P&C IFRS 17 benchmarks across 60+ Asian insurers

Three years into IFRS 17, a set of actual choices now exists to look at. In 2025 the Casualty Actuarial Society published a research paper — "Actuarial Considerations Associated with IFRS 17 Implementation for General Insurers in Asia — Part 1: Premium Allocation Approach," by Delvin Cai, Leo Lee and Stephen Dong of PwC — that surveys more than 60 general insurers across Hong Kong, Singapore and Malaysia, all of which adopted IFRS 17 on 1 January 2023. The full paper is available on the CAS website, with a companion article in Actuarial Review.

One honest framing note up front: this is a practitioner survey of emerging market practice, compiled by PwC teams working in those markets. It is not an independent regulatory study or an audit of financial statements. It tells you what firms are doing, not what regulators have approved, and the sample is three markets in one region. With that understood, it is one of the most detailed windows into actual PAA implementation choices that currently exists.

This post works through the findings section by section — measurement model, eligibility, discounting, risk adjustment, tooling — and closes with the pattern they collectively suggest.

1. Measurement model: the PAA is the market default

The first question IFRS 17 forces is which measurement model applies. For general insurance, the choice is almost always between the Premium Allocation Approach and the General Measurement Model (GMM).

The survey answer is clear: 86% of respondents will use only the PAA; just 5% only the GMM. The remaining roughly 9% use both, presumably where a small portion of the book — longer-duration liability or speciality products — does not qualify for the PAA.

The PAA is IFRS 17's simplified model. Under para 53, it is available where, at group inception, either the coverage period of each contract in the group is one year or less (para 53(b)), or the entity reasonably expects the PAA would produce a liability for remaining coverage (LRC) not materially different from the GMM (para 53(a)). For the short-duration P&C business that dominates these three markets — motor, property, marine — the coverage-period limb makes most of the book automatically eligible. The materiality limb under para 53(a) matters for annual-plus products such as engineering contracts or long-tail liability, where a full GMM-vs-PAA comparison is the alternative.

The dominance of the PAA is not a surprise, but the near-absence of the GMM is worth noting. In life markets, the GMM is the default; here it is the exception.

2. Eligibility: cleared by judgement

Where contracts exceed one year, the eligibility question under para 53(a) can be settled in two ways: run a full GMM and compare, or conclude by judgement that the difference is not material. The survey shows the market has taken the judgement route almost universally.

Three findings together tell the story:

• 65% pass PAA eligibility for groups with contracts over one year on materiality grounds, without running a quantitative GMM comparison.
• 80% use some form of qualitative metric to support the argument that the PAA would not be materially different from the GMM.
• 83% apply both a percentage threshold and an absolute-amount threshold when defining "not materially different."

Para 53(a) sets the test by reference to the LRC specifically: the PAA must not produce a liability for remaining coverage that is materially different from the GMM liability. Para 54 adds that the materiality limb is not met where significant variability in fulfilment cash flows is expected to affect the LRC during the coverage period. Most qualitative arguments in this market rest on the absence of that variability — stable short-tail products where discount effects and risk-adjustment differences between the two models are demonstrably small.

The dual-threshold approach (percentage plus absolute amount) is a robust way to bound materiality, preventing individually small but large-in-aggregate books from qualifying without proper analysis.

3. Discounting: where 86% draw the line

Under the PAA the liability divides into two: the LRC for unexpired risk, and the LIC for incurred claims. IFRS 17 treats their discounting differently, and the survey shows the market has largely followed the standard's implied hierarchy.

86% discount the LIC only; 10% discount both LRC and LIC.

The split is coherent with the standard's mechanics. Para 56 provides that an entity need not discount the LRC if, at initial recognition, it expects the time between providing each part of the services and the related premium due date to be no more than one year. The test is about the interval from service delivery to premium receipt — not about whether the overall contract period is under one year. For most annual P&C contracts that condition is met comfortably, so the LRC exemption applies and discounting the LRC adds no information at material cost.

The LIC is different. Under para 59(b), the LIC must be measured at fulfilment cash flows — which include discounting and a risk adjustment for non-financial risk — unless the claims are expected to be paid or received within one year of the claim date. Short-tail classes often meet that one-year settlement test, so the exemption still applies. Long-tail classes — liability, workers compensation, marine — do not, and the LIC there requires both discounting and a risk adjustment.

On the discount rate itself, 98% use the bottom-up approach. Para B80 describes it: start from the liquid risk-free yield curve and add a liquidity adjustment to reflect the fact that insurance liabilities are less liquid than the reference instruments. The alternative — top-down (para B81), stripping non-insurance risks from asset yields — is almost unused.

Within the bottom-up approach: 71% reference government bond rates as the risk-free base; 27% use swap rates. Government bonds are the simpler, more data-rich reference in these markets. Swap rates offer a more complete term structure in jurisdictions where the government curve is thin at longer maturities, but in HK, Singapore and Malaysia the government curve is liquid enough for most P&C purposes.

On the illiquidity premium: roughly seven in ten apply no explicit illiquidity premium, most citing materiality. This is defensible for short-tail business where the LIC discount period is brief, but it is a judgement that auditors can and will test as books grow and longer-tail classes become more material.

4. Risk adjustment: the 75th percentile converges

The risk adjustment (RA) for non-financial risk is one of the most principle-based elements of IFRS 17. The standard does not prescribe a method or a confidence level (para B91); it requires only that every entity disclose the confidence-level equivalent of its chosen technique (para 119). VaR, Tail VaR, and Cost of Capital are all used in practice.

