📑 Table of Contents
- Introduction
- Confidence in the model is confidence in the capital
- We’re not modeling the world we live in. We’re modeling the world we used to have
- Liquidity is not resilience
Introduction
Hurricane Melissa made landfall in southwestern Jamaica as a devastating Category 5 storm and is the strongest to hit the island since record-keeping began. Officials estimate the damage at between 28 % and 32 % of Jamaica’s previous-year GDP.
The storm inflicted catastrophic impact on infrastructure, homes, and the tourism-centric economy in s a region still recovering from Hurricane Beryl in July 2024. Beryl, though less intense than Melissa, left millions without power, blocked major roads, destroyed homes and disrupted livelihoods.
Agricultural regions were particularly vulnerable: crops still rebuilding from Beryl’s shock were pummeled anew. The tourism sector (which contributes roughly 30 % of GDP and supports about 175,000 workers) now faces not just recovery from physical damage, but a broader financial gap in reconstruction, capital allocation and resilience planning.
This RMN Briefing discusses the post-event outlook for risk, market and models with:
- Sridhar Manyem, AM Best’s Senior Director, Industry Research, Analytics and Public Relations. Sridhar is the head of the industry research team and his responsibilities include publishing AM Best’s perspectives of topical issues relating to the insurance industry and possible implications to Best’s Credit Ratings.
- Dr. Josh Hacker, Jupiter Intelligence’s Chief Science Officer and Co-Founder. Josh is an atmospheric scientist focusing on lower atmosphere prediction and predictability across many time and space scales and drives science innovation for Jupiter’s ClimateScore platform the underlying foundation for all of Jupiter’s products and solutions.
- Professor Paula Jarzabkowski teaches Strategic Management at The University of Queensland and City St George’s, University of London and is a expert in public-private mechanisms, globally, to address the disaster insurance protection gap and government programs to increase homeowner and community disaster resilience.
"Confidence in the model is confidence in the capital"
Risk Market News: What was your first takeaway from Hurricane Melissa and its financial implications?
Sridhar Manyem: For reinsurers and investors, the first question isn’t how big was the storm? It’s how well did the model perform? And are the losses within our established risk limits?
Melissa reinforces how dependent capital on confidence in the modeling pro-cess.
When the modeled loss diverges widely from the actual loss, investors hesitate to redeploy capital. That’s the essence of the market right now: confidence in the model reflects the confidence in the capital. Investors may express this uncertainty through a number of ways including required spread, pricing, and allocations.
RMN: How much of that confidence problem is about data versus methodology?
Manyem: Mostly data. Methodology evolves, but if the exposure data and the stochastic event sets are weak, even the best model can mislead.
Regional calibration in places like the Caribbean is still developing. The event sets are smaller, the validation data scarcer. That translates directly into model uncertainty.
Better exposure mapping, more consistent claims data, will go a long way toward narrowing the confidence gap.
RMN: Does that mean investors are over-reacting to uncertainty?
Manyem: No. I’d call it rational caution, not fear. Capital markets are disciplined. When they can’t quantify risk precisely, they price in an uncertainty premium. That’s not moral hazard. It’s a market function.
After a season like 2025, when you’ve had Melissa on the heels of Beryl, investors look at the calibration of the regional models.
RMN: How does AM Best incorporate these modeling is-sues into its own analysis?
Manyem: We view catastrophe loss as a severe threat to the balance sheet of a P/C insurer because the impact could be significant, rapid, and unexpected.
In our Enterprise Risk Management (ERM) evaluation, we consider an insurer’s approach to stress test-ing. Companies with effective ERM use accurate and comprehensive data to manage catastrophe risks, while understanding model limitations. We look at model parameters including demand surge, storm surge, adjustment expenses etc, in addition to data quality.
In our Best’s Capital Adequacy Ratio (BCAR) model, we include a capital require-ment to capture catastrophic risk in the model. In addition to the capital requirement, we conduct a stress test to give us an insight on the state of a company’s balance sheet after an event.
We are agnostic about the models insurers use but in our ERM process and other considerations, we look at management’s understanding of a respective model, conservativeness in parameterization, and the ownership of model outputs to make risk management decisions.
RMN: Some experts argue that the models themselves need to be rebuilt from scratch. Do you agree?
Manyem: Rebuilt? No. Refined? Absolutely.
Catastrophe models have matured enormously in the last two decades. The next phase is incremental—better data assimilation, more transparency, stronger validation.
What we need instead is evolution with accountability: climate conditioning, improved hazard footprints, and consistent documentation of assumptions.
