow replication, and settlement liquidity are different financial objectives.
Insurance balance-sheet regimes
International standards do not impose one worldwide numerical duration target. The IAIS framework on insurer ALM recognizes scenario-based management of asset-liability interactions. It supplies a global supervisory vocabulary but does not turn a particular national reserve basis or risk-free curve into a universal standard.
In the European Union, the Solvency II insurance framework addresses risk management, prudent investment, capital, and technical provisions. Article 132's prudent-person approach concerns the security, quality, liquidity, and profitability of the portfolio as a whole. An internal duration match can support prudential risk management without proving that the entire investment portfolio satisfies every obligation. Solvency II valuation and capital calculations have their own assumptions, discount rates, and stress design; a purely accounting or market-value duration calculation may differ.
The Solvency II review directive was adopted as Directive (EU) 2025/2, with amendments scheduled to apply from January 30, 2027, subject to the directive's implementation mechanism. As of this monograph's November 14, 2026 cutoff, the 2027 application date is future. The review should be documented as a prospective change when evaluating the durability of an insurer's models, not silently incorporated into the rules then in effect.
In the United Kingdom, the Prudential Regulation Authority's matching-adjustment supervisory statement, SS7/18, records the version published October 23, 2025 and effective October 27, 2025, alongside a July 29, 2026 version scheduled for December 31, 2026. As of the research cutoff, the latter is a published future version. Matching-adjustment rules concern eligible assets and liabilities, portfolio features, cash-flow matching, and regulatory discounting; ordinary modified-duration matching does not itself confer eligibility for the adjustment. This is a separate legal and valuation question.
For U.S. actuarial services, the ASOP No. 22 asset-adequacy standard concerns statements of actuarial opinion based on specified asset adequacy analysis of life, annuity, or health reserves and other liabilities and is effective from June 1, 2022. This professional-standard framework relies on legally specified reserve and actuarial practice requirements. A company may have satisfactory economic duration matching but fail a prescribed reserve adequacy or interim cash-flow test. Conversely, a statutory reserve result is not automatically a complete market-value rate-risk hedge.
Financial reporting introduces another basis. The IFRS 17 insurance accounting standard combines current measurement of future insurance cash flows with insurance service and finance components, and has been effective since 2023. The IFRS 9 financial-instruments standard classifies and measures assets according to business model and cash-flow characteristics. An economically matched asset-liability pair may still show accounting volatility because the two sides have different recognition, measurement, or presentation requirements. Reporting classifications should be reconciled but should not redefine the economic risk measure.
What international supervisory evidence actually shows
The 2024 EIOPA insurance stress test findings assessed 48 undertakings in 20 European jurisdictions, representing approximately 75% of EU market assets under the defined test scope. The adverse scenario combined higher market risk premia and rates with other economic shocks, examined capital and liquidity, and permitted different assumptions about management actions. EIOPA reported an initial aggregate solvency ratio of 221.8% and a drop in excess assets and eligible own funds in the stressed scenario. These are modeled outcomes for a particular selected sample and scenario, not a controlled experiment identifying duration mismatch as a single causal factor.
The October 2026 insurance risk dashboard uses Q2 2026 quarterly company data and 2025 year-end annual data. It reports that the duration mismatch measure was unchanged at the end of 2025, while market and credit risk indicators depend on other observation dates and methods. The dashboard's note states that the duration mismatch indicator uses modified duration of assets and liabilities. A period-to-period “unchanged” aggregate metric should not be interpreted as proof of sufficient matching at any individual insurer or under any curve movement.
These two sources are complementary. The stress test provides evidence of joint vulnerability to market changes and simulated management actions; the dashboard gives ongoing aggregate monitoring under prescribed definitions. Neither establishes a universal premium for positive convexity, a universal optimum duration gap, or the ability to fund all policyholder cash needs. Their scientifically valid use is to test whether the insurer's internal analytical conclusions are consistent with broader prudential concerns and to identify additional variables requiring scenario analysis.
