(Re)in Summary
- Country-level model recalibration and granular hazard data could close blind spots in catastrophe modelling and support new cross-border risk pooling in ASEAN markets.
- Insurers should recalibrate frequency and severity models with country-specific data, rather than relying on one regional model, and adopt open, transparent statistical frameworks that can be recalibrated using their own underwriting data.
- New entrants can use national-loss benchmarks for initial pricing and early-stage capital stress testing, experts said.
- Multiple insurers can use new statistical frameworks to explore cross-country risk pooling, even between markets with very different risk profiles, to reduce volatility and value-at-risk.
- Insurers can also layer portfolio data with granular hazard data to close timeliness, coverage and actionability gaps in underwriting.
Insurers should supplement and close gaps in cat models by using granular, faster hazard data, and recalibrate frequency and severity models with country-specific statistical rigour, experts said at the Asia Actuarial Conference (AAC) 2026 held in Singapore.
Rather than using a single, undifferentiated regional model that misses real variation between countries, insurers should adopt country-specific calibration to capture spatial data sharing, said Dr Xiao Xu, a research manager at the Society of Actuaries (SoA).
The frequency and severity of nat cat events are often uncorrelated across countries. At the same time, some countries cluster statistically, while others don’t, undercutting assumptions that ASEAN countries have a uniform risk profile.
“The highest frequency countries are not necessarily the countries that get the most economic losses,” Dr Xiao said. “We need a more accurate model to not only capture the regional characteristics but also for individual country characteristics.”
Dr Xiao proposed that insurers move away from proprietary, region-wide “black box” catastrophe models and instead adopt an open statistical framework, combining a country-specific frequency model with a log-linear severity model tied to population density, that insurers can recalibrate using their own underwriting data.
Established insurers can substitute internal metrics like sums insured for the framework’s public proxies, while new entrants can use national-loss benchmarks for initial pricing and stress testing.
“If you have no experience in these markets… you can also use a statistical framework to develop initial pricing assumptions and some early capital stress testing results.”
Cross-country pool
Using her statistical framework, Dr Xiao found that pooling flood or storm risk across ASEAN countries with differing risk profiles reduced volatility and value-at-risk for participants, even when their exposures varied widely.
In a simulated Malaysia–Philippines flood pool, with expected coverage of $38.7 million and $37.6 million respectively, both countries saw reduced standard deviation and value-at-risk, with an 81% probability that no payout would be needed and a 16% probability the pool would fall short.
A four-country storm pool consisting of the Philippines, Vietnam, Myanmar and Thailand could lead to Myanmar’s value-at-risk falling 62% while it contributed just 6% of expected coverage, according to Dr Xiao, with the Philippines seeing only a modest 1.9% rise in expected loss and a reduction in overall volatility despite being the dominant risk.
Multiple insurers can collaborate using the framework to pool predefined liability portions across complementary ASEAN markets, Dr Xiao said. Insurers can also use the framework to design equitable premium contributions and payout mechanisms to support transparent risk governance.
“The economic growth in this area is significant but also comes with high economic exposure to climate risk,” she said. “Increasing losses from disasters also highlight the need for a more robust and scalable disaster financing solution.”
Cat model gap
Carriers are facing timeliness, coverage and actionability gaps in catastrophe modelling, said Dennis Tay, Principal, P&C Actuarial Consulting at Oliver Wyman.
Standard cat models, good for pricing, reinsurance and capital, are slow to update but hard to run for every single risk, Tay said. “You cannot run a cat model every single quarter… every risk that comes through, you’re not going to rerun the cat model.”
Underinvestment in cat models has also led to less understanding and sophistication when it comes to underwriting, creating an “actionability” gap, said Tay. “Either I just write (the risk) and hope that my retentions are sufficient, or my reinsurance is sufficient, or that you know, let’s hope that it doesn’t happen.”
Tay proposes a framework that layers an insurer’s own portfolio data with granular, location-specific hazard data like wind speed, shaking intensity or rainfall at a single point, rather than headline figures like storm category or earthquake magnitude. At the same time, he suggests adding a “decision layer” that translates that combination into practical guidance for underwriters, claims teams and executives, without replacing the catastrophe models insurers already rely on for pricing and capital.
“Is this information useful? I think it helps you to contextualise for every risk that comes onto your table, every risk in your portfolio,” he added. “How bad was it when that thing happened? And likewise, it then contextualises a loss.”
As an example, an underwriter weighing a risk directly in a storm’s path can use historical intensity data from past events, plus pre-loaded assumptions about how wind speed decays over terrain, to judge how severe the exposure is, Tay said.
“There’s a bit of comfort, there’s a bit of data, there’s a bit of an anchor point that I can provide, and the underwriter feels maybe a bit more comfortable with looking at the risk.”
To quickly and efficiently go through nat cat predictions and understand their impacts on portfolios, Tay suggests that actuaries can run portfolios through historical events to generate heat maps of exposure by location, so decision-makers can quickly see accumulation, or track how a portfolio has “de-risked” from a past event’s footprint over time, without rerunning the full cat model each time.
“Decision makers just need data points,” he said. “They need to be able to refer to things that help contextualise what they’re doing.”
The data that’s needed are already largely available to insurers, and can be built into basic dashboards, rather than waiting for a full platform, said Tay. “I don’t need a precise number, but I need a way to benchmark against something that has happened before. I just need to be able to defend my number.”
The same data can also pay off beyond underwriting, helping insurers give stakeholders quick, defensible figures. “You can also tell people, well, we’re in a bit of trouble because we’ve got $100 million of exposure within a wind field of 250 kilometres per hour and higher,” he said. “That’s a bad number, but at least you can stay immediate.”
The same framework can also let claims or customer service teams reach out proactively to policyholders within an affected radius, rather than waiting for calls to come in.
“A lot of people talk about insurance being quite reactive,” Tay said. “Can we empower claims teams and/or customer service representatives to be a bit more proactive in reaching out to affected people?”
Tay hopes that starting with an imperfect, in-house first version of this framework will build momentum and investment over time, so that hazard-based insights become part of everyday underwriting decisions.
“We could start to build towards a position where people are more comfortable with the risk that they’re writing, the reporting,” he added. “Maybe some of the issues that we see with underinsurance, with protection gaps, and things like that might start to resolve themselves.”