So I’m more in the “continuous calibration” camp. The science doesn’t need to be reinvented; it needs to be verified and trusted.
RMN What regulatory or rating developments could strengthen that trust?
Manyem: Disclosure and comparability.
The goal isn’t to micromanage the science; it’s to ensure everyone knows what’s under the hood. Once investors can line up those assumptions across models, they can make informed capital decisions.
Transparency builds confidence; confidence attracts capacity. It’s that simple.
RMN: How long will it take before we have a clear sense of the industry losses from Hurricane Melissa and how the models performed?
Manyem: I think it will take time for the final loss amounts from Melissa to be known. Verisk estimates have the losses around $2.2 Billion - $4.2 Billion.
Once the storm settles down and loss adjusters assess losses we will know better on how the models have performed. Modeling is one key component of a risk management exercise in addition to stress testing, establishing risk tolerances and consideration of exogenous factors (interest rates, spreads, capital allocation, and risk appetite to name a few) that will determine what to do with a risk (avoid, retain, mitigate through reinsurance).
Modeling is a core pillar of that but the models provide insight for action.
RMN: What’s the key message for the reinsurance and ILS markets post-Melissa?
Manyem: Modeling remains the bridge between risk and capital. Strengthen that bridge—through better data, validation, and transparency—and the capital will follow.
As I like to say: confidence in the model is confidence in the capital. The market is also influenced by interest rates, spreads, investors’ risk appetite and capital allocation.
"We’re not modeling the world we live in. We’re modeling the world we used to have"
Dr. Josh Hacker, Jupiter Intelligence’s Chief Science Officer and Co-Founder
Risk Market News: How would you summarize what Melissa and Beryl revealed about the state of catastrophe modeling today?
Josh Hacker: They exposed how fragile our assumptions still are. The models assume independence between seasons; the climate doesn’t. They assume a stable baseline; the climate’s shifting under our feet.
We’re not modeling the world we live in. We’re modeling the world we used to have.
The good news is that we can change that. We just have to accept that the data are no longer a static archive of the past — they’re an evolving system. Once we treat them that way, the models, and the capital that depend on them, will evolve too.
RMN: Hurricane Melissa struck Jamaica just over a year after Hurricane Beryl. What stands out to you about these back-to-back events?
Hacker: It’s the kind of compounding risk that’s becoming the new normal. You had Beryl last year, and now Melissa, and the recovery window between the two just wasn’t long enough. Some of the agricultural regions couldn’t have recovered that fast, and now they’re set back again. You’re going to see a fair bit of food insecurity and export disruption because of that.
What’s interesting is that this isn’t just about intensity, it’s about frequency and sequence. When storms cluster like this, they stretch the whole recovery timeline. That’s a huge challenge for anyone trying to model or price the actual economic impact.
RMN: Are today’s catastrophe models equipped to capture that kind of sequential behavior?
Hacker: Not really.
Most of the models are built to simulate a single year, not a sequence of years. They’re designed for a one-year catalog of events that assumes a kind of statistical independence between successive years.
But the physics of the system are changing. We’re seeing plausible years now that have never been observed before. Kerry Emanuel at MIT has shown this: you can have a series of major storms that fit within the current climate state but have no historical precedent. The damage is nonlinear and the loss curve steepens faster than the hazard curve.
So, the idea that you can represent risk with one “average” catalog doesn’t hold up anymore. There are combinations of events that we’ve never experienced but that are now physically possible.
RMN: Many modelers say they’re “climate-conditioning” their catalogs to account for that. Does that help?
Hacker: It helps a little, but it’s a halfway measure. You’re taking an existing stochastic catalog and “beefing up the tails”, trying to adjust probabilities to match a warmer climate. But that doesn’t change the underlying structure of the model.
What you really want is to model the heck out of the current year and simulate the storms that are physically plausible today, not just statistically plausible based on the last fifty years. When you’re climate-conditioning a catastrophe model, you don’t get all the way there.
The vendors know this, but they’re constrained by regulation and the economics of their business. They have a defensible position now; the U.S. regulatory market is huge and doesn’t reward rebuilding a model from the ground up. So, there’s not much incentive to take the leap.
RMN: How does this limitation affect the way insurers and reinsurers use those models?
Hacker: They still use them because they have to. The models are the currency of the industry, You need them to justify capital allocations and regulatory filings. But there’s a growing recognition that the outputs don’t capture everything.