Rate Shock Approximations — Stress Testing, Model Limits, and Verification
Shock design, exact repricing, and approximation error
The usual quadratic price approximation is informative only if the rate change is defined. A “100-basis-point shock” could mean adding 100 basis points to every zero rate, raising a single yield-to-maturity, imposing a regulator's stressed curve, or shifting selected key rates. The same words can therefore produce different changes in present value and different measured convexities. A reproducible test records which curve nodes are shocked, how intermediate tenors are interpolated, whether assets and liabilities use identical movements, and what happens to projected cash flows.
For fixed cash flows under a common scalar shift, exact repricing is simple and should serve as the accuracy benchmark. At a shift h, compare exact with . Record absolute approximation error in currency and relative error as a percentage of initial value. Test both +h and −h because third-order effects cause asymmetry. Vary h to ensure that the estimated derivatives are stable around the valuation point rather than artifacts of a shock too large for local analysis or too small for reliable numerical precision.
Where discontinuities occur—for example, an option exercise boundary or management action that changes sharply—the derivative may be ill-defined or materially unstable. Scenario repricing remains meaningful even when a local Taylor expansion is inappropriate. Large financial market shocks require full revaluation under a range of scenarios, including basis, spread, and volatility changes. The second-order approximation is a diagnostic and communication tool, not a replacement for the insurer's risk engine.
An actuarial stress test should also examine the evolution of the position after the shock. A favorable current surplus movement may coexist with lower future reinvestment returns, altered policyholder behavior, expensive hedging, or adverse future cash-flow coverage. Time zero economic value, interim liquidity, statutory own funds, and future earnings should be reported in separate lines.
Stochastic and behavioral dependence
In a stochastic model, interest rates evolve along paths. Discount factors can correlate with mortality-sensitive behavior, lapses, inflation, asset credit returns, or the insurer's investment decisions. A scenario-specific effective duration may be calculated as the price sensitivity of a model-consistent value to a curve perturbation, but that estimate is conditional on the model, calibrated parameters, and method of perturbation. Differentiating the modeled expectation may require addressing pathwise payoff discontinuities or numerical Monte Carlo noise.
The 1978 Boyle term-structure paper already links immunization to assumptions about rate dynamics. Modern insurance application deepens that issue: the insurer does not hold a purely deterministic bond-liability pair when the obligations may be contingent and policyholders can respond to economic conditions.
Risk-factor selection deserves explicit attention. An insurer can have low sensitivity to one curve, high sensitivity to another, and economically important exposure to their spread. A stochastic model should therefore identify whether its factors reproduce historical and forward-looking curve twists, negative or near-zero yields where relevant, yield volatility, credit spread widening, policyholder options, and regime changes. Calibration uncertainty is part of the interpretation of duration and convexity because finite-difference Greeks are derivatives of a calibrated model, not measurements independent of the model.
The more ambitious the behavioral model, the more essential independent checking becomes. Historical surrender observations may come from a rate regime unlike the one being stressed. Extrapolating lapse elasticity far outside the observed data range can distort the sign and magnitude of effective convexity. Governance should record alternative behavioral assumptions and explain which conclusions depend on them.
Liquidity, capital, accounting, and governance interfaces
An asset-liability portfolio can have a stable mark-to-market surplus yet need cash during a shock. Derivative margin, forced asset sales, surrender outflows, and delayed reinsurance recoveries have specific dates. A duration gap measured as a derivative of present value does not encode those dates. Liquidity ladders, collateral forecasts, short-horizon cash-flow stress, and contingency funding analysis are necessary complements.
The IAIS asset-liability risk-management guidance links economic values with the ability to meet cash flows and with the insurer's wider enterprise risk framework. The U.S. cash-flow analysis requirements in ASOP No. 7 similarly emphasize interim values and documented scenarios. Both perspectives support treating a duration gap as one specified exposure within a governance system, rather than a final assessment of insurer solvency.
Solvency and accounting consequences can diverge from economic value. A change in yield curves may alter the prudential value of liabilities, market value of assets, eligible own funds, and required capital by different mechanisms. IFRS 17 and IFRS 9 can recognize or present related movements differently. The insurer should distinguish (a) economic surplus sensitivity under a consistent valuation basis; (b) regulatory capital sensitivity under applicable rules; (c) accounting earnings or other comprehensive income; and (d) actual settlement liquidity. These outputs need reconciliation, not forced equality.
