IFRS 17 changed the accounting treatment of insurance contracts more than any change in a generation. It did not change what a reserve is. A reserve remains an estimate of the present value of future cash flows from past events, plus an explicit allowance for uncertainty. What IFRS 17 did do is make that decomposition visible — separately disclosing the estimate of future cash flows, the risk adjustment for non-financial risk, the contractual service margin and, for the direct-participation approach, the financial-services margin.
This article treats reserving through the lens of the IFRS 17 building blocks. It assumes familiarity with the standard's general model. For a related treatment of catastrophe severity distributions see Climate and Catastrophe Risk; for the capital view see Solvency II, ORSA and Economic Capital.
The building blocks
Under the general measurement model, the fulfilment cash flows have three components:
- Estimates of future cash flows — probability-weighted, current, unbiased.
- Discounting — reflecting the time value of money and financial risks.
- Risk adjustment for non-financial risk — the compensation the entity requires for bearing uncertainty about non-financial cash flows.
The contractual service margin (CSM) then captures the unearned profit, recognised in P&L as services are provided. Each building block is a modelling exercise in its own right, with its own uncertainty.
Triangles: the empirical spine
Reserving for non-life business — and for the incurred-but-not-reported (IBNR) portion of life reinsurance — starts from claims triangles. Origin period (accident year, underwriting year or reporting year) forms the rows; development period forms the columns.
The chain ladder method projects incremental or cumulative claims to ultimate using age-to-age development factors:
f_j = Σ_i C_{i,j+1} / Σ_i C_{i,j}
Multiplying successive factors gives the tail from the latest observed cumulative to ultimate. Chain ladder is the industry's workhorse because it is data-driven, transparent and easy to validate.
Bornhuetter–Ferguson (BF) blends chain ladder with an a priori ultimate — typically an expected loss ratio applied to earned premium — using the chain-ladder-implied percentage developed:
Ultimate_BF = Reported + (Expected Ultimate × (1 − % developed))
BF is more stable than chain ladder when the origin period has limited development, and is standard for immature years and lines with slow tails.
Cape Cod and Benktander are refinements that estimate the a priori loss ratio from the data itself, reducing dependence on external premium expectations.
Beyond the point estimate: the three uncertainties
IFRS 17 does not prescribe a method for the risk adjustment, but it does require disclosure of the confidence level to which the risk adjustment corresponds. That places the uncertainty question at the centre of the reserving process. Three uncertainties should be distinguished.
Process uncertainty
Process uncertainty is the irreducible randomness of the underlying claims process — the fact that, even if the model and its parameters were known perfectly, the realised claims would still vary. For a compound-Poisson-lognormal aggregate, process variance is analytically computable; for empirical triangle-based projections it can be estimated by bootstrapping the residuals of the chain-ladder model (Mack, 1993; England & Verrall, 2002).
Parameter uncertainty
Parameter uncertainty is the estimation error in the development factors (or the a priori loss ratios, or the tail parameters). It shrinks as data accumulates but never disappears. Bootstrap resampling of the chain-ladder residuals, or Bayesian posterior distributions on the parameters, yields quantified parameter uncertainty.
Model uncertainty
Model uncertainty is the risk that the family of models is wrong. A pure chain ladder and a paid-plus-incurred hybrid can produce ultimates that differ by 5–15% on the same triangle; that gap is not process or parameter uncertainty — it is model risk. Reporting a single method as if the choice itself were free of uncertainty is the most common weakness in reserving practice.
Risk adjustment: methods and disclosure
Common risk-adjustment methods:
- Confidence-level method: the risk adjustment equals the difference between a chosen quantile (say the 75th percentile) of the ultimate loss distribution and the mean. The disclosure requirement fits this method directly.
- Cost-of-capital method: consistent with Solvency II's risk margin. The risk adjustment is the present value of the cost of holding required capital over the runoff of the portfolio.
- Conditional tail expectation (CTE) method: expected loss above a chosen quantile; a coherent risk measure and easier to explain than VaR.