For insurers, that means relying more on scenario analysis and custom modeling. For reinsurers, it’s about diversification and exposure management. For capital mar-kets — cat bonds, ILS, sovereign pools — it’s about trusting the trigger logic even when you know the hazard data might be incomplete.
The risk is that if the models are under-representing the tails, then the structures being priced — the cat bonds, the reinsurance layers — are too thin. You end up with instruments that look fine on paper but don’t really cover what’s coming.
RMN: That brings up the moral-hazard argument and that some players may prefer not to know the full extent of the risk. Do you see that happening?
Hacker: I wouldn’t call it a conspiracy, but yes, there’s an incentive not to know. If you acknowledge that the tails are fatter than you’ve priced for, everything gets more ex-pensive — reinsurance, bonds, collateral.
So, there’s a quiet bias toward maintaining the illusion of precision. The cat model gives you a number, and everyone downstream treats it as a truth — even though it’s really a fragile estimate.
That’s why I say the problem isn’t just data; it’s the system of belief around the data. The whole pricing chain depends on treating the output as definitive when it’s re-ally conditional.
RMN: Where does Jupiter fit in this ecosystem?
Hacker: We’re complementary right now. We provide climate and hazard analytics that help clients see the risk landscape differently. Our clients include insurers, reinsurers, banks, and large asset owners.
Banks and lenders, in particular, are using our data to get a second view of risk. They’re also looking at uninsurable losses. Things that fall outside the traditional policy framework and they want to understand how climate exposure affects asset value, loan pricing, and long-term portfolio risk.
For insurers, it’s more about client advisory and risk engineering. Some buy a cat model and buy us too. They use the cat model for capital and regulatory work, and they use us for decision support. It’s a multiple-views-of-risk strategy.
RMN: You mentioned the banking side. Are financial institutions moving faster than insurers in integrating climate data?
Hacker: In some ways, yes. The leading banks are fully convinced that climate risk is real and material. They’re not arguing about the science anymore, they’re trying to figure out how to gain a competitive advantage from knowing it first.
There’s a widening gap between leaders and laggards. The leaders are building in-house analytics teams and integrating climate risk into credit and asset-management decisions. The laggards are waiting for regulation to force them.
We’re seeing the same divide in the insurance sector. Some carriers are experimenting with physical-risk data at the portfolio level; others are still outsourcing all their climate risk modeling. The gap is stretching.
RMN: How does that relate to sovereign mechanisms like CCRIF in the Caribbean?
Hacker: CCRIF is an interesting case. It’s a well-designed liquidity mechanism; fast pay-outs, clear triggers, strong governance. But it’s still using the same basic cat-model architecture as everyone else.
If that model doesn’t include events like Melissa, sequential or compounding storms, then the payouts can’t keep up with the actual economic damage. You’re not going to be writing a bond that’s big enough.
That’s the structural challenge. The mechanism works, but it’s constrained by the data feeding it. Liquidity is great, but it’s not resilience. Without better models, the system is always catching up to the next surprise.
RMN: What do you think will drive the next wave of innovation in modeling?
Hacker: Two things: physics-based modeling and data fusion. We need to stop thinking of climate models and cat models as separate worlds. They’re parts of the same continuum; one tells you how the atmosphere behaves; the other translates that into losses.
And then there’s the data revolution. With remote sensing, IoT, and AI, we have the ability to observe exposures and hazards at resolutions we couldn’t touch before. The challenge is integrating that information into financial frameworks that still expect a single annual loss curve.
Eventually, the regulatory environment will catch up, but it’s going to take time. The incentive structure isn’t there yet. Until it is, most climate-risk innovation will mostly happen outside the core cat-model vendors.
"Liquidity is not resilience"
Paula Jarzabkowski, Professor at The University of Queensland
Risk Market News: How did Hurricane Melissa change the conversation around protection-gap entities in the Caribbean?
Paula Jarzabkowski: Melissa really underscored the strain on small-island economies. Jamaica, for ex-ample, had barely recovered from 2024 Beryl before another system hit.
That exposes the limits of what instruments like CCRIF (the Caribbean Catastrophe Risk Insurance Facility) can actually do.
CCRIF isn’t designed to rebuild entire economies; it’s designed for liquidity. It gets cash out the door quickly so governments can keep services running. That’s its strength. But it was never meant to solve the long-term recovery problem.
Liquidity is not resilience. It buys time, not transformation.
RMN: So CCRIF’s structure works as intended, but the need goes beyond it?
Jarzabkowski: Exactly. It’s a success within its mandate, but the underlying problem is the insurance penetration gap.