U.K. matching-adjustment eligibility, European Solvency II risk-based valuation, and U.S. actuarial cash-flow testing each impose jurisdiction-specific constraints. Applying a locally accurate duration formula does not establish that an asset is admissible for a designated backing portfolio, that a discount adjustment is permitted, or that the firm meets reserve opinion requirements. Date-controlled compliance interpretation is an independent part of responsible ALM work.
Validation and interpretation of evidence
Model validation should include analytic unit tests for simple positive cash-flow portfolios, comparison of finite differences with closed-form derivatives, bump-size convergence, verification of reprojected optional cash flows, reproducibility of interpolation, and reconciliation to changes reported by the economic valuation model. A risk-control group should scrutinize how a liability value was projected, which management actions were included, and whether the same cash flow was counted on both sides of a transaction.
The auditor's record should explain each material derivative's curve, base valuation, value denominator, sign, units, scenario horizon, cash-flow policy, and frequency of refresh. Dollar measures should be expressed in reporting currency with conversion rules. For a material change to the asset mix or liability assumptions, recalculating only the asset duration while keeping an old liability duration is not adequate; both need compatible valuation dates and current cash-flow projections.
Empirical analysis must respect the observation level. EIOPA's 2024 stress-test population is a supervised sample, and the 2026 risk dashboard indicator is an aggregate reporting measure. The numerical examples below are purpose-built to test the mathematical relationships; they are not evidence about the distribution of insurer duration gaps. Keeping those categories separate prevents a plausible analytical statement from becoming a false empirical generalization.
Worked ALM Calculations — Integrated Scientific Interpretation and Implications
Equal duration with unequal convexity
Construct a liability with two guaranteed payments. Its present values at an initial continuously compounded flat rate of 4% are 50 at year 2 and 50 at year 10, in arbitrary monetary units. The actual promised amounts are therefore 50e^{0.08} = 54.1643533837 at year 2 and 50e^{0.40} = 74.5912348821 at year 10. Total liability present value is L = 100. By the continuous-discount identities, liability duration is (2×50 + 10×50)/100 = 6 years, and liability convexity is (2²×50 + 10²×50)/100 = 52 years².
Now construct an asset consisting of a single year-6 zero-coupon payment with present value 100. Its nominal payment is 100e^{0.24}; its value at the initial rate is A = 100, duration D_A = 6 years, and convexity C_A = 36 years². The asset and liability values and durations match exactly, so S(0)=0 and S'(0)=0. However, A C_A − L C_L = 3,600 − 5,200 = −1,600 monetary-units-years². The local quadratic change in surplus is −800(Δy)².
Reprice exactly under parallel shifts Δy:
| Parallel shock | Asset value | Liability value | Exact surplus | Quadratic surplus estimate |
|---|---|---|---|---|
| −200 basis points | 112.749685 | 113.110677 | −0.360991 | −0.320000 |
| −100 basis points | 106.183655 | 106.268613 | −0.084958 | −0.080000 |
| +100 basis points | 94.176453 | 94.251805 | −0.075351 | −0.080000 |
| +200 basis points | 88.692044 | 88.976010 | −0.283966 | −0.320000 |
The exact values follow from A(Δy)=100e^{−6Δy} and L(Δy)=50e^{−2Δy}+50e^{−10Δy}. Both rising and falling rates reduce surplus locally because the liability cash flows are more temporally dispersed than the asset payment. The magnitudes differ for equal positive and negative shifts because of higher-order terms. This calculation verifies the central proposition: equality of present value and duration can leave a quantitatively important second-order surplus exposure.
If the asset instead replicates the same present-value cash flows at years 2 and 10, then A(Δy)=L(Δy) for every parallel shift under the assumed fixed cash flows and common discounting. More strongly, identical dated cash flows provide a perfect deterministic replication under any discount curve assigned consistently to those dates. The replication conclusion depends on the contractual ability to deliver those asset payments and on the omission of default, expenses, and contingency uncertainty. It illustrates the difference between moment matching and actual cash-flow matching.
