Physical Address
304 North Cardinal St.
Dorchester Center, MA 02124
For centuries, mathematics has given us a remarkable ability to make decisions under uncertainty. Insurance companies estimate future claims from historical losses. Banks estimate the probability of default from past borrower behavior. Engineers design bridges to withstand events expected to occur once every hundred years. Investors price financial assets by estimating the likelihood of future outcomes. Although these activities appear very different, they all rely on the same mathematical idea. We observe the past, estimate the probability distribution that generated those observations, and use that distribution to reason about the future.
The success of modern statistics, actuarial science, and financial mathematics rests on a simple premise. Although the future is uncertain, it is not fundamentally different from the past. Yesterday’s observations remain informative because they are assumed to have been generated by essentially the same process that will produce tomorrow’s observations. For many problems, this assumption works remarkably well. Human mortality evolves slowly. Manufacturing processes are designed to remain stable. Even financial markets, despite their volatility, often exhibit statistical regularities that can be estimated over sufficiently long periods.
Climate risk presents a different challenge. The problem is not simply that extreme events are uncertain. Uncertainty has always existed. The problem is that the processes generating those uncertainties may themselves be changing. If the probability distribution evolves over time, then history becomes a moving target rather than a stable reference point. This distinction may appear subtle, but it has profound mathematical consequences.
Most statistical methods begin with assumptions. Some assume observations are independent. Others assume relationships are linear or that errors follow particular probability distributions. One of the most important assumptions is that the statistical properties of a process remain sufficiently stable for historical observations to reveal something meaningful about the future. Statisticians refer to this property as stationarity. Stationarity does not imply that nothing changes. Daily temperatures fluctuate. Financial markets rise and fall. Rivers flood and recede. Rather, it means that the underlying statistical mechanism generating those observations remains sufficiently stable that concepts such as averages, variances, and return periods retain their meaning over time.
Many of the quantitative methods used throughout finance, engineering, insurance, and economics depend upon this stability. Historical observations are treated as repeated samples from the same underlying process. As more observations accumulate, uncertainty about that process decreases and statistical estimates become increasingly reliable. But what happens if the process itself changes? Suppose that extreme rainfall becomes progressively more intense over several decades. Suppose that wildfire seasons lengthen as temperatures and vegetation patterns evolve. Suppose that coastal flooding changes as sea levels rise. In each case, the question is not merely whether extreme events occur. The question is whether the probability distribution governing those events remains the same as the one that generated the historical record.
If it does not, then adding more historical observations may not improve our estimates of future risk. In some cases, it may actually make them worse by increasing confidence in a distribution that no longer exists. This is one of the defining mathematical challenges of climate risk. The objective is no longer to estimate a fixed probability distribution. Instead, we seek to understand how probability distributions themselves evolve through time.
The implications extend far beyond weather. Insurance premiums are based on expected future losses. Infrastructure standards rely on estimates of rare events. Banks assess credit exposures using historical relationships. Asset managers estimate future returns and volatility from past market behavior. Governments estimate future fiscal liabilities arising from disasters, infrastructure investment, and economic growth. If the underlying hazards become non stationary, every one of these financial models inherits that uncertainty. Climate risk therefore propagates through balance sheets, insurance markets, capital markets, and public finance, not simply because hazards change, but because the mathematical assumptions used to quantify those hazards become less reliable.
None of this means that classical mathematics has failed. Probability theory remains as rigorous as ever. Statistical inference remains indispensable. Extreme value theory, stochastic processes, Bayesian statistics, causal inference, and decision theory continue to provide the foundation for quantitative risk analysis. What changes is how these tools are used. Instead of estimating parameters that remain constant, models may allow them to evolve through time. Instead of assuming that historical extremes are representative of future extremes, extreme value models may incorporate changing physical drivers or other explanatory variables. Instead of relying exclusively on historical observations, models increasingly combine observational data with physical climate models, Bayesian updating, simulation, and uncertainty quantification. The mathematics becomes more adaptive because the system being modeled is more adaptive.
This shift also changes the philosophy of quantitative analysis. Traditional statistical modeling often asks, What is the most likely future? Climate risk analysis increasingly asks a different question. How should we make decisions when the system generating future outcomes is itself evolving? Prediction remains valuable, but it is no longer sufficient. Decision makers increasingly require models that quantify uncertainty, make their assumptions explicit, update as new evidence becomes available, and remain transparent enough to be independently validated. The goal is no longer to produce a single answer. It is to construct mathematical frameworks that continue learning as the world changes.
This perspective shapes the work presented throughout Arctica Lab. The essays that follow examine the statistical and mathematical foundations needed to study climate risk in non stationary systems. We explore how probability theory changes when distributions evolve, how Bayesian methods support learning under uncertainty, how extreme events can be modeled when historical frequencies are no longer stable, and how uncertainty propagates through financial institutions and public balance sheets.
Climate change does not invalidate mathematics. Rather, it reminds us that mathematics is always conditional on the assumptions from which it begins. The central challenge for climate finance is therefore not to invent entirely new mathematics, but to understand which mathematical assumptions remain appropriate, which begin to fail, and how quantitative methods can evolve alongside the changing systems they seek to describe.
