How Insurance Companies Use Mathematics to Price Risk

An insurance company has no idea whether your house will burn down this year. For one policy, the outcome is close to a coin flip weighted by unknowable luck. But ask the same company how many of 10,000 similar houses will have a major claim, and it can answer with surprising precision.

That contrast is the entire insurance business. Individual risk is unpredictable. Aggregate risk is not. The bridge between the two is a handful of ideas from probability and statistics.

More Customers Means Less Risk

You might expect that insuring more homes means taking on more risk. In total dollars, it does. But the uncertainty per policy shrinks as the pool grows.

Suppose each home has a 1% chance of a $200,000 loss in a year:

• Expected loss per home: 0.01 × $200,000 = $2,000
• Standard deviation for one home: about $19,900

For a single home, the actual outcome is either $0 or $200,000, never $2,000. Now pool 10,000 independent homes. The standard deviation of the average loss per home becomes:

σ_average = σ_single / √n = 19,900 / √10,000 ≈ $199

Uncertainty per policy drops from about $19,900 to about $199, a 100-fold reduction. That's the law of large numbers at work, and it's why insurers want to be big.

Step 1: Expected Value

The foundation of every premium is expected value, the probability-weighted average of outcomes:

E[Loss] = Σ (probability of outcome × size of outcome)

For auto insurance, an actuary might model:

That $595 is the pure premium, the break-even cost of coverage before any expenses or profit.

Step 2: Loading the Premium

No insurer charges the pure premium. The actual price adds a loading:

Gross premium = (Pure premium + Expenses + Risk margin) / (1 − Profit and tax share)

The industry watches this through the combined ratio: claims plus expenses, divided by premiums earned. A combined ratio below 100% means an underwriting profit. U.S. property-casualty insurers often run close to 100%, and they make much of their money by investing premiums before claims are paid. That float is a form of time value of money.

Step 3: Life Tables and Mortality

Life insurance rests on the life table, which lists the probability of dying at each age. The first scientific life table is usually credited to Edmond Halley, the astronomer of comet fame. In 1693 he published an analysis of births and deaths in the city of Breslau and used it to price annuities.

A simplified modern version:

Mortality risk roughly doubles every 8 years in adulthood, a pattern known as the Gompertz law (Benjamin Gompertz, 1825). That's exponential growth, and it's why term life premiums rise so sharply with age.

Step 4: Classifying Risk

Insurers split customers into rating classes so that each class has a similar expected loss. Age, driving record, location and building materials all shift the probability or severity of a claim.

The statistical tool of choice today is the generalized linear model (GLM). It predicts claim frequency and claim severity from many variables at once. Pricing is essentially regression with a probability distribution attached.

Step 5: Planning for Catastrophes

The law of large numbers assumes losses are independent. Hurricanes, earthquakes and pandemics break that assumption: thousands of policies can claim at once.

For those correlated risks, insurers use catastrophe models and buy reinsurance, insurance for insurers. Regulators in the European Union, under Solvency II, require insurers to hold enough capital to survive a 1-in-200-year loss event. That's the 99.5th percentile of the annual loss distribution.

Two Concepts Worth Knowing

Variance and Standard Deviation

Variance measures how spread out outcomes are around the expected value. Its square root, the standard deviation, is in the same units as the loss. The formulas are on our statistics formulas page.

The Normal Distribution and Z-Scores

When you add up many independent losses, the total tends toward a normal distribution, thanks to the central limit theorem. An insurer can then ask how many standard deviations above the mean it can afford, and read the probability off a standard normal table or a z-score calculator.

Quick Answer: How Do Insurance Companies Calculate Premiums?

Insurers estimate the expected loss per policy (probability × severity, summed across events), then add expenses, a risk margin and profit. Because they pool thousands of independent policies, the law of large numbers makes the average loss highly predictable, even though individual losses aren't.

Why You Still Buy Insurance

If premiums are always higher than expected losses, is insurance a bad deal? Not necessarily. Mathematically you lose money on average. But you're trading a small, certain cost for protection against a large, ruinous one.

Economists call this risk aversion, and it's modeled with concave utility functions: losing $200,000 hurts far more than 100 times losing $2,000. Insurance is one of the few trades where both sides can come out ahead in terms of what they value.

Try Them Yourself

• Z-Score Calculator: how unusual is a given loss?
• Standard Normal Table: probabilities for the bell curve
• Z Tables: positive and negative z-score lookups
• Statistics Formulas: expected value, variance and standard deviation
• Random Number Generator: simulate your own insurance pool
• Sampling Techniques Unveiled: how data drives decisions

Try this: generate 100 random numbers from 1 to 100 and count how many are "1" (a claim). Then try 1,000 and 10,000. Watch the claim rate settle toward 1%.