Dollar-duration mismatch when assets exceed liabilities
Consider a hypothetical insurer with A = 110 and L = 100, measured in millions, under a single common parallel yield shock. Let asset duration be D_A = 5, liability duration D_L = 8, asset convexity C_A = 32, and liability convexity C_L = 75 under the same convention. Initial economic surplus is 10 million. Dollar duration of assets is 550 million-years, and dollar duration of liabilities is 800 million-years, giving S'(0)=250 million-years per unit positive rate change. The insurer benefits from a sufficiently small rise in rates and loses from a sufficiently small fall.
Dollar convexity is 3,520 for assets and 7,500 for liabilities, in million-years². Thus S''(0)=−3,980. At a +100-basis-point shift, the first-order approximation predicts +2.5 million, and the second-order adjustment is −0.199 million, giving ΔS ≈ +2.301 million. At −100 basis points, the analogous approximation is −2.5 −0.199 = −2.699 million. These figures are Taylor approximations from the stated sensitivity statistics; they are not exact repricings of a specified portfolio, and they do not imply a unique set of actual cash flows.
To remove the first derivative while retaining A = 110 and L = 100, an asset duration of D_A = 800/110 ≈ 7.272727 years would be required, assuming liability duration stays eight and the same valuation basis applies. This is neither equal to the liability duration nor a sufficient condition for adequate capital or liquidity. Convexity, yield-curve shape, and contingent cash-flow risks remain.
A different use of the duration-gap idea divides the derivative by an asset or liability base to form a normalized gap. Such ratios can be helpful for comparisons, but the analyst must declare the denominator. The currency derivative S'(0) is the unambiguous first-order quantity for the specified surplus function.
Nonparallel curve movements
Return to the first example. A year-6 zero-coupon asset is set against payments at years 2 and 10. Assume a deliberately simplified three-node curve in which the year-2, year-6, and year-10 zero rates can be shocked independently, and each dated payment is affected only by its own node. The asset sensitivities to the three nodes are (0, −600, 0) monetary-units-years, and the liability sensitivities are (−100, 0, −500). Therefore the surplus sensitivity vector is (+100, −600, +500). The sum is zero, matching the parallel-duration result.
Under a +100-basis-point shock to the ten-year node alone, the asset value remains 100 and the year-10 liability component becomes 50e^{−0.10}=45.241871. Liability total is then 95.241871 and surplus rises to +4.758129. Under a +100-basis-point shock to the six-year node alone, the asset value becomes 100e^{−0.06}=94.176453, while liability value remains 100; surplus falls to −5.823547. Under a +100-basis-point shock to the two-year node alone, the year-2 liability component becomes 50e^{−0.02}=49.009934, generating surplus +0.990066. These exact results follow directly from the deliberately segmented node assumptions.
The simplified curve is a pedagogical device, not a claim that a market curve may be moved without interpolation interactions at other maturities. With a bootstrapped and interpolated curve, repricing all cash flows under a specified node perturbation gives the appropriate key-rate derivatives. The key point persists: a zero parallel-duration gap means the components of a sensitivity vector sum to zero, and cannot ensure that the vector itself is zero.
A mixed curve twist can therefore cause a loss much larger than the convexity effect of a small parallel shift. A company with long annuities, callable assets, and short-dated collateral demands must test all three together, rather than report one aggregate duration statistic as if it measured them.
Behavioral liability scenario and hedge governance
Consider an insurer offering a guaranteed crediting rate and a surrender option. At the valuation date, the baseline projection may put most benefits far into the future. If market rates rise sufficiently, some policyholders may move to alternative investments, bringing payments forward. A deterministic liability duration calibrated to the original timing overstates the persistence of those payments under a behavioral shock. Conversely, a fall in rates can make guarantees more valuable and alter lapse incentives, increasing liability value more than a conventional duration-only approximation would predict.
A model-based hedge requires both the first derivative of discount factors and the derivative of cash-flow behavior. The earlier expression for L'(y) separates the two. If the insurer hedges only the discounted baseline projection, the change in cash-flow amounts and timing remains unhedged. If it hedges the effective projection under a fitted surrender function, the result depends on the validity of that fitted function and can change across rate regimes.
A defensible analysis therefore reports a baseline deterministic duration, a behavioral effective duration, and a range of alternative effective durations under credible lapse assumptions. It also reports policy cash payments under upward and downward scenarios, asset sales needed to fund them, and collateral requirements from hedging instruments. The comparison is a stress design, not a precise prediction of human behavior.
