What Does an Actuary Do? Putting a Price on Things That Have Not Happened

An insurer takes money now in exchange for a promise to pay later if something happens, which means it is selling something whose cost is unknown at the moment of sale. Actuarial work exists to close that gap. The core technique is to model the frequency of events and the severity of each event separately, then combine them to get a distribution of total losses rather than an average, because the average is not what determines whether a company survives. From that distribution come the numbers a business runs on: the expected cost, which sets the technical price; the variability, which sets how much the price must be loaded above expectation; and the tail, the rare enormous outcomes, which determines how much capital must be held and how much reinsurance to buy. The same structure applies to a pension scheme, where the events are people surviving, retiring, leaving service and dying, and the severity is the benefit payable, discounted back to today at an assumed rate of return.

The profession splits into areas that share mathematics and differ in what they are uncertain about:

• •Life insurance and annuities, dominated by mortality and longevity, where the risk is symmetrical in an awkward way since an insurer loses if people die early and an annuity provider loses if they live long
• •General insurance, covering motor, property, liability and everything else, where claims are volatile, delayed and frequently unknown in amount for years after the event
• •Pensions, which combine demographic assumptions with long-horizon investment assumptions and where a small change in the discount rate moves the liability enormously
• •Health insurance, where the drivers are utilisation, medical inflation and selection effects
• •Capital modelling and risk management, which aggregates every exposure a firm has and asks what happens in the bad tail
• •Increasingly, data science and predictive modelling roles, along with climate and catastrophe risk, where the historical record is a weak guide to the future

Three difficulties recur and are the reason the work is not simply statistics. The first is reserving: an insurer must state today what it owes for accidents that have already happened, including claims reported but not yet settled and claims incurred but not yet reported, and the estimate must be made before anyone knows the answer, using development patterns from past years that may no longer apply. The second is the discount rate, since any long-dated liability must be expressed as a present value, and the choice of rate is simultaneously a technical question, an accounting question and a political one, because a higher assumed return makes a pension deficit disappear on paper without changing a single future payment. The third is that the tail is where the danger lives and the data is thinnest, so extreme outcomes must be modelled with theory and judgement rather than observation, and the standard warning is that a model calibrated to a quiet period will always understate the storm.

Insurance works by pooling people whose risks are independent, and it is undermined when the people who buy know more about their own risk than the insurer does. That is adverse selection, and its consequence is that a price set for the average attracts the above-average risks, pushing the price up and driving the good risks out, a spiral that can collapse a market entirely. The traditional defence is underwriting, charging different prices for different measured characteristics, and the more accurate that becomes, the more it collides with fairness. Regulators have intervened repeatedly: the European Union prohibited gender-based pricing in insurance from 2012, several jurisdictions restrict the use of genetic test results, and the use of postcodes, credit data and machine learning models has drawn scrutiny for reproducing discrimination through proxies. The unresolved tension is that perfect risk pricing and risk pooling are opposites, and insurance is valuable precisely because it does not charge everyone exactly their own expected loss.

An actuary prices and reserves for events that have not happened, by modelling how often something occurs and how much it costs, then combining those into a distribution rather than an average. The tail of that distribution sets capital requirements and reinsurance, not the mean. The recurring difficulties are estimating claims that are already incurred but unknown, choosing a discount rate that moves long-dated liabilities enormously, and modelling extremes where data is thinnest. Better risk pricing steadily works against the pooling that makes insurance worth having.