Whichever method is chosen, the confidence-level disclosure requires mapping the chosen method to an equivalent quantile. That mapping is itself a modelling exercise and should be documented.
Discounting
IFRS 17 discount rates reflect the time value of money and the financial characteristics of the cash flows — currency, liquidity — but not risks that are already captured elsewhere. The bottom-up approach starts with a risk-free curve and adds an illiquidity premium. The top-down approach starts with reference-portfolio yields and deducts credit and market risks not shared with the cash flows.
The choice between top-down and bottom-up has material P&L implications, particularly for long-duration life business. Sensitivity of the fulfilment cash flows to the discount curve — and reconciliation to the Solvency II curve — should be part of standard reporting.
Numerical intuition
Consider a general-liability portfolio with 10 accident years of history and 15 years of expected development. A well-behaved chain ladder produces an ultimate estimate; bootstrap resampling produces a distribution around it. For a book of this shape:
- Process uncertainty (given true parameters) is typically 5–15% coefficient of variation on the total reserve.
- Parameter uncertainty adds another 5–10% CV, more for immature years.
- Model uncertainty — comparing chain ladder against BF, Cape Cod, incurred-only versus paid-plus-incurred — commonly widens the credible interval by another 5–10%.
The combined CV drives the risk-adjustment magnitude and the disclosure. Institutions that present only the point estimate are hiding the number that matters most for solvency and pricing.
Implementation pitfalls
- Distorted triangles from large losses. Chain ladder is sensitive to a small number of large claims. Standard practice is to cap and separate large claims, projecting them under their own severity model (see Frequency–Severity Modelling).
- Trend and calendar-year effects. Chain ladder assumes no calendar-year effect (inflation, legal environment changes). When present, they need explicit modelling — often a stochastic-inflation overlay.
- Reinsurance treatment. Gross versus net triangles produce different development patterns because reinsurance recoveries follow their own timing. Best practice is to project gross triangles and apply reinsurance programme mechanics.
- Data granularity. Aggregating heterogeneous business into a single triangle destroys the segmentation that reserving methods rely on. Splitting is better than blending when data supports it.
Validation and governance
Reserving is subject to independent actuarial review under most local regulatory frameworks — the Solvency II actuarial function, the appointed-actuary regime, or their equivalents. IFRS 17 adds an audit-firm review of the accounting numbers. The validation checks:
- Method appropriateness by line and maturity.
- Data reconciliation to the source system and to the exposure base.
- Sensitivity of ultimate estimates to method, tail assumptions and discount rates.
- Diagnostics — residual plots, backtests of prior estimates against realised runoff.
- Uncertainty disclosure consistent with the risk-adjustment method chosen.
Governance chains should ensure that reserve-committee minutes record the range of methods considered and the reasons for the selected point estimate.
Limitations
Reserving is unavoidably backward-looking; the future can differ from the past in ways triangles cannot capture. Emerging risks — social inflation, secondary-peril climate loss growth, cyber accumulation — sit outside the historical development factors until they show up as unfavourable actuals. Explicit overlays, informed by exposure analysis and scenario testing, close part of the gap; humility about the remainder is the honest disclosure.
Conclusion
IFRS 17 rewards reserving processes that make uncertainty visible. A single point estimate with a risk adjustment expressed to a stated confidence level is a specific disclosure — and one that puts the reserving actuary's method choice, uncertainty decomposition and governance under scrutiny. Done rigorously, the discipline improves not just accounting but pricing, capital and reinsurance strategy.
Related reading
References and further reading
- IFRS Foundation, IFRS 17 Insurance Contracts.
- Mack, T., Distribution-free calculation of the standard error of chain ladder reserve estimates, ASTIN Bulletin, 1993.
- England, P., and Verrall, R., Stochastic Claims Reserving in General Insurance, Institute of Actuaries, 2002.
- International Actuarial Association, Risk Adjustments for Insurance Contracts under IFRS 17.
About the author
Part of an ongoing series on actuarial and financial reporting — more about the author.