In Part 2, we flagged that insurance companies need a different kind of earnings normalization. Here’s exactly how to do it.
In Part 2, we built a complete DCF for First River Bank. At the end, I flagged a problem we didn’t solve: a bank’s earnings can be distorted by credit cycles, but an insurance company’s earnings face a more extreme version of the same challenge. A single hurricane season, a pandemic, or a multi-billion-dollar catastrophe can wipe out years of underwriting profit in one quarter. You cannot plug last year’s net income into a model and call it a day.
This post is about how to deal with that. The concept is called earnings normalization — deriving a baseline that reflects what an insurer earns in an average year, stripped of the noise that catastrophes and reserve swings introduce.
We’ll cover why GAAP earnings mislead for insurers, how to measure underwriting profitability, how multi-year catastrophes distort the picture, how to normalize investment income, and finally how to build a normalized net income figure that you can actually put into a model.
Why GAAP Earnings Mislead for Insurers
GAAP stands for Generally Accepted Accounting Principles — the standardized rules that govern how U.S. companies report their financial results. Every public company uses GAAP for its income statement.
For most companies, GAAP earnings are a reasonable proxy for economic earnings. For insurers, they are often badly distorted — in both directions.
Two things cause most of the distortion.
First: catastrophe years. When a hurricane hits, an insurer pays out claims. Massive claims show up as losses in the income statement immediately, in the quarter they occur. A company that earned million in each of the prior three years might report a loss of .5 billion in a single catastrophe year. If you look only at that year, you’d conclude the business is a money-loser. You’d be wrong. The .5 billion loss doesn’t tell you what the business earns in a normal year — it tells you what happened when a large storm hit Florida.
Second: reserve development. Insurers must estimate their future claims liability — the money they expect to pay out on policies already written but not yet fully settled. This estimate sits on the balance sheet as a reserve. The estimate is always uncertain, and it gets revised each year as more information comes in.
When reserves prove too high — the insurer set aside more than it needed — the excess flows back through the income statement as income. This is called favorable prior-year development. When reserves prove too low — the insurer underestimated future claims — the shortfall hits the income statement as an additional expense. This is called adverse prior-year development.
Neither development item reflects current-year underwriting performance. Both appear on the income statement anyway. A company with strong underwriting in the current year can look mediocre because adverse development from three years ago is running through the books. A company with weak underwriting can look excellent because favorable reserve releases are padding reported earnings.
When you value an insurer from raw GAAP earnings without stripping these items out, you are valuing the accounting noise along with the business. We need to build a number that represents the business.
The Combined Ratio — How Underwriting Profitability Is Measured
The core profitability measure for an insurer is not earnings per share. It is the combined ratio.
The combined ratio measures what the insurance operation itself earns, independent of investment income. It is calculated as:
Combined Ratio = (Losses Incurred + Operating Expenses) ÷ Premiums Earned
Let’s define each piece.
Losses incurred — the claims the insurer paid or expects to pay for policies in force during the period. This is the fundamental cost of the insurance business: policyholders file claims, and the insurer pays them.
Operating expenses — everything it costs to run the insurance operation: agent commissions, underwriting overhead, policy administration, and so on. These are sometimes split into the loss ratio (losses only) and the expense ratio (operating costs only), with their sum being the combined ratio.
Premiums earned — the revenue the insurer has “earned” by providing coverage. When a policyholder pays a premium in advance, only the portion covering the period that has already elapsed counts as earned. The rest sits on the balance sheet as unearned premium.
The combined ratio tells you, for every dollar of premiums earned, how many cents went out the door in claims and expenses.
- A combined ratio below 100% means the insurance operation turned an underwriting profit. For every dollar of premiums, less than a dollar went out.
- A combined ratio above 100% means the insurance operation ran an underwriting loss. For every dollar of premiums, more than a dollar went out.
- A combined ratio at 100% means the insurance operation broke even.
Here is an example:
| Item | Amount |
|---|---|
| Losses Incurred | M |
| Operating Expenses | M |
| Total Costs | M |
| Premiums Earned | ,000M |
| Combined Ratio | 83% |
An 83% combined ratio means the insurer earned 17 cents of underwriting profit for every dollar of premium. That is a genuinely strong result.
A combined ratio in the low-to-mid 90s is considered healthy for a diversified property and casualty insurer in a normal year. Some specialty lines run lower; some lines with high claims frequency run higher. Context matters.
The cat-year distortion: In a severe catastrophe year, losses incurred spike. A company running at 93% in most years might post a 115% combined ratio in a hurricane year. That spike tells you something about the company’s catastrophe exposure — it is not a reflection of normalized underwriting profitability. If you use the 115% to build your model, you will dramatically undervalue a business that earns 93% in seven out of ten years.
Loss Development — How Reserve Swings Distort the Picture
Reserve development complicates the combined ratio the same way it complicates GAAP earnings.