Despite that freedom, the survey finds near-total convergence: 95% calibrate the RA to approximately the 75th percentile of the distribution of outcomes — the equivalent of holding enough margin to be 75% confident of covering the undiscounted liability. That convergence reflects market convention rather than a regulatory mandate: firms chose this level because neighbouring peers and auditors anchor there.

91% set a higher RA percentage on the LRC than on the LIC. This is the expected pattern, and para B91 explains why: the RA should reflect the uncertainty remaining in the fulfilment cash flows. For unexpired risk (the LRC), that uncertainty covers both whether a claim will occur and how severe it will be. For incurred claims (the LIC), the event has already happened; only development uncertainty remains. So a higher margin on the LRC is natural — and most pronounced for short-tail classes, where the LIC resolves quickly once the claim occurs.

The survey notes that the LRC RA is commonly set as a gross-up of the LIC RA — in the market observed, a roughly 25% uplift on long-tail lines, with a larger gross-up on short-tail classes where the LIC resolves faster and hence carries less residual uncertainty. That 25% benchmark has a lineage: it was first documented in a 2001 Institute of Actuaries of Australia paper on risk margins.

75% apply an explicit diversification benefit when calibrating the RA — reducing the aggregate margin to reflect that risks across groups are not perfectly correlated. The remaining 16% apply no diversification benefit, treating each group's RA in isolation.

76% did not adopt the other comprehensive income (OCI) option for discount-rate changes. Where that option is not elected, the financial effect of changing discount rates passes through profit or loss as insurance finance income or expense; electing OCI defers it and reduces income-statement volatility. The market's preference for P&L routing is notable — it increases income statement exposure to rate movements.

5. Tooling: three years in, the spreadsheet stays

Two tooling findings are worth separating.

The first is IFRS 17 software versus Excel for the insurance finance income or expense calculation (IFIE): 49% use dedicated IFRS 17 software; 49% use Excel. An even split after three years of live reporting suggests that neither approach is clearly dominant. Accounting-led IFRS 17 platforms handle the liability roll-forward and present-value mechanics but may not carry the cash-flow modelling granularity that more complex books need.

The second finding is more pointed. 63% still use Excel for the loss component calculation and onerous-contract testing.

The loss component test requires judgement that is hard to systematise in most IFRS 17 software: pricing data, revised loss estimates, and expected claims development must be combined at the group level, often using data that lives outside the accounting platform. IFRS 17 paragraphs 47 to 52 govern the loss component under the GMM; under the PAA the equivalent test is in paragraphs 57 to 58, where the question is whether the fulfilment cash flows for the remaining coverage exceed the carrying amount of the LRC. Until insurers integrate that fulfilment-cash-flow estimate directly into their accounting systems, the spreadsheet stays in the loop.

Onerous testing granularity: 42% test at the statutory class level, 41% at the reserving class level, and just 2% at the individual contract level. The grouping rules in IFRS 17 paragraphs 14 to 22 define the minimum granularity — portfolios of similar risk (para 14), divided by profitability band (para 16), within annual cohorts (para 22) — but the survey shows most firms are working at class rather than contract level. The underlying cash-flow model drives the granularity available; firms running reserving triangles at the reserving class level will naturally test onerous groups at that level.

6. The pattern: pragmatic convergence

Across almost every question, the survey shows the market converging — and converging on simplifications. The PAA almost universally replaces the GMM. Eligibility is argued by judgement rather than modelling. The LRC is not discounted, the bottom-up rate is the universal method, and the illiquidity premium is typically omitted on materiality grounds. The RA clusters at the 75th percentile. Excel handles the most judgement-intensive calculation.

That convergence happened quickly, within the first reporting cycle. It reflects a rational response to a principle-based standard: where the standard permits a simplification, and where materiality arguments support it, firms took it. The operational savings are real — running parallel GMMs for eligibility, building a liquid yield curve with an explicit illiquidity adjustment, and replacing the spreadsheet-based loss component test with something fully systemic are all expensive.

The authors are explicit about the risk: whether these simplifications can be sustained in the long term remains to be seen. Regulators and auditors in all three markets are building experience with IFRS 17 reporting, and the scrutiny on eligibility arguments, materiality of the illiquidity premium, and onerous-testing granularity is increasing each reporting cycle. The choices that were defensible in the first year under a new standard face a higher evidential bar in year four.

IFRS 17 was designed to improve the transparency and comparability of insurance reporting. The survey shows a market that moved quickly toward a shared set of practices. Whether those shared practices produce comparable financial statements — or shared simplifications that limit the standard's objectives — is the open question the next few years of audit challenge and regulatory review will answer.

Using this as a benchmark

A survey like this is most useful as a calibration point for your own choices. If your PAA eligibility argument uses a qualitative approach without a quantitative GMM comparison, 65% of the surveyed market does the same — but that does not make the argument robust; it makes it common. If your RA sits below the 75th percentile, you can explain why; if it sits at the 75th because that is where the market sits, the explanation needs more than that.

The useful exercise is to compare your implementation choices against the survey findings, identify where your choices differ from the majority, and verify that the difference is documented and defensible rather than a gap. The areas most likely to attract attention are the illiquidity premium (where "materiality" needs quantified support), the onerous-testing granularity (where reserving-class-level testing needs to be shown to cover all onerous groups), and the LRC RA gross-up (where the 25% benchmark needs to be explained in terms of the specific risk profile, not just cited as a market convention).

For the methodology, the engine configuration, or the review framework on any of these points, the IFRS 17 work on this site covers PAA implementation, risk adjustment calibration, and onerous-contract testing.

The CAS paper is here — the full survey splits, the authors' methodology, and the source for every figure in this post.