An insurer holding long credit assets can have apparently favorable rate sensitivity while experiencing spread losses and illiquidity during a market selloff. If the insurer relies on derivatives to offset asset duration, its counterparty collateral agreement may require cash payment before any long-term asset-liability offset is economically realized. Economic surplus hedging and liquidity management must be tested jointly in scenarios but reported as separately identifiable outcomes.
Reproducibility notes and integrated conclusions
The first case assumes fixed positive cash flows, continuously compounded 4% annual initial yield, common parallel yield shocks, no credit or option risk, and exact exponential discounting. All displayed exact balances are rounded to six decimals after computation. Initial promised cash flows can be recovered by multiplying each stated present-value component by the relevant 4% discount reversal. The quadratic estimate uses D_A = D_L = 6 and C_A − C_L = −16 with A = L = 100. The rate-shock response can be independently reproduced in a spreadsheet, a simple calculator, or any programming language with the exponential function.
The second case consists of internally consistent point-in-time values and rate derivatives selected for illustrative calculation; because no specific underlying cash-flow portfolio is given, exact repricing is intentionally not claimed. The third case explicitly chooses independent segmented curve nodes rather than a continuous market interpolation. These scope distinctions prevent numerical precision from suggesting unsupported realism.
The integrated results support five analytical conclusions. First, matching ordinary duration ratios is not the same as matching the first derivative of economic surplus when values differ. Second, matching dollar duration removes only the local first-order term for a specified rate shock. Third, a negative dollar-convexity gap can generate losses for shocks in either direction even under exact first-order matching. Fourth, nonparallel curve movements can dominate parallel-shift convexity effects. Fifth, contingent insurance payment behavior, legal eligibility, and actual settlement liquidity introduce important dimensions outside the direct scope of fixed-cash-flow Greeks.
These conclusions follow from established mathematics and verified institutional constraints. They constitute an original integrated exposition, not a claim of a newly invented actuarial theory.
Conclusion
Duration and convexity are derivatives of an explicitly defined value function. Their application to insurance extends beyond bond mathematics because assets and liabilities can differ in amount, timing, market factors, cash-flow conditionality, regulatory valuation, and immediate funding needs. The correct first-order rate exposure of economic surplus is the difference between the dollar-duration exposures of assets and liabilities. The correct second-order exposure under the same shock is the difference between their dollar-convexity exposures. Both depend on using a consistent valuation basis and shock definition.
The monograph's fully specified example establishes that assets and liabilities can have equal value and identical duration while liability convexity exceeds asset convexity. Local surplus then falls for a positive or negative rate shift. Exact repricing confirms the effect and reveals higher-order asymmetry. A second example shows why an insurer with a positive initial surplus may require an asset duration different from liability duration to remove its first-order exposure. The third example establishes that a zero parallel-duration gap can conceal substantial key-rate mismatches.
The scientific interpretation is consequently conditional and operational. Analysts must state the liability cash-flow basis, curve variables, rate convention, options and behavioral assumptions, shock magnitude, and revaluation process. They must examine nonparallel and non-rate exposures and connect present-value sensitivities to capital, accounting, liquidity, and legal admissibility without conflating those measures. Primary actuarial and insurance-supervisory standards reinforce this integrated work, while contemporary European monitoring provides time- and sample-bound context.
The resulting research position is a disciplined use of established ALM mathematics: compute consistent surplus derivatives, verify approximations against full repricing, test the shocks and behaviors excluded by the local measure, and interpret every reported duration or convexity figure within its institutional definition. The article contributes comparative synthesis, transparent derivations, and independently recalculated illustrations. No new formal theory, doctrine, or model is asserted, and no external peer review or public publication is claimed.
