Reading the Polo study: from physical causes to economic losses

1. A new study reopens the link between cause and loss

World Weather Attribution (WWA) released its September Hurricane Polo study on October 15, 2026: warming–El Niño comparisons for potential intensity, but unquantified warming influence on rainfall because models and observations disagree. [1]

For businesses assessing insurance, investment and supply chains, that combination raises two questions. Which assets, and which planning horizon, warrant additional protection when knowledge of weather conditions changes? How should budgets be allocated when evidence is stronger for one variable than another? Predicting damage from only the strongest result, or discarding protection because another result remains uncertain, loses information needed for the decision.

The entry point into international markets is not determined by a storm-intensity number alone. Transmission depends on the production and trade supported by exposed factories, routes and power supplies. An insurer and a manufacturer reading the same study need different further information. One asks about losses across an insured portfolio; the other asks how long a necessary component might be unavailable and how production could recover. A study of physical causes can inform both without supplying the same answer.

2. Why construct a world that did not happen?

Counterfactuals compare with a world lacking the factor of interest. The Intergovernmental Panel on Climate Change (IPCC) describes comparing distributions at fixed magnitude or fixed probability, with results depending on conditioning. [2]

Investment decisions contain a similar fork. Is the comparison between losses with and without an added flood barrier? Or between losses for unchanged facilities under different sea-level or rainfall conditions? The first asks about protection; the second asks about environmental change. Allocating the reduction in one comparison as a share of the cause in the other overlooks the fact that the facility and its environment are different interventions.

Changing the comparison need not be manipulation; different questions require different interventions. A number stripped of its question, however, is easily misused. Holding assets in the same location makes a weather change easier to interpret, while leaving population movement or new construction outside the answer. A long-horizon loss estimate must set out changes in weather and assets separately, then examine their combinations.

In practice, complete the left side of a comparison table before reading its result: the target, conditions kept unchanged, factor removed and unit of the answer. That reduces the chance of mistaking an explanation of cause for a prediction of damage. When aligning data over time, specifying how information available then differs from later information likewise separates a decision that was possible before the event from its subsequent explanation.

3. Two ratios are not shares of a cause

A likelihood ratio needs its numerator and reference world read together. Each panel below sets its own reference to 1. A longer bar means greater likelihood relative to that panel’s reference, not a share of a total combining the panels.

Numbers partitioning one total can support a sum or composition. Here the comparison worlds differ. Adding the ratios as a total effect, or dividing one by the other as a contribution share, answers neither research question. A pie chart assigning the smaller ratio to natural variability and the larger to human influence would invent a common denominator.

The mistake is especially easy when results enter financial presentations. A striking ratio survives an internal briefing or investor deck more readily than its conditions. Capital allocators, however, need to know how their prospective losses change. Use of a research ratio requires its target variable and comparison world beside the number; translating it into future frequency or damage to owned assets requires separate assumptions.

4. Between potential intensity and realised wind

Potential intensity (PI) represents a theoretical thermodynamic upper bound on storm intensity given ocean and atmospheric conditions. Gilford’s 2021 methods paper explains its theory and calculation. This differs from a record of wind reaching a facility. A stronger environmental ceiling still needs to be connected through a storm’s track and development before it describes forces on a factory wall or port crane. [5]

For investment planning, the sequence is metric, site, component and operational effect. Even when comparing resistance to wind, a building’s overall design specification differs from the state of doors, roofing and power connections. A vulnerable component can halt work without major damage to the main structure. Examining which force on which component interrupts production supports comparison of protections better than substituting a research metric for a factory shutdown rate.

There is room here for loss to grow more slowly than the hazard. Replacing components, relocating critical equipment or keeping recovery spares can change the response to the same force. Conversely, expanding production around a weak connection can increase potential loss faster than the weather changes. Asset modifications need to sit beside revised hazard assessments. Only then can an owner ask whether investment is keeping pace with the changed hazard.

5. An unquantified rainfall result still carries information

When observations and models do not show the same trend, an estimate linking the observed change to a particular cause becomes weaker. Observations alone may mix natural variability, data characteristics and other changes. Model failure to reproduce them cannot erase the observations either. The boundary between what can and cannot be estimated helps identify the additional regional evidence that would improve a decision.

Local flood protection does not become meaningless until a causal share is precisely known. An operator can still investigate the rainfall at which drainage, floor elevation and water-entry routes cause a facility to fail. Conversely, a refined attribution ratio may do little to compare construction benefits if water movement around the site remains unknown. Research uncertainty and missing information for protection design can reside in different parts of the problem.