Insurers must estimate their loss reserves — the amount they expect to pay out on claims that have been filed but not yet fully resolved, plus claims that have occurred but not yet been reported at all. For long-tailed lines of business — such as workers’ compensation or liability coverage — claims can take years or even decades to fully settle. The reserve estimates for those lines are inherently uncertain.
Each year, the actuarial team revisits those prior-year estimates. If claims are coming in lighter than expected, the reserve is released: the surplus flows back through the income statement, reducing reported losses for the current year. This is favorable development. If claims are coming in heavier, additional reserves must be established: the additional cost flows through the income statement as a charge. This is adverse development.
From a reporting standpoint, development from prior years appears in the current-year combined ratio. A company that had 5 percentage points of favorable development will report a combined ratio 5 points better than its true current-year underwriting performance. A company with 5 points of adverse development will report a combined ratio 5 points worse.
This is why analysts who follow insurance companies always look at the accident-year combined ratio alongside the calendar-year combined ratio. The accident-year ratio reflects only the current year’s business, stripped of prior-year development. The calendar-year ratio is what you find on the income statement — current year plus any favorable or adverse development from prior years.
When normalizing earnings, you want accident-year data, or you want to strip development explicitly. The goal is a combined ratio that represents what this insurer earns when it writes policies in a normal year, under normal conditions, with no unusual development running through the books.
Normalizing the Combined Ratio — The Multi-Year Average
The most practical method for normalizing underwriting profitability is a multi-year average combined ratio, with outlier years handled carefully.
Here is the approach:
Step 1: Gather accident-year combined ratios for 5–10 years. Use accident-year data where available — most large insurers disclose this in their annual reports and earnings supplements. If accident-year data is not available, use calendar-year and note the limitation.
Step 2: Identify and strip catastrophe-year outliers. Review each year for major catastrophe events. Cat years are typically identified by the company itself — they disclose cat losses separately, and you can find the cat load in their disclosures. A year with an unusually high cat load (say, more than 3–4 percentage points above average) is a candidate for adjustment. You can either exclude that year from the average or replace the cat load with a long-run average cat assumption.
Step 3: Average the remaining years. The result is a normalized combined ratio that represents sustainable underwriting profitability across conditions.
Let’s work through an example. Here is seven years of hypothetical combined ratios for Granite Mutual Insurance:
| Year | Combined Ratio | Cat Load | Note |
|---|---|---|---|
| 2017 | 92.1% | 3.2 pts | Normal year |
| 2018 | 94.5% | 5.1 pts | Normal year |
| 2019 | 111.3% | 17.8 pts | Major hurricane — outlier |
| 2020 | 93.0% | 2.9 pts | Normal year |
| 2021 | 95.2% | 4.4 pts | Normal year |
| 2022 | 108.9% | 14.1 pts | Wildfire season — outlier |
| 2023 | 91.8% | 3.0 pts | Normal year |
Two years stand out: 2019 (hurricane) and 2022 (wildfires). We strip those years and average the remaining five:
Normalized Combined Ratio = (92.1 + 94.5 + 93.0 + 95.2 + 91.8) ÷ 5
= 466.6 ÷ 5
= 93.3%
Granite Mutual’s normalized combined ratio is 93.3%. This is the baseline we’ll use.
Now we can derive normalized underwriting income:
Normalized Underwriting Income = Premiums Earned × (1 − Normalized Combined Ratio)
= ,000M × (1 − 0.933)
= ,000M × 0.067
= M
For every dollar of premiums, Granite earns about 6.7 cents of underwriting profit in a normal year.
Normalizing Investment Income — Float and the Rolling Yield
Insurance companies do not just earn underwriting income. They also earn investment income on their float.
Float is one of the most valuable concepts in insurance. Here is how it works: a policyholder pays their annual premium in January. The insurer won’t pay out any claims until later — maybe months from now, maybe never. In the meantime, the insurer holds the premium as an asset and invests it. That pool of policyholder money sitting with the insurer, earning investment returns while awaiting claims, is the float.
A large, well-run insurer can maintain billions of dollars in float for decades. Warren Buffett famously described Berkshire Hathaway’s insurance float as “free money” — an enormous pool of investable assets that policyholders essentially lend to the insurer at no explicit cost. (Whether it truly costs nothing depends on the combined ratio — a company consistently running above 100% is paying for the privilege of holding that float via underwriting losses.)
Investment income for an insurer has two components that create distortion when measured on a single-year basis:
Realized gains and losses — when an insurer sells a bond or stock from its portfolio, the gain or loss hits the income statement immediately. A company that needed to rebalance its portfolio in a falling-rate environment might book large realized gains. A company that sold bonds during a crisis might book large realized losses. Neither reflects the ongoing earning power of the investment portfolio.
Single-year yield fluctuations — interest rates move. The yield a billion bond portfolio earned in 2021 was very different from what it earned in 2024. A single year’s investment income is heavily influenced by where rates happened to be when the portfolio was assembled.
