Modern insurance rests upon a simple but powerful premise: that the statistical properties governing natural hazards remain sufficiently stable for historical experience to provide a meaningful guide to future risk. This premise is known as stationarity.
Stationarity does not imply that disasters occur with perfect regularity, nor does it suggest that every year resembles the last. Floods, hurricanes, wildfires, droughts, and other natural hazards have always exhibited considerable variability. Rather, stationarity assumes that while individual events remain unpredictable, the broader statistical processes governing their occurrence remain sufficiently stable through time that probabilities can be estimated with reasonable confidence. This seemingly technical assumption quietly underpins much of modern insurance.
Historical loss records inform catastrophe models. Premiums are calculated using estimates of expected future losses derived from historical observations. Reinsurance treaties are negotiated based upon modeled probabilities of extreme events. Capital reserves are established to satisfy solvency requirements derived from expected loss distributions. Regulators oversee financial resilience using stress scenarios that implicitly assume historical relationships remain broadly informative. None of these activities require perfect prediction.
Insurance has never depended upon knowing precisely when the next hurricane will strike or where the next wildfire will ignite. Instead, it depends upon confidence that although individual disasters cannot be predicted, the statistical environment within which those disasters occur remains sufficiently stable for probabilities themselves to retain meaning. Climate change increasingly challenges that premise.
As atmospheric conditions evolve, oceans warm, precipitation patterns shift, sea levels rise, and wildfire behavior changes, the statistical characteristics of many natural hazards also begin to change. Historical observations continue to provide valuable information, but they no longer describe a system whose future necessarily resembles its past. The underlying probability distributions themselves gradually evolve.
This distinction is more important than simply observing that disasters are becoming more frequent or more severe.
Insurance has always existed to absorb catastrophe losses. Large disasters alone do not invalidate the business model. What becomes increasingly difficult is estimating future losses using assumptions derived from historical experience when the environment generating those losses is itself changing. The challenge is therefore not merely one of increasing risk, but of increasing uncertainty regarding how that risk behaves.
The consequences extend throughout the insurance ecosystem.
Catastrophe models become more dependent upon forward-looking assumptions rather than historical calibration alone. Premiums become more difficult to justify to regulators and policyholders. Reinsurers demand additional compensation for uncertainty rather than simply expected loss. Investors supplying catastrophe capital become increasingly concerned not only with the size of potential losses, but with whether existing models adequately capture changing distributions. Capital requirements themselves become more difficult to calibrate as confidence in historical volatility declines.
The implications extend even further beyond insurance.
Mortgage lending assumes that collateral can be evaluated using reasonably stable hazard estimates. Infrastructure investments often rely upon historical design standards. Municipal governments prepare long-term capital plans using expectations regarding future environmental conditions. Sovereign fiscal planning increasingly incorporates assumptions regarding disaster frequency and recovery costs. Across these institutions, historical observations remain essential, but their interpretation becomes increasingly complex when the statistical properties of the underlying system continue to evolve.
Recognizing the end of stationarity does not imply that risk has become impossible to model or manage. Financial institutions have always adapted to changing information, improved models, and new forms of uncertainty. The challenge is that adaptation itself requires acknowledging that one of the foundational assumptions underlying modern risk pricing has become less reliable than it once was. Historical data remains indispensable, but it can no longer be interpreted as though the statistical environment itself were fixed.
For insurance markets, this represents more than an actuarial challenge. It becomes a structural challenge affecting pricing, diversification, capital allocation, regulation, and ultimately the broader financial architecture responsible for absorbing climate risk. As confidence in stationarity weakens, the remaining assumptions supporting modern insurance—diversification, correlation, tail behavior, and private capital capacity—come under increasing pressure as well.
Understanding the end of stationarity is therefore not simply an academic exercise in probability theory. It marks the beginning of a broader shift in how financial institutions must think about climate risk. The remaining essays in this series examine how the erosion of additional assumptions propagates through insurance markets and ultimately into public balance sheets.