Applying one safety factor to every variable is therefore crude. Stronger hazard evidence calls for comparison of measures effective against that hazard. Strong regional differences call for observations, terrain and damage records, tested across conditions. Separating modest measures useful in many states from large investments valuable only under a narrow assumption helps avoid both inaction and excessive fixed cost.

6. SG Group View: inspect three connections

SG Group emphasises three connections: from comparison world to hazard, from hazard to loss, and from loss to its bearer. The first preserves the research question. The second adds location, assets and response. The third adds contracts, finance and time. Skipping a connection turns a meaningful scientific number into a misapplied economic forecast. Locating the information needed at each connection also makes it clearer which decision should change when the science improves.

UNDRR’s definition of disaster risk includes hazard, exposure, vulnerability and capacity. Exposure concerns people or assets in harm’s way; vulnerability concerns their susceptibility to damage from a given hazard. A larger weather metric cannot close a loss estimate without what is present and what it can withstand. This is a foundation for moving from physical hazard to financial damage. [3]

The framework can support different rational actions from the same study. A company concentrating production at one coastal site may investigate replacement capacity; one with dispersed sites may investigate shared power or transport dependencies. Insurers examine contract concentration and local authorities examine breaks in evacuation or essential services. These differences arise from the objects connected to the hazard, rather than agreement or disagreement with the study. Investment comparison starts by identifying whose loss the expenditure seeks to reduce.

7. When a local disaster reaches distant trade

Cross-border effects require a link between the damaged location and overseas production or demand. A shutdown at a supplier of hard-to-replace components can stop a distant assembly plant after inventories run out. Where suppliers are easily substituted, or demand is weak and stocks ample, the same local damage may have a smaller international effect. A large storm and a strike on an economic bottleneck are separate conditions.

A list of supplier addresses can miss shared weaknesses. Separate factories using the same port, road or substation may be less diversified than their count suggests. Conversely, sites in one region may shorten interruption if they can switch transport and power routes. The relevant comparison is the set of functions lost simultaneously to one hazard, rather than the number of company names. Shared dependencies similarly change how supply chains connect lead times and inventories.

Inferring a global shortage of a particular commodity from the Polo study would require evidence about its suppliers and transport capacity. The hazard characteristics identified by research help narrow the search. Wind directs attention to equipment resistance; rain to inundation and road access; successive hazards to the possibility of replacement routes failing together. This specifies which supply conditions could transmit into markets, rather than predicting a commodity price level.

8. Read the contract before translating research into insurance prices

An insurance premium is not simply the price of a disaster occurring at a location. Suppose the policyholder bears losses up to an amount and the insurer pays a layer above it. A hazard change reaches the insurer’s payment when it affects that layer. More small losses and a greater possibility of very large losses can affect the same contract differently, even with a similar increase in average hazard. Multiplying a premium by a research ratio omits the range of payment.

Within one company, building, inventory and interruption cover may have different conditions, so covered losses and payment timing need not align. The comparison must ask whether the intended cover includes a period when repairs are complete but orders have not returned, or when equipment survives but an external road remains unusable. These are questions for comparing contracts, not statements of any particular policy’s terms. Actual terms govern the answer.

Underwriters must assess simultaneous losses as well as site-level hazards. If dispersed contracts respond to a shared climate state, treating them as independent may understate large simultaneous payments and short-term funding needs. Dependence alone does not change the sum of expected losses. Conversely, a narrow damage footprint with little portfolio overlap can leave an insurer with limited payments despite a prominent disaster. For equity analysis too, retained regions and loss layers determine the route to earnings more directly than the storm headline.

Underwriting capacity also requires separating average annual payments from payments concentrated in a short period. An arrangement profitable on average can create liquidity pressure if claims arrive before assets can be monetised. Ample immediate cash, conversely, does not establish the economics of carrying losses over many years. Separating frequency, amount and concentration in time makes capacity more concrete than a comparison of premiums alone.

9. The mismatch in a payment tied to an observation

Consider a contract paying on rainfall or wind measured at a predefined station, rather than assessing each actual loss. Even if designed for rapid determination, a station located away from the asset can fail to reach the payment condition while the asset suffers damage. It can also trigger without major damage. Using one number as scientific metric, payment trigger and business loss hides this mismatch.

A buyer needs to test which observation represents its difficulty. Wind damage to roofing makes measurement and location relevant; delayed cargo makes the conditions interrupting roads or ports relevant. Convenient publication does not prove suitability. Comparing historical triggers with interruptions, retaining both losses without payment and payments without interruption, identifies the gap the contract actually fills.

Payment timing matters too. Even with the same eventual compensation, an early cash shortage can delay parts or temporary workspace and prolong interruption. An observation-linked payment supporting initial recovery can have value distinct from complete indemnification. Expected speed, however, does not remove the risk of a missed trigger. Comparing timing and fit separately provides a practical connection between a scientific metric and a financial contract.