Eighty to ninety-five percent of residential properties across the Caribbean are, reportedly, uninsured or underinsured. You can’t expect a government liquidity product to cover reconstruction of these private assets; that’s unrealistic.
What CCRIF has done well is evolve its coverage. After earlier hurricanes it added an excess-rainfall model, recognizing that storms are lingering longer and dumping more water. That’s model evolution in response to experience.
Still, these are liquidity mechanisms. They keep economies afloat in the first
weeks after a disaster, not for the years it takes to rebuild infrastructure and livelihoods.
RMN: Some argue the real constraint is scientific. The models themselves. Do you agree?
Jarzabkowski: No, I don’t think the model is the main problem. Any parametric product carries some basis risk; that’s inherent.
The larger issue is capital; how much you can afford to buy and what price private investors demand.
The more capital you want, the more money you’ve got to put into it.
That’s tough for small, indebted economies. Even with perfect data, if the premium burden is too high, governments can’t sustain coverage. So the constraint is socio-economic as much as technical.
RMN: How do you see the balance between public and private capital in these mechanisms?
Jarzabkowski: It has to be a partnership. Protection-gap entities like CCRIF sit between state and market. They exist because private capital alone won’t enter these thin markets, and governments alone can’t carry the risk.
The public side, development banks, aid agencies, provides concessional financing and model credibility. The private side brings discipline and speed.
But there’s always a tension: the public mandate is to keep people insured; the private mandate is to make a return. PGEs are about managing that tension rather than resolving it.
RMN: There are some market criticisms that suggest that many catastrophe models are structurally outdated. How do you respond?
Jarzabkowski: I understand this point, but I’d frame it differently. Models can and should keep improving, but they’re only one piece of the system.
Even a perfect model can’t fix affordability or governance. The Caribbean has developed some of the best sovereign modeling in the world through CCRIF, donor countries, and the World Bank. The challenge now is turning modeled knowledge in-to fiscal planning and resilience investment, not just into faster payouts.
You can refine the model forever, but if the social system can’t act on its results, the gap stays open.
RMN: Where does regulation fit in? Does it help or hinder innovation?
Jarzabkowski: Regulation is essential. It ensures that reserves exist and that payouts happen. Without it, trust collapses.
But regulation alone can’t drive innovation. What we need is a stronger dialogue between public and private actors: governments, reinsurers, multilaterals, and modelers sitting at the same table.
When that happens, you get innovation that’s still credible. CCRIF’s rainfall product is a good example—it was created precisely because governments and scientists collaborated after seeing gaps in existing coverage.
RMN: Do you see the global reinsurance market as supportive of these multi-country pool structures?
Jarzabkowski: Yes, but within limits. The big reinsurers participate because CCRIF and similar pools have excellent governance and clear triggers. They trust the framework. But re-insurers are also facing their own global load—fires in California, storms in Australia, floods in Europe. Capital is global; risk is local.
That means small markets are competing for attention and capacity. When uncertainty rises, capital migrates to where data are stronger and returns more predictable. So again, it’s about data credibility and governance confidence. Those two things determine where capacity goes.
RMN: Is there a path for CCRIF or other PGEs to move beyond emergency liquidity into long-term resilience?
Jarzabkowski: Possibly, but it would require new mandates and blended capital. Right now these multi-country PGEs “bridge the response gap.” They deliver short-term financial relief. To reduce the reconstruction gap, they’d need to fund reconstruction and incentivize risk reduction. That means working with ministries of finance, housing, and planning, not just treasury departments.
It’s also a question of scale. PGEs are small relative to the losses they address. They can catalyze, but they can’t substitute for sustained investment in resilient infrastructure or social insurance.
RMN: If you had to define the single biggest misconception about these mechanisms, what would it be?
Jarzabkowski: That they’re meant to replace traditional insurance. They’re not. They’re complementary. Tools for liquidity and fiscal stability.
The bigger misconception is that modeling equals resilience. It doesn’t. Model-ing is the map, not the territory. Without policy measures that act on those insights, you just have better estimates of the same vulnerability.
RMN: After Melissa, what is the key lesson for risk and capital markets?
Jarzabkowski: That speed is not the same as stability. Fast payouts save lives and keep systems running, but long-term resilience demands affordable coverage, diversified capital, and stronger social safety nets.
The region’s challenge is to link data, finance, and policy so that liquidity be-comes a stepping-stone to rebuilding better.
And this is for all PGEs, not simply the multi-country ones doing disaster liquidity,, because if every payout simply restores what was lost, we’re not reducing risk, we’re just resetting it.