References
Actuarial Standards Board. (2025). Life or Health Cash Flow Analysis (Actuarial Standard of Practice No. 7; revised December 2025, effective June 1, 2026). American Academy of Actuaries. https://www.actuarialstandardsboard.org/asops/life-or-health-cash-flow-analysis/
Actuarial Standards Board. (2021). Statements of Actuarial Opinion Based on Asset Adequacy Analysis for Life Insurance, Annuity, or Health Insurance Reserves and Other Liabilities (Actuarial Standard of Practice No. 22; effective June 1, 2022). American Academy of Actuaries. https://www.actuarialstandardsboard.org/asops/asop-no-22-statements-of-actuarial-opinion-based-on-asset-adequacy-analysis-for-life-insurance-annuity-or-health-insurance-reserves-and-other-liabilities/
Boyle, P. P. (1978). Immunization under stochastic models of the term structure. Journal of the Institute of Actuaries, 105(2), 177–187. Cambridge University Press journal record. https://www.cambridge.org/core/journals/journal-of-the-institute-of-actuaries/article/abs/immunization-under-stochastic-models-of-the-term-structure/6267820175C6E7634E4F71F202E1639B
European Insurance and Occupational Pensions Authority. (2024). Insurance Stress Test 2024: Scope, methodological approach, results, and recommendations [Official study webpage, December 17, 2024 results]. https://www.eiopa.europa.eu/insurance-stress-test-2024_en
European Insurance and Occupational Pensions Authority. (2026). Insurance Risk Dashboard: October 2026 (EIOPA-BoS-26-495; quarterly data through Q2 2026 and annual data through 2025 year-end). https://www.eiopa.europa.eu/assets/insurance-risk-dashboard/EIOPA-BoS-26-495%202026%20October%20Insurance%20Risk%20Dashboard.html
European Insurance and Occupational Pensions Authority. (2026). Risk-Free Interest Rate Term Structures [Monthly technical information and shifted term structures for financial-stability reporting]. https://www.eiopa.europa.eu/tools-and-data/risk-free-interest-rate-term-structures_en
European Parliament and Council. (2009). Directive 2009/138/EC on the taking-up and pursuit of the business of Insurance and Reinsurance (Solvency II) [Consolidated 2024 version used for source-specific Article 132 interpretation]. Official Journal of the European Union. https://eur-lex.europa.eu/legal-content/EN/TXT/?uri=CELEX%3A02009L0138-20240109
European Parliament and Council. (2025). Directive (EU) 2025/2 of November 27, 2024 amending Directive 2009/138/EC and related directives (Official Journal, January 8, 2025; specified application from January 30, 2027). https://eur-lex.europa.eu/legal-content/en/ALL/?uri=CELEX%3A32025L0002
Fisher, L., & Weil, R. L. (1971). Coping with the risk of interest-rate fluctuations: Returns to bondholders from naive and optimal strategies. The Journal of Business, 44(4), 408–431. https://ideas.repec.org/a/ucp/jnlbus/v44y1971i4p408-31.html
IFRS Foundation. (2014, as amended). IFRS 9 Financial Instruments [Issued-standard official description; effective 2018]. https://www.ifrs.org/issued-standards/list-of-standards/ifrs-9-financial-instruments/
IFRS Foundation. (2017, as amended). IFRS 17 Insurance Contracts [Issued-standard official description; effective 2023]. https://www.ifrs.org/issued-standards/list-of-standards/ifrs-17-insurance-contracts/
International Association of Insurance Supervisors. (2024, maintained online). Insurance Core Principles and ComFrame, including ICP 16.5: Asset-Liability Management Policy. https://www.iais.org/icp-online-tool/
Macaulay, F. R. (1938). Some Theoretical Problems Suggested by the Movements of Interest Rates, Bond Yields and Stock Prices in the United States since 1856. National Bureau of Economic Research. https://www.nber.org/books-and-chapters/some-theoretical-problems-suggested-movements-interest-rates-bond-yields-and-stock-prices-united
Prudential Regulation Authority, Bank of England. (2018, subsequent versions). SS7/18: Solvency II: Matching Adjustment [Version published October 23, 2025 and effective October 27, 2025; future version published July 29, 2026 and effective December 31, 2026]. https://www.bankofengland.co.uk/prudential-regulation/publication/2018/solvency-2-matching-adjustment-ss
Redington, F. M. (1952). Review of the principles of life-office valuations. Journal of the Institute of Actuaries, 78(3), 286–340. Cambridge University Press journal record. https://www.cambridge.org/core/journals/journal-of-the-institute-of-actuaries/article/abs/review-of-the-principles-of-lifeoffice-valuations/42DE39108ADCAD8D476B8E633F8E606F