10. Which loss is the investment meant to reduce?

Separate measures reducing physical damage from measures accelerating recovery after interruption. A project blocking water entry acts on direct damage; an agreement for alternate workspace acts on interruption after damage. Both may reduce business loss but work in different states. Applying one disaster ratio equally to them produces a superficial comparison of construction and contract costs.

Start with the workflow without protection and identify when revenue or cost changes. Counting only repair costs can omit lost contracts or expediting expense during interruption. Counting revenue that can be delivered later as permanently lost can exaggerate benefits. Whether an alternate process exists and customers can wait feeds back into the value of the project.

For long-lived facilities, a sequence useful across several environments may be more informative than one assumed future. Moving critical equipment upward first, with perimeter construction available as observations develop, preserves different options from committing everything immediately. Flexibility can be valuable, while waiting leaves exposure in place. The cost of keeping options and the cost of delay belong in the same comparison.

Match evaluation horizons as well. Placing an upfront construction cost beside an annual service charge overlooks how service life can reverse the ranking. Include maintenance, replacement and exit costs, then compare the outcome protected over the same period. Assumptions converting future benefits into today’s money are separate from hazard estimates. A long assumed service life and unchanged hazard throughout that life should not be bundled into one certain fact.

11. What can a low-loss event teach?

Low loss in an event does not necessarily demonstrate resistance to the same hazard. An absence of critical assets in the most severe footprint differs from assets protected by effective measures. Crediting the first to protection can count a fortunate track as an engineering success. Ignoring protection in the second can turn preparedness that worked into a reason to cut its budget.

To assess protection, record assets that survived comparable hazards as well as damaged assets. Matching structure, terrain, use and prior response helps investigate which differences changed recovery. Collecting only dramatic damage photographs misses protected facilities and responses that shortened interruption. Even with the same reported total loss, where the reduction occurred changes the next allocation of expenditure.

Nor should later detailed weather information be treated as known before the event. Each forecast update needs to be matched to available options and their execution time. Judging only by the outcome rewards decisions based on unavailable information and can mislabel reasonable caution as waste. Guidance on time shifts and causation helps separate information arriving first from information improving a decision.

12. The clock for explanation and the clock for action

Attribution defines an event that happened and compares changed causal factors. Daily forecasts and warnings support decisions about where and what may happen next. The two can inform each other without serving the same decision at the same time. A manager deciding whether to stop a factory before a storm cannot wait for an attribution study. A manager assessing long-lived equipment cannot evaluate its entire service life from one short-range forecast.

The World Meteorological Organization describes moving from forecasts of weather conditions toward forecasts and warnings incorporating their impacts on people and infrastructure. Connecting forecasts to action involves coordination with exposed users and disaster managers, as well as forecasters. Between the existence of scientific information and its timely use by the people who need it lies the design of a working process. [4]

Businesses need to pair a response trigger with the time needed to act. Moving materials takes time; customer communication, rerouting and worker safety have sequences too. Waiting for a more detailed forecast can improve precision while reducing feasible responses. Comparing the cost of early action with options lost through delay turns a demand for perfect prediction into a specification of information sufficient for a needed action.

13. Market effects also depend on the comparison

Economic impact also changes with what is held fixed. Suppose output in one producing area falls while spending on substitutes rises. Lost production, the buyer’s extra spending and revenue earned by substitute suppliers are different quantities. A seller’s higher revenue does not show that the affected region suffered no loss; nor can all extra spending simply be added as a net loss to the whole economy. The sign of an effect can appear different depending on whose outcome is compared.

Trade without that storm and trade exposed to that type of hazard without warming are different comparisons too. One asks about the economic difference from an individual event; the other about the difference under a changed environment. Moving a weather comparison into the latter estimate of spending or profit requires assumptions about business responses. The answer changes with whether output and prices are fixed or sourcing and protection can adjust.

If publication changes a business outlook, the effect of that information should also be examined separately from the disaster’s loss. Revised expectations of maintenance or insurance terms concern prospective receipts and costs; past building damage has not increased on the release date. Treating a publication-day market move as a measure of disaster loss mixes other news and previously known hazards. Identifying the assumption the study newly changes is a starting point for connecting scientific updates to market reassessment.

14. Change conditions and compare the ranking of responses

The first case combines a stronger hazard with improved protection and recovery. Loss can then grow less than hazard. The condition is not construction alone: information, people able to act and available repair materials must work together. Recording damage, interruption time and delivery effects separately helps identify which capabilities should be retained when explaining protected revenue.