The normalization approach for investment income uses a rolling average yield applied to the average float balance:
Normalized Investment Income = Rolling Average Portfolio Yield × Average Float
Rolling average portfolio yield: Take the portfolio yield (net investment income ÷ average investable assets) for the most recent 5 years and average them. This smooths out year-to-year rate effects and strips out realized gains entirely (use only net investment income from the income statement, which excludes realized gains/losses).
Average float: Use the average float over the same 5-year window. Float tends to grow steadily with premium volume, so averaging over several years removes the effect of unusually large or small years.
Continuing with Granite Mutual:
| Year | Portfolio Yield | Float |
|---|---|---|
| 2019 | 3.9% | .1B |
| 2020 | 3.2% | .4B |
| 2021 | 2.8% | .7B |
| 2022 | 3.5% | .0B |
| 2023 | 4.1% | .2B |
Rolling Average Yield = (3.9 + 3.2 + 2.8 + 3.5 + 4.1) ÷ 5
= 17.5 ÷ 5
= 3.5%
Average Float = (.1B + .4B + .7B + .0B + .2B) ÷ 5
= .4B ÷ 5
= .68B
Normalized Investment Income = 3.5% × .68B
= .8M
Putting It Together: Normalized Net Income
We now have both components of insurance earnings normalized. The final step is straightforward:
Normalized Pre-Tax Income = Normalized Underwriting Income + Normalized Investment Income Normalized Net Income = Normalized Pre-Tax Income × (1 − Effective Tax Rate)
For Granite Mutual (using a 21% effective tax rate):
| Component | Amount |
|---|---|
| Normalized Underwriting Income | .0M |
| Normalized Investment Income | .8M |
| Normalized Pre-Tax Income | .8M |
| Taxes (21%) | .5M |
| Normalized Net Income | .3M |
This .3M is the number you put into your valuation model. It represents what Granite Mutual earns in a normal year — not a catastrophe year, not a year with large reserve releases, not a year with unusual realized gains.
From here, you would apply the same FCFE framework we used for First River Bank: project normalized net income forward using a sustainable growth rate, retain an appropriate amount to maintain surplus capital (the insurance equivalent of bank regulatory capital), discount the remaining FCFE at the cost of equity, and add a terminal value.
Why This Matters: The 73% CAGR Problem
Consider a company that reported the following EPS:
| Year | EPS |
|---|---|
| 2019 | .20 |
| 2020 | /bin/zsh.45 (hurricane year) |
| 2021 | .40 (reserve releases, benign year) |
| 2022 | .10 |
| 2023 | .85 |
If you look only at the 2020–2023 period, EPS grew from /bin/zsh.45 to .85. That is a 73% compound annual growth rate.
Is this a 73%-growth business? Absolutely not. The 2020 trough was a catastrophe year that depressed earnings to almost nothing. The 2021 spike included large favorable reserve development from prior years. The “growth” is almost entirely a recovery from an artificial low to a normalized level.
A 73% EPS CAGR headline number like this will sometimes appear in stock screens, in analyst reports, and even in investor presentations. It is a number that looks extraordinary and tells you almost nothing about the business’s actual earning trajectory.
The normalized earnings approach cuts through this noise. If you normalize Granite Mutual’s earnings correctly, you find that it earns about M in a typical year. From a normalized baseline of, say, M in 2020 (what the business would have earned without the catastrophe), the 2020–2023 growth rate is much more modest — and much more meaningful.
This is the core skill for analyzing insurance companies: separating what the business earns from what the accounting period happened to record.
A Note to Luca and Lili
Insurance companies are some of the most misunderstood businesses in investing, and I think there are two reasons for that.
The first is that the numbers are genuinely confusing. A company posts a .5 billion loss one year and billion in profits the next. What is the “real” earning power? The answer isn’t obvious from the income statement, and most investors don’t know how to ask the question correctly. So they either skip insurance companies entirely — the way I skipped banks early on — or they make the mistake of treating whatever the last year showed as the truth.
The second is that insurance companies, when they are excellent, are quietly extraordinary. A business that gets paid premiums upfront, invests the money for years before ever paying a claim, runs a combined ratio below 95%, and compounds its float at 4–5% annually can generate remarkable returns on equity without much leverage or risk. The best insurance businesses in the world look boring on the surface and are anything but boring underneath.
What I want you to take from this post isn’t just the mechanics — though the mechanics matter. It’s the habit of asking: what does this business earn when nothing unusual is happening? That question will serve you across every industry, not just insurance. The accounting period is always imperfect. The analyst’s job is to see through it.
— Papa
The One-Sentence Summary
To normalize insurance earnings, average the combined ratio across 5–10 years excluding catastrophe outliers to derive normalized underwriting income, apply a rolling average portfolio yield to average float to derive normalized investment income, and tax-effect the sum — the result is the number that actually belongs in your valuation model.
Next: Financial Firms, Part 4 — Risk and the Discount Rate. We’ve built the earnings number. Now we need the right discount rate. Part 4 explains why financial firms use cost of equity instead of WACC — and how to arrive at the right number.
— Jim