The second case protects the factory building while transport or power remains interrupted. More investment inside the factory has limited incremental value if it does not change restart time. Funding shared infrastructure or contracting some work elsewhere may become relatively more valuable. A fallback exposed to the same hazard cannot be credited merely because it is called an alternative.

The third case identifies suitable protection but lacks initial funding or execution time. The project with the highest benefit on paper need not solve a post-disaster cash shortage. This is a reason to combine initial liquidity with long-run damage reduction. Without inventing scenario probabilities, separating conditions the company controls from external conditions allows rankings to be updated when observations or contracts change.

15. Reducing loss and moving its burden

Insurance can change who pays without repairing a damaged asset or restoring lost time. Risk sharing and funding play a different role from works that reduce physical damage. If a company buys more cover while spending less on protection, the combination needs to be assessed for effects on insurers, counterparties and local recovery. A lower accounting loss need not mean a lower loss to society.

For households and small businesses, a short interruption may quickly reach living expenses or payment deadlines. Work a large company can transfer elsewhere may remain stopped without another site or funding. Recovery can differ under a shared hazard because means of replacing income differ. Support allocation therefore has reason to consider when bridging funds are needed as well as damage size. This is a perspective for comparing burdens, not an inference about a particular region’s support programme.

Road or drainage improvements may benefit several firms and residents while the region paying differs from the region benefiting. Without considering upstream and downstream or industrial and residential areas together, shifting hazard can look like success. Ranking works solely by protected asset value can also undervalue places essential to life but carrying low valuations. An attribution ratio alone cannot choose what should be protected.

16. Examine the argument for waiting under uncertainty

One objection is that a wide range of model results favours waiting for better research rather than spending now. Waiting can be rational for short-lived equipment, small exposed assets and easily revised choices. Better information can avoid locking money into ineffective works. But waiting also needs a comparison: what might be lost during the interval, and whether the same works remain feasible later.

The opposite assertion—that evidence of increasing hazard justifies rushing every protection project—is also insufficient. Building the wrong capability yields little benefit even when the hazard explanation is correct. Allocating additional funds among observations, evacuation and structural works depends on local weaknesses. Confidence in science does not automatically prove a particular project’s economics. The reason to act and the reason to choose that action both need to be shown.

Nor do firms with low historical losses necessarily have superior protection. Exposure counts, land conditions and insurance scope can differ. Valuing a short accident-free record alone can direct capital toward firms fortunate enough to avoid hazards while missing firms with effective protection. Matching experienced hazards with losses and interruptions moves the comparison closer to the effect of the response.

A method for retaining hypotheses and falsifiers is useful for recording such objections. If the hypothesis is that the equipment stops only when external power fails, a shutdown without a power failure is contrary evidence. If works are meant to accelerate recovery, cases dominated by waiting for parts deserve examination. Breaking the issue into decisions that can change identifies the next evidence more clearly than a broad declaration for or against the research.

17. What additional evidence would change this assessment?

On the scientific side, the relevant question is whether other observations and models retain the conclusion for the same event definition. A large change from a modest adjustment to region or period narrows the scope for economic use. Independent materials supporting the same relationship strengthen the basis for comparing long-run protection. Revised results need to be read for the conditioning that changed, not just whether the number became larger or smaller.

On the asset side, decisions change with linked records of local weather, damage, lost functions and recovery time. Comparing protected and unprotected locations still requires adjustment for hazard differences. Dates of insurance or support receipts alongside restart dates help investigate whether money accelerated recovery or another constraint remained. Time and process records serve investment selection better than aggregate loss alone.

On the market side, examine actual shipments, substitute supply and delivered-price differences. Stability in these weakens a short-run shortage thesis even if hazard research changes. Orders concentrating on limited substitutes, with transport or quality constraints appearing, make the route from a local event to wider costs more concrete. Examining the intervening functions tests a causal claim better than placing a published ratio beside a market move.

18. Connecting research to the use of capital

The study’s economic value lies less in choosing a winner between causes than in showing how the answer depends on the comparison. Preserving the question removing warming and the question changing natural variability allows the long-run environment and a contemporaneous state to be considered separately. Adding exposed assets and payment contracts then specifies the decisions a result can inform. Information can be used to reduce the conditions driving loss, rather than to announce one loss forecast.

Businesses and investors should take away a process for naming the comparison world and connecting local weaknesses to effective measures, rather than a formula transferring a ratio into a cost cell. Resident safety, business continuity, insurance payment and public recovery protect different outcomes. The same budget can be used differently depending on the outcome chosen. Matching the precision a decision needs to the time left for action, without pretending uncertainty has disappeared, is a starting point for putting research